CHAPMAN & HALL/CRC

Boca Raton London New York Washington, D.C.

CHAPMAN & HALL/CRC
Monographs and Surveys in
Pure and Applied Mathematics 120
ABSTRACT
CAUCHY
PROBLEMS:
Three Approaches
IRINA V. MELNIKOVA
ALEXEI FILINKOV
2001 CRC Press LLC

This book contains information obtained from authentic and highly regarded sources. Reprinted material
is quoted with permission, and sources are indicated. A wide variety of references are listed. Reasonable
efforts have been made to publish reliable data and information, but the author and the publisher cannot
assume responsibility for the validity of all materials or for the consequences of their use.
Apart from any fair dealing for the purpose of research or private study, or criticism or review, as permitted
under the UK Copyright Designs and Patents Act, 1988, this publication may not be reproduced, stored
or transmitted, in any form or by any means, electronic or mechanical, including photocopying, micro-
lming, and recording, or by any information storage or retrieval system, without the prior permission
in writing of the publishers, or in the case of reprographic reproduction only in accordance with the
terms of the licenses issued by the Copyright Licensing Agency in the UK, or in accordance with the
terms of the license issued by the appropriate Reproduction Rights Organization outside the UK.
The consent of CRC Press LLC does not extend to copying for general distribution, for promotion, for
creating new works, or for resale. Specic permission must be obtained in writing from CRC Press LLC
for such copying.
Direct all inquiries to CRC Press LLC, 2000 N.W. Corporate Blvd., Boca Raton, Florida 33431.

Trademark Notice:

Product or corporate names may be trademarks or registered trademarks, and are
used only for identication and explanation, without intent to infringe.

Visit the CRC Press Web site at www.crcpress.com

2001 by Chapman & Hall/CRC
No claim to original U.S. Government works
International Standard Book Number 1-58488-250-6
Library of Congress Card Number 2001017069
Printed in the United States of America 1 2 3 4 5 6 7 8 9 0
Printed on acid-free paper

Library of Congress Cataloging-in-Publication Data

Filinkov, A.I. (Aleksei Ivanovich)
Abstract Cauchy problems : three approaches / Alexei Filinkov, Irina
V. Melnikova
p. cm. (Chapman & Hall/CRC monographs and surveys in pure and
applied mathematics ; 120)
Includes bibliographical references and index.
ISBN 1-58488-250-6 (alk. paper)
1. Cauchy problem. I. Melnikova, I.V. (Irina Valerianovna) II.
Title. III. Series.
[DNLM: 1. Hepatitis B virus. QW 710 G289h]
QA377 .F48 2001
515



.35dc21 2001017069
CIP

C2506-disclaimer Page 1 Wednesday, February 14, 2001 2:53 PM
2001 CRC Press LLC
To our teacher
V.K. Ivanov
2001 CRC Press LLC
2001 CRC Press LLC
Contents
Preface
Intro duction
0 Illustration and Motivation
0.1 Heat equation
0.2 The reversed Cauchy problem for the
Heat equation
0.3 Wave equation
1 Semigroup Metho ds
1.1 C
0
-semigroups
1.1.1 Denitions and main properties
1.1.2 The Cauchy problem
1.1.3 Examples
1.2 Integrated semigroups
1.2.1 Exponentially bounded integrated semigroups
1.2.2 (n, )-well-posedness of the Cauchy problem
1.2.3 Local integrated semigroups
1.2.4 Examples
1.3 -convoluted semigroups
1.3.1 Generators of -convoluted semigroups
1.3.2 -convoluted Cauchy problem
1.4 C-regularized semigroups
1.4.1 Generators of C-regularized semigroups
1.4.2 C-well-posedness of the Cauchy problem
1.4.3 Local C-regularized semigroups
1.4.4 Integrated semigroups and
C-regularized semigroups
1.4.5 Examples
1.5 Degenerate semigroups
1.5.1 Generators of degenerate semigroups
2001 CRC Press LLC
2001 CRC Press LLC 2001 CRC Press LLC
2001 CRC Press LLC
1.5.2 Degenerate 1-time integrated semigroups
1.5.3 Maximal correctness class
1.5.4 (n, )-well-posedness of a degenerate
Cauchy problem
1.5.5 Examples
1.6 The Cauchy problem for inclusions
1.6.1 Multivalued linear operators
1.6.2 Uniform well-posedness
1.6.3 (n, )-well-posedness
1.7 Second order equations
1.7.1 M, N-functions method
1.7.2 Integrated semigroups method
1.7.3 The degenerate Cauchy problem
2 Abstract Distribution Metho ds
2.1 The Cauchy problem
2.1.1 Abstract (vector-valued) distributions
2.1.2 Well-posedness in the space of distributions
2.1.3 Well-posedness in the space of
exponential distributions
2.2 The degenerate Cauchy problem
2.2.1 A-associated vectors and degenerate
distribution semigroups
2.2.2 Well-posedness in the sense of distributions
2.2.3 Well-posedness in the space of
exponential distributions
2.3 Ultradistributions and new distributions
2.3.1 Abstract ultradistributions
2.3.2 The Cauchy problem in spaces of
abstract ultradistributions
2.3.3 The Cauchy problem in spaces of
new distributions
3 Regularization Metho ds
3.1 The ill-posed Cauchy problem
3.1.1 Quasi-reversibility method
3.1.2 Auxiliary bounded conditions (ABC) method
3.1.3 Carassos method
3.2 Regularization and regularized
semigroups
3.2.1 Comparison of the ABC and
the quasi-reversibility methods
3.2.2 Dierential and variational
methods of regularization
2001 CRC Press LLC
2001 CRC Press LLC
3.2.3 Regularizing operators and local C-regularized semi-
groups
3.2.4 Regularization of slightly ill-posed problems
Bibliographic Remark
Bibliography
Glossary of Notation

2001 CRC Press LLC
2001 CRC Press LLC
Preface
Many mathematical models in physics, engineering, nance, biology, etc.,
involve studying the Cauchy problem
u

(t) = Au(t) +F(t), t [0, T), T , u(0) = x,

where A is a linear operator on a Banach space X and F is an X-
valued function that represents the deterministic or stochastic inuence of a
medium. The rst and main step in solving such problems consists of study-
ing the homogeneous Cauchy problem
u

(t) = Au(t), t [0, T), T , u(0) = x. (CP)

The problem (CP) has been comprehensively studied in the case when the
operator A generates a C
0
-semigroup. A C
0
-semigroup generated by a
bounded operator A is nothing but an exponential operator-function
e
At
=

k=1
A
k
t
k
k!
.
In general, C
0
-semigroups inherit some main properties of exponential func-
tions. As it turns out, the generation of a C
0
-semigroup by the operator
A is closely related to the uniform well-posedness of the Cauchy problem
(CP). However, many operators that are important in applications do not
generate C
0
-semigroups.
The focus of this book is the Cauchy problem (CP) with operators A,
which do not generate C
0
-semigroups. We focus on three approaches to
treating such problems:
semigroup methods
abstract distribution methods
regularization methods
2001 CRC Press LLC
2001 CRC Press LLC
The strategic concept of the rst two approaches is the relaxation of
the notion of well-posedness so that a Cauchy problem that is not well-
posed in the classical sense becomes well-posed in some other sense. In
the rst approach, using semigroups more general than C
0
-semigroups
integrated, C-regularized, and -convoluted semigroups one can construct
solution operators on some subsets of D(A). The corresponding solutions
are stable with respect to norms that are stronger than the norm of X.
Using distribution semigroups in the second approach, one can construct
a family of generalized solution operators on X. The essence of the third
approach is the following: assuming that for some x there exists a solution
u() of (CP), for given x

(x x

) we construct a solution u

() of a
well-posed (regularized) problem depending on a regularization parameter
and choose = () in a way that u

u as 0. So u

() can be taken
as an approximate solution and the operator R
,t
dened by R
,t
x

:= u

(t)
is a regularizing operator.
The motivation for writing this book was twofold: rst, to give a self-
contained account of the above-mentioned three approaches, which is acces-
sible to nonspecialists. Second, to demonstrate the profound connections
between these seemingly quite dierent methods. In particular, we demon-
strate that integrated semigroups are primitives of generalized solution op-
erators, and many regularizing operators coincide with C

-regularized semi-
groups.
The books three chapters are devoted respectively to semigroup meth-
ods, abstract distribution methods, and regularization methods. We dis-
cuss not only the Cauchy problem (CP), but also the following important
generalizations: the degenerate Cauchy problem, the Cauchy problem for
inclusion and the Cauchy problem for the second order equation.
Acknowledgments
This project was supported by the Australian Research Council. The
rst author is grateful to the Department of Pure Mathematics of Adelaide
University for their hospitality during her visits to Adelaide. The rst
author was also partially supported by grant RFBR N 99-01-00142 (Russian
Federation).
The authors thank Professor Alan Carey, Professor Mike Eastwood,
Dr. John van der Hoek and Dr. Nick Buchdahl for their support and
encouragement.
The authors thank Dr. Maxim Alshansky, Dr. Uljana Anufrieva, Dr.
Alexander Freyberg and Dr. Isna Maizurna for participation in some parts
of this project
The authors also thank Cris Carey for editing the manuscript and Ann
Ross for technical support.
Irina Melnikova
Alexei Filinkov
2001 CRC Press LLC
2001 CRC Press LLC
Intro duction
We aim to present semigroup methods, abstract distribution methods, and
regularization methods for the abstract Cauchy problem
u

(t) = Au(t), t [0, T), T , u(0) = x, (CP)

where A is a linear operator on a Banach space X, u() is an X-valued
function and x X. We also demonstrate the connections among these
methods.
In Chapter 0 we use Heat and Wave equations to illustrate some of the
main ideas, notions, and connections from the discussion below.
In Section 1.1 we give a brief account of the theory of C
0
-semigroups.
Here the reader will nd a proof of the following fundamental result of this
theory.
Theorem 1.1.1 Suppose that A is a closed densely dened linear operator
on X. Then the following statements are equivalent:
(I) the Cauchy problem (CP) is uniformly well-posed on D(A): for any
x D(A) there exists a unique solution, which is uniformly stable
with respect to the initial data;
(II) the operator A is the generator of a C
0
-semigroup {U(t), t 0};
(III) Miyadera-Feller-Phillips-Hille-Yosida (MFPHY) condition holds for
the resolvent of operator A: there exist K > 0, R so that
_
_
_R
(k)
A
()
_
_
_
K k!
(Re )
k+1
,
for all C with Re > , and all k = 0, 1, . . . .
In this case the solution of (CP) has the form
u() = U()x, x D(A),
and sup
t[0,T]
u(t) Kx for some constant K = K(T).
2001 CRC Press LLC
2001 CRC Press LLC
Condition (III) is usually used as the criterion for the uniform well-
posedness of (CP).
Section 1.2 is devoted to n-times integrated semigroups connected with
the well-posedness of (CP) on subsets from D(A
n+1
). Here is the main
result for exponentially bounded semigroups.
Theorem 1.2.4 Let A b e a densely dened linear op erator on X with
nonempty resolvent set. Then the following statements are equivalent:
(I) A is the generator of an exp onentially b ounded n-times integrated semi-
group {V (t), t 0};
(I I) the Cauchy problem (CP) is (n, )-well-p osed: for any x D(A
n+1
)
there exists a unique solution such that
K > 0, R : u(t) Ke
t
x
A
n,
where x
A
n = x +Ax +. . . +A
n
x.
Statements (I), (II) of this theorem are equivalent to the MFPHY-type
condition
_
_
_
_
_
R
A
()

n
_
(k )
_
_
_
_

K k!
( )
k +1
for all > , and k = 0, 1, . . . .
In the particular case when A generates a C
0
-semigroup U, the n-times
integrated semigroup generated by A is the n-th order primitive of U. In
general (and in contrast to C
0
-semigroups), integrated semigroups may be
not exponentially bounded, may be locally dened, degenerate, and their
generators may be not densely dened. In Section 1.2 we consider non-
degenerate exponentially bounded and local integrated semigroups, their
connections with the Cauchy problem, and discuss several examples. We
also show that an n-times integrated semigroup with a densely dened
generator A can be dened as a family of bounded linear operators V (t)
satisfying
V (t)x
t
n
n!
x =
_
t
0
V (s)Axds
=
_
t
0
AV (s)xds, 0 t < T, x D(A),
and
V (t)x
t
n
n!
x = A
_
t
0
V (s)xds, x X.
Replacing function t
n
/n! by a continuously dierentiable function (t), one
arrives at the notion of -convoluted semigroup, where =

.
2001 CRC Press LLC
2001 CRC Press LLC
Denition 1.3.1 Let A b e a closed op erator and (t) b e a continuous
function on [0, T), T . If there exists a strongly continuous op erator-
family {S

(t), 0 t < T} such that S

(t)Ax = AS

(t)x for x D(A),

t [0, T), and for all x X
S

(t)x = A
_
t
0
S

(s)xds + (t)x, 0 t < T,

where (t) =
_
t
0
(s)ds, then S

is called a -convoluted semigroup gener-
ated by A, and A is called the generator of S

.
In the particular case (t) = t
n
/n! , the -convoluted semigroup S

is
an n-times integrated semigroup. If A is the generator of a C
0
-semigroup
U, then
S

(t)x =
_
t
0
(t s)
n1
(n 1)!
U(s)xds
is an n-times integrated semigroup and a -convoluted semigroup. This
is the reason that S

is called a -convoluted semigroup, but not a -

convoluted semigroup.
The corresponding Cauchy problem
v

(t) = Av(t) + (t)x, 0 t < T, v(0) = 0,

is called the -convoluted Cauchy problem. If there exists a solution of the
Cauchy problem
u

(t) = Au(t), 0 t < T, u(0) = x,

then, as usual for a nonhomogeneous equation, we have v = u.
The results of Section 1.3 connect the existence of a -convoluted semi-
group and well-posedness of a -convoluted Cauchy problem with the be-
haviour of the resolvent R
A
(). In contrast to the case of an exponentially
bounded and local n-times integrated semigroup, where the resolvent has
polynomial estimates in a half-plane or in a logarithmic region respectively,
in the convoluted case the resolvent exists in some smaller region and is
allowed to increase exponentially:
R
A
() Ke
M()
.
Here a real-valued function M(), C, grows not faster than and is
dened by the order of decreasing of (), the Laplace transform of (t). In
Chapter 3 we demonstrate that a (CP) with a generator of a -convoluted
semigroup takes an intermediate place between slightly ill-posed problems
(with a generator of an integrated semigroup) and essentially ill-posed
problems (with A generating a C
0
-semigroup).
2001 CRC Press LLC
2001 CRC Press LLC
In Section 1.4 we consider the Cauchy problem (CP) with A being the
generator of an exponentially bounded C-regularized semigroup {S(t), t
0}, where C is some bounded invertible operator on X. It is shown that
for x CD(A) there exists a solution of this problem:
u() = U()x = C
1
S()x.
Solution operators U(t) are not necessarily bounded on X, therefore in
general the solution u() is not stable in X with respect to variation of
x. The well-posedness of the Cauchy problem can be restored in a certain
subspace of X with a norm stronger than the norm of X. Namely, if A is the
generator of a C-regularized semigroup, then for any x CD(A), u() =
C
1
S()x is a unique solution of (CP), and
sup
0tT
u(t) KC
1
x
for some constant K. There are operators A (for example, those having ar-
bitrary large positive eigenvalues) that generate C-regularized semigroups,
dened only for t from some bounded subset of R. Such semigroups corre-
spond to the local C-well-posedness of the Cauchy problem, and are also
considered in Section 1.4.
In Section 1.5 we study degenerate C
0
-semigroups and integrated semi-
groups connected with the degenerate Cauchy problem
Bu

(t) = Au(t), t 0, u(0) = x, (DP)

where B, A : X Y are linear operators in Banach spaces X, Y , and
ker B = {0}. We assume that the set

B
(A) :=
_
C

R() := (B A)
1
B is a bounded operator on X
_
is not empty. This set is called the B-resolvent set of operator A and the
operator R() is called the B-resolvent of operator A.
We prove the equivalence of the uniform well-posedness of (DP) on the
maximal correctness class, the existence of a degenerate C
0
-semigroup with
A and B being the pair of generators, and MFPHY-type estimates together
with the decomposition of X:
X = X
1
ker B, X
1
:= D
1
, D
1
:= R()X,
B
(A).
This decomposition plays a role similar to the density condition for the
generator of a C
0
-semigroup. We also prove results about (n, )- and n-
well-posedness of the degenerate Cauchy problem.
2001 CRC Press LLC
2001 CRC Press LLC
Both degenerate Cauchy problems
Bu

(t) = Au(t), t [0, T), T , (DP)

u(0) = x, ker B = {0}
and
d
dt
Bv(t) = Av(t), t [0, T), T , (DP1)
Bv(0) = x,
can be considered as particular cases of the Cauchy problem for an inclusion
with a multivalued linear operator A
u

(t) Au(t), t [0, T), T , u(0) = x. (IP)

If we set A = B
1
A for problem (DP), or A = AB
1
and u() := Bv()
for (DP1), then u() is a solution of the Cauchy problem (IP). Conversely,
if u() is a solution of (IP) with A = B
1
A or A = AB
1
, then u() is the
solution of (DP) or any v() from the set Bv() = u() is the solution of
(DP1), respectively.
In Section 1.6 we use the technique of degenerate semigroups with
multivalued generators to study the uniform well-posedness and the (n, )-
well-posedness of the Cauchy problem (IP). As a consequence, we obtain
MFPHY-type necessary and sucient conditions for the well-posedness of
(DP) and for the existence of a solution of (DP1).
Section 1.7 is devoted to semigroup methods for abstract Cauchy
problems
u

(t) = Au

(t) +Bu(t), t 0, u(0) = x, u

(0) = y, (CP2)
where A, B : X X, and
Qu

(t) = Au

(t) +Bu(t), t 0, u(0) = x, u

(0) = y, (DP2)
where Q, A, B : X Y are linear operators, ker Q = {0}, and X, Y are
Banach spaces. In this section we use semigroup methods for studying
the well-posedness of these problems. First, we construct M, N-functions,
the solution operators for problem (CP2), which generalize semigroups and
cosine/sine operator-functions. Second, we study (CP2), (DP2) by reducing
them to a rst order Cauchy problem in a product space. Using the results
of Sections 1.2 and 1.6 on integrated semigroups we obtain necessary and
sucient conditions for the well-posedness of (CP2), (DP2) in terms of
resolvent operators
R
r
(
2
) := (
2
I AB)
1
and R
d
(
2
) := (Q
2
AB)
1
,
2001 CRC Press LLC
2001 CRC Press LLC
respectively.
The relation between the operators A, B plays a very important role
in studying well-posedness of second order problems. We show that for
the equivalence of the well-posedness of (CP2) and the existence of an
integrated semigroup for the Cauchy problem
v

(t) = v(t), t 0, v(0) = v

0
,
=
_
0 I
B A
_
, v(t) =
_
u(t)
u

(t)
_
, v
0
=
_
x
y
_
,
operators A, B have to be biclosed. With the help of the theory of integrated
semigroups we prove that the MFPHY-type condition:
K > 0, 0 :
_
_
_
_
d
k
d
k
R
r
(
2
)
_
_
_
_
,
_
_
_
_
d
k
d
k
R
r
(
2
)
_
I A
_
_
_
_
_

K k!
(Re )
k +1
,
for all C with Re > and all k = 0, 1, . . . ,
is necessary and sucient for well-posedness of (CP2) with biclosed op-
erators A, B. Using the M, N-functions theory, we obtain the following
MFPHY-type well-posedness condition
K > 0, 0 :
_
_
_
_
d
k
d
k
R
r
(
2
)
_
_
_
_
,
_
_
_
_
d
k
d
k
_
I A
_
R
r
(
2
)
_
_
_
_

K k!
(Re )
k +1
,
for all C with Re > and all k = 0, 1, . . . ,
for problem (CP2) with commuting operators A, B. Using the theory of
degenerate integrated semigroups we prove MFPHY-type necessary and
sucient conditions for well-posedness of the degenerate Cauchy problem
(DP2).
Chapter 2 is devoted to the abstract distributions methods. In Section
2.1 for any x X, we consider the Cauchy problem (CP) with A : D(A)
X X, in the space of (Schwartz) distributions. Our main aim here is to
obtain necessary and sucient conditions for well-posedness in the space of
distributions in terms of the resolvent of A. We show that they are the same
as in the case when A is the generator of a local integrated semigroup. Thus,
only the Cauchy problem with a generator of such a semigroup can be well-
posed in the space of distributions. In this case, a distribution semigroup
is the solution operator distribution, and an integrated semigroup is the
continuous function whose existence is guaranteed by the abstract structure
theorem for this distribution.
2001 CRC Press LLC
2001 CRC Press LLC
In Section 2.2, we prove the necessary and sucient conditions for well-
posedness of (DP) and (IP) in spaces of distributions, generalizing the con-
ditions obtained in Section 2.1. Naturally, the semigroups connected with
these problems are degenerate. The important fact is that an n-times in-
tegrated semigroup with a pair of generators A, B or with the multivalued
generator A is degenerate on ker B = ker R(), but the corresponding so-
lution operator distribution and the distribution semigroup are degenerate
on ker R
n+1
:= ker R
n+1
(). Here R() is the pseudoresolvent that is equal
to (BA)
1
B or (I A)
1
, respectively. In this section we assume that
X admits the decomposition
X = X
n+1
ker R
n+1
, X
n+1
:= D
n+1
, D
n+1
:= R
n+1
()X.
First, we describe the structure of ker R
n+1
, which is used in the main
theorem, where we discuss the structure of solution operator distributions
and distribution semigroups on ker R
n+1
. We prove that (DP) is well-
posed in the space of distributions if and only if

A (the part of operator A
in X
n+1
) is the generator of a distribution semigroup, or equivalently,

A is
the generator of a local k-times integrated semigroup for some k. Finally,
we consider (DP) in the space of exponential distributions. The distinctive
feature of the exponential case is that the parameter k of the integrated
semigroup coincides with the parameter in the estimate for pseudoresolvent
R().
Studying the Cauchy problem in the spaces of distributions we prove
that polynomial estimates for the resolvent of A (or for the pseudoresolvent
of a multivalued operator A) in a certain logarithmic region of a complex
plane is the criterion for well-posedness of the Cauchy problem with such an
operator A (or A, respectively) in the sense of distributions. This domain
and resolvent estimates are connected with the order of a corresponding
distribution semigroup and solution operator distribution.
In Section 2.3 we consider the Cauchy problem with operator A having
the resolvent in a certain region smaller than logarithmic. We investigate
well-posedness of such (CP) in the space of (Beurling) ultradistributions.
Utradistributions are dened as elements of dual spaces for spaces of in-
nitely dierentiable functions with a locally convex topology. In these
spaces well-posedness of the Cauchy problem is studied in a more general
sense than the spaces of Schwartz distributions. We show that the exis-
tence of a solution operator ultradistribution is equivalent to exponential
estimates for the resolvent in some domain . As is proved in Section 1.3,
the latter fact is equivalent to the existence of a unique solution of the
convoluted equation
v(t)

= Av(t) + (t)x, t 0, v(0) = 0,

where function (t) is dened by the resolvent of A.
2001 CRC Press LLC
2001 CRC Press LLC
Spaces of test functions for ultradistributions are spaces of innitely
dierentiable functions dened in terms of estimates on their derivatives,
depending on some sequence M
n
. The same sequence allows one to esti-
mate coecients of dierential operators of innite order, which are called
ultradierential operators. The spaces of ultradistributions are invariant
under ultradierential operators just as spaces of distributions are invari-
ant under dierential operators. In this section, rstly we discuss some
properties of ultradistributions. Then we investigate the well-posedness of
(CP) in the spaces of ultradistributions. The limit case for (CP) well-posed
in the spaces of ultradistributions is the case of the Cauchy problem with
operator A having no regular points in a right semiplane. For example, the
reversed Cauchy problem for the heat equation is equivalent to the local
Cauchy problem (CP) with A = d
2
/dx
2
in the corresponding space X.
It is proved that such a (CP) is well-posed in spaces of new distributions.
Spaces of test elements x X for construction of spaces of new distribu-
tions are dened in terms of the behaviour of A
n
x for any n, and their
dual spaces of new distributions are subspaces of X

. Section 2.3, together

with Section 1.6 on -convoluted semigroups, demonstrates the transition
from uniformly well-posed problems to essentially ill-posed problems via
slightly ill-posed problems.
Thus, in Chapters 1 and 2 we investigate Cauchy problems that are
not uniformly well-posed. The technique of integrated, -convoluted, and
C-regularized semigroups presented in Chapter 1 allows one to construct
a solution of (CP) and local (CP) for initial values x from various subsets
of D(A), stable in X with respect to x in corresponding graph-norms.
The technique of distributions, ultradistributions, and new distributions
presented in Chapter 2 allows the construction of a generalized solution for
any x X, stable in a space of distributions.
In Chapter 3 we consider regularizing operators which allow one to
construct an approximate solution of (CP), which is stable in X for concor-
dant parameters of regularization and error. The general theory of ill-posed
problems usually treats ill-posed problems in the form of the equation of
the rst kind
u = f, : U F, (OP)
where, in general, operator
1
either does not exist or is not bounded.
The main regularization methods for (OP) are variational:
(a) Ivanovs quasivalues method,
(b) the residual method,
(c) Tikhonovs method.
In Section 3.1 we study regularization methods for the ill-posed local
Cauchy problem
u

(t) = Au(t), 0 t < T, T < , u(0) = x,

2001 CRC Press LLC
2001 CRC Press LLC
with operator A : D(A) X X generating a C
0
-semigroup or satisfy-
ing some similar conditions.
These methods employ the dierential nature of this problem and allow
us to construct a regularized solution of local (CP) as a solution of a well-
posed regularized problem, without reducing it to the form (OP). They are
the following:
1) the quasi-reversibility metho d, where the regularized solution is a
solution of the Cauchy problem
u

(t) = (AA
2
)u

(t), 0 < t T,
u

(0) = x, > 0;
2) the auxiliary b oundary conditions metho d, where the regularized
solution is a solution of the boundary value problem
u

(t) = A u

(t), 0 < t < T +, > 0,

u

(0) + u

(T +) = x, > 0;
3) Carassos metho d of reducing an ill-posed problem to a well-posed
Dirichlet problem.
In Section 3.2, we obtain error estimates for these regularization meth-
ods and compare them. We discuss connections between dierential meth-
ods and variational methods. Furthermore, we establish the connection
between regularization methods for the ill-posed Cauchy problem and the
C-regularized semigroup method.
2001 CRC Press LLC
2001 CRC Press LLC
Chapter 0
Illustration and
Motivation
In this chapter we use very basic model examples to illustrate some of the
main ideas, notions, and connections from the discussion above.
0.1 Heat equation
For simplicity, let = (0, 1). In general, can be an open set from R
n
.
Consider the Cauchy-Dirichlet problem on X = L
2
():
u(x, t)
t


2
u(x, t)
x
2
= 0, t [0, T], x , (0.1.1)
u(0, t) = u(1, t) = 0,
u(x, 0) = u
0
(x).
Note that L
2
() is a separable Hilbert space. Consider operator A =
d
2
dx
2
in L
2
() with the domain
D(A) = H
2
() H
1
0
(),
where H
2
() and H
1
0
() are the following classical Sobolev spaces
H
1
0
() =
_
u L
2
()

u
x
L
2
(), u(0, t) = u(1, t) = 0
_
,
H
2
() =
_
u L
2
()

2
u
x
2
L
2
()
_
.
The operator A has a self-adjoint compact inverse and therefore its spec-
trum consists of discrete eigenvalues. Eigenfunctions and eigenvalues of A
2001 CRC Press LLC
2001 CRC Press LLC
can be obtained by solving
d
2
e
k
dx
2
=
k
e
k
, e
k
(0) = e
k
(1) = 0, k N,
which gives

k
= k
2

2
> 0, e
k
=

2 sinkx, k N.
Since e
k

k=1
forms an orthonormal basis in L
2
(), any f L
2
() can be
written in the form
f =

k=1
f
k
e
k
, where f
k
= (f, e
k
)
L
2
()
,
and we have
|f|
L
2
()
=
_

k=1
[f
k
[
2
_
1/2
.
Assuming that u
0
L
2
() and u(t) L
2
() for each t [0, T], we write
u(t) =

k=1
u
k
(t)e
k
, and u
0
=

k=1
u
0
k
e
k
.
Then the equation in (0.1.1) becomes

k=1
_
du
k
dt
+
k
u
k
_
e
k
= 0.
By uniqueness of expansions in L
2
(), we deduce
du
k
dt
+
k
u
k
= 0, k N,
u
k
(0) = u
0
k
,
and the unique solution of this problem is
u
k
(t) = u
0
k
e

k
t
.
Now we show that the unique solution of (0.1.1) is given by
u(t) =

k=1
u
k
(t)e
k
=

k=1
e

k
t
u
0
k
e
k
. (0.1.2)
Since we assumed that u
0
L
2
(), we have

k=1
[u
k
(t)[
2
=

k=1
[u
0
k
[
2
e
2
k
t

k=1
[u
0
k
[
2
< , (0.1.3)
2001 CRC Press LLC
2001 CRC Press LLC
that is for any t [0, T], u(t) L
2
(). Moreover we have some regularity
with respect to t, namely u C
_
[0, T], L
2
()
_
, the Banach space of
continuous functions over [0, T] with values in L
2
() equipped with the
norm
|f|
C
_
[0,T], L
2
()
_
= sup
t[0,T]
_
_
1
0
[f(x, t)[
2
dx
_
1/2
.
To verify this we note that the estimate
sup
t[0,T]
_
m+p

k=m
[u
k
(t)[
2
_
1/2

_
m+p

k=m
[u
0
k
[
2
_
1/2
, m 1, p > 0,
implies that the sequence of partial sums of the series

k=1
u
k
(t) is a
Cauchy sequence in C
_
[0, T], L
2
()
_
. Now we have to nd out in what
sense the Cauchy problem (0.1.1) is satised. First, since
[u
k
(t) u
0
k
[
2
[u
0
k
[
2
_
1 e

k
t
_
2
,
we can write

k=1
[u
k
(t) u
0
k
[
2

_
1 e

N
t
_
2
N

k=1
[u
0
k
[
2
+

k=N+1
[u
0
k
[
2
,
where N is large enough. This implies that
lim
t0
(u(t) u
0
) = 0 in L
2
(),
and therefore u(0) = u
0
in L
2
().
Secondly, we demonstrate that u L
2
_
[0, T], H
1
0
()
_
, which clearly
implies that the boundary condition u(0, t) = u(1, t) = 0 is satised almost
everywhere in t [0, T]. It is well-known that v H
1
0
() if and only if

k=1

k
[v
k
[
2
< , where v
k
= (v, e
k
).
Since
[u
k
(t)[
2
[u
0
k
[
2
e
2
k
t
, k N,
then
_
T
0
[u
k
(t)[
2
dt
1

k
[u
0
k
[
2
,
which implies
_
T
0
_
m+p

k=m

k
[u
k
(t)[
2
_
dt
m+p

k=m
[u
0
k
[
2
, m 1, p > 0,
2001 CRC Press LLC
2001 CRC Press LLC
and therefore
_
T
0
_

k=1

k
[u
k
(t)[
2
_
dt < ,
that is u L
2
_
[0, T], H
1
0
()
_
for any u
0
L
2
().
Note that if u
0
H
1
0
(), then the series

k=1

k
[u
0
k
[
2
is convergent,
and since
sup
t[0,T]
_
m+p

k=m

k
[u
k
(t)[
2
_
1/2

_
m+p

k=m

k
[u
0
k
[
2
_
1/2
, m 1, p > 0,
we have u C
_
[0, T], H
1
0
()
_
.
Finally, we discuss in what sense the equation in (0.1.1) is satised. We
set
a(u, v) =
_

du
dx
dv
dx
dx, u, v H
1
0
(),
U
m
(t) =
m

k=1
u
k
(t)e
k
, U

m
(t) =
m

k=1
du
k
dt
e
k
.
Then, since
du
k
dt
+
k
u
k
= 0, k N,
we have
_
U

m
(t), e
k
_
+a
_
U
m
(t), e
k
_
= 0, 1 k m. (0.1.4)
Now x k < m and let T(0, T) (the space of Schwartz test functions).
Multiplying (0.1.4) by , and integrating by parts, we obtain

_
T
0
_
U
m
(t), e
k
_

(t)dt +
_
T
0
a
_
U
m
(t), e
k
_
(t)dt = 0.
As U
m

m
U uniformly over [0, T], we have that the function t
_
U
m
(t), e
k
_

(t) converges to the function t

_
u(t), e
k
_

(t), and therefore

lim
m
_
T
0
_
U
m
(t), e
k
_

(t)dt =
_
T
0
_
u(t), e
k
_

(t)dt.
Since u L
2
_
[0, T], H
1
0
()
_
and

_
T
0
a
_
U
m
(t) u(t), e
k
_
(t)dt

K
_
T
0
|U
m
(t) u(t)|
2
dt,
we have
lim
m
_
T
0
a
_
U
m
(t), e
k
_
(t)dt =
_
T
0
a(u(t), e
k
)(t)dt.
2001 CRC Press LLC
2001 CRC Press LLC
Thus, for any e
k
, we have

_
T
0
(u(t), e
k
)

(t)dt +
_
T
0
a(u(t), e
k
)(t)dt = 0,
and since the system e
k

k=1
is dense in H
1
0
(), we obtain

_
T
0
(u(t), v)

(t)dt +
_
T
0
a(u(t), v)(t)dt = 0, (0.1.5)
for all v H
1
0
() and T(0, T).
So we conclude that for any u
0
L
2
(), the function u, dened by
formula (0.1.2), satises the equation in (0.1.1) in the following weak sense
d
dt
_
u(), v
_
+a
_
u(), v
_
= 0, for all v H
1
0
(), (0.1.6)
in the sense of the distributions from T

(0, T), where the distribution

_
u(), v
_
T

(0, T) is dened by
_
_
u(), v
_
,
_
:=
_
T
0
_
u(t), v
_
(t)dt, for any T(0, T),
and similarly
_
a
_
u(), v
_
,
_
:=
_
T
0
a
_
u(t), v
_
(t)dt, for any T(0, T).
Also, we can rewrite the variational equation (0.1.6) in terms of distribu-
tions both in temporal and spatial variables. Since T() is dense in H
1
0
(),
we can choose v T() , and (0.1.6) is equivalent to

T
u(x, t)v(x)

(t)d(x, t)
_

T
u(x, t)v

(x)(t)d(x, t) = 0, (0.1.7)
where
T
= (0, T) and d(x, t) = dxdt. Set = v := v(x)(t), then
T(
T
) and (0.1.7) becomes

T
u

t
d(x, t)
_

T
u

x
2
d(x, t) = 0, (0.1.8)
for all T(
T
) with v T(), T(0, T). As the set of linear combi-
nations of functions v with v T(), T(0, T), is dense in T(
T
),
we have that (0.1.8) holds for all T(
T
):

u
t


2
u
x
2
,
_
= 0, T(
T
), (0.1.9)
2001 CRC Press LLC
2001 CRC Press LLC
that is the equation in (0.1.1) is satised in the sense of distributions in
T
.
Thus, we have demonstrated that the variational formula (0.1.6) implies the
distributional formula (0.1.9). The converse is not true in general, but it is
true in this example since we know that u had values in H
1
0
().
Now if we assume some regularity for u
0
, say u
0
H
1
0
() (or D(A)),
then
u
t
L
2
_
[0, T], L
2
()
_
and Au L
2
().
Therefore, in this case the equation in (0.1.1)
u
t
= Au
is satised in L
2
_
[0, T], L
2
()
_
and we say that u is a strong solution of
the Cauchy problem (0.1.1).
Next, we rephrase this discussion using the semigroup terminology.
Firstly, we rewrite (0.1.1) in the following abstract form on X = L
2
():
u

(t) = Au(t), t [0, T], u(0) = u

0
, (0.1.10)
where, as before
A =
d
2
dx
2
with D(A) = H
2
() H
1
0
(),
is an unbounded linear operator on L
2
(). Note that v D(A) if and only
if

k=1

2
k
[v
k
[
2
< , where v
k
= (v, e
k
). For each t 0, dene a linear
operator on L
2
() by
U(t)v :=

k=1
e

k
t
v
k
e
k
, v L
2
().
We note that, due to (0.1.3), U(t) is a bounded operator for each t 0.
Furthermore, since we have
_
_
U(t)v
_
_
2
L
2
()
=

k=1

k
t
v
k

k=1

v
k

2
= |v|
2
,
then
_
_
U(t)v
_
_
L
2
()
1 for each t 0. Now we show that the semigroup
property
U(t)U(s) = U(t +s)
is satised for all t, s 0. Let v X. Then, using the self-adjointness of
U(t), we obtain
U(t)U(s)v = U(t)
_
U(s)v
_
=

k=1
e

k
t
_
U(s)v, e
k
_
e
k
2001 CRC Press LLC
2001 CRC Press LLC
=

k=1
e

k
t
_
v, U(s)e
k
_
e
k
=

k=1
e

k
t
_
v, e

k
s
e
k
_
e
k
=

k=1
e

k
(t+s)
_
v, e
k
_
e
k
= U(t +s)v.
Thus, U is a C
0
-semigroup of contractions. Now we consider
u(t) := U(t)u
0
, t 0, u
0
X.
The estimate (0.1.3) implies that u is continuous in t 0 and u(0) = u
0
in
X. Let u
0
D(A). Then

k=1
[
k
e

k
t
u
0
k
[
2

k=1
[
k
u
0
k
[
2
<
implies that
u

(t) =

k=1

k
e

k
t
u
0
k
e
k
exists for all t 0 and is continuous in t 0. Furthermore,

k=1
[
k
_
U(t)u
0
, e
k
_
[
2
=

k=1
[
k
e

k
t
u
0
k
[
2
<
for all t 0, hence u(t) D(A) and
Au(t) =

k=1

k
_
U(t)u
0
, e
k
_
e
k
=

k=1

k
e

k
t
u
0
k
e
k
= u

(t).
Therefore, if u
0
D(A), then the function u() = U()u
0
is a (strong)
solution of the Cauchy problem (0.1.10). Also it is not dicult to show
that
A = lim
h0+
U(h) I
h
,
which means that the operator A is the generator of the semigroup U.
Now again let u
0
D(A) and consider a (Schwartz) distribution | from
T

dened by
|() := u()H(), ) =
_

0
u(t)(t)dt, T T(, T),
where u is dened by (0.1.2) and
H(t) =
_
1, t 0
0, t < 0
2001 CRC Press LLC
2001 CRC Press LLC
is the Heaviside function. Then | satises the equation in (0.1.1) in the
follwing sense
|

A| = u
0
,
where is the Dirac distribution (delta function), or
|

, ) A|, ) = u
0
, ), T,
The operator-valued distribution S dened by
S()u
0
S, )u
0
:= |, )
=
_
T
0
u(t)(t)dt =
_
T
0
(t)

k=1
e

k
t
u
0
k
e
k
dt
is a distribution semigroup.
0.2 The reversed Cauchy problem for the
Heat equation
In the same setting as above, we consider the ill-posed Cauchy problem
u(x, t)
t
+

2
u(x, t)
x
2
= 0, t [0, T], x ,
u(0, t) = u(1, t) = 0,
u(x, 0) = u
0
(x),
which can be written in the abstract form
u

(t) = Bu(t), t [0, T], u(0) = u

0
(0.2.11)
with B = A. Formally, we can write the solution of (0.2.11) at t = T as
u(T) = U(T)u
0
=

k=1
e

k
T
u
0
k
e
k
,
where operators U(t), t 0, are in general unbounded, and therefore they
do not form a C
0
-semigroup. A solution to problem (0.2.11) can be found
using the quasi-reversibility method, which consists of solving the following
well-posed Cauchy problem
u

(t) = (B B
2
)u

(t), t [0, T], > 0, (0.2.12)

u

(0) = u
0
,
2001 CRC Press LLC
2001 CRC Press LLC
and then showing that u

(T) u(T) in some sense. The unique solution

of (0.2.12) is given by
u

(T) = U

(T)u
0
=

k=1
e
(
k

2
k
)T
u
0
k
e
k
,
and here U

is a C-regularized semigroup with the operator C dened by

Cv =

k=1
e

2
k
T
v
k
e
k
, v X.
Alternatively, one can use the auxiliary boundary conditions method, which
consists of solving the well-posed boundary value problem
u

(t) = Bu

(t), t [0, T +], > 0, > 0,

u

(0) +u

(T +) = u
0
,
The solution to this problem is given by
u

(T) = U

(T)u
0
=

k=1
e

k
T
1 +e

k
(T+)
u
0
k
e
k
,
where again U

is a C-regularized semigroup with the operator C dened

by
Cv =

k=1
1
1 +e

k
(T+)
v
k
e
k
, v X.
0.3 Wave equation
Consider the problem

2
u(x, t)
t
2


2
u(x, t)
x
2
= 0, t [0, T], x = (0, 1),
(0.3.13)
u(0, t) = u(1, t) = 0,
u(x, 0) = u
0
(x),
u
t
(x, 0) = u
1
(x),
in L
2
(). Let
u
0
=

k=1
u
0
k
e
k
, u
1
=

k=1
u
1
k
e
k
,
2001 CRC Press LLC
2001 CRC Press LLC
where e
k

k=1
is an orthonormal basis in L
2
() consisting of eigenvectors
e
k
=

2 sinkx, k N, of the operator A =
d
2
dx
2
with the domain
D(A) = H
2
() H
1
0
() L
2
().
We are looking for a L
2
() solution of (0.3.13) by setting
u(t) =

k=1
u
k
(t)e
k
.
The unique solution of each problem
d
2
u
k
dt
2
+
k
u
k
= 0, k N, (0.3.14)
u
k
(0) = u
0
k
, u

k
(0) = u
1
k
,
is
u
k
(t) = u
0
k
cos (

k
t) +u
1
k
sin (

k
t)

k
, k N. (0.3.15)
Thus, we can formally write
u(t) =

k=1
u
k
(t)e
k
(0.3.16)
=

k=1
cos (

k
t) u
0
k
e
k
+

k=1
sin (

k
t)

k
u
1
k
e
k
.
Now we investigate the convergence of the series in (0.3.16).
Case 1. u
0
L
2
(), u
1
H
1
().
Here H
1
() = L(H
1
0
(), R) is the dual space of H
1
0
(), and we have the
following continuous injections
H
1
0
() L
2
() H
1
().
Recall that u
0
L
2
() and u
1
H
1
() if and only if

k=1
[u
0
k
[
2
< and

k=1
[u
1
k
[
2

k
< ,
respectively. Under these assumptions, the sequence
U
m
() =
m

k=1
u
k
()e
k
2001 CRC Press LLC
2001 CRC Press LLC
is a Cauchy sequence in C
_
[0, T], L
2
()
_
since
sup
t[0,T]
[u
k
(t)[
2
2
_
[u
0
k
[
2
+
[u
1
k
[
2

k
_
, k N.
Further, since
u

k
(t) =

k
u
0
k
sin (

k
t) +u
1
k
cos (

k
t),
we have
1

k
sup
t[0,T]
[u

k
(t)[
2
2
_
[u
0
k
[
2
+
[u
1
k
[
2

k
_
, k N,
therefore, for all m N and all p 0
sup
t[0,T]
_
_
_U

m+p
(t) U

m
(t)
_
_
_
2
H
1
()
= sup
t[0,T]
m+p

k=m+1
[u

k
(t)[
2

k
2
m+p

k=m+1
_
[u
0
k
[
2
+
[u
1
k
[
2

k
_
,
and
U

m
() =
m

k=1
u

k
()e
k
is a Cauchy sequence in C
_
[0, T], H
1
()
_
. Thus we have shown that U
m
and U

m
converge to
u C
_
[0, T], L
2
()
_
and
u
t
C
_
[0, T], H
1
()
_
,
respectively. Also taking into account
U
m
(0) =
m

k=1
u
0
k
e
k
, U

m
(0) =
m

k=1
u
1
k
e
k
,
we have
u(0) = u
0
and
u
t
(0) = u
1
.
So, if u
0
L
2
(), u
1
H
1
(), then the function u dened by (0.3.16),
can satisfy the equation in (0.3.13) in some very weak sense, which we will
discuss after the next case.
Case 2. u
0
H
1
0
(), u
1
L
2
().
2001 CRC Press LLC
2001 CRC Press LLC
In this case we have

k=1

k
[u
0
k
[
2
< ,

k=1
[u
1
k
[
2
< ,
and
sup
t[0,T]

k
[u
k
(t)[
2
2
_

k
[u
0
k
[
2
+[u
1
k
[
2
_
, k N,
sup
t[0,T]
[u

k
(t)[
2
2
_

k
[u
0
k
[
2
+[u
1
k
[
2
_
, k N,
which implies that
u C
_
[0, T], H
1
0
()
_
and
u
t
C
_
[0, T], L
2
()
_
,
with u(0) = u
0
,
u
t
(0) = u
1
. Note that

2
u
t
2
C
_
[0, T], H
1
()
_
,
as
1

k
sup
t[0,T]
[u

k
(t)[
2

k
sup
t[0,T]
[u
k
(t)[
2
(0.3.17)
2
_

k
[u
0
k
[
2
+[u
1
k
[
2
_
, k N.
In a fashion similar to the Heat equation (see (0.1.4)), from (0.3.14) we
write
_
U

m
(t), e
k
_
+a
_
U
m
(t), e
k
_
= 0, 1 k m.
Now let T(0, T), then for any xed k < m, we have

_
T
0
_
U

m
(t), e
k
_

(t)dt +
_
T
0
a
_
U
m
(t), e
k
_
(t)dt = 0.
The uniform convergence of U
m
to u in C
_
[0, T], H
1
0
()
_
implies that
lim
m
_
T
0
a
_
U
m
(t), e
k
_
(t)dt =
_
T
0
a(u(t), e
k
)(t)dt,
and uniform convergence of U

m
to
u
t
in C
_
[0, T], L
2
()
_
gives
lim
m
_
T
0
_
U

m
(t), e
k
_

(t)dt =
_
T
0
_
u
t
(t), e
k
_

(t)dt.
2001 CRC Press LLC
2001 CRC Press LLC
We conclude that u satises the equation in (0.3.13) in the following sense

_
T
0
(
u
t
(t), v)

(t)dt +
_
T
0
a(u(t), v)(t)dt = 0,
for all v H
1
0
() and T(0, T). In other words, the equation
d
dt
_
u
t
(), v
_
+a
_
u(), v
_
= 0 (0.3.18)
is satised in T

(0, T) for all v H

1
0
().
Now we return to the Case 1. From (0.3.14) we obtain
_
T
0
_
U
m
(t), e
k
_

(t)dt
_
T
0
_
U
m
(t), Ae
k
_
(t)dt = 0,
for all xed k < m and for all T(0, T), therefore
_
T
0
(u(t), e
k
)

(t)dt
_
T
0
(u(t), Ae
k
)(t)dt = 0,
for any k N and for all T(0, T). Since e
k

k=1
forms a Hilbert basis
in D(A), we obtain
_
T
0
(u(t), v)

(t)dt
_
T
0
(u(t), Av)(t)dt = 0,
for all v D(A), T(0, T), or in other words, the equation
d
2
dt
2
_
u(), v
_
=
_
u(), Av
_
(0.3.19)
is satised in T

(0, T) for all v D(A).

Case 3. u
0
D(A), u
1
H
1
0
().
Here we have

k=1

2
k
[u
0
k
[
2
< ,

k=1

k
[u
1
k
[
2
< .
Therefore (0.3.17) implies the uniform convergence of U

m
to

2
u
t
2
, and of
AU
m
to Au, in the space C
_
[0, T], L
2
()
_
. Thus, the function u dened
by (0.3.16), satises the equation

2
u
t
2
= Au(t)
2001 CRC Press LLC
2001 CRC Press LLC
in L
2
().
Now we describe these three cases using the semigroup terminology.
Problem (0.3.13) can be written in the abstract form
u

(t) = Au(t), t [0, T], (0.3.20)

u(0) = u
0
, u

(0) = u
1
,
where
A =
d
2
dx
2
with D(A) = H
2
() H
1
0
() L
2
() X
is an unbounded linear operator on L
2
(). Taking into account the dis-
cussion of Case 1, we dene the following bounded linear operators on
X:
((t)v :=

k=1
cos (

k
t) v
k
e
k
, o(t)v :=

k=1
sin(

k
t)

k
v
k
e
k
,
which are the cos- and sin-operator-functions corresponding to our prob-
lem. The unique solution of (0.3.13) is given by
u(t) = ((t)u
0
+o(t)u
1
,
where the equation in (0.3.13) is satised as described in Cases 13 ac-
cording to the choice of the initial data.
Now we write the problem (0.3.20) in the form of the rst order equation
in a product space
w

(t) = w(t), t [0, T], w(0) =

_
u
0
u
1
_
, (0.3.21)
where
w(t) =
_
u(t)
u

(t)
_
L
2
() L
2
(),
=
_
0 I
A 0
_
, D() = D(A) L
2
().
The solution operator of (0.3.21) can be formally written as
w(t) = U(t)
_
u
0
u
1
_
:=
_
((t) o(t)
(

(t) o

(t)
__
u
0
u
1
_
=
_
((t)u
0
+o(t)u
1
(

(t)u
0
+o

(t)u
1
_
, t 0, u
0
, u
1
L
2
(),
2001 CRC Press LLC
2001 CRC Press LLC
and noting that for any v L
2
(), o

(t)v = ((t)v, we write U in the form

U(t) =
_
((t) o(t)
(

(t) ((t)
_
, t 0.
From the denition of ( it is clear that ( is not necessarily dierentiable
in t on L
2
(), implying that the operators U(t) are in general unbounded
on L
2
() L
2
(), and therefore they do not form a C
0
-semigroup on this
space.
Let us consider a smaller space H
1
0
() L
2
(), and let
=
_
0 I
A 0
_
, with D() = D(A) H
1
0
().
From the calculations in Case 2 we deduce that the operators U(t) are
bounded on H
1
0
() L
2
(), and they form a C
0
-semigroup generated by
. Further, if u
0
D(A), u
1
H
1
0
(), then U gives the unique strong
solution of (0.3.21), and if u
0
H
1
0
(), u
1
L
2
(), then U gives the
unique solution of (0.3.21) in the variational sense.
On the initial space L
2
() L
2
() we consider the integral of the
operator-function U():
V (t) =
_
o(t)
_
t
0
o(s)ds
((t) I o(t)
_
, t 0.
The operators V (t) are bounded and strongly continuous on L
2
()L
2
().
Moreover, they form an integrated semigroup generated by , which gives
the unique solution of the integrated problem (0.3.21):
v(t) t
_
u
0
u
1
_
=
_
t
0
v(s)ds, t [0, T], (0.3.22)
v =
_
v
1
v
2
_
, u
0
, u
1
L
2
(),
which is equivalent to
v
1
(t) tu
0
=
_
t
0
v
2
(s)ds,
v
2
(t) tu
1
=
_
t
0
Av
1
(s)ds.
This solution is equivalent to the solution obtained in Case 1 in the
following sense: a function v() is a solution of (0.3.22) if and only if
u() = v
2
() + u
0
is a solution of (0.3.19) in T

(0, T) for all v D(A).

2001 CRC Press LLC
2001 CRC Press LLC
Note that in this case v
1
(t) =
_
t
0
u(s)ds. Also if we take
_
u
0
u
1
_
D(
2
),
then V ()
_
u
0
u
1
_
is dierentiable in t, and V

()
_
u
0
u
1
_
is the unique
solution of (0.3.21).
Finally, we note here that the operator is the generator of the distri-
bution semigroup 1 dened by
1()v =
_

0

(t)V (t)vdt, T, v L
2
() L
2
(),
which can be regarded as the distributional derivative of the integrated
semigroup V .
2001 CRC Press LLC
2001 CRC Press LLC
Chapter 1
Semigroup Methods
1.1 C
0
-semigroups
We begin with a brief summary of very basic notions and results from the
theory of C
0
-semigroups.
1.1.1 Denitions and main properties
Let X be a Banach space.
Denition 1.1.1 A one-parameter family of bounded linear operators
U(t), t 0 is called a C
0
-semigroup (or semigroup of class C
0
) if the
following conditions hold
(U1) U(t +h) = U(t)U(h), t, h 0;
(U2) U(0) = I;
(U3) U(t) is strongly continuous with respect to t 0.
Denition 1.1.2 The operator dened by
U

(0)x := lim
h0
h
1
(U(h) I)x,
D(U

(0)) =
_
x X

lim
h0
U(h) I
h
x
_
,
is called the (innitesimal) generator of the semigroup U.
The properties of C
0
-semigroups are established in the following
proposition.
2001 CRC Press LLC
2001 CRC Press LLC
Proposition 1.1.1 Let U(t), t 0 be a C
0
-semigroup and A be its
generator. Then
1) U(t)U(h) = U(h)U(t) for all t, h 0;
2) U is exponentially bounded:
K > 0, R : t 0,
|U(t)| Ke
t
; (1.1.1)
3) D(A) = X and operator A is closed;
4) U

(t)x = U(t)Ax = AU(t)x for all x D(A);

5) C : Re > , (I A)
1
=: R
A
() and
R
A
()x =
_

0
e
t
U(t)xdt, x X. (1.1.2)
Proof 1) The commutativity of U(t) and U(h), t, h 0, follows from
property (U1).
2) An arbitrary t > 0 can be written in the form t = n + , where
n N, 0 < 1. Then we have U(t) = U
n
(1)U(), and
|U(t)| |U(1)|
n
K = Ke
nln U(1)
,
where K = sup
01
|U()| < .
If ln|U(1)| 0, then |U(t)| K, that is (1.1.1) holds for = 0. If
ln|U(1)| > 0, then
|U(t)| Ke
(n+) ln U(1))
= Ke
t ln U(1)
= Ke
t
,
where = ln|U(1)|.
3) To prove that D(A) = X, we consider the set
| :=
_
v
a,b
=
_
b
a
U()ud, x X, b > a > 0
_
.
We show that | D(A):
h
1
_
U(h) I
_
v
a,b
= h
1
_
b
a
[U(h +) U()] xd
= h
1
_
_
b+h
a+h
U(t)xdt
_
b
a
U()xd
_
= h
1
_
_
b+h
b
U()xd
_
a
a+h
U()xd
_

_
U(b) U(a)
_
x as h 0.
2001 CRC Press LLC
2001 CRC Press LLC
Now we will show that | = X. Suppose that there exists f X

such that
f(|) = 0, f , 0. Then
b > a > 0, f(v
a,b
) =
_
b
a
f(U()x)d = 0.
Hence
x X, f(U()x) = 0, > 0,
and f(U()x)
0
f(x) = 0, that is f 0. This contradiction proves the
equality | = X, therefore D(A) = X.
To prove the closedness of the generator A, we take a sequence x
n

D(A) such that x
n
x and Ax
n
v. We show that Ax exists and is
equal to v :
h
1
_
h
0
U()U

(0)x
n
d = h
1
_
h
0
dU()x
n
d
d = h
1
_
U(h) I
_
x
n
.
Let n , then we have
h
1
_
h
0
U()vd = h
1
_
U(h) I
_
x.
Let h 0, then the left-hand side converges to v, that is x D(U

(0)) and
v = U

(0)x = Ax.
4) For x D(A) it follows from Denition 1.1.2 that
U(t)Ax = U(t) lim
h0
h
1
_
U(h) I
_
x
= lim
h0
h
1
_
U(h) I
_
U(t)x = AU(t)x,
U

(t)x = lim
h0
h
1
_
U(t +h) U(t)
_
x
= lim
h0
h
1
_
U(t)U(h) U(t)
_
x
= U(t) lim
h0
h
1
_
U(h) I
_
x = U(t)Ax = AU(t)x.
5) In view of estimate (1.1.1) the integral in the right-hand side of
(1.1.2):
_

0
e
t
U(t)xdt,
exists for all x X and C with Re > . Integrating by parts and
taking into account the fact that A is closed, we have for x D(A)
_

0
e
t
U(t)Axdt = A
_

0
e
t
U(t)xdt =
_

0
e
t
U

(t)xdt
= x +
_

0
e
t
U(t)xdt.
2001 CRC Press LLC
2001 CRC Press LLC
After extending the equality by continuity to X = D(A) we obtain
(I A)
_

0
e
t
U(t)xdt = x, x X. (1.1.3)
On the other hand,
_

0
e
t
U(t)(I A)xdt = x, x D(A). (1.1.4)
Equalities (1.1.3), (1.1.4) imply the existence of a bounded on X operator
R
A
() = (I A)
1
and relation (1.1.2).
Proposition 1.1.2 Let U be a strongly continuous operator-function such
that
K > 0, R : |U(t)| Ke
t
, t 0.
Let
R() :=
_

0
e
t
U(t)dt , Re > .
Then R() satises the resolvent identity
( )R()R() = R() R(), Re , Re > , (1.1.5)
if and only if U satises (U1).
Proof Let Re , Re > . The result follows from the uniqueness theorem
for Laplace transforms given the following observations
R()R() =
_

0
e
t
_

0
e
t
U(s)U(t)dsdt,
and
R() R()

=
_

0
e
()t
R()dt
_

0
( )
1
e
()t
e
t
U(t)dt
=
_

0
e
()t
_

0
e
s
U(s)dsdt
_

0
e
()t
_
t
0
e
s
U(s)dsdt
=
_

0
e
()t
_

t
e
s
U(s)dsdt
=
_

0
e
t
_

t
e
(st)
U(s)dsdt
=
_

0
e
t
_

0
e
s
U(s +t)dsdt.
2001 CRC Press LLC
2001 CRC Press LLC
1.1.2 The Cauchy problem
Consider the Cauchy problem
u

(t) = Au(t), t 0, u(0) = x, (CP)

where A is a linear operator with D(A) X.
Denition 1.1.3 A function
u() C
1
_
[0, ), X
_
C
_
[0, ), D(A)
_
is called a solution of the Cauchy problem (CP) if u(t) satises the equation
for t 0 and the initial condition for t = 0.
Denition 1.1.4 The Cauchy problem (CP) is said to be uniformly well-
posed on E X, (where E = X) if
(a) a solution exists for any x E;
(b) the solution is unique and for any T > 0 is uniformly stable for
t [0, T] with respect to the initial data.
Remark 1.1.1 According to Hadamards denition of correctness, De-
nition 1.1.4 includes the condition of stability of a solution with respect
to the initial data. Now we show that the Cauchy problem (CP) has the
following special property.
Lemma 1.1.1 If for any x D(A) there exists a unique solution of (CP)
and the resolvent set (A) is not empty, then this solution is stable with
respect to the initial data.
Proof Let u(t) = U(t)x, t 0, be a unique solution of the Cauchy problem
with initial value x D(A), where
U() : [D(A)] C
_
[0, T], [D(A)]
_
are the solution operators for any T > 0. Here [D(A)] is the Banach space
_
D(A), |x|
A
= |x| +|Ax|
_
.
We show that all U(t) are closed. Suppose x
n
x in [D(A)] and U(t)x
n
=
u
n
(t) v(t) in C
_
[0, T], [D(A)]
_
. Then u

n
(t) = Au
n
(t) Av(t) in X
uniformly in t, therefore from
u
n
(t) = x
n
+
_
t
o
u

n
()d
2001 CRC Press LLC
2001 CRC Press LLC
we obtain
v(t) = x +
_
t
o
Av()d,
that is v is a solution of the Cauchy problem with initial value x, continu-
ously dierentiable on [0, T]. Since a solution is unique, v(t) = U(t)x, t 0.
Thus, operators U(t) are everywhere dened on [D(A)] and closed, there-
fore by the Banach theorem the operator-function U() is continuous and
sup
t[0,T]
|u(t)|
A
K|x|
A
, (1.1.6)
which implies that the Cauchy problem is uniformly well-posed in the space
[D(A)].
Since the resolvent set (A) of operator A is not empty, consider

0
(A), x D(A) and y = R(
0
)x, then we have y D(A
2
) and
U(t)x = U(t)Ay +
0
U(t)y = AU(t)y +
0
U(t)y,
therefore
|U(t)x| |AU(t)y| +[
0
[|U(t)y| K
0
|U(t)y|
A
.
From (1.1.6) we obtain
|U(t)x| K
1
|y|
A
= K
1
_
|y| +|Ay|
_
= K
1
_
|y| +|x
0
y|
_
K
2
_
|x| +|y|
_
K
2
_
|x| +|R(
0
)| |x|
_
K|x|, t [0, T],
that is (CP) is uniformly well-posed in the space X.
Theorem 1.1.1 Suppose that A is a closed densely dened linear operator
on X. Then the following statements are equivalent:
(I) the Cauchy problem (CP) is uniformly well-posed on D(A);
(II) the operator A is the generator of a C
0
-semigroup U(t), t 0;
(III) Miyadera-Feller-Phillips-Hille-Yosida (MFPHY) condition holds for
the resolvent of operator A: there exist K > 0, R so that
_
_
_R
(k)
A
()
_
_
_
K k!
(Re )
k+1
, (1.1.7)
for all C with Re > , and all k = 0, 1, . . . .
2001 CRC Press LLC
2001 CRC Press LLC
In this case the solution of (CP) has the form
u() = U()x, x D(A).
Proof (I) = (II). The solution u(t), t 0, of the Cauchy problem (CP)
exists and is unique for any x D(A). Denote it by U()x. Since for any
T > 0 the solution is uniformly stable with respect to t [0, T], dened
in such a way linear operators U(t) are uniformly bounded with respect to
t [0, T] on D(A). In view of the condition D(A) = X, this implies that
U(t) can be extended by continuity to the whole space with preservation
of the norm estimate. We show that the obtained family of linear bounded
operators U(t), t 0 is a C
0
-semigroup.
Since U(t)x satises the equation U

(t)x = AU(t)x for all x D(A)

and t 0, we have U(t)x D(A) whenever x D(A). For x D(A)
the functions U(t + h)x and U(t)U(h)x are solutions of (CP) with initial
condition U(h)x. The uniqueness of the solution gives us the equality
x D(A), U(t +h)x = U(t)U(h)x, t, h 0,
which can be extended to the whole X.
Thus (U1) is proved, and (U2) follows from the initial condition. Since
|U(t)| is uniformly bounded with respect to t [0, T], and U(t)x is continu-
ous with respect to t 0 on D(A) (D(A) = X), then the operator-function
U() is strongly continuous with respect to t 0 , so (U3) holds true.
Furthermore for x D(A) we have
lim
h0
h
1
_
U(h) I
_
x = U

(0)x = AU(0)x = Ax,

that is D(A) D(U

(0)) = D, and U

(0) = A on D(A). To prove that

D D(A), we show that the resolvent
R
U

(0)
() =
_

0
e
t
U(t)dt, Re > ,
is equal to the resolvent of operator A. For x D(A) D we have
(I A)R
U

(0)
()x = (I A)
_

0
e
t
U(t)xdt
=
_

0
e
t
AU(t)xdt +
_

0
e
t
U(t)xdt
= x.
Since A is closed, this equality can be extended to the whole X. The oper-
ator R
U

(0)
() maps X on D, therefore D D(A), and A is the generator
of the C
0
-semigroup U(t), t 0.
2001 CRC Press LLC
2001 CRC Press LLC
(II) = (III). The estimates (1.1.7) follow from (1.1.2) and (1.1.1). In
fact,
|R
A
()| =
_
_
_
_
_

0
e
t
U(t)dt
_
_
_
_

_

0
[ e
t
[ |U(t)|dt
K
_

0
e
[Re ]t
dt =
K
Re
,
_
_
_
_
d
d
R
A
()
_
_
_
_
=
_
_
_
_
_

0
te
t
U(t)dt
_
_
_
_
K
_

0
te
[Re ]t
dt
=
K 1!
(Re )
2
and so on
_
_
_
_
d
k
d
k
R
A
()
_
_
_
_
=
_
_
_
_
_

0
t
k
e
t
U(t)dt
_
_
_
_
K
_

0
t
k
e
[Re ]t
dt
=
K k!
(Re )
k+1
.
(III) = (I). First we construct the solution of (CP) for suciently
smooth initial data. Noting that D(A
3
) = X, let x D(A
3
) and
> max, 0. Dene
u(t, x) = x+tAx+
t
2
2
A
2
x+
1
2i
_
+i
i

3
e
t
R
A
()A
3
xd, t 0. (1.1.8)
The integral in (1.1.8) vanishes for t = 0, consequently u(0, x) = x. It
is obvious that | u(t, x)| =
t
O(e
t
). The possibility of dierentiation
under the integral sign implies the existence of a continuous derivative of
u(t, x), t 0. On the other hand, in view of the closedness of operator A,
u(t, x) D(A) and
u

(t, x) A u(t, x)
=
t
2
2
A
3
x +
1
2i
_
+i
i

3
e
t
(I A)R
A
()A
3
xd = 0,
for t 0. Hence, u(t, x), t 0, x D(A
3
), is a solution of (CP). In the
way similar to that in the proof of equality (1.1.2) in Proposition 1.1.1, we
have
_

0
e
t
u(t, x)dt = R
A
()x. (1.1.9)
Using the Widder-Post inversion formula for Laplace transform, we obtain
u(t, x) = lim
n
(1)
n
n!
_
n
t
_
n+1
d
n
d
n
R
A
()x

=
n
t
, (1.1.10)
2001 CRC Press LLC
2001 CRC Press LLC
therefore, taking into account inequalities (1.1.7) we have
| u(t, x)| K|x| lim
n
_
1
t
n
_
(n+1)
= Ke
t
|x|, (1.1.11)
for t 0. Denote

U(t)x u(t, x), x D(A

3
), t 0.
In view of (1.1.11) we can extend

U()x and equality (1.1.9) by continuity
to the whole of X so that the extension U() is strongly continuous with
respect to t 0 and
|U(t)| Ke
t
, t 0.
To complete the proof we show that any solution u() of (CP) can be rep-
resented in the form
u() = U()u(0). (1.1.12)
In fact, it follows from the denition of U that for x D(A
3
) we have the
equality R
A
()U(t)x = R
A
()

U(t)x =

U(t)R
A
()x, t 0, which can be
extended to whole X. Therefore
U(t)Ax = AU(t)x, x D(A), t 0,
function U()x is dierentiable for x D(A), and for any solution u() of
(CP) we have
d
ds
U(t s)u(s) = AU(t s)u(s) +U(t s)Au(s) = 0, 0 s t.
Hence
U(0)u(t) = U(t)u(0) = u(t), t 0.
Remark 1.1.2 Since we proved in Proposition 1.1.1 that the generator of
a C
0
-semigroup is densely dened, we could include the assumption of the
density of D(A) in statements (I) and (III) only.
Remark 1.1.3 Using the equality
d
k
d
k
R
A
()x = (1)
k
k!R
k+1
A
()x, x X, (1.1.13)
we can rewrite the MFPHY condition (1.1.7) in the following form:
K > 0, R:
|R
k
A
()|
K
(Re )
k
, (1.1.14)
2001 CRC Press LLC
2001 CRC Press LLC
for all C with Re > , and all k = 1, 2 . . . .
Remark 1.1.4 The proof of the implication (III) = (I) (via (II)) is given
in Theorem 1.1.1 due to Fattorini [84] and is based on the inversion formula
for Laplace transform. Now we outline alternative proofs of the implication
(III) = (II), which exploit some dierent ideas used in various generaliza-
tions of C
0
-semigroups. We refer to these later on.
First, we note that the right-hand side of (1.1.10) (the inversion formula
for Laplace transform) can be rewritten in the form
lim
n
_
t
n
_
n
R
n
A
_
t
n
_
= lim
n
_
I
t
n
A
_
n
,
and (1.1.10) is an operator generalization of the formula
e
ta
= lim
n
_
1
t
n
a
_
n
for the numerical exponential function.
Using this formula one can construct a semigroup U generated by an
operator A as a strong limit of the following functions
U
n
(t) =
_
I
t
n
A
_
n
, t 0, n = 1, 2, . . . .
See [130] for one of the possible proofs.
Now we give another important alternative proof. Consider
bounded operators
A
n
=
n
I +
2
n
R
A
(
n
),
where
n
R and
n
. We show that lim
n
A
n
x = Ax for any
x D(A). Consider x D(A), then
A
n
x =
n
R
A
(
n
)Ax.
Noting that lim
n

n
R
A
(
n
)x = x for any x X, we obtain
A
n
x Ax, x D(A).
Since operators A
n
are bounded, we can dene a semigroup generated by
each operator A
n
in the form of the following series
e
tA
n
= e

n
t+
2
n
R
A
(
n
)t
= e

n
t

k=0
(
2
n
t)
k
R
A
(
n
)
k
k!
,
where
|e
tA
n
| e

n
t

k=0

2k
n
t
k
K
k!(
n
)
k
= Ke

n
t
e

2
n
t

= Ke
t
n

Ke
2t
,
n
> 2. (1.1.15)
2001 CRC Press LLC
2001 CRC Press LLC
Consider the identity
e
tA
n
e
tA
m
=
_
t
0
e
(ts)A
n
(A
n
A
m
)e
sA
m
ds,
then, for x D(A), we have
|e
tA
n
x e
tA
m
x|
_
t
0
|e
(ts)A
n
| |e
sA
m
| |(A
n
A
m
)x|ds,
therefore, in view of (1.1.15), functions e
tA
n
x converge uniformly in
t [0, T]. Taking into account density of D(A) and uniform boundness
of e
tA
n
on [0, T], we obtain that e
tA
n
x converge uniformly in t [0, T] for
any x X. We denote
U(t)x := lim
n
e
tA
n
x,
then U is obviously a C
0
-semigroup with the estimate |U(t)| Ke
2t
,
which follows from the inequality (1.1.15) by a passage to the limit as
n .
Let us prove that the operator A is the generator of this semigroup.
The identity
e
tA
n
x x =
_
t
0
e
sA
n
A
n
xds, x D(A),
implies that
U(t)x x =
_
t
0
U(s)Axds, x D(A),
and hence U

(0)x = Ax, that is U

(0) A, and
_
I U

(0)
_
R
A
() =
_
I A
_
R
A
() = I.
In view of Proposition 1.1.1 the operator U

(0) has a resolvent for C

with Re > , therefore R
A
() = R
U

(0)
() and U

(0) = A.
We also note that this construction of a C
0
-semigroup implies that the
condition
K > 0, R:
_
_
_R
(k)
A
()
_
_
_
K k!
( )
k+1
,
for all > and all k = 0, 1, . . . ,
is equivalent to condition (1.1.7) and is also referred to as the MFPHY
condition.
2001 CRC Press LLC
2001 CRC Press LLC
Remark 1.1.5 It follows from the equality (1.1.13) that the condition
0 : |R
A
()|
1
Re
, Re >
or
0 : |R
A
()|
1

, > ,
which is called the Hille-Yosida condition, is sucient for condition (1.1.7)
to hold with K = 1, and therefore for uniform well-posedness of (CP).
In general, the following sucient condition for the unique solvability
of (CP) holds.
Theorem 1.1.2 Suppose that for a linear operator A the following condi-
tion holds
K > 0, 0 : |R
A
()| K(1 +[[)

, Re > , (1.1.16)
for some 1. Then for any x D(A
[[]]+3
) the function

J(t) =
_
J(t) = v.p.
1
2i
_
+i
i
e
t
R
A
()xd, t > 0, >
J(+0), t = 0
yields the unique solution of (CP).
Proof Any x D(A
[[]]+3
) we can write in the form
x = R
A
()
[[]]+3
x
1
,
where x
1
= (I A)
[[]]+3
x. Then, using the resolvent identity (1.1.5), we
obtain
R
A
()x =
_
R
A
() R
A
()
_
R
A
()
[[]]+2

x
1
=
R
A
()
( )
[[]]+3
x
1

R
A
()
[[]]+3

x
1

R
A
()
[[]]+2
( )
2
x
1
. . .
R
A
()
( )
[[]]+3
x
1
for Re , Re > , ,= . By Jordans lemma we have
J(t)x = v.p.
1
2i
_
+i
i
e
t
R
A
()xd
= v.p.
1
2i
_
+i
i
e
t
R
A
()
( )
[[]]+3
x
1
d, t > 0.
2001 CRC Press LLC
2001 CRC Press LLC
It follows from (1.1.16) that the latter integral converges absolutely and
uniformly in t 0. After dierentiating under the integral sign we obtain
the integral
v.p.
1
2i
_
+i
i
e
t
R
A
()
( )
[[]]+3
x
1
d,
which also converges absolutely and uniformly in t 0, and therefore is
equal to J

(t) for t > 0. It is not dicult to check that J(t) satises the
equation in (CP) for t > 0. The existence of lim
t0
J

(t) :=

J

(0) and the

closedness of A imply that

J

(0) = A

J(0). By the Cauchy theorem we have

J(0) = R
A
()
[[]]+3
x
1
= x,
i.e.,

J(t) is a solution of the Cauchy problem (CP).
Now we show the uniqueness of the solution. Suppose that u() is a solu-
tion of (CP). Since u() is continuously dierentiable in t 0,
_
t
0
e

u()d
can be integrated by parts
_
t
0
e

u()d =
1
_
u(0) u(t)e
t
_
+
1
_
t
0
e

Au()d.
Suppose u(0) = 0, then taking into account closedness of A, we have
(AI)
_
t
0
e

u()d = u(t)e
t
,
therefore

_
t
0
e
(t)
u()d = R
A
()u(t),
which, together with the condition (1.1.16), implies that u() = 0 in [0, t],
i.e. the solution is unique.
Remark 1.1.6 The following result guarantees the uniqueness of the solu-
tion of (CP) under weaker than (1.1.16) conditions on operator A. It can
be proved similar to the uniqueness in the theorem above.
Lyubichs Uniqueness Theorem Suppose that a linear operator A has
a resolvent for all > and
limsup
+
ln|R
A
()|

= 0,
then the Cauchy problem (CP) has at most one weak solution (that is a solu-
tion that is dened and continuous for t 0 and continuously dierentiable
for t > 0).
2001 CRC Press LLC
2001 CRC Press LLC
In Section 1.2, using the technique of integrated semigroups, we show
that conditions of the type (1.1.16) on the resolvent of A, are necessary and
sucient for the existence of the solution and its stability with respect to
small variations of initial data measured in the graph-norm of A
n
, where n
depends on .
Remark 1.1.7 Another important class of operators closely related to a
well-posed (CP) is the generators of analytic semigroups. Let A be a densely
dened operator, 0

2
. Then A is the generator of a semigroup
analytic in
_
C

[arg [ , ,= 0
_
and strongly continuous in
_
C

[arg [
_
if and only if for every 0 <

< there exists a real number = (

)
such that
|R
A
()|
K
[ [
(1.1.17)
for from the region
_
C

[arg ( )[

+

2
, ,=
_
with K = K

.
We note that the estimate (1.1.17) need be assumed only in the half-
plane Re > . (For proofs see [84] Section 4.2)
Remark 1.1.8 One of the remarkable advantages of using semigroup meth-
ods for treating dierential equations in applications is the presence of the
following perturbation result.
Let A (closed densely dened linear operator) be the generator of a
C
0
-semigroup U and let B be a closed densely dened operator such that
D(B) D(A) and
|BU(t)x| (t)|x|, x D(A), t > 0,
where is a nite, measurable function from L
1
(0, 1). Then the operator
A + B with domain D(A + B) = D(A) is also the generator of a C
0
-
semigroup.
In particular, if B is everywhere dened and bounded, then A + B
generates a C
0
-semigroup U
1
with
|U
1
(t)| Ke
(+B)t
, t 0,
2001 CRC Press LLC
2001 CRC Press LLC
given that
|U(t)| Ke
t
, t 0.
For proofs see [84] Chapter 5 and [130] Chapter 9, where one can also
nd perturbation results for m-dissipative operators, essentially self-adjoint
operators and operators generating analytic semigroups.
1.1.3 Examples
Example 1.1.1
In Chapter 0 we used the Fourier method to construct various semi-
groups related to the Heat and Wave equations. Now we use a very simple
rst-order equation to illustrate the semigroup methods.
Consider the Cauchy problem
u(x, t)
t
+
u(x, t)
x
= 0, t 0, x R, (1.1.18)
u(x, 0) = f(x)
on the space X = L
2
(R). We can write it in the abstract form
u

(t) = Au(t), t 0, u(0) = f,

where A = d/dx with the domain
D(A) =
_
u L
2
(R)

L
2
(R)
_
,
To nd the resolvent of A, we solve the equation
(I A)g = g +g

= f, g D(A), (1.1.19)
assuming that f X is given. If > 0 then the solution is
g(x) = (R
A
()f) (x) =
_
x

e
(xs)
f(s)ds, x R.
Using Fourier transforms, it is not dicult to verify that the Hille-Yosida
condition:
|R
A
()|
1

,
holds for > 0. Hence (1.1.18) is uniformly well-posed on D(A). Further-
more, A is the generator of a C
0
-semigroup dened by
(U(t)f)(x) := f(x t), x R, t 0,
2001 CRC Press LLC
2001 CRC Press LLC
and for any f D(A), the function
u(x, t) = (U(t)f)(x), t 0, x R
is the unique solution of (1.1.18), which is stable with respect to f.
Now we consider the Cauchy problem (1.1.18) on the space
X = L
2
[0, ). In this case,
D(A) =
_
u L
2
[0, )

L
2
[0, ), u(0) = 0
_
,
and
(R
A
()f) (x) =
_
x
0
e
(xs)
f(s)ds, > 0, x [0, ).
The C
0
-semigroup generated by A is dened by
(U(t)f)(x) =
_
f(x t) , x t
0 , 0 x t
.
Finally, if X = L
2
(, 0] and
D(A) =
_
u L
2
(, 0]

L
2
(, 0], u(0) = 0
_
,
then all > 0 do not belong to the resolvent set of A. In this case, (1.1.18)
is solvable only for f 0.
Note that these results agree with the well-known geometric interpre-
tation: the solution is constant on the characteristics x = t + C, C R.
Example 1.1.2 (A class of operators generating C
0
-semigroups)
Let
X =
_
L
p
(R) L
p
(R), |u| = |u
1
|
L
p +|u
2
|
L
p
_
, where u =
_
u
1
u
2
_
.
Consider the operator A dened by
Au =
_
g f
0 g
_
u
with
D(A) =
__
u
1
u
2
_
X

gu
1
+fu
2
L
p
(R), gu
2
L
p
(R)
_
,
where g(x) = 1 +[x[, f(x) = [x[

, > 0.
2001 CRC Press LLC
2001 CRC Press LLC
We show that if (0, 1], then A generates a C
0
-semigroup on X. For
the formally dened operator-function
U(t) := e
At
= I +tA+
t
2
A
2
2!
+. . .
we can write
U(t)u = e
tg
_
1 tf
0 1
_
u
and
|U(t)| = max
xR
_
(1 + t[x[

)e
t(1+|x|)
_
= (1 +

t
1
)e
t
(
1+

t
)
=
t0
O(t
1
).
Let (0, 1], then U(t) is bounded as t 0. Now we show that the
operator (I A)
1
, > 0, is the resolvent of A and we verify the MFPHY
condition for R
A
(). Since
_
(I A)
1
_
k
u =
1
( +g)
2k
_
( +g)
k
k( +g)
k1
f
0 ( +g)
k
_
u
for k = 1, 2, . . ., then
|(I A)
k
u| |( +g)
k
u
1
|
L
p + | k( +g)
(k+1)
fu
2
|
L
p
+ |( +g)
k
u
2
|
L
p.
If > 0 then we have the following estimates
|( +g)
k
u
i
|
L
p =
_
_

0
[[
kp

1 +
g

kp
[u
i
[
p
dx
_
1/p

k
|u
i
|
L
p, i = 1, 2,
and
|k( +g)
k+1
fu
2
|
L
p =
__

kfu
2
( +g)
k+1

p
dx
_
1/p

1
[[
k
__

k
p
[1 +
g

[
kp

f
+g

p
[u
2
[
p
dx
_
1/p

1
[[
k
__

[f[
p
[u
2
[
p
[g[
p
dx
_
1/p

k
|u
2
|
L
p.
2001 CRC Press LLC
2001 CRC Press LLC
Therefore (I A)
1
= R
A
() and
|R
k
A
()u|
1
[[
k
|u
1
|
L
p +
K + 1
[[
k
|u
2
|
L
p
K
1

k
|u|
for any (0, 1] and > 0. Thus, if (0, 1], then A is the generator of
a C
0
-semigroup U. In general, we have
(I A)
1
u =
1
( +g)
2
_
( +g) f
0 ( +g)
_
u.
If 1 < 2, then the operator (I A)
1
, > 0, is bounded and therefore
|(I A)
1
| = |R
A
()| K, > 0,
but the resolvent R
A
() does not satisfy the MFPHY condition. If > 2,
then the operator (I A)
1
is unbounded.
1.2 Integrated semigroups
In Section 1.1 it was proved that the Cauchy problem
u

(t) = Au(t), t 0, u(0) = x, (CP)

with A being the generator of a C
0
-semigroup, is uniformly well-posed.
Now we begin to consider families of bounded operators connected with
problems (CP) that are not uniformly well-posed. The rst of these families
are n-times integrated semigroups. They are related to (CP) well-posed on
D(A
n+1
). If A generates a C
0
-semigroup U, then the n-times integrated
semigroup generated by A is the n-th order primitive of U. In contrast to
C
0
-semigroups, integrated semigroups may be not exponentially bounded,
may be locally dened, and their generators may be not densely dened.
In this section we consider nondegenerate exponentially bounded and local
integrated semigroups, and their connection with the Cauchy problem.
1.2.1 Exponentially bounded integrated semigroups
Denition 1.2.1 Let n N. A one-parameter family of bounded linear
operators V (t), t 0 is called an exponentially bounded n-times inte-
grated semigroup if the following conditions hold
(V1) V (t)V (s) =
1
(n 1)!
_
s
0
_
(s r)
n1
V (t +r) (t +s r)
n1
V (r)
_
dr,
for s, t 0;
2001 CRC Press LLC
2001 CRC Press LLC
(V2) V (t) is strongly continuous with respect to t 0;
(V3) K > 0, R : |V (t)| Ke
t
, t 0.
Semigroup V is called nondegenerate if
(V4) if V (t)x = 0 for all t 0, then x = 0.
A C
0
-semigroup is called 0-times integrated semigroup.
It follows from (V1), (V4) that V (0) = 0.
Suppose U is a C
0
-semigroup. Consider the integrals
V
1
(t) =
_
t
0
U(s) ds , . . .
V
n
(t) =
_
t
0
V
n1
(s)ds =
_
t
0
(t s)
n1
(n 1)!
U(s)ds , t 0.
For the generator A we have

1
R
A
() =
_

0
e
t
V
1
(t)dt , . . .

n
R
A
() =
_

0
e
t
V
n
(t)dt , Re >
or
R
A
() =
_

0

n
e
t
V
n
(t)dt .
We show that the operator R
A
(), which is dened by this equality, satises
the resolvent identity if and only if V
n
satises (V1).
Proposition 1.2.1 Let n N and V be a strongly continuous operator-
function such that
K > 0, R : |V (t)| Ke
t
, t 0.
Let
R() :=
_

0

n
e
t
V (t)dt , Re > .
Then R() satises the resolvent identity
( )R()R() = R() R()
for , C with Re , Re > , if and only if V satises (V1).
2001 CRC Press LLC
2001 CRC Press LLC
Proof Let Re > Re > . From the resolvent identity we have

n
R()
n
R() =
R() R()
( )
n

n
(1.2.1)

n
R()
n
R()
( )
n

R()(
n

n
)
( )
n
.
The result follows from the uniqueness theorem for Laplace transforms and
from the following observations:

n
R()
n
R() =
_

0
e
t
_

0
e
s
V (t)V (s)dsdt
and

n
R()
n
R()

=
_

0
e
()t

n
R()dt +
_

0
( )
1
e
t
V (t)dt
=
_

0
e
()t
_

0
e
s
V (s)dsdt
_

0
e
()t
_
t
0
e
s
V (s)dsdt
=
_

0
e
()t
_

t
e
s
V (s)dsdt
=
_

0
e
t
_

t
e
(st)
V (s)dsdt
=
_

0
e
t
_

0
e
s
V (t +s)dsdt .
Integrating n times by parts we obtain the rst term on the right-hand side
of (1.2.1):

n
R()
n
R()
( )
n
=
_

0
e
t
_

0
e
s
_
s
0
1
(n 1)!
(s r)
n1
V (r +t)drdsdt .
The second term:
R()(
n

n
)
( )
n
=
_
n1

k=o
1

k+1

nk
_
R()

n
=
n1

k=0

(k+1)
_

0

kn
e
s
V (s)ds
=
n1

k=0

(k+1)
_

0
e
s
_
s
0
(s r)
nk1
(n k 1)!
V (r)drds
2001 CRC Press LLC
2001 CRC Press LLC
=
n1

k=0
_

0
e
t
t
k
k!
dt
_

0
e
s
_
s
0
(s r)
nk1
(n k 1)!
V (r)drds
=
_

0
e
t
_

0
e
s
_
s
0
1
(n 1)!

_
n1

k=0
_
n 1
k
_
(s r)
nk1
t
k
_
V (r)drds
=
_

0
e
t
_

0
e
s
1
(n 1)!
_
s
0
(t +s r)
n1
V (r)drdsdt .
If the semigroup V (t), t 0 is nondegenerate, then the operator
R() is invertible. It follows from the resolvent identity that I R()
1
is independent of , that is there exists a unique operator A such that
R()
1
= (I A) for Re > . This operator A := I R()
1
is called
the generator of n-times integrated semigroup V (t), t 0.
It follows from Proposition 1.1.2 that for a C
0
-semigroup this denition
of generator coincides with the generally accepted denition by means of
the semigroups derivative at zero.
Now we clarify the connection between integrated semigroups and the
Cauchy problem (CP).
Proposition 1.2.2 Let A be the generator of n-times integrated semigroup
V (t), t 0 (where n N 0). Then
1) for x D(A), t 0
V (t)x D(A), AV (t)x = V (t)Ax,
and
V (t)x =
t
n
n!
x +
_
t
0
V (s)Ax ds; (1.2.2)
2) for x D(A), t 0
_
t
0
V (s)xds D(A),
and
A
_
t
0
V (s)xds = V (t)x
t
n
n!
x; (1.2.3)
3) for x D(A
n
), n N
V
(n)
(t)x = V (t)A
n
x +
n1

k=0
t
k
k!
A
k
x; (1.2.4)
2001 CRC Press LLC
2001 CRC Press LLC
4) for x D(A
n+1
)
d
dt
V
(n)
(t)x = AV
(n)
(t)x = V
(n)
Ax. (1.2.5)
Proof Let (A) , then for Re > , x X we have
_

0
e
t
V (t)R
A
()xdt =
n
R
A
()R
A
()x =
n
R
A
()R
A
()x
=
_

0
e
t
R
A
()V (t)xdt.
Hence R
A
()V (t) = V (t)R
A
(), that implies AV (t) = V (t)A, x D(A).
Let x D(A), then for Re > we have

n+1
n!
_

0
e
t
t
n
xdt
= x = R
A
() R
A
()Ax
=
_

0

n+1
e
t
V (t)xdt
_

0

n
e
t
V (t)Axdt
=
_

0

n+1
e
t
V (t)xdt
_

0

n+1
e
t
_
t
0
V (s)Axdsdt,
from where (1.2.2) follows by the uniqueness theorem for Laplace trans-
forms. Since A is closed, (1.2.3) follows from (1.2.2); dierentiation of
(1.2.2) gives (1.2.4); (1.2.5) follows from (1.2.4).
Theorem 1.2.1 (Arendt-Widder) Let a 0 and r : (a, ) X be an
innitely dierentiable function. For K > 0, (, a ] the following
statements are equivalent:
(I)
|r
(k)
()|
K k!
( )
k+1
, > a, k = 0, 1, . . . ;
(II) there exists a function V : [0, ) X satisfying V (0) = 0 and
lim
0
sup
h
h
1
|V (t +h) V (t)| Ke
t
, t 0, (1.2.6)
such that
r() =
_

0
e
t
V (t)dt, > a.
Moreover, r() has an analytic extension to
_
C

Re >
_
.
2001 CRC Press LLC
2001 CRC Press LLC
The Arendt-Widder theorem and Proposition 1.2.1 imply
Theorem 1.2.2 Let n 0N, R, K > 0. A linear operator A is the
generator of an (n + 1)-times integrated semigroup V satisfying condition
(1.2.6) if and only if there exists a max, 0 such that (a, ) (A)
and
_
_
_
_
_
R
A
()

n
_
(k)
_
_
_
_

K k!
( )
k+1
(1.2.7)
for all > a, and k = 0, 1, . . . . In this case
R
A
() =
_

0

n+1
e
t
V (t)dt, > a.
Hence, for n = 0 we have the equivalence of existence of an integrated
semigroup and MFPHY-type condition. At rst sight, this equivalence is
weaker than in Theorem 1.1.1, but the fact is that integrated semigroups,
in contrast to C
0
-semigroups, may have not densely dened generators. If
D(A) = X, then the following extension of Theorem 1.1.1 holds.
Theorem 1.2.3 Let A be a densely dened linear operator with (a, )
(A), where a 0, K > 0, (, a]. Then condition (1.2.7) is
equivalent to the statement: A is the generator of an n-times integrated
semigroup V (t), t 0, such that
|V (t)| Ke
t
.
Proof Suppose that condition (1.2.7) is fullled, then A is the generator of
an (n + 1)-times integrated semigroup V
n+1
(t), t 0. In view of (1.2.6)
for V
n+1
, the set F
1
of all x X such that V
n+1
()x C
1
(0, ), X is
closed. By Proposition 1.2.2, D(A) F
1
, hence D(A) = X F
1
and
F
1
= X. Then the relation V (t)x = V

n+1
(t)x, x X, denes an n-times
integrated semigroup V (t), t 0 generated by A.
1.2.2 (n, )-well-posedness of the Cauchy problem
We consider the Cauchy problem
u

(t) = Au(t), t 0, u(0) = x, (CP)

where A is a linear closed densely dened operator on a Banach space X.
We denote by [ D(A
n
) ] the Banach space
_
D(A
n
), |x|
A
n = |x| +|Ax| +. . . +|A
n
x|
_
.
Denition 1.2.2 Let n N. The Cauchy problem (CP) is said to be
(n, )-well-posed on E if for any x E D(A
n+1
)
2001 CRC Press LLC
2001 CRC Press LLC
(a) there exists a unique solution
u() C
_
[0, ], D(A)
_
C
1
_
[0, ], X
_
;
(b) K > 0, R : |u(t)| Ke
t
|x|
A
n.
If E = D(A
n+1
), then we say that the problem (CP) is (n, )-well-posed.
Theorem 1.2.4 Let A be a densely dened linear operator on X with
nonempty resolvent set. Then the following statements are equivalent:
(I) A is the generator of an n-times integrated semigroup V (t), t 0;
(II) the Cauchy problem (CP) is (n, )-well-posed.
Proof (I) = (II). Let x D(A
n+1
), consider the function V ()x. By
Proposition 1.2.2, it is (n+1)-times continuously dierentiable, V
(n)
(t)x
D(A), and
V
(n)
(t)x = V (t)A
n
x +
n1

k=0
t
k
k!
A
k
x,
d
dt
V
(n)
(t)x = AV
(n)
(t)x.
Let u(t) := V
(n)
(t)x, then u(0) = x and u(t) D(A) for t 0. Further-
more, |u(t)| Ke
t
|x|
A
n, and u

(t) = Au(t).
We now show that u() is unique. Let v() be a solution of (CP), then for
(A), R
n
A
()v() is the solution of (CP) with the initial value R
n
()x
D(A
n+1
). As in the proof of uniqueness in Theorem 1.1.1, we have
d
ds
V
n
(t s)R
n
A
()v(s)
= AV
n
(t s)R
n
A
()v(s) +V
n
(t s)AR
n
A
()v(s) = 0,
for 0 s t. Hence
V
n
(0)R
n
A
()v(t) = R
n
A
()V
n
(0)v(t) = R
n
A
()V
n
(t)v(0),
and v() = V
n
()x.
(II) = (I). Let x D(A
n+1
), then there exists a unique solution
u() of (CP) such that |u(t)| Ke
t
|x|
A
n. For (A) the function
w() := R
A
()u() is the solution of the Cauchy problem with the initial
value R
A
()x and with the estimate |w(t)| K
1
e
t
|x|
A
n1. Let
u
1
(t) =
_
t
0
u(s)ds,
2001 CRC Press LLC
2001 CRC Press LLC
then
u
1
(t) = w(t)
_
t
0
w(s)ds R
A
()x.
Therefore, |u
1
(t)| K
2
e
t
|x|
A
n1. By induction we obtain an n-times
integrated solution
u
n
(t) =
_
t
0
1
(n 1)!
(t s)
n1
u(s)ds, t 0,
which is exponentially bounded. For t 0 we dene an operator-function
V (t) : D(A
n+1
) X by the relation V (t)x = u
n
(t), where u
n
(t) is
the unique n-times integrated solution of (CP) with the initial value
x D(A
n+1
). Since D(A) = D(A
n+1
) = X, we can extend V to the
whole X. The operator-function V (t), t 0, is exponentially bounded
and for x D(A
n+1
), V (t)x is continuous in t, therefore V () is strongly
continuous. In addition, it is not dicult to show that the operator
R() =
_

0

n
e
t
V (t)dt coincides with the resolvent R
A
() of operator A.
Hence V (t), t 0 is an n-times integrated semigroup with the generator
A.
Remark 1.2.1 If we do not assume that D(A) = X, then after con-
structing u
1
(t), ..., u
n
(t) in the proof of the implication (II) = (I), we can
extend u
1
(t) to D(A
n
), ... , u
n
(t) to D(A), and as a result, construct an
(n + 1)-times integrated semigroup V (t)x := u
n+1
(t) on X.
1.2.3 Local integrated semigroups
Denition 1.2.3 Let n N, T (0, ). A one-parameter family of
bounded linear operators V (t), 0 t < T is called a local n-times inte-
grated semigroup in X if (V1) in Denition 1.2.1 is fullled for t, s [0, T)
such that s +t [0, T), and (V2) holds for t [0, T).
If V (t), t 0 is an exponentially bounded n-times integrated semigroup
then its generator A is dened from the equality
(I A)
1
x =
_

0

n
e
t
V (t)x dt, x X, > . (1.2.8)
For a local n-times integrated semigroup V (t), t [0, T) this integral
may not exist. For this reason the innitesimal generator A
0
of a local
semigroup V (t) is dened in the following way
A
0
x := lim
t0
t
1
_
V
(n)
(t)x x
_
2001 CRC Press LLC
2001 CRC Press LLC
D(A
0
) =
_
x
_
0<<T
C
n
()

lim
t0
t
1
_
V
(n)
(t)x x
_
exists
_
,
where
C
n
() =
_
x X such that
V (t)x : [0, ) X is n-times continuously dierentiable
_
.
It is proved in [138] that A
0
is closable and we call A
0
the complete innites-
imal generator or the generator of a local n-times integrated semigroup
V (t), t [0, T). For a densely dened operator, this denition of the
complete innitesimal generator of an exponentially bounded semigroup is
equivalent to the denition of a generator via formula (1.2.8).
We list below the main properties of local integrated semigroups which
are similar to those in Proposition 1.2.2.
Proposition 1.2.3 Let n N and let A be the generator of a local n-times
integrated semigroup V (t), t [0, T), then
1) for x D(A), t [0, T)
V (t)x D(A) and AV (t)x = V (t)Ax;
2) for x D(A), t [0, T)
V (t)x =
t
n
n!
x +
_
t
0
V (s)Axds; (1.2.9)
3) if D(A) = X, then for x X, t [0, T)
_
t
0
V (s)xds D(A), A
_
t
0
V (s)xds = V (t)x
t
n
n!
x;
4) D(A) = X if and only if C
n
(T) = X.
Denition 1.2.4 The local Cauchy problem
u

(t) = Au(t), t [0, T), u(0) = x, (CP)

is said to be n-well-posed if for any x D(A
n+1
) there exists a unique
solution satisfying
sup
t[0,][0,T)
|u(t, x)| K

|x|
A
n (1.2.10)
for some constant K

.
2001 CRC Press LLC
2001 CRC Press LLC
The following result is similar to Lemma 1.1.1.
Lemma 1.2.1 If for any x D(A
n+1
) there exists a unique solution of
local (CP) and (A) ,= , then this solution satises (1.2.10).
Proof Let us consider the solution operator
S : [D(A
n+1
)] C
_
[0, ], [D(A)]
_
which is dened everywhere on [D(A
n+1
)] by
Sx = u, x [D(A
n+1
)],
and is closed. Let x
j
x and u
j
(t) y, then we have
u

j
(t) = Au
j
(t) Ay(t) and u
j
(t) x
j
y(t) x =
_
t
0
Ay(t)dt,
hence y

(t) = Ay(t), y(0) = x. By the Banach theorem, the operator S is

continuous, that is
sup
t[0,]
|u(t)|
A
K

|x|
A
n+1.
Furthermore, as in the proof of Lemma 1.1.1, we have
sup
t[0,]
|u(t)| K

|x|
A
n.
Thus, this solution is stable with respect to the initial data in the corre-
sponding n-norm.
Proposition 1.2.4 If A is the generator of a local n-times integrated semi-
group V (t), t [0, T), then local (CP) is n-well-posed.
Proof Let k N with 1 k n. Then, from (1.2.9) we have
V
(k)
(t) =
k

i=1
t
ni
(n i)!
A
ki
x +V (t)A
k
x
for x D(A
k
) and 0 t < T. Now for x D(A
n+1
) we dene u(t) :=
V
n
(t)x, t [0, T). Using (1.2.9) one can easily show that u() is a solution
of local (CP). In Theorem 1.2.4 it is proved that every solution of this
problem has the form V
n
()x.
Theorem 1.2.5 Let A be the generator of a local n-times integrated semi-
group V (t), t [0, T) and D(A) = X. Then there exists region
=
_
C

Re >
n

log(1 +[[) +
1

log
C

_
(A),
2001 CRC Press LLC
2001 CRC Press LLC
(0, T), C > 0, 0 < < 1,
such that
K > 0 : , |R
A
()|
K[[
n
log(1 +[[)
.
Proof If A is the generator of a local n-times integrated semigroup V (t),
0 t < T, then the function
R(, ) =
_

0

n
e
t
V (t)dt
is dened for any (0, T). Taking into account (1.2.9), for x D(A) we
have
R(, )Ax
=
n
e

_

0
V (s)Axds +
_

0

n+1
e
t
_
t
0
V (s)Axdsdt
=
n
e

_
V ()x

n
n!
x
_
+
_

0

n+1
e
t
_
V (t)x
t
n
n!
x
_
dt,
R(, )(I A)x
=
_

0

n+1
e
t
V (t)xdt
n
e

_
V ()x

n
n!
x
_

_

0

n+1
e
t
_
V (t)x
t
n
n!
x
_
dt
=
n
e

V ()x +
n
e

n
n!
x +
_

0

n+1
e
t
t
n
n!
xdt.
Since
_

0

n+1
e
t
t
n
n!
xdt =
n
e

n
n!
x +
_

0

n
e
t
t
n1
(n 1)!
xdt
= . . . =
n

k=0
()
k
k!
e

x +x,
then
R(, )(I A)x = (I A)R(, )x = (I G())x, x D(A),
where
G()x =
n
e

V ()x +
n1

k=0
()
k
k!
e

x,
2001 CRC Press LLC
2001 CRC Press LLC
and
|G()| C(1 +[[)
n
e
Re
, C = C(, n).
We also have that G() commutes with R(, ) on X and with A on D(A).
Using this estimate for |G()|, we can nd a region C such that
|G()| < 1 for any . After taking logarithms of the inequality
C(1 +[[)
n
e
Re
< < 1,
we obtain that the estimates
|G()| < ,
_
_
(I G())
1
_
_
<
1
1
hold in the region
=
_
C

Re >
n

log(1 +[[) +
1

log
C

_
.
Therefore, there exists the operator (I A)
1
, which is bounded on
D(A) = X, and
K > 0 : , |(I A)
1
|
K[[
n
log(1 +[[)
.
Since D(A) = X, we have (I A)
1
= R
A
().
Remark 1.2.2 The converse result that polynomial estimates on the resol-
vent of operator A in some region imply that A is the generator of some
local integrated semigroup is proved later in Theorem 2.1.5.
The following necessary and sucient condition in terms of estimates
on powers of the resolvent is the generalization of (1.1.14).
Theorem 1.2.6 If for an operator A
: (0, T),
sup
_
_
_

k
R
k
A
()x
_
_

0
k

, > , k = 0, 1, . . .
_
C

|x|
A
n (1.2.11)
for some constant C

, then A is the generator of a local n-times integrated

semigroup V (t), t [0, T). Conversely, if A is the generator of a local n-
times integrated semigroup V (t), t [0, T) and D(A) = X or (A) ,= ,
then (1.2.11) holds.
2001 CRC Press LLC
2001 CRC Press LLC
Proof If A is the generator of a local n-times integrated semigroup V (t),
t [0, T), then by Proposition 1.2.4 for any x D(A
n+1
) there exists a
unique solution of the local Cauchy problem (CP):
u() = V
(n)
()x =: U()x
such that
|U(t)x| K

|x|
A
n, 0 t < T. (1.2.12)
For U(t)x we have
A
_

0
e
t
U(t)xdt = e

U()x x +
_

0
e
t
U(t)xdt.
Hence by Theorem 1.2.5,
R
A
()x =
_

0
e
t
U(t)xdt +e

R
A
()U()x, (A).
By the Cauchys integral formula

k
[R
A
()]
k
x =
(1)
k1

k
(k 1)!
R
(k1)
A
()x
=

k
(k 1)!
_

0
e
t
t
k1
U(t)xdt
+
1
2i
_

_
1

_
k
R
A
()U()xd
I
1
+I
2
,
for some contour , which is described in [226]. The following estimates are
obtained in [226] for these integrals
|I
1
| C
1
|x|
A
n, x D(A
n
), > 0,
: |I
2
| C
2
|x|
A
n, x D(A
n
),
k

[0, ], > .
This and (1.2.12) imply (1.2.11).
Conversely, it is proved in [213] that if (1.2.11) is fullled for operator
A, then
U(t)x := lim
k
_
I
t
k
A
_
k
x, t [0, T),
is dened for x D(A
n
). For x D(A
n+1
), u() = U()x is the unique
solution of the local Cauchy problem with the stability property (1.2.12).
In this case (see [255]) A is the generator of a local n-times integrated
semigroup V (t), 0 t < T.
2001 CRC Press LLC
2001 CRC Press LLC
Remark 1.2.3 We showed that if A is the generator of an n-times inte-
grated semigroup V (t), 0 t < T, T (dened as the closure of
operator V
(n+1)
(0)), then we have (1.2.9):
V (t)x
t
n
n!
x
=
_
t
0
V (s)Axds =
_
t
0
AV (s)xds, 0 t < T, x D(A),
and if A is densely dened, then
V (t)x
t
n
n!
x = A
_
t
0
V (s)xds, x X. (1.2.13)
Now we prove that if A and a family of bounded linear operators
V (t), 0 t < T satisfy (1.2.9), (1.2.13), then V satises the semi-
group relation (V1) from the denition of an n-times integrated semigroup.
Hence, an n-times integrated semigroup with a densely dened generator
A may be equivalently dened as follows:
Denition 1.2.5 A family of bounded linear operators V (t), 0 t < T
is called a local n-times integrated semigroup generated by operator A, if
for x D(A), AV (t)x = V (t)Ax, and
A
_
t
0
V (s)xds = V (t)x
t
n
n!
x, x X.
The operator A is called the generator of V .
If A is the generator of V then
D(A) = D :=
_
x X

y : V (t)x =
t
n
n!
x +
_
t
0
V (s)yds, t [0, T)
_
,
where y = Ax. In fact, let x D(A), then x D and y = Ax. Conversely,
if x D, then for (A)
V (t)R
A
()x
t
n
n!
R
A
()x =
_
t
0
V (s)R
A
()yds =
_
t
0
V (s)AR
A
()x.
Hence x D(A) and y = Ax. Note that it was proved in Theorem 1.2.5,
that for such a family (A) ,= .
Lemma 1.2.2 Let > 0 and A be the generator of a local n-times inte-
grated semigroup V (t), 0 t < T in the sense of Denition 1.2.5. Then
for any continuous function H : [0, ) C and for any x X
A
_
t
0
(H V )(s)xds = (H V )(t)x (H F)(t)x, 0 t < .
2001 CRC Press LLC
2001 CRC Press LLC
Moreover, if H C
1
[0, ), C, then for any x X and 0 t <
A(H V )(t)x = (H

V )(t)x (H

F)(t)x +H(0)
_
V (t)x F(t)x
_
,
where
(H V )(t)x :=
_
t
0
H(s)V (t s)xds, F(t) :=
t
n
n!
.
Proof Since operator A is closed, we have
_
t
0
(H V )(s)xds D(A), x X , 0 t < .
From Denition 1.2.5 we obtain
A
_
t
0
(H V )(s)xds = A
_
t
0
__
s
0
H(r)V (s r)xdr
_
ds
=
_
t
0
H(r)
_
A
_
t
r
V (s r)xds
_
dr =
_
t
0
H(r)
_
A
_
tr
0
V (s)xds
_
dr
=
_
t
0
H(r) [V (t r)x F(t r)x] dr = (H V )(t)x (H F)(t)x,
x X.
Since
(H V )(t)x =
_
t
0
H(s)V (t s)xds =
_
t
0
H(t s)V (s)xds, x X,
then the second statement of the lemma follows from the rst one after
integration by parts.
Lemma 1.2.3 Let > 0 and A be the generator of a local n-times inte-
grated semigroup V (t) , 0 t < in the sense of Denition 1.2.5. Then
for any x X and 0 t, s <
V (t)V (s)x
=
_
t+s
s
(t +s r)
n1
(n 1)!
V (r)xdr
_
t
0
(t +s r)
n1
(n 1)!
V (r)xdr.
Proof Let h > 0, dene
E
h
(t) := E(t +h), where E(t) =
t
n1
(n 1)!
.
For 0 s < and xed x, dene
w(t) :=
_
t+s
s
E(t +s r)V (r)xdr
_
t
0
E(t +s r)V (r)xdr,
2001 CRC Press LLC
2001 CRC Press LLC
where x X, 0 t < . We have
w(t) =
_
t+s
0
E(t +s r)V (r)xdr

_
t
0
E(t +s r)V (r)xdr
_
s
0
E(t +s r)V (r)xdr
= (E V )(t +s)x (E
s
V )(t)x (E
t
V )(s)x, x X.
Using Lemma 1.2.2, we obtain
A
_
t
0
w(r)dr
= A
_
t
0
(E V )(s +r)xdr A
_
t
0
(E
s
V )(r)xdr
A
_
t
0
(E
r
V )(s)xdr
= A
_
t+s
0
(E V )(r)xdr A
_
s
0
(E V )(r)xdr
A
_
t
0
(E
s
V )(r)xdr A
_
(F
t
V )(s) (F V )(s)
_
= (E V )(t +s)x (E F)(t +s)x (E V )(s)x + (E F)(s)x
(E
s
V )(t)x + (E
s
F)(t)x (E
t
V )(s)x + (E
t
F)(s)x
F(t)V (s)x +F(t)F(s)x + (E V )(s)x (E F)(s)x
= (E V )(t +s)x (E
s
V )(t)x (E
t
V )(s)x F(t)V (s)x
+
_
F(t)F(s) (E F)(t +s) + (E
s
F)(t) + (E
t
F)(s)
_
x
= w(t) F(t)V (s)x, x X,
where
F(t) =
_
t
0
E(s)ds =
t
n
n!
.
Here we used the following relation
F(t)F(s)x
= (E F)(t +s)x (E
s
F)(t)x (E
t
V )(s)x
=
_
t+s
s
E(t +s r)F(r)xdr
_
t
0
E(t +s r)F(r)xdr
2001 CRC Press LLC
2001 CRC Press LLC
for x X, which can be easily proved using integration by parts. Therefore
v() = w() is a solution of the problem
v

(t) = Av(t) +F(t)V (s)x, 0 t < , (1.2.14)

v(0) = 0 , v C
_
[0, ), D(A) C
1
[0, ), X
_
.
Since V (t)V (s)x, t, s [0, ), is also a solution of this problem, then by
uniqueness we obtain that w(t) = V (t)V (s)x, t, s [0, ).
1.2.4 Examples
Example 1.2.1 (An integrated semigroup with the generator that is non-
densely dened)
Consider the operator A = d/dx on X = C[0, ) with
D(A) =
_
u C[0, )

C[0, ), u(0) = 0
_
.
Since D(A) ,= X then A cannot be a generator of a C
0
-semigroup (compare
this with Example 1.1.1). In this case A is the generator of the 1-times
integrated semigroup V dened by
(V (t)f)(x) =
_
_
_

_
xt
x
f(s) ds, x t
_
o
x
f(s) ds, 0 x t
.
Note that A is the generator of a C
0
-semigroup in the space C
0
[0, ) of
continuous on [0, ) functions vanishing at zero.
Example 1.2.2 (A class of operators generating integrated semigroups)
As in Example 1.1.2 let
X =
_
L
p
(R) L
p
(R), |u| = |u
1
|
L
p +|u
2
|
L
p
_
, where u =
_
u
1
u
2
_
.
Consider the operator A dened by
Au =
_
h f
0 h
_
u
with
D(A) =
__
u
1
u
2
_
X

hu
1
+fu
2
L
p
(R), hu
2
L
p
(R)
_
,
2001 CRC Press LLC
2001 CRC Press LLC
where h(x) = 1 + [x[, f(x) = [x[

, > 0. For the formally dened

operator-function
V (t) =
_
t
0
e
As
ds, t 0,
we can write
V (t)u =
1
h
_
1 e
ht
tfe
ht
+ (e
ht
1)f/h
0 1 e
ht
_
u, u X.
If (1, 2], then the family of bounded linear operators V (t), t 0
satises conditions (V1) (V4) for a 1-time integrated semigroup. The
operator A is the generator of this semigroup since
I R()
1
= I
__

0
e
t
V (t)dt
_
1
= I
_
1
+h

f
(+h)
2
0
1
+h
_
1
= A.
Furthermore, if 2, then the operator-functions V
k
(t), t 0, dened by
V
k
(t)u =
_
t
0
V
k1
(s)uds, u X, k 2, V
1
= V,
form k-times integrated semigroups on X. In particular,
V
2
(t)u =
_
t
0
V (s)uds
=
1
h
_
t
1e
ht
h
tfe
ht
+
(1e
ht
)f
h
+
(1e
ht
)f
h
2

tf
h
0 t
1e
ht
h
_
u,
u X,
denes a 2-times integrated semigroup V
2
on X. We also note that if > 2,
then (see Example 1.1.2) all > 0 do not belong to the resolvent set of
A, and therefore for any n, the operator A cannot generate an n-times
integrated semigroup on X.
Example 1.2.3
The dierential operator
A =

||k
a

,
where
D

=
_

x
1
_

1
. . .
_

x
n
_

n
, [[ =
n

j=1

j
, max [[ > n/2,
2001 CRC Press LLC
2001 CRC Press LLC
and its symbol
p(x) =

||k
a

(ix)

is an elliptic polynomial with Re p(x) < , is the generator of ([[n/2]] +2)-

times integrated semigroup in the spaces C
0
(R
n
), L
p
(R
n
), 1 p .
Example 1.2.4 (An integrated semigroup associated with the second order
Cauchy problem)
This example generalizes the Cauchy problem for Wave equation which
was discussed in Chapter 0. Consider the second order Cauchy problem
u

(t) = Bu(t), t 0, u(0) = x, u

(0) = y (1.2.15)
in a Banach space X, with linear operator B generating cosine and sine
operator-functions (() and o(). In Section 1.7 we study the properties of
/, ^-functions, which in particular give the properties of (, o-functions.
The unique solution of (1.2.15) has the form
u(t) = ((t)x +o(t)y.
The problem (1.2.15) can be reduced to the Cauchy problem for the rst
order system
w

(t) = w(t), w(0) =

_
x
y
_
,
where
=
_
0 I
B 0
_
, w(t) =
_
u(t)
u

(t)
_
.
Using cosine and sine operator-functions, we can rewrite w in the following
form
w(t) =
_
((t)x +o(t)y
(

(t)x +((t)y
_
U(t)
_
x
y
_
,
where operator U(t) is not dened everywhere on X X for all t 0,
because function (() is not necessarily dierentiable on X.
The operator is the generator of the integrated semigroup
V (t) =
_
o(t)
_
t
0
o()d
((t) I o(t)
_
.
Conditions (V1) (V4) are satised due to the properties of (, o-functions.
Example 1.2.5 (An integrated semigroup that is not exponentially
bounded)
2001 CRC Press LLC
2001 CRC Press LLC
Let X = l
2
and
Ax := a
m
x
m

m=1
, a
m
= m+ie
2m
2
m
2

1/2
, > 1,
then the operators
U(t)x = e
a
m
t
x
m

m=1
form an unbounded semigroup. The operators
V (t)x =
_
t
0
U(s)xds =
__
e
a
m
t
a
m
1
_
x
m
_
b
m
x
m

are bounded for any t 0 since

|V (t)| = sup
m
[b
m
[ = sup
m
e
mtm
2
= e
t
2
/4
,
and form an integrated semigroup that is not exponentially bounded.
Example 1.2.6 (Local n-times integrated semigroup)
Let X = l
2
, T > 0,
Ax := a
m
x
m

m=1
, a
m
=
m
T
+i
_
e
2m
m
2

m
2
T
2
_
1/2
,
D(A) =
_
x l
2

Ax l
2
_
.
For operator A we have
(A) =
_
C

= a
m
, m N
_
and since Re a
m
=
m
T
, then for any R there exists (A) such
that Re > . The operators T(t)x := e
a
m
t
x
m

m=1
form an unbounded
semigroup. Integrating e
a
m
t
we obtain the factor me
m
, which on the n-th
step makes the product bounded for t < nT, and, as a result, we obtain a
local n-times integrated semigroup V
n
(t) :
V
n
(t)x =
__
t
0
(t s)
n1
(n 1)!
e
a
m
s
x
m
ds
_

m=1
=
__
a
n
m
e
a
m
t

p=1
(a
m
)
p
t
np
(n p)!
_
x
m
_
.
2001 CRC Press LLC
2001 CRC Press LLC
Since [e
a
m
t
[ = e
mt
T
and [a
m
[ =
e
m
m
, it follows that
m
n
e
m(
t
T
n)

p=1
m
p
e
pm
t
np
(n p)!
[
_
t
0
(t s)
n1
(n 1)!
e
a
m
s
ds[
m
n
e
m(
t
T
n)
+
n

p=1
m
p
e
pm
t
np
(n p)!
,
therefore
|V
n
(t)| = sup
mN
_

_
t
0
(t s)
n1
(n 1)!
e
a
m
s
ds

_
is bounded if and only if 0 t < nT.
1.3 -convoluted semigroups
In the previous section we showed that an n-times integrated semigroup
with the densely dened generator A can be dened as a family of bounded
linear operators V (t), t [0, T) satisfying
V (t)x
t
n
n!
x =
_
t
0
V (s)Axds =
_
t
0
AV (s)xds,
0 t < T, x D(A),
and
V (t)x
t
n
n!
x = A
_
t
0
V (s)xds, x X.
A similar approach is used in this section for -convoluted semigroups.
1.3.1 Generators of -convoluted semigroups
Denition 1.3.1 Let A be a closed operator and () be a continuous func-
tion on [0, T), T . If there exists a strongly continuous operator-family
S

(t), t [0, T) such that S

(t)Ax = AS

(t)x for x D(A), t [0, T),

and for all x X
_
t
0
S

(s)xds D(A)
and
S

(t)x = A
_
t
0
S

(s)xds + (t)x, 0 t < T, (1.3.1)

2001 CRC Press LLC
2001 CRC Press LLC
where (t) =
_
t
0
(s)ds, then S

is called a -convoluted semigroup gener-

ated by A, and A is called the generator of S

.
In the particular case (t) = t
n
/n! , the -convoluted semigroup S

is
an n-times integrated semigroup. If A is the generator of a C
0
-semigroup
U, then
S

(t)x =
_
t
0
(t s)
n1
(n 1)!
U(s)xds
is an n-times integrated semigroup and a -convoluted semigroup. This
is the reason that S

is called a -convoluted semigroup, but not a -

convoluted semigroup.
The corresponding Cauchy problem
v

(t) = Av(t) + (t)x, 0 t < T, v(0) = 0 (1.3.2)

is called the -convoluted Cauchy problem. If there exists a solution of the
Cauchy problem
u

(t) = Au(t), 0 t < T, u(0) = x, (1.3.3)

then, as usual for a nonhomogeneous equation, we have v = u.
The following results connect the existence of a -convoluted semigroup
and the well-posedness of a -convoluted Cauchy problem with the be-
haviour of the resolvent R
A
(). In contrast to the case of an exponentially
bounded and local n-times integrated semigroup, where the resolvent has
polynomial estimates in a half-plane or in a logarithmic region, respectively,
in the convoluted case the resolvent exists in some smaller region and is
allowed to increase exponentially:
|R
A
()| Ke
M()
.
Here, a real-valued function M(), C, grows not faster than and is
dened by the order of decreasing of (), the Laplace transform of (t).
Theorem 1.3.1 Let () be an exponentially bounded function such that
[ ()[ =
||
O
_
e
M()
_
.
Suppose that A is the generator of a -convoluted semigroup S

(t),
0 t < T, then there exists a region
=
_
C

Re >
M()
T
+
_
such that |R
A
()| Ke
M()
for any and some K > 0.
2001 CRC Press LLC
2001 CRC Press LLC
Proof Let [(t)[ Ke
t
. Consider the operator
R(, t) :=
1
_
t
0
e
s
S

(s) ds,
where
1
= 1/ (). As in the case of integrated semigroups (Theorem
1.2.6) we now show that R(, t) is nearly the resolvent of A:
R
A
() = R(, t)
_
I B
t
()
_
1
, ,
where B
t
() are linear operators such that |B
t
()| < 1 for t [0, T)
and . Since
_
t
0
e
s
S

(s) dsx D(A) for any x X, we can apply

operator (I A) to R(, t)x
(I A)R(, t)x =

1
__
t
0
e
s
S

(s) ds A
_
t
0
e
s
S

(s) ds
_
x, x X.
Using the equation for the -convoluted semigroup for the rst term in the
right-hand side and integrating by parts the second term, we obtain
(I A)R(, t)
=
1
_

_
t
0
e
s
_
A
_
s
0
S

() d + (s)I
_
ds
A
_
t
0
e
s
_
s
0
S

()d ds Ae
t
_
t
0
S

()d
_
=
1
_

_
t
0
e
s
(s)Ids e
t
A
_
t
0
S

(s)ds
_
=
1
__
t
0
e
s
(s)I ds e
t
S

(t)
_
= I
1
_
e
t
S

(t) +
_

t
e
s
(s)I ds
_
=: I B
t
().
We estimate norms of operators B
t
() for Re
1
> :
|B
t
()| [
1
[
_
e
Re t
|S

(t)| +
_

t
e
Re s
[(s)[ ds
_
K[
1
[
_
e
Re t
+
_

t
e
sRe s
ds
_
K[
1
[ e
(Re )t
_
1 +
1

_
Ke
M()(Re )t
.
2001 CRC Press LLC
2001 CRC Press LLC
Let Ke
M()(Re )t
< 1, then
M() (Re )t ln

K
0.
For t = T we obtain
Re
M() ln

K
T
+ =:
M()
T
+,
and for these
|B
t
()| .
Since for x D(A) we have
(I A)R(, t)x = R(, t)(I A)x = (I B
t
())x,
then the operator (I A)
1
= R
A
() exists in the region
=
_
C

Re
M()
T
+
_
,
and for these
|R()| [
1
[

_
t
0
e
s
S

(s) ds

1
1
Ke
M()
.
Remark 1.3.1 If, instead of the condition
[ ()[ =
||
O
_
e
M()
_
we take
[ ()[ =
||
O
_
e
M()
_
for some , then for any

T,,
:=
_
C

Re >
M()
T
+
_
we have
|R
A
()| Ke
M()
.
Remark 1.3.2 Note that the equality (1.3.1) is considered on X, not on
D(A). This is the reason that, in contrast to Theorem 1.2.5 for integrated
semigroups, here we do not assume that D(A) = X.
2001 CRC Press LLC
2001 CRC Press LLC
Theorem 1.3.2 Suppose

,,
, |R
A
()| Ke
M()
(1.3.4)
for some K, , , > 0, and
[ ()[ =
||
O
_
e
M()
_
(1.3.5)
with some > 0 such that (/ 1) > 0. Then A is the generator of a
-convoluted semigroup on [0, T
1
) = [0, (

1)).
Proof Let R
A
() satisfy (1.3.4) and () satisfy (1.3.5) with some >
0 such that (/ 1) > 0. Consider the inverse Laplace transform of
()R
A
()
S
1

(t) :=
_

e
t
()R
A
() d,
where =
,,
is the boundary of
,,
. Then
|S
1

(t)|
_

Ke
Re t+M()M()
[d[.
Since
e
Re t+M()M()
= e
(t+

)Re
on , the operators S
1

(t) are dened for

t +

< 0 or t <
_

1
_
=: T
1
.
Now we show that operators S
1

(t) satisfy the equation (1.3.1) for the -

convoluted semigroup:
S
1

(t)x = A
_
t
0
S
1

(s)xds + (t)x, x X.
We have
A
_
t
0
S
1

(s)xds
= (AI)
_
t
0
_

e
s
R
A
() ()xdds
=
_
t
0
_

e
s
R
A
() ()xdds
_
t
0
_

e
s
()xdds
=
_

R
A
() ()
_
t
0
e
s
xdsd
_
t
0
(s)xds
= S
1

(t)x
_

R
A
() ()xd (t)x.
2001 CRC Press LLC
2001 CRC Press LLC
Since on we have
|R
A
() ()| Ke
M()M()
= Ke
(1

)Re
,
and (1

) < 0, then the integral

_

R
A
() ()xd is equal to zero. Hence
for any x X, we have
A
_
t
0
S
1

(s)xds = S
1

(t)x (t)x.
Combining these results we obtain the necessary and sucient condi-
tions for A to be the generator of a -convoluted semigroup. They are also
sucient for well-posedness of the -convoluted Cauchy problem (1.3.2).
1.3.2 -convoluted Cauchy problem
Theorem 1.3.3 Let M(), C, be a real-valued function growing at
innity not faster than
p
, p < 1. Then the following conditions are equiv-
alent:
(K) A is the generator of a -convoluted semigroup on [0, T) for some
T > 0, with such that
[ ()[ =
||
O
_
e
M()
_
for some > 0;
(R)
, , > 0 :
,,
, |R
A
()| Ke
M()
;
(K1) for any T > 0, A is the generator of a -convoluted semigroup on
[0, T) with such that
[ ()[ =
||
O
_
e
M()
_
for some > 0;
(R1)
> 0 , > 0 :
,,
, |R
A
()| Ke
M()
.
In this case the -convoluted Cauchy problem has the unique solution
v(t) =
_
t
0
S

(s)xds, t [0, T).

Proof From Theorem 1.3.1 and Remark 1.3.1 we have (K) = (R) with
= T, = and (K1) = (R1) with = T, = . From Theorem 1.3.2
we have (R) = (K1) with dened from the equality T = (

1), and
(R1) = (K) with T = , = .
It follows from the denition of a -convoluted semigroup that v(t) =
_
t
0
S

(s)xds, x X, t [0, T), is a solution of (1.3.2). By the Lyubichs

2001 CRC Press LLC
2001 CRC Press LLC
theorem (Remark 1.1.6), the uniqueness of this solution follows from the
condition on M() and the estimate (1.3.4):
lim

ln|R
A
()|
[[
lim

[
p
[
[[
= 0.
Using the uniqueness of a solution of the -convoluted Cauchy problem we
can obtain for a -convoluted semigroup its semigroup relation:
S

(t)S

(s) = S

(s)S

(t) (1.3.6)
=
_
t+s
s
(t +s r)S

(r)dr
_
t
s
(t +s r)S

(r)dr.
Note that (V1) from Denition 1.2.1 is the particular case of (1.3.6) with
(t) = t
n1
/(n 1)! , and (1.3.6) was proved in Lemma 1.2.3 .
Theorem 1.3.4 Let S

(t), 0 t < T be a -convoluted semigroup, then

for any 0 t < T the equality (1.3.6) holds.
Pro of See the proof of Lemma 1.2.3 with E(t) = (t).
In the Section 2.3, devoted to Cauchy problems in the spaces of (ab-
stract) ultradistributions, we show that the well-posedness of the Cauchy
problem in a space of ultradistributions is equivalent to the condition (R),
where M() is the function associated with a sequence M
n
that denes the
space of ultradistributions. Hence, any of the conditions (K), (K1), (R1)
with such function M() are equivalent to well-posedness of the Cauchy
problem in the space of ultradistributions.
1.4 C-regularized semigroups
We continue the study of families of bounded operators connected with
Cauchy problems that are not uniformly well-posed. C-regularized semi-
groups are related to problems that are C-well-posed (see Denition 1.4.4)
on CD(A) for some bounded invertible operator C.
1.4.1 Generators of C-regularized semigroups
Suppose C is an injective, bounded operator on a Banach space X.
Denition 1.4.1 A one-parameter family of bounded linear operators
S(t), t 0 is called an exponentially bounded C-regularized semigroup if
(C1) S(t +h)C = S(t)S(h), t, h 0, S(0) = C;
(C2) S(t) is strongly continuous with respect to t 0;
2001 CRC Press LLC
2001 CRC Press LLC
(C3) K > 0, R : |S(t)| Ke
t
, t 0.
It follows from (C1) at t = 0 that S(h) commutes with C for all h 0:
(C4) S(h)C = CS(h), h 0.
If C = I, then the C-regularized semigroup is a C
0
-semigroup.
Dene the operator
L()x =
_

0
e
t
S(t)xdt, x X. (1.4.1)
If C = I, then L() = R
A
(), where A is the generator of the corresponding
C
0
-semigroup.
Property (C3) implies that L() is a bounded operator for all C
with Re > . We show that this operator, called the C-resolvent, is
invertible and satises the regularized resolvent identity.
Proposition 1.4.1 Let S(t), t 0 be an exponentially bounded
C-regularized semigroup. Then for all , C with Re , Re >
( )L()L() = L()C L()C, (1.4.2)
and there exists a closed operator Z dened by
Zx :=
_
I L()
1
C
_
x, (1.4.3)
D(Z) =
_
x X

Cx ranL()
_
.
Proof Let x X and consider , C with Re , Re > . Then
L()L()x
=
_

0
_

0
e
(s+t)
S(s +t)Cxdtds
=
_

0
_

0
e
((t)+t)
S()Cxdtd
= ( )
1
_ _

0
e

S()Cxd
_

0
e

S()Cxd
_
= ( )
1
_
L()Cx L()Cx
_
,
i.e. (1.4.2) holds. We now show that L() is injective for Re > . It
follows from (1.4.2) that if L() = 0, then CL()x = 0 and L()x = 0,
that is ker L() = ker L(). From the denition of L() and the uniqueness
theorem for Laplace transform we obtain that if L()x = 0, then S(0)x =
2001 CRC Press LLC
2001 CRC Press LLC
Cx = 0 and x = 0. Hence the operator L() is invertible for Re > . Let
Cx ranL(). From (1.4.2) we have
L() [L() C ] x = L() [L() C ] x, Re , Re > ,
and
L()
1
[L() C ] x = L()
1
[L() C ] x.
Therefore, the operator Z dened by
Zx := L()
1
[L() C ] x = [I L()
1
C ]x
is independent of and is closed. Then we have
(I Z)
1
= C
1
L(), (1.4.4)
D
_
(I Z)
1
_
=
_
x X

L()x ranC
_
and
L()x = C(I Z)
1
x, x D
_
(I Z)
1
_
.
Denition 1.4.2 The operator Z dened by (1.4.3) is called the generator
of a C-regularized semigroup S(t), t 0.
Along with the generator Z we introduce the innitesimal generators A
and G by
Ax := C
1
lim
t0
t
1
[S(t)x Cx] ,
D(A) =
_
x X

lim
t0
t
1
[S(t)x Cx] ranC
_
and
Gx := lim
t0
t
1
_
C
1
(S(t)x Cx)

, (1.4.5)
D(G) =
_
x ranC

lim
t0
t
1
_
C
1
(S(t)x Cx)

_
.
Proposition 1.4.2 Let S(t), t 0 be a C-regularized semigroup. Then
for operators G, A, Z the following relation holds
G A = Z. (1.4.6)
If ranC = X, then D(G) = X.
Proof Let x D(G), then
C lim
t0
C
1
S(t)x Cx
t
= lim
t0
S(t)x Cx
t
ranC
2001 CRC Press LLC
2001 CRC Press LLC
and
Ax = C
1
lim
t0
S(t)x Cx
t
= lim
t0
C
1
S(t)x x
t
= Gx.
Hence D(G) D(A) and G A. In order to show A Z, let x D(A).
Then
dS(t)Cx
dt
= lim
h0
S(t)(S(h) C)x
h
= S(t) lim
h0
S(h)x Cx
h
= S(t)CAx,
and
CL()(I A)x = L()C(I A)x
= L()Cx
_

0
e
t
S(t)CAx dt
= L()Cx
_

0
e
t
dS(t)Cx
dt
dt = C
2
x.
Hence L()(I A)x = Cx, i.e. Cx ranL and
Zx =
_
I L
1
()C

x = Ax.
Now let x D(Z), then there exists y such that Cx = L()y. We consider
S(h)x Cx
h
=
S(h)C
1
L()y L()y
h
=
1
h
_
C
1
S(h)
_

0
e
t
S(t)ydt
_

0
e
t
S(t)ydt
_
=
1
h
_ _

0
e
t
S(t +h)ydt
_

0
e
t
S(t)ydt
_
=
e
h
1
h
_

h
e
t
S(t)ydt
1
h
_
h
0
e
t
S(t)ydt

h0
L()y Cy = C(x y) ranC.
Hence x D(A), D(A) = D(Z) and Ax = Zx.
Now suppose ranC = X. Taking into account that C is injective,
we obtain that ranC
2
is also dense in X. Therefore, in order to prove
D(G) = X, we show that
x ranC
2
, > 0, x

D(G) : |x x

| < .
2001 CRC Press LLC
2001 CRC Press LLC
It follows from properties of C-regularized semigroups that the operators
U(t) := C
1
S(t), t 0
with
D(U(t)) =
_
x X

S(t)x ranC
_
,
form a C
0
-semigroup on ran C
2
. Put x = C
2
y and
x(t) :=
_
t
0
U()xd = C
_
t
0
S()yd ranC.
We have |x(t) x|
t0
0. Let us show that x(t) D(G). Since
U(h)x(t) = U(h)C
_
t
0
S()yd =
_
t
0
S(h +)Cyd,
we obtain
h
1
[U(h)x(t) x(t)]
= h
1
_
t
0
S(h +)Cyd h
1
_
t
0
S()Cyd
= h
1
_
t+h
h
S()Cyd h
1
_
t
0
S()Cyd
= h
1
_
t+h
t
S()Cyd h
1
_
h
0
S()Cyd

h0
S(t)Cy C
2
y.
Hence, x(t) D(G), Gx(t) = S(t)Cy C
2
y, and D(G) = X.
Denition 1.4.3 The closure of operator G dened by (1.4.5) is called
the complete innitesimal generator of the C-regularized semigroup S(t),
t 0.
For the complete innitesimal generator we have the inclusion G Z = A.
Proposition 1.4.3 Let S(t), t 0 be a C-regularized semigroup. If
(G) ,= , then G = A = Z.
Proof Let us show the equality
Ax = C
1
GCx = C
1
GCx, x D(A), (1.4.7)
which together with the assumption (G) ,= implies that G = A. Take
x D(A), then Cx D(G) :
GCx = lim
h0
h
1
_
C
1
S(h)Cx Cx

= lim
h0
h
1
[S(h)x Cx] = CAx.
2001 CRC Press LLC
2001 CRC Press LLC
Therefore GCx = CAx for all x D(A). Applying C
1
to both sides of
this equality, we obtain (1.4.7).
Let (G) and x D(A), put y = (I A)x, z = (I G)
1
y.
Then z D(G) and (I G)z = (I A)z = y = (I A)x. In view of
invertibility of (I G), the operator
(I A) = C
1
(I G)C
is also invertible. Hence z = x D(G) and D(A) D(G), that is A = G.
The following theorem makes clear the connection between a generator
of a C-regularized semigroup and the Cauchy problem (CP).
Theorem 1.4.1 Let Z be the generator of an exponentially bounded C-
regularized semigroup S(t), t 0. Then
for x X
S(t)x Cx = Z
_
t
0
S()xd; (1.4.8)
for x D(Z)
S(t)x Cx =
_
t
0
S()Zxd, (1.4.9)
dS(t)
dt
x = S(t)Zx = ZS(t)x, S(0)x = Cx; (1.4.10)
for x CD(Z) we have
dU(t)
dt
x = ZU(t)x, U(0)x = x, (1.4.11)
where U(t)x := C
1
S(t)x.
Proof Keeping in mind that A = Z, we prove that if x X, then
_
t
0
S()xd D(A) and (1.4.8) holds true. Let x X, then
S(h)
_
t
0
S()xd C
_
t
0
S()xd
h
=
_
t
0
S( +h)Cxd C
_
t
0
S()xd
h
=
C
_
t+h
0
S()xd
_
h
0
S()Cxd
h

h0
C [S(t) C] x.
Thus
A
_
t
0
S()xd = S(t)x Cx.
2001 CRC Press LLC
2001 CRC Press LLC
Since the operator A = Z is closed, we have (1.4.9):
_
t
0
S()Axd =
_
t
0
S()Zxd = S(t)x Cx
which implies (1.4.10) (1.4.11).
Now we can summarize all properties of the generator of a C-regularized
semigroup.
Theorem 1.4.2 Let Z be the generator of an exponentially bounded C-
regularized semigroup S(t), t 0, then Z is closed. If, in addition,
ranC = X, then Z is densely dened. For all C with Re > the
following conditions hold
(Z1) (I Z)
1
Cx = C(I Z)
1
x for all x ran(I Z);
(Z2) K > 0,
|(I Z)
m
C|
K
(Re )
m
, m = 1, 2, . . . .
Proof In Proposition 1.4.1 we proved that the operator
(I Z)
1
= C
1
L()
exists on
ran(I Z) =
_
x X

L()x ranC
_
.
Hence C(I Z)
1
x = L()x for x ran(I Z). On the other hand,
for all x X we have L()Cx = CL()x, therefore (Z1) holds. Let us
show that ranC D((I Z)
m
), m = 1, 2, . . . . If x X, then
L()Cx = CL()x ranC and the following operators are dened
(I Z)
1
Cx = C
1
L()Cx = C
1
_

0
e
t
1
S(t
1
)Cxdt
1
=
_

0
e
t
1
C
1
S(t
1
)Cxdt
1
=
_

0
e
t
1
S(t
1
)xdt
1
,
(I Z)
2
Cx =
_

0
C
1
e
t
2
S(t
2
)
_

0
e
t
1
C
1
S(t
1
)Cxdt
1
dt
2
=
_

0
_

0
e
(t
1
+t
2
)
S(t
1
+t
2
)xdt
1
dt
2
and so on. Hence we obtain the estimates (Z2):
|(I Z)
m
Cx| = |(C
1
L()) . . . (C
1
L())Cx|
2001 CRC Press LLC
2001 CRC Press LLC

_

0
. . .
_

0
e
Re (t
1
+...+t
m
)
Ke
(t
1
+...+t
m
)
|x|dt
1
. . . dt
m

K
(Re )
m
|x|.
Property (Z2) generalizes the MFPHY condition to the case of C-
regularized semigroups, and Theorem 1.4.2 indicates that the MFPHY-type
condition on the C-resolvent is necessary for operator Z to be the generator
of a C-regularized semigroup. The next theorem gives sucient conditions
for operator Z to be the generator of a C-regularized semigroup. Theorem
1.4.2 together with Theorem 1.4.3 are the generalization of the MFPHY
theorem to the case of C-regularized semigroups with ranC = X.
Theorem 1.4.3 Let Z be a densely dened closed linear operator satisfying
(Z1), (Z2) for C with Re > . Then there exists an exponentially
bounded C-regularized semigroup S(t), t 0 with the generator Z.
This theorem can be proved in a fashion similar to the proof in Remark
1.1.4. First we dene operators
P

x =
2
(I Z)
1
x x, x ran (I Z),
such that lim

Cx = ZCx. Next dene

S

(t)x := e
t

n=1
t
n

2n
n!
(I Z)
n
Cx, x X.
These operators are bounded and have all the properties of C-regularized
semigroups. Finally, we set
S(t)x = lim

(t)x, x X.
1.4.2 C-well-posedness of the Cauchy problem
We consider the Cauchy problem
u

(t) = Au(t), t 0, u(0) = x, (CP)

where A is the generator of an exponentially bounded C-regularized semi-
group S(t), t 0. It is shown in Theorem 1.4.1 that for x CD(A)
there exists a solution to this problem: u(t) = U(t)x = C
1
S(t)x, t 0.
The semigroup U is not necessarily bounded on X, therefore, in general,
the solution u() is not stable in X with respect to variation of x. We now
show that the well-posedness of the Cauchy problem can be restored in a
certain subspace of X with a norm stronger than the norm of X.
2001 CRC Press LLC
2001 CRC Press LLC
Theorem 1.4.4 Let S(t), t 0 be an exponentially bounded
C-regularized semigroup. Then the following statements are equivalent:
(I) A is the generator of S(t), t 0;
(II) a) A is a closed operator and it commutes with C:
ACx = CAx, x D(A);
b) there exists a Banach space with the norm | |

such that
ranC X, and the following estimates hold
K
1
> 0 : x , |x| K
1
|x|

,
K
2
> 0 : x ranC, |x|

K
2
|C
1
x|.
Furthermore, the part of the operator A in is the generator of
a C
0
-semigroup.
Proof (I) = (II). Let S(t), t 0 be an exponentially bounded C-
regularized semigroup satisfying (C1) (C3), then a) holds by virtue of
(Z1). Let b > , dene in X the linear manifold
=
_
x X

C
1
S(t)x is continuous with respect to t 0,
and lim
t
e
bt
|C
1
S(t)x| = 0
_
with the norm
|x|

= sup
t0
e
bt
|C
1
S(t)x|.
Taking into account that C
1
S(t)S(h)x = S(t +h)x, we have ran S(h)
for h 0. If x ranC = ranS(0), then
|x|

sup
t0
Ke
(b)t
|C
1
x| = K|C
1
x|.
For x we have
|x| =
_
e
bt
|C
1
S(t)x|
_
t=0
|x|

.
It is not dicult to show that with the norm | |

is a Banach space. It
is invariant with respect to operators C
1
S(t) as
C
1
S()C
1
S(t)x = C
1
S( +t)x, x .
2001 CRC Press LLC
2001 CRC Press LLC
Therefore, the operators U(t) = C
1
S(t) map into . Let us show
that A

(the part of operator A in ) is the generator of a C

0
-semigroup
U(t), t 0 in . We have U(t+) = U(t)U() and U(0) = I. Moreover,
|U(t)x|

= sup
0
e
b
|C
1
S( +t)x|
sup
0
e
b(+t)
e
bt
|C
1
S( +t)x|
= e
bt
sup
t
e
bt
|C
1
S()x|
e
bt
|x|

and
|U(h)x x|

= sup
t0
e
bt
|C
1
S(t +h)x C
1
S(t)x|
max
_
sup
t[0,T]
e
bt
|C
1
S(t +h)x C
1
S(t)x|,
sup
tT
e
bt
|C
1
S(t)x| +e
bh
sup
tT
e
b(t+h)
|C
1
S(t +h)x|
_
maxa
T
(h), c
T
(h).
By denition of we have
> 0, T
0
: h 0, c
T
0
(h) <

2
.
Also, since the continuous function C
1
S()x is uniformly continuous on
any closed interval, we have
> 0, : h < , a
T
0
(h) <

2
.
Hence
|U(h)x x|

h0
0,
i.e., the operators U(t) form a C
0
-semigroup on . Let

A be its generator,
then for x D(

A)
_
_
_t
1
_
C
1
S(t)x x

Ax
_
_
_
K
1
_
_
_t
1
_
C
1
S(t)x x

Ax
_
_
_

t0
0.
Therefore x D(A) and Ax =

Ax , that is x D(A

) and
D(

A) D(A

). To prove D(A

) D(

A), we take x D(A) such that

2001 CRC Press LLC
2001 CRC Press LLC
Ax , then it follows from (1.4.9) that
U(t)x x =
_
t
0
U()Axd.
Hence, taking into account that for any x the function U()Ax is
continuous in > 0, we obtain
_
_
_t
1
_
C
1
S(t)x x

Ax
_
_
_

t0
0.
Thus x D(

A) and A

=

A.
(II) = (I). Let U be a C
0
-semigroup generated by the operator A

. It
is not dicult to verify that the operator-function S() dened by S(t)x =
U(t)Cx for all x X and t 0, satises (C1) (C3). Let Z be the
generator of S, then C
1
ZCx = Zx, x D(Z), and
(I Z)
1
Cx =
_

0
e
t
S(t)xdt, x X, > .
On the other hand, for the C
0
-semigroup U we have
(I A

)
1
Cx =
_

0
e
t
U(t)Cxdt, x X, > ,
therefore
(I A

)
1
Cx = (I Z)
1
Cx, x X, > . (1.4.12)
Hence
(I Z)
1
C(I A)x = (I A

)
1
C(I A)x
= (I A

)
1
(I A

)Cx = Cx
for x D(A). Consequently Ax = C
1
ZCx = Zx, and A Z. In the
same way, it follows from (1.4.12) that Z A, so A = Z.
By Theorem 1.4.4, we obtain that (CP) with operator A being the
generator of a C-regularized semigroup is well-posed in the space . The
fact that A

generates a C
0
-semigroup in implies that for any x D(A

)
there exists the unique solution of (CP): u() = U()x such that
sup
0tT
|u(t)|

K|x|

. (1.4.13)
Denition 1.4.4 The Cauchy problem (CP) is said to be C-well-posed if
for any x CD(A) there exists a unique solution, that is stable with respect
to the C
1
-graph-norm:
T < , K > 0 :
sup
0tT
|u(t)| K|x|
C
1 K
_
|x| +|C
1
x|
_
.
2001 CRC Press LLC
2001 CRC Press LLC
If A is the generator of a C-regularized semigroup, then for any x
CD(A) D(A

), u() = C
1
S()x is the unique solution of (CP). Taking
into account b), we obtain from (1.4.13) that
sup
0tT
|u(t)| K
1
KK
2
|C
1
x|.
This implies C-well-posedness of (CP) with operator A being the generator
of a C-regularized semigroup.
1.4.3 Local C-regularized semigroups
There are operators A (for example, those having arbitrary large positive
eigenvalues) that generate C-regularized semigroups, dened only for t
from some bounded subset of R. Such semigroups correspond to the local
C-well-posedness of the Cauchy problem.
Let T (0, ) and let C : X X be an injective bounded linear
operator with dense range: ranC = X.
Denition 1.4.5 A one-parameter family of bounded linear operators
S(t), t [0, T) is called a local C-regularized semigroup in X if
(LC1) S(t +s)C = S(t)S(s) for s, t 0, s +t < T and S(0) = C,
(LC2) for every x X, S(t)x : [0, T) X is continuous in t.
In the case T = a local C-regularized semigroup is a C-regularized
semigroup. If there exist K > 0 and R such that |S(t)| Ke
t
for
t 0, then S is an exponentially bounded C-regularized semigroup.
Let S be a local C-regularized semigroup. Dene its generator G by
Gx := lim
h0
h
1
(C
1
S(h)x x), x D(G),
D(G) =
_
x ranC

lim
h0
h
1
(C
1
S(h)x x) exists
_
It is proved in [138] that G is a densely dened, closable linear operator. Its
closure Gis called the complete innitesimal generator of S. The connection
of G with the local Laplace transform
L

()x :=
_

0
e
t
S(t)xdt, R, x X, (1.4.14)
and with the local Cauchy problem (CP) is described in the following
proposition.
Proposition 1.4.4 Let G be the complete innitesimal generator of a local
C-regularized semigroup S(t), t [0, T) and let L

() be as in (1.4.14).
Then G is densely dened and the following hold:
2001 CRC Press LLC
2001 CRC Press LLC
1) for x D(G), 0 t < T we have S(t)x D(G),
GS(t)x = S(t)Gx and Gx = S

(t)x;
2) for x D(G), GL

()x = L

()Gx;
3) for x X, we have L

()x D(G) and

(I G)L

()x = Cx e

S()x;
4) for x X,
L

()L

()x = L

()L

()x
and
K

:
_
_
_
_
d
n1
d
n1
L

()
_
_
_
_
K

(n 1)!

n
(1.4.15)
for > 0, n N.
From properties 2) 4) we conclude that if > 0 is suciently large, then
the operator L

() is similar to the C-resolvent.

Denition 1.4.6 Let A be a closed linear operator in X, and let (0, T).
A family of bounded linear operators L

(), > is called an asymptotic

C-resolvent of A if
(I) for x X, L

()x is an innitely dierentiable function of and

L

()L

()x = L

()L

()x;
(II) for x X, L

()x D(A) and

(I A)L

()x = Cx +V

()x,
where V

satises the following estimates:

K

:
_
_
_
_
d
n1
d
n1
V

()x
_
_
_
_
K

n1
e

|x|
for > , n N;
(III) for x D(A), AL

()x = L

()Ax.
As in Theorems 1.2.5, 1.2.6 for local integrated semigroups, the following
theorem gives conditions for an operator to be the complete innitesimal
generator of a C-regularized semigroup.
2001 CRC Press LLC
2001 CRC Press LLC
Theorem 1.4.5 A closed linear operator A in X is the complete innites-
imal generator of a local C-regularized semigroup S(t), t [0, T) if and
only if
(I) D(A) is dense in X;
(II) for every (0, T) there exists an asymptotic C-resolvent L

() of
A such that (1.4.15) is fullled for
n

[0, ], > , n N;
(III) CD(A) is a core for A, i.e.
A[
CD(A)
= A.
Proof The proof of necessity follows from Proposition 1.4.4. The proof of
suciency is based on the following analogy with the case of C
0
-semigroups.
If G is the complete innitesimal generator of a semigroup

S(t), t > 0,
then

S(t) can be represented in the form

S(t)x = lim
n
_
I
t
n
G
_
n
Cx, x X.
From the equality for the resolvent of operator G :
d
n
R
G
()
d
n
= (1)
n
n! R
n+1
G
()
similarly to (1.1.9) we have

S(t)x = lim
n
_
n
t
_
n
R
n
G
_
n
t
_
Cx =
(1.4.16)
= lim
n
_
n
t
_
n
(1)
n1
(n 1)!
d
n1
d
n1
R
G
()Cx

=
n
t
.
So we can see that a local C-regularized semigroup can be presented in
the form (1.4.16) where C-resolvent R
G
(
n
t
)C is replaced by the asymptotic
C-resolvent L

(
n
t
).
Let us x (0, T). For an operator A, satisfying the conditions of
the theorem and for n > (here constant is from the denition of
asymptotic C-resolvent), we dene the family of bounded linear operators
S
n,
(t), 0 t r in X:
S
n,
(t) =
_

_
(1)
n1
(n1)!
_
n
t
_
n
L
(n1)

_
n
t
_
for 0 < t ,
C for t = 0.
2001 CRC Press LLC
2001 CRC Press LLC
Let S

(t) = lim
n
S
n,
(t). It is proved in [133] that the operators
S

(t) exist, do not depend on , and form a local C-regularized semigroup

generated by A.
Consider the local Cauchy problem
u

(t) = Au(t), t [0, T), u(0) = x (CP)

Denition 1.4.7 The local problem (CP) is called C-well-posed on [0, T)
if for any x CD(A) there exists a unique solution u() such that
|u(t)| K(t)|C
1
x|,
where K(t) is bounded on every compact interval from [0, T).
The following theorem on well-posedness of local (CP) can be proved in a
fashion similar to the corresponding theorems on C-well-posedness of (CP).
Theorem 1.4.6 Suppose A is a densely dened closed linear operator in
X satisfying
a) x D(A), Cx D(A) and ACx = CAx,
b) A[
CD(A)
= A.
Then A is the complete innitesimal generator of a local C-regularized semi-
group S(t), t [0, T) if and only if the local (CP) is C-well-posed on
[0, T).
1.4.4 Integrated semigroups and
C-regularized semigroups
The following two theorems clarify the connection of C-regularized semi-
groups with integrated semigroups.
Theorem 1.4.7 Let A be a linear operator in X with (A) ,= . Let
(A) and n 0 be an integer. The following statements are equivalent:
(I) A is the generator of (n +1)-times integrated semigroup V (t), t 0
in X, satisfying
K > 0, R : |V (t +h) V (t)| Khe
(t+h)
for all t, h 0;
(II) A is the generator of a C-regularized semigroup S(t), t 0 with
operator C = R
n+1
A
(), satisfying
K > 0, R : |S(t +h) S(t)| Khe
(t+h)
for all t, h 0.
2001 CRC Press LLC
2001 CRC Press LLC
In this case, we have
V (t)x = (I A)
n+1
_
t
0
_
t
1
0
. . .
_
t
n
0
S(t
n+1
)xdt
n+1
. . . dt
1
for all x X and t 0.
In the case of a densely dened operator A, this theorem and Theorem 1.2.3
lead to the following result.
Theorem 1.4.8 Let A be a densely dened linear operator in X, (A)
and n N. Then the following statements are equivalent:
(I) A is the generator of an n-times integrated exponentially bounded semi-
group V (t), t 0;
(II) A is the generator of an exponentially bounded C-regularized semi-
group S(t), t 0 with C = (I A)
n
.
Proof We prove the following implications: (R) = (II) = (I) = (R),
where
(R) K > 0, a R : (a, ) (A),
_
_
(I A)
m
(I A)
n
_
_

K
( a)
m
,
for all > a and m = 1, 2, . . . .
The implication (R) = (II) follows from Theorem 1.4.3.
Now we prove (II) = (I). We dene the operators S
k
(t), k 0, by
S
0
(t) = S(t),
S
k
(t)x =
_
t
0
. . .
_
t
2
0
S(t
1
)xdt
1
. . . dt
k
, x X, t 0, k 1.
It is shown in the proof of Theorem 1.4.1 that
_
t
0
S()xd D(A) and
S(t)x Cx = A
_
t
0
S()xd, x X, t 0.
Then by induction in k, we have
S
k
(t)x D(A
k
),
_
t
0
(I A)
k1
S
k1
()xd D(A),
and
K
k
> 0 : |(I A)
k
S
k
(t)| K
k
e
t
.
2001 CRC Press LLC
2001 CRC Press LLC
Furthermore, for x X, (I A)
k
S
k
(t)x is continuous in t 0.
We dene V (t)x := (I A)
n
S
n
(t)x, x X. Then V satises (V2),
(V3) and (V4) (see Denition 1.2.1). Let us prove (V1). It follows from
Proposition 1.2.1 that it is sucient to show that
_

0

n
e
t
V (t)xdt = (I A)
1
x, x X.
Integrating by parts we obtain
_

0

n
e
t
V (t)xdt =
_

0

n
e
t
(I A)
n
S
n
(t)xdt
= (I A)
n
_

0
e
t
S(t)xdt
= (I A)
n
(I A)
1
Cx = (I A)
1
x.
(I) = (R). By Proposition 1.2.1 for the generator of a nondegenerate
n-times integrated semigroup V we have the inclusion (, ) (A) and
(I A)
1
x =
_

0

n
e
t
V (t)xdt, x X, > . (1.4.17)
Taking into account (1.2.6):
V
(n)
(t)x = V (t)A
n
x +
n1

k=0
t
k
k!
A
k
x,
we integrate (1.4.17) by parts
(I A)
1
x =
_

0
e
t
_
n1

k=0
t
k
k!
A
k
+V (t)A
n
_
xdt, x D(A
n
).
Dierentiating this equality (m1)-times with respect to , we obtain
(m1)! |(I A)
m
(I A)
n
x|

_

0
t
m1
e
t
_
n1

k=0
t
k
k!
|A
k
(I A)
n
|
+Ke
t
|A
n
(I A)
n
|
_
|x| dt
(m1)! K
1
_

0
t
m1
e
t
e
t
+e
t
2
dt|x|
(m1)! K
1
( a)
m
|x|
for x X, > a, m 1. Here the constant K is from (V3), and
a = max 1, ,
K
1
= 2 max
k=1,...,n1
_
|A
k
(I A)
n
|, K|A
n
(I A)
n
|
_
.
2001 CRC Press LLC
2001 CRC Press LLC
1.4.5 Examples
Example 1.4.1 The operator
A =
_
0 I
B 0
_
with D(A) =
__
x
y
_
X X

x, y D(B)
_
,
which was considered as the generator of an integrated semigroup in Ex-
ample 1.2.4, is the generator of an exponentially bounded C-regularized
semigroup S dened by
S(t)
_
x
y
_
=
_

0
((t) +Bo(t) ((t) +
0
o(t)

0
(

(t) +B((t) (

(t) +
0
((t)
_
R
B
(
2
0
)
_
x
y
_
with C = R
A
(
0
),
0
> , where is dened by the (-function. Here
R
B
(
2
) = (
2
I B)
1
. Note that S() is dened for any x, y X since
range of R
B
(
2
0
) is D(B). The C-resolvent of operator A has the form
R
A
(
0
)R
A
() =
_

0
I
B
0
_
R
B
(
2
0
)
_
I
B
_
R
B
(
2
)
=
_
B +
0
+
0
B( +
0
) B +
0
_
R
B
(
2
) R
B
(
2
0
)

2
0
.
Example 1.4.2 (An exponentially bounded C-regularized semigroup that
is not an n-times integrated semigroup)
As in Examples 1.1.2 and 1.2.2, let
X =
_
L
p
(R) L
p
(R), |u| = |u
1
|
L
p +|u
2
|
L
p
_
, where u =
_
u
1
u
2
_
.
Consider the operator A dened by
Au =
_
g f
0 g
_
u
with
D(A) =
__
u
1
u
2
_
X

gu
1
+fu
2
L
p
(R), gu
2
L
p
(R)
_
,
where g(x) = 1+[x[, f(x) = [x[

, > 0. In Example 1.2.1 we demonstrated

that for any (0, 1] the operator A is the generator of a C
0
-semigroup U
dened by
U(t)u = e
tg
_
1 tf
0 1
_
u, t 0 , u X.
2001 CRC Press LLC
2001 CRC Press LLC
Recall that for any 0 we have
|U(t)| = max
xR
_
(1 + t[x[

)e
t(1+|x|)
_
= (1 +

t
1
)e
t
(
1+

t
)
=
t0
O(t
1
).
Note that for any (1, 2], A is the generator of an integrated semi-
group (see Example 1.2.2), therefore by Theorem 1.4.8, A generates a C-
regularized semigroup with C = R
A
().
Now let > 2. We have
(I A)
1
u = ( +g)
2
_
( +g) f
0 ( +g)
_
u, > 0,
therefore, the operator (I A)
1
is not bounded for > 0. Hence, the
resolvent of A does not exist for > 0 and A cannot be a generator of an
n-times integrated semigroup for any n. Since |U(t)| = O(t
1
) as t 0,
then the semigroup U has a singularity of order = 1 at t = 0. This
semigroup satises the following denition of a semigroup of growth (see
[84] and references therein):
(a) X
0
:=

t>0
U(t)[X] is dense in X,
(b) U is non-degenerate,
(c) |t

U(t)| is bounded as t 0.
Let n = [[]]. Dene the operator C in X by
Cu =
1
n!
_

0
t
n
e
t
U(t)udt, u X, > 0,
then C is injective and R(C) = X. It is not dicult to verify that the
family S(t) := U(t)C, t 0 is an exponentially bounded C-regularized
semigroup. If n = 0, then U has an integrable singularity. In this case,
C = (I A)
1
, > 0, is the resolvent of A and A is the generator of an
integrated semigroup.
Now let n 1. If (I A)
1
exists and is equal to the resolvent of
A, then C = (1)
n
R
n+1
A
() and A is the generator of an (n + 1)-times
integrated semigroup. Otherwise (which is the case for > 2), A is the
generator of a C-regularized semigroup and is not a generator of an inte-
grated semigroup.
Finally, consider the operator C dened by
Cu := (1 +[f[)
1
u, u X,
2001 CRC Press LLC
2001 CRC Press LLC
then we have that C is injective and R(C) = X. In this case, S(t) =
U(t)C, t 0 is an exponentially bounded C-regularized semigroup. We
note that this semigroup remains exponentially bounded even for functions
like f(x) = e
(1+|x|)
that grow faster than [x[

for any > 0.

Example 1.4.3 (A C-regularized semigroup that is not exponentially
bounded)
Consider the operator A dened by
Af(z) = zf(z), D(A) =
_
f X = L
2
(C)

zf(z) X
_
,
which is the generator of the C-regularized semigroup
S(t)f := e
zt|z|
2
f(z), f X,
with Cf = e
|z|
2
f, f X. We have
|S(t)| = sup
zC
_
e
zt|z|
2
_
= e
t
2
4
,
that is, S is not exponentially bounded. Note that in this example the
operator (I A) is not invertible for any C and (A) = C.
Example 1.4.4 (Local semigroups corresponding to the regularization
methods)
Let
Au =
d
2
u
ds
2
on L
2
[0, l] with
D(A) =
_
u L
2
[0, l]

d
2
u
ds
2
L
2
[0, l], u(0) = u(l) = 0
_
.
then
n
= (
n
l
)
2
are the eigenvalues for this operator and its eigenvectors

n
(s) = sin
n
l
s form an orthogonal basis in L
2
[0, l].
The operators
S
1
(t)x :=

k=1
e

k
t
2
k
T
x
k

k
and
S
2
(t)x :=

k=1
e

k
t
1 +e

k
T
x
k

k
,
where x
k
= (x,
k
), t < T, > 0, are local C-regularized semigroups on
(0, T). The asymptotic C-resolvent for S
1
is
L

()x =

k=1
(e
(
k
)
1)e

2
k
T

k
x
k

k
.
2001 CRC Press LLC
2001 CRC Press LLC
1.5 Degenerate semigroups
In this section we consider degenerate integrated and C
0
-semigroups con-
nected with the degenerate Cauchy problem
Bu

(t) = Au(t), t 0, u(0) = x, ker B ,= 0, (DP)

where B, A : X Y are linear operators in Banach spaces X, Y .
We assume that the set

B
(A) :=
_
C

R() := (B A)
1
B is a bounded operator on X
_
is not empty. This set is called the B-resolvent set of operator A and the
operator R() is called the B-resolvent of operator A.
We prove the equivalence of the uniform well-posedness of (DP) on the
maximal correctness class
:=
_
x D(A)

Ax ranB
_
= R()X,
the existence of a degenerate C
0
-semigroup and MPFHY-type estimates
together with the decomposition of X:
X = X
1
ker B, (1.5.1)
where X
1
:= D
1
and D
1
:= R()X with some
B
(A). The decompo-
sition (1.5.1) plays a role similar to the condition of density of a generator
of a C
0
-semigroup. We also consider the well-posedness of the degenerate
Cauchy problem on subsets of R
n
()X.
1.5.1 Generators of degenerate semigroups
Denition 1.5.1 A one-parameter family of bounded linear operators
U(t), t 0 on X is called a degenerate C
0
-semigroup if the semigroup
relation (U1) and the strong continuity condition (U3) of Denition 1.1.1
hold, and the operator U(0) (and hence U(t) for any t 0) has a nonempty
kernel.
Denition 1.5.2 Let n N. An n-times integrated semigroup V (t),
t [0, T) is called a degenerate n-times integrated semigroup if conditions
(V1) (V3) of Denition 1.2.1 hold and (V4) does not.
Remark 1.5.1 The exponential boundedness of a degenerate semigroup
U(t), t 0:
K > 0, R : |U(t)| Ke
t
, t 0,
2001 CRC Press LLC
2001 CRC Press LLC
follows from property (U1). Note that a degenerate n-times integrated
semigroup, similarly to a nondegenerate semigroup, may be not exponen-
tially bounded.
Remark 1.5.2 Property (U1) implies that the operator U(0) is a projector
on X, generating the following decomposition of X
X = ranU(0) ker U(0).
In this case, the restriction of U() on ranU(0) is a nondegenerate C
0
-
semigroup.
Let V () be a strongly continuous operator-function such that
K > 0, R : |V (t)| Ke
t
,
for all t 0, and
r() =
_

0

n
e
t
V (t)dt
for some n N 0. It was proved in Propositions 1.1.2, 1.2.1 that r()
satises the (pseudo) resolvent identity
( )r()r() = r() r(), Re , Re > ,
if and only if V satises (V1) for n 1 or (U1) for n = 0. Such function
r() is called the pseudoresolvent. Moreover, if V is nondegenerate, then
r() is invertible and is the resolvent of the generator A dened by
A := I r
1
().
Obviously, when dealing with single-valued operators, this denition of the
generator cannot be used in the case of a degenerate semigroup V . We
introduce the notion of a pair of generators A, B, which allows us to link
degenerate semigroups with the degenerate Cauchy problem (DP).
Denition 1.5.3 Let n N 0. Linear operators A, B : X Y are
called the generators of an exponentially bounded degenerate n-times inte-
grated semigroup V (t), t 0 on X if A is closed, B is bounded and
(B A)
1
B =
_

0

n
e
t
V (t)dt
for all C with Re > .
From Arendt-Widders Theorem 1.2.1 we obtain the following result char-
acterizing the generators of degenerate (n+1)-times integrated semigroups.
Theorem 1.5.1 Let A, B : X Y be linear operators, A be closed and B
be bounded. Then the following statements are equivalent:
2001 CRC Press LLC
2001 CRC Press LLC
(I) the operators A and B are the generators of a degenerate (n+1)-times
integrated semigroup V (t), t 0 satisfying the condition
lim
h0
sup
1
n
|V (t +h) V (t)| Ke
t
, t 0; (1.5.2)
(II) the estimates
K > 0, R :
_
_
_
_
_
d
k
d
k
_
(B A)
1
B

n
_
_
_
_
_
_

K k!
( )
k+1
(1.5.3)
hold for all > and k = 0, 1, . . . .
In the next subsection we use degenerate 1-time integrated semigroups and
degenerate C
0
-semigroups in the investigation of the uniform well-posedness
of the Cauchy problem (DP).
1.5.2 Degenerate 1-time integrated semigroups
Let A and B be the generators of a degenerate 1-time integrated semigroup
V (t), t 0 satisfying (1.5.2). Then the B-resolvent set
B
(A) is not
empty. We consider the set D
1
:= R()X, where > . As in the case of
domains of generators of nondegenerate semigroups, the introduced set does
not depend on > . To show this we consider , > and x = R()z
with z X. Then (B AB)x = Bz and
x = R()z = (B A)
1
B(z ( )x) (B A)
1
Bz
1
= R()z
1
.
Proposition 1.5.1 Let A and B be the generators of a degenerate 1-time
integrated semigroup V (t), t 0 satisfying (1.5.2). Then ker B X
1
=
0 and ker B X
1
is a subspace in X.
Proof Recall that X
1
:= D
1
. Consider the family of linear operators
R(), > . By (1.5.3) they are bounded on X. We show that for any
x X
1
R()x

x. (1.5.4)
Let x D
1
, then there exists y X such that x = R()y for some
R
B
(A). Using the resolvent identity for R() and estimates (1.5.3)
with n = 0, we obtain
|R()x x| = |R()R()y R()y|
=
_
_
_
_

( )
R()y

( )
R()y R()y
_
_
_
_

|R()| +


|R()y|

0.
2001 CRC Press LLC
2001 CRC Press LLC
Furthermore, (1.5.3) implies that operators R() are uniformly bounded,
and by the Banach-Steinhaus Theorem, (1.5.4) holds for any x X
1
.
Hence,
ker B X
1
= 0.
The uniformly bounded family of operators R() converges on the set
ker B X
1
as . Hence, by the Banach-Steinhaus theorem, it con-
verges on ker B X
1
to a bounded linear operator P such that Px = x for
x X
1
. Furthermore, Px = 0 for x ker B, ranP = X
1
and P
2
= P.
Hence, P is a projector in ker B X
1
and
ker B X
1
= ker P ranP = ker B X
1
.
The connection between the degenerate Cauchy problem (DP) and the
generators of degenerate semigroups is established with the help of the
following theorem.
Theorem 1.5.2 Let A and B be the generators of a degenerate 1-time
integrated semigroup V (t), t 0 satisfying (1.5.2), then
R()V (t) = V (t)R() (1.5.5)
for all C with Re > and all t 0. Further
tBx = BV (t)x A
_
t
0
V (s)xds (1.5.6)
for all x ker B X
1
and t 0.
The operator-function V

() is a degenerate C
0
-semigroup on the sub-
space ker B X
1
that coincides with
F :=
_
x X

V (t)x C
1
(0, ), X
_
,
and
BV

(t)x = AV

(t)x (1.5.7)
for all x R()X
1
and t 0.
Proof Let Re , Re > . Then for any x X,
_

0
e
t
V (t)R()xdt = R()R()x = R()R()x
=
_

0
e
t
R()V (t)xdt.
Using the uniqueness of the Laplace transform, we obtain (1.5.5).
2001 CRC Press LLC
2001 CRC Press LLC
Now let x X and > , then
_

0

2
e
t
t Bxdt = Bx = (B A)(B A)
1
Bx (1.5.8)
=
_

0

2
e
t
BV (t)xdt A
_

0
e
t
V (t)xdt.
Take x D
1
, then x = (
0
B A)
1
By for some y X. For such x we
have V (t)x D(A) and
AV (t)x = AV (t)(
0
B A)
1
By = A(
0
B A)
1
BV (t)y
=
0
(
0
B A)
1
BV (t)y BV (t)y
=
0
BV (t)x BV (t)y.
Hence,
|AV (t)x| K|B| e
t
_
[
0
[ |x| +|y|
_
.
As AV ()x is exponentially bounded for x D
1
, the Laplace transform of
AV ()x exists. Since the operator A is closed, by (1.5.8) we have
_

0

2
e
t
tBxdt =
_

0

2
e
t
BV (t)xdt
_

0
e
t
AV (t)dt
=
_

0

2
e
t
_
BV (t)x
_
t
0
AV (s)xds
_
dt.
Hence,
tBx = BV (t)x
_
t
0
AV (s)xds
for all x D
1
and t 0. Due to the closedness of A, this equality holds
for all x X
1
= D
1
. Moreover, the equality holds for x ker B, since
V (t)x = 0 for all x ker B and t 0. (1.5.6) is proved.
Since property (1.5.2) guarantees that V

(t) is bounded:
M > 0, x F, |V

(t)x| Me
t
|x|, t 0,
we conclude that the linear manifold F is closed. Let x F. Dierentiating
the equality (V1) with n = 1, we obtain
V (t +s)x V (s)x = V (t)V

(s)x, t, s 0. (1.5.9)
The left-hand side of this equality is dierentiable with respect to t, hence
for any x F we have V

(s)x F, i.e. for any s 0, V

(s) is a bounded
operator on F. Dierentiating (1.5.9) we have
V

(t +s)x = V

(t)V

(s)x, t, s 0, x F, (1.5.10)
2001 CRC Press LLC
2001 CRC Press LLC
which implies the semigroup property (U1) for V

. We now show that V

is a degenerate semigroup on F. From the denition of a generator of V ,

we have
(B A)
1
Bx =
_

0
e
t
V

(t)xdt (1.5.11)
for all x F and all C with Re > . From (1.5.10) and (1.5.11) we
have [V

(0)]
2
= V

(0). Now consider x ker B, then the uniqueness of the

Laplace transform guarantees that V

(t)x = 0 for t 0. If x ker V

(0),
then (1.5.10) implies V

(t)x = 0 for t 0. Hence, by (1.5.11), x ker B.

Therefore, ker V

(0) = ker B.
Thus, V

(0) is a projector, V

(t), t 0 is a degenerate C
0
-semigroup
on F, and F is decomposed into the direct sum:
F = ker V

(0) ranV

(0) = ker B ranV

(0).
We now show that ranV

(0) = X
1
. By Remark 1.5.2 we have that the set
F
1
:=
_
x X

(t)x exists for t 0

_
contains the domain of the generator of the nondegenerate semigroup
V

(t)

ran V

(0)
, which is dense in ranV

(0). Hence F
1
ranV

(0). Taking
into account (1.5.8) and that V (t)x = 0 for x ker B, we have the inclusion
ker B F
1
and hence F
1
F.
As F
1
F = F, we have F
1
= F. Let x F
1
, then
R()x =
_

0
e
t
V

(t)xdt
= V

(0)x +
_

0
e
t
V

(t)xdt.
As , the limit of the second term in the right-hand side of this
equality is equal to zero, hence
V

(0)x = lim

R()x.
Since the operators R() are uniformly bounded, the equality holds for
any x F. Hence, ranV

(0) D
1
= X
1
.
Now we consider the set R()X
1
for some C with Re > . It is
not dicult to verify that R()X
1
does not depend on . Let x = R()y,
where y X
1
. We apply the operator (B A)
1
to the equality (1.5.6)
with y. Using the property (1.5.5) and the equality
(B A)
1
A = (B A)
1
B I,
2001 CRC Press LLC
2001 CRC Press LLC
which holds on D(A), we obtain
tx = V (t)x
_
t
0
V (s)
_
(B A)
1
B I

yds. (1.5.12)
Hence, x F and R()X
1
F. From (1.5.4) we have R()X
1
= X
1
,
therefore,
X
1
F = F, and ker V

(0) X
1
= ker B X
1
F.
This inclusion, together with already proved inclusions:
ranV

(0) X
1
and
F = ker B ranV

(0) ker B X
1
,
gives the equality F = ker B X
1
.
To prove (1.5.7) we dierentiate (1.5.12), where y X
1
and x = R()y:
x = V

(t)x V (t) [R() I] y.

Since [R() I]y X
1
, the second term in the right-hand side is dier-
entiable in t. Hence V

()x is also dierentiable and

V

(t)x = V

(t)[R() I]y.
Applying operator B to both sides of this equality, we obtain
B
d
dt
V

(t)x = B[R() I]V

(t)y = AR()V

(t)y
= AV

(t)R()y = AV

(t)x.
From Theorems 1.5.1, 1.5.2 we obtain the result about the connection
of the well-posedness of the degenerate Cauchy problem and estimates for
the B-resolvent of A.
Theorem 1.5.3 Let A, B : X Y be linear operators. Suppose that A is
closed, B is bounded, and estimates (1.5.3) are fullled for n = 0. Then
(DP) is uniformly well-posed on R()X
1
, Re > .
1.5.3 Maximal correctness class
Denition 1.5.4 If, for any x E X, a Cauchy problem (DP) has a
unique stable solution, then the set E is called the correctness class for this
problem.
2001 CRC Press LLC
2001 CRC Press LLC
The set
:=
_
x D(A)

Ax ranB
_
is the maximal correctness class for the Cauchy problem (DP) because if
u() is a solution of (DP), then u(t) for t 0, and the condition x
is necessary for the existence of a solution.
Proposition 1.5.2 Let A, B : X Y be linear operators such that

B
(A) ,= , then = D
1
:= R()X.
Proof Let x D
1
, then there exists y X such that x = R()y. Hence,
Ax = By + Bx = B(x y), and x M. Conversely, if x M, then
there exists z X such that Ax = Bz. Then (B A)x = B(x +z) and
x = R()(x z) D
1
.
Theorem 1.5.4 Let A, B : X Y be linear operators, A be closed, B be
bounded, and the B-resolvent set of A be nonempty. Then the following
statments are equivalent:
(I) the Cauchy problem (DP) is uniformly well-posed on D
1
;
(II) the operators A and B are the generators of a degenerate
C
0
-semigroup;
(III) the estimates (1.5.3) hold with n = 0 and X admits the decomposition
X = ker B X
1
.
Proof (I) =(II). Let u() be a solution of (DP). Dene on the solution
operators

U(t)x := u(t) for x D
1
and t 0. The uniform well-posedness
of (DP) implies that operators

U(t) are bounded on D
1
and therefore they
can be extended to X
1
= D
1
. As in the nondegenerate case (Theorem
1.1.1) the operators

U(t), t 0 form a C
0
-semigroup. This semigroup
_

U(t), t 0
_
is nondegenerate and is exponentially bounded on X
1
:
K > 0, R :
_
_
U(t)
_
_
K e
t
, t 0.
Let G : X
1
X
1
be the generator of this semigroup, then G is a closed,
densely dened linear operator on X
1
. The function

U(t)x, t 0, is
dierentiable with respect to t 0 if and only if x D(G). For any
x D(G),

U()x is the solution of the Cauchy problem
u

(t) = Gu(t), t 0, u(0) = x.

Since the semigroup

U is exponentially bounded on X
1
, the Laplace trans-
form of

U()x is dened for Re > and
_

0
e
t

U(t)xdt = (I G)
1
x, x X
1
.
2001 CRC Press LLC
2001 CRC Press LLC
By construction of

U we have D
1
D(G). Let x D
1
, then integrating
by parts we have

_

0
e
t

U(t)xdt = x +
_

0
e
t

(t)xdt.
for C with Re > . Applying B to this equality and taking into
account that

U() is a solution of (DP) and that A is closed, we obtain
B
_

0
e
t

U(t)xdt = Bx +
_

0
e
t
A

U(t)xdt, x D
1
,
and
(B A)
_

0
e
t

U(t)xdt = Bx, x X
1
. (1.5.13)
Now we prove that (B A) is invertible for C with Re > . Let
x ker(B A) and consider w(t) := e
t
x, then
Bw

(t) = Bw(t) = Aw(t), w(0) = x.

Since x D
1
and (DP) is well-posed on D
1
, we have w() = U()x and
|w(t)| Ke
t
|x| for t 0. Therefore, if 0 ,= x ker(B A), then
Re . If Re > , then the operator (B A) is invertible. Hence,
from (1.5.13) we obtain
R()x = (B A)
1
Bx =
_

0
e
t

U(t)xdt, x X
1
, Re > ,
and
R()x = (I G)
1
x, x X
1
, Re > . (1.5.14)
Let
0

B
(A) and dene the operator
Px := (
0
I G)R(
0
)x.
Since D
1
D(G), the domain of P coincides with X. This operator is
closed since R(
0
) is bounded and (
0
I G) is closed. Hence, P is a
bounded operator on X. From (1.5.14) we have P
2
= P, i.e. P is a
projector on X.
Let U(t) :=

U(t)P for t 0. Clearly, the family of bounded linear
operators U(t), t 0 is a degenerate C
0
-semigroup and for any x X
we have the equality
R()x =
_

0
e
t
U(t)xdt, Re > ,
which means that operators A and B are the generators of U(t), t 0.
2001 CRC Press LLC
2001 CRC Press LLC
(II) = (III). If A and B are the generators of a degenerate C
0
- semi-
group U(t), t 0, then it is not dicult to verify that A and B are the
generators of the degenerate 1-time integrated semigroup V (t) :=
_
t
0
U(t)dt,
t 0, which satises (1.5.2). Hence, by Theorem 1.5.1 the estimates (1.5.3)
hold for n = 0, and by Theorem 1.5.2
X = F =
_
x X

V ()x C
1
_
[0, ), X
_
_
= ker B X
1
.
(III) = (I). If X = ker B X
1
, then the equality R()X
1
= R()X
holds. Hence, by Theorem 1.5.3 the Cauchy problem (DP) is uniformly
well-posed on D
1
= R()X.
1.5.4 (n, )-well-posedness of a degenerate
Cauchy problem
Let A, B : X Y be linear operators, A be closed, and B be bounded.
Let the B-resolvent set of A be nonempty and consider
B
(A). We
introduce the sets D
i
= R
i
()X, i = 1, 2, . . ., which are obviously linear
manifolds. At the begining of this section we proved that D
1
does not
depend on
B
(A). Similarly, one can prove that D
i
do not depend on
for all i 1. We denote by [D
n
] sets D
n
, n N, equipped with the norm
|x|
R
n := inf
y:y=R
n
()x
|y|
for all x D
n
.
Denition 1.5.5 The Cauchy problem (DP) is called (n, )-well-posed on
D [D
n
] if for any x D there exists a unique solution u() satisfying
|u(t)| Ke
t
|x|
R
n
for some n N, K > 0 and R.
To obtain a sucient condition for the (n, )-well-posedness of (DP) we
need the following properties of degenerate integrated semigroups that can
proved in a way similar to Proposition 1.2.2 and Theorem 1.5.2.
Theorem 1.5.5 Let n 0 N. Let A and B be the generators of a
degenerate (n +1)-times integrated semigroup V (t), t 0 satisfying the
condition (1.5.2). Then
R()V (t) = V (t)R(), t 0, Re >
and
t
n+1
(n + 1)!
Bx = BV (t)x A
_
t
0
V (s)xds, t 0, x X
1
= D
1
.
2001 CRC Press LLC
2001 CRC Press LLC
The family of operators V

(t), t 0 is a degenerate exponentially bounded

n-times integrated semigroup on the subspace F = R()X
1
. For all x
R
k
()F, k = 0, 1, . . . , n, we have
BV
n+1
(t)x =
t
nk
(n k)!
Bx AV
k
(t)x, t 0,
For all x R
n+1
()F we have
B
d
dt
V
n+1
(t)x = AV
n+1
(t)x, t 0.
Proofs of this theorem and the next one can be obtained from Theorem
1.6.4.
Theorem 1.5.6 Let A, B : X Y be linear operators, A be closed, B be
bounded and estimates (1.5.3) be fullled. Then (DP) is (n, )-well-posed
on
D = ((B A)
1
B)
n+1
(B A)
1
B(X
1
) D
n+1
.
In the Section 1.6, assuming a decomposition of X, we prove the necessary
and sucient conditions for the (n, )-well-posedness and for the n-well-
posedness of the degenerate Cauchy problem (DP).
1.5.5 Examples
Non-densely dened operators that generate integrated semigroups provide
examples of degenerate semigroups.
Let A be the generator of an integrated semigroup V (t), t 0 in X.
Suppose that D(A) is complemented in X, i.e. X = D(A) Y for some
subspace Y . Let B = P
D(A)
be the projector from X to D(A). We have
ker B = Y , ranB = D(A) and B
2
= B. Then
(I A)x = (B A)x, x D(A),
and
(B A)
1
x = (I A)
1
x = R
A
()x, x D(A).
Hence, if MFPHY condition is fullled for R
A
(), then it is also fullled
for (B A)
1
B.
Example 1.5.1 Let X = C[0, ) and
A =
d
dx
with D(A) =
_
f X

X, f(0) = 0
_
.
Then Y =
_
f X

f const
_
, Bf = f(x) f(0) and
D
1
=
_
f X

Af ranB
_
=
_
f X

X, f(0) = 0, f

(0) = 0
_
.
2001 CRC Press LLC
2001 CRC Press LLC
The Cauchy problem
u(x, t)
t

u(0, t)
t
+
u(x, t)
x
= 0, x [0, ), t 0,
u(x, 0) = f(x),
is uniformly well-posed on D
1
. For f D
1
we have
u(x, t) = (U(t)f)(x) =
_
f(x t) , x t
0 , x t
.
We refer the reader to G. Da Prato and E. Sinestrari [50], where several
examples of non-densely dened operators can be found. Many examples of
degenerate Cauchy problems are discussed in A. Favini and A. Yagi [102].
1.6 The Cauchy problem for inclusions
The degenerate Cauchy problem
Bu

(t) = Au(t), t 0, u(0) = x, ker B ,= 0, (DP)

which we studied in the previous section, and the degenerate Cauchy prob-
lem
d
dt
Bv(t) = Av(t), t 0, Bv(0) = x, (DP1)
can be considered as particular cases of the Cauchy problem for an inclusion
u

(t) /u(t), t 0, u(0) = x, (IP)

with a multivalued linear operator /. If we set / = B
1
A for problem
(DP) or / = AB
1
and u := Bv for (DP1), then u() is a solution of the
Cauchy problem (IP). Conversly, if u() is a solution of (IP) with / = B
1
A
or / = AB
1
, then u() is the solution of (DP) or any v() from the set
Bv = u is the solution of (DP1), respectively.
In this section, we use the technique of degenerate semigroups with mul-
tivalued generators for studying the well-posedness of the Cauchy problem
(IP). As a consequence, we obtain a criterion for the well-posedness of (DP)
and a criterion for the existence of a solution for (DP1).
1.6.1 Multivalued linear operators
Let X be a complex Banach space. For subsets from X we dene addition
and scalar multiplication
F +G :=
_
f +g

f F, g G
_
,
C, F :=
_
f

f F
_
.
2001 CRC Press LLC
2001 CRC Press LLC
Denition 1.6.1 A map / : X 2
X
is called a multivalued linear oper-
ator on X if
D(/) =
_
u X

/u ,=
_
is a linear manifold in X and
/u +/v /(u +v) (1.6.1)
for any , C and u, v D(/).
In the particular case of = = 0 we have 0 /0.
By /(X) we denote the space of all multivalued linear operators on X.
Proposition 1.6.1 Let / /(X), then
(a) u, v D(/), /u +/v = /(u +v),
(b) u D(/), ,= 0, /u = /(u).
Proof Since / is a linear operator, it is sucient to prove the inclusions
/(u +v) /u +/v and /(u) /u. From (1.6.1) we have
/(u +v) /v (/(u +v) v) = /u,
therefore,
/(u +v) /u +/v.
Also
/(u) = (
1
/(u)) /u.
Corollary 1.6.1 For any u, v D(/) and any , with [[ + [[ ,= 0, we
have
/u +/v = /(u +v).
Proposition 1.6.2 Let / /(X), then /0 is a linear manifold in X
and /u = f +/0 for any u D(/) and f /u.
Proof By Proposition 1.6.1 we have /0 +/0 = /0 and /0 = /0, hence
/0 is a linear manifold. Let u D(/) and f /u, then
f +/0 /u +/0 = /u.
On the other hand, for any g /u we have g f /u /u = /0, hence
g = f + (g f) f +/0.
This proves the inverse inclusion /u f + /0, and hence the equality
/u = f + /0. In particular, / is a single-valued operator if and only if
/0 = 0.
2001 CRC Press LLC
2001 CRC Press LLC
Denition 1.6.2 An operator /
1
is called the inverse of an operator /
if
D(/
1
) := ran/ =
_
uD(A)
/u
and
/
1
f :=
_
u D(/)

f /u
_
.
It is easy to see that /
1
is a linear and (in general) multivalued operator
on X.
Denition 1.6.3 The set
(/) :=
_
C

(I /)
1
is a single-valued bounded linear operator on X
_
is called the resolvent set of the operator /.
Proposition 1.6.3 Let / /(X) and (/) ,= , then for any
(/), /0 = ker (I /)
1
.
Proof By the denition of kernel we have that f ker (I /)
1
if and
only if (I /)
1
f = 0, that is f (I /)0 = /0.
Proposition 1.6.4 Let / /(X) and suppose that the decomposition
X = /0 Y holds for some subspace Y X. Then

/u := /u Y is a
single-valued linear operator with D(

/) = D(/).
Proof First we show that D(

/) = D(/). From the denition of

/ we
have D(

/) D(/). Let u D(/), then there exists f X such that
u = f +/0 and f = f
1
+f
2
, where f
1
/0, f
2
Y . We have

/u = /u Y = (f
2
+/0) Y f
2
,= ,
that is u D(

/). Hence D(

/) D(/) and D(

/) = D(/).
Now let u, v D(

/) = D(/) and , C, then

/u +

/v
=
_
(/u) Y
_
+
_
(/v) Y
_
/(u +v) Y
=

/(u +v).
Since

/0 = /0 Y = 0 , we have that

/ is a single-valued linear
operator.
The operator

/ is called the single-valued branch of /.
2001 CRC Press LLC
2001 CRC Press LLC
Denition 1.6.4 We say that an operator / /(X) is closed if for any
sequences u
n
D(/) and f
n
/u
n
such that
lim
n
u
n
= u and lim
n
f
n
= f,
we have u D(/) and f /u.
It is clear that / is closed if and only if

/ is closed. Furthermore, for a
closed operator / we have
u
n
D(/), /( lim
n
u
n
) = lim
n
/u
n
:= lim
n

/u
n
+/0
if all limits exist, and
u C
_
[0, T], D(/)
_
,
/
_
t
0
u()d =
_
t
0
/u()d =:
_
t
0
f()d +/0, t [0, T),
if integrals exist.
1.6.2 Uniform well-posedness
Denition 1.6.5 The Cauchy problem
u

(t) /u(t), t 0, u(0) = x, (IP)

is called uniformly well-posed on E X, if
(a) for any x E, there exists a unique solution
u() C
1
_
[0, ), X
_
C
_
[0, ), D(/)
_
(b) for any T > 0, sup
0tT
|u(t)| |x|.
The following theorem connects the well-posedness of (IP) with the well-
posedness of (CP) with operator

/, the single-valued branch of /:
u

(t) =

/u(t), t 0, u(0) = x, (1.6.2)
and gives the MFPHY-type condition for the uniform well-posedness of
(IP). Note that in contrast to Theorem 1.5.1, here we do not assume that
(/) ,= .
Theorem 1.6.1 Let / be a multivalued linear operator on X. Let
X
1
:= D(/),

/u := /u X
1
, D(

/) :=
_
u X

/u ,=
_
.
Then the following statements are equivalent:
2001 CRC Press LLC
2001 CRC Press LLC
(I) the Cauchy problem (IP) is uniformly well-posed on D(/);
(II) the operator

/ is a single-valued branch of / and it is the generator
of a C
0
-semigroup U(t), t 0 on X
1
;
(III) the operator

/ is a single-valued branch of / and the MFPHY con-
dition holds for R

A
= (I

/)
1
:
K > 0, R :
_
_
_
_
d
k
d
k
(I

/)
1
_
_
_
_

K k!
( )
k+1
for all > and k = 0, 1, . . . .
In this case for any x D(/), u() = U()x is the unique solution of (IP).
Proof (I) =(II). Dene on D(/) the solution operators U(t) by U(t)x :=
u(t), t 0, where u() is the solution of (IP). From the well-posedness of
(IP) we have that for any t 0, U(t) is a bounded operator on D(/).
Hence it can be extended to X
1
= D(/).
In a way similar to the nondegenerate case (Theorem 1.1.1), one can
prove that the operators U(t) form a C
0
-semigroup on X
1
. Let G : X
1
X
1
be the generator of this semigroup. Then G is a closed, densely dened
linear operator with the property
U()x C
1
_
[0, ], X
_
x D(G).
Since for any x D(/) we have u() = U()x C
1
_
[0, ], X
_
, then
D(/) D(G). Since the operator / is closed, we have
x D(/),
U(t)x x =
_
t
0
U

()xd
_
t
0
/U()xd = /
_
t
0
U()xd.
Using the closedness of / we can extend the obtained inclusion to X
1
U(t)x x /
_
t
0
U()xd, x X
1
,
and
U(t)x x
t

1
t
/
_
t
0
U()xd, t > 0, x X
1
. (1.6.3)
Consider the domain of the generator
D(G) :=
_
x X
1

lim
t0
U(t)x x
t
_
,
2001 CRC Press LLC
2001 CRC Press LLC
then (1.6.3) implies D(G) D(/), i.e. D(G) = D(/) and for any x
D(/), Gx /x. Since Gx X
1
, we have
x D(/) = D(G), Gx

/x := /x X
1
, (1.6.4)
hence D(/) D(

/). By denition of

/, we have D(

/) D(/), and
therefore D(/) = D(

/).
Now we show that

/ is a single-valued operator. Let y

/0 := /0X
1
and z := (I G)
1
y with (G), Re > . Note that since G is the
generator of a C
0
-semigroup, then any C with Re > belongs to
(G). For these z D(G) = D(/) we have
(I G)z = y
and
z = y +Gz /0 +Gz /0 +/z = /z,
that is z /z. Hence (ze
t
)

= ze
t
/(ze
t
) and u(t) = ze
t
, t 0,
is a solution of (IP) with the initial value z for any with Re > . From
(1.6.4) we have
x D(G) = D(

/), Gx

/x,
and since

/ is a single-valued operator, we have G =

/.
(II) = (III) = (I). By assumption,

/ is the generator of a C
0
-
semigroup. By Theorem 1.1.1 it is equivalent to the MFPHY condition
for R

A
and to the uniform well-posedness on D(

/) of the Cauchy problem
(1.6.2). Hence (IP) is uniformly well-posed on D(/) = D(

/).
Corollary 1.6.2 If, in Theorem 1.6.1, we do not assume that the operator
/ is closed, then the result holds true for the single-valued branch of /,
the closure of /.
Let us show that under the condition (/) ,= the estimates for the
resolvent of / are fullled. In Denition 1.5.1 we dened degenerate semi-
groups and now we need the notion of a generator of a degenerate semi-
group.
Denition 1.6.6 An operator / is called the generator of a degenerate
(exponentially bounded) C
0
-semigroup U(t), t 0 if it satises the
relation
(I /)
1
x =
_

0
e
t
U(t)xdt (1.6.5)
for all C with Re > .
2001 CRC Press LLC
2001 CRC Press LLC
It follows from (1.6.5) that ker U = ker(I /)
1
. On the other hand, by
Proposition 1.6.3, we have ker(I /)
1
= /0. That is, the generator of
any degenerate semigroup is a multivalued operator, and conversely, if a
multivalued operator is the generator of a semigroup, then this semigroup
is degenerate.
Theorem 1.6.2 Let / be a multivalued linear operator in a Banach space
X with nonempty resolvent set. Then the following is equivalent:
(I) the Cauchy problem (IP) is uniformly well-posed on D(/);
(II) the MFPHY condition holds for R
A
() and X admits the
decomposition
X = /0 X
1
. (1.6.6)
Proof (I) = (II). In Theorem 1.6.1 it was proved that

/ is the single-
valued branch of / and the generator of a C
0
-semigroup U(t), t 0.
Consider the semigroup | equal to U on X
1
and zero on /0. It is a
degenerate exponentially bounded C
0
-semigroup with the generator /:
(I /)
1
x =
_

0
e
t
|(t)xdt, Re > ,
hence the MFPHY condition holds for R
A
().
Let x X, y := (I /)
1
x, (/) and z := x (I

/)y.
We have y D(/) = D(

/) and (I

/)y X
1
. Since for any f X
1
,
(I /)
1
f = (I

/)
1
f, we obtain
(I /)
1
z = (I /)
1
x (

I /)
1
(I

/)(I /)
1
x.
Hence z ker(I /)
1
= /0 and for x X we have the decomposition
x = z + (

I

/)y /0 +X
1
.
As

/ is a single-valued operator, we have

/0 = /0 X
1
= 0. Hence, we
obtain the decomposition of X into the direct sum X = /0 X
1
.
(II) = (I). Since we have the decomposition X = /0 X
1
, then

/ is
a single-valued closed linear operator with D(

/) = D(/) (see Proposition
1.6.4), and the MFPHY conditions for R
A
() and R

A
() are equivalent. By
Theorem 1.1.1, the Cauchy problem (1.6.2) with operator

/ is uniformly
well-posed on D(

/). Let us show that this problem is well-posed on D(

/)
if and only if (IP) is well-posed on D(/). If
u

(t) =

/u(t), t 0, u(0) = x D(

/),
2001 CRC Press LLC
2001 CRC Press LLC
then
u

(t) /u(t), t 0, u(0) = x D(/).

Conversely, if u

(t) /u(t), then u

(t) D(/) = X
1
. Hence, u

(t)

/u(t) = /u(t) X
1
, and therefore u

(t) =

/u(t). Thus, the Cauchy
problem (IP) is uniformly well-posed on D(/).
The following two theorems are corollaries of Theorem 1.6.2.
Theorem 1.6.3 Let X
1
= D
1
, where
D
1
=
_
x D(A)

Ax ranB
_
.
Consider the Cauchy problem (DP) with linear operators A and B such
that AB
1
is closed (for example, one of them is closed and the other is
bounded). Suppose

B
(A) :=
_
C

(B A)
1
B is bounded
_
,= .
Then the following conditions are equivalent:
(I) the Cauchy problem (DP) is uniformly well-posed on D
1
;
(I I) there exists a C
0
-semigroup U(t), t 0 on X
1
such that
(B A)
1
Bx =
_

0
e
t
U(t)xdt
for all x X
1
and C with Re > ;
(III) X = X
1
ker B and
K > 0, R:
_
_
_
_
d
k
d
k
(B A)
1
B
_
_
_
_

K k!
( )
k+1
for all > and k = 0, 1, . . . .
And, in this case, the solution of (DP) is u(t) = U(t)x, t 0, x D
1
.
If B is bounded and A is closed, then this result gives the same well-
posedness conditions as Theorem 1.5.4. In Section 1.5 we used a degenerate
C
0
-semigroup with generators A, B, and here we use a C
0
-semigroup with
a generator equal to a single-valued branch of operator AB
1
. In the case
of ker B = 0 we obtain Theorem 1.1.1.
2001 CRC Press LLC
2001 CRC Press LLC
Theorem 1.6.4 Let X
1
= T
1
, where
T
1
:=
_
x X

x ranB(B A)
1
_
.
Consider the Cauchy problem (DP1) with linear operators A and B such
that B
1
A is closed (for example both operators are bounded). Suppose
_
C

B(B A)
1
is bounded
_
,= .
Then the following conditions are equivalent:
(I) for any x T
1
there exist solutions v of the Cauchy problem (DP1)
such that Bv is uniquely dened and is stable with respect to x;
(II) there exists a C
0
-semigroup U(t), t 0 on X
1
such that
B(B A)
1
x =
_

0
e
t
U(t)xdt
for all x X
1
and C with Re > ;
(III) X = X
1
Aker B, and
K > 0, R:
_
_
_
_
d
k
d
k
B(B A)
1
_
_
_
_

K k!
( )
k+1
for all > and k = 0, 1, . . . .
And in this case the solution of (DP1) is Bv(t) = U(t)x, t 0, x T
1
.
1.6.3 (n, )-well-posedness
Denition 1.6.7 The Cauchy problem
u

(t) /u(t), t 0, u(0) = x, (IP)

is called (n, )-well-posed on E X if for any x E there exists a unique
solution
u() C
1
_
[0, ), X
_
C
_
[0, ), D(/)
_
such that
|u(t)| Ke
t
|x|
A
n,
for some K > 0, where
|x|
A
n :=
n

k=0
|/
k
x|.
2001 CRC Press LLC
2001 CRC Press LLC
Here
|/
k
x| = inf
fA
k
x
|f|
X
is the factor-norm of /
k
x in X//
k
0.
If the resolvent set (/) is not empty, then the norm |x|
A
n is equivalent
to
|x|
R
n := inf
y:R
n
A
()y=x
|y|, (/).
Now we investigate the connections between the (n, )-well-posedness of
(IP) and the existence of an exponentially bounded degenerate k(n)-times
integrated semigroup.
Denition 1.6.8 An operator / is called the generator of an exponentially
bounded degenerate n-times integrated semigroup V (t), t 0 if it satises
the relation
(I /)
1
x =
_

0

n
e
t
V (t)xdt (1.6.7)
for all C with Re > .
As in the case of degenerate C
0
-semigroups, it follows from (1.6.7) that
ker V = ker(I /)
1
= /0, and an exponentially bounded n-times inte-
grated semigroup has a multivalued generator if and only if the semigroup
is degenerate.
Theorem 1.6.5 Let n N, R, and let / be a closed linear multivalued
operator on X. Suppose that (/) ,= is in the half-plane Re > . If the
Cauchy problem (IP) is (n, )-well-posed on D(/
n+1
), then the MFPHY-
type condition
K > 0 :
_
_
_
_
d
k
d
k
(I /)
1

n+1
_
_
_
_

K k!
( )
k+1
(1.6.8)
holds for all > and k = 0, 1, . . . .
Proof Let x D(/
n+1
) =: D
n+1
and let u(t), t 0, be the unique
solution of (IP) corresponding to x. By the denition of the (n, )-well-
posedness for (IP), u(t) satises the estimate
|u(t)| Ke
t
|x|
R
n. (1.6.9)
Now we introduce the solution operators U(t) on D
n+1
by
U(t)x := u(t), t 0.
2001 CRC Press LLC
2001 CRC Press LLC
Due to (1.6.9) we can extend U(t) to the space [D
n+1
]
n
, the closure of D
n+1
in the norm | |
R
n. As U()x = u() is a solution of (IP), we have
U

(t)x /U(t)x = U(t)x (I /)U(t)x. (1.6.10)

Now we show that R
A
()U()x = U()R
A
()x, x D
n+1
, is a solution of
(IP) with the initial value R
A
()x. Let y = R
A
()x D
n+1
. Then from
(n, )-well-posedness of (IP) we have U

(t)R
A
()x /U(t)R
A
()x. So
U()R
A
()x is a solution of (IP) satisfying the estimate
|U(t)R
A
()x| Ke
t
|R
A
()x|
R
n K
1
e
t
|x|
R
n1.
Applying R
A
() to (1.6.10), we obtain
R
A
()U

(t)x = R
A
()/U(t)x = R
A
()U(t)x U(t)x. (1.6.11)
As U(t)x D(/), we have
U(t)x = R
A
()(I /)U(t)x (I /)R
A
()U(t)x.
Hence,
R
A
()/U(t)x = R
A
()U(t)x U(t)x /R
A
()U(t)x,
R
A
()U

(t)x = (R
A
()U(t)x)

/R
A
()U(t)x,
and R
A
()U(0)x = R
A
()x. That is, R
A
()U()x, x D
n+1
is a so-
lution of (IP). The equality R
A
()U(t)x = U(t)R
A
()x follows from the
uniqueness of the solution. Integrating (1.6.11) from 0 to t, we obtain
R
A
()U(t)x R
A
()x =
_
t
0
R
A
()U(s)xds
_
t
0
U(s)xds,
or
_
t
0
U(s)xds
= R
A
()U(t)x +R
A
()x +
_
t
0
R
A
()U(s)xds
= U(t)R
A
()x +R
A
()x +
_
t
0
U(s)R
A
()xds, x D
n+1
.
On D
n
we dene
U
1
(t)x := U(t)R
A
()x +R
A
()x +
_
t
0
U(s)R
A
()xds.
2001 CRC Press LLC
2001 CRC Press LLC
Applying R
A
() to both sides of this equality, we obtain
R
A
()U
1
(t)x
= R
A
()U(t)R
A
()x +R
A
()R
A
()x +R
A
()
_
t
0
U(s)R
A
()xds.
Commutativity of R
A
() and U(t) on D
n+1
implies
R
A
()U
1
(t)x = U
1
(t)R
A
()x, x D
n
.
Hence, U
1
(t) commutes with R
A
() on D
n
, satises the estimate
|U
1
(t)x| Ke
t
|x|
R
n1,
and can be extended to [D
n
]
n1
. In general,
U
k
(t)x := (1.6.12)
U
k1
(t)R
A
()x +
t
k1
(k 1)!
R
A
()x +
_
t
0
U
k1
(s)R
A
()xds.
The operators U
k
(t), t 0, k = 1, . . . , n , are dened on D
n+1k
, where
they commute with R
A
() and satisfy the estimate
|U
k
(t)x| Kt
k1
e
t
|x|
R
nk Ke

1
t
|x|
R
nk
for any
1
> . Therefore, we can extend U
k
(t) to [D
n+1k
]
nk
. In
particular, the operators U
n
(t) commute with R
A
() on D(/), satisfy the
estimate
|U
n
(t)x| Ke

1
t
|x|,
and can be extended to D(A). The operators U
n+1
(t) commute with R
A
()
on the whole space X and satisfy
|U
n+1
(t)x| Ke

1
t
|x|. (1.6.13)
Now denote V (t)x := U
n+1
(t)x and we show that V (t), t 0 is an
(n + 1)-times integrated semigroup with the generator /. From (1.6.12)
we have that for any x X, V (t)x is continuous in t 0, so (V2) holds.
We have (V3) from (1.6.13). To show that V satises (V1), by Proposition
1.2.1 it is sucient to show that
_

0

n+1
e
t
V (t)dt
satises the pseudoresolvent identity. Hence, it is sucient to prove the
equality
R
A
()x =
_

0

n+1
e
t
V (t)xdt (1.6.14)
2001 CRC Press LLC
2001 CRC Press LLC
for C with Re > . Multiplying (1.6.12) by
k
e
t
with k = n + 1
and integrating from 0 to we get
_

0

n+1
e
t
U
n+1
(t)xdt = R
A
()x =
_

0

n+1
e
t
V (t)xdt.
This equation holds for all from some open set where R
A
() exists by
assumption. Using the resolvent identity for the function
_

0

n+1
e
t
V (t)xdt,
the analytical extension of R
A
() to the half-plane Re > , we obtain
(1.6.14) for Re > . Thus, V (t), t 0 is an (n + 1)-times degenerate
integrated semigroup generated by /, and hence the estimate (1.6.8) holds.

Theorem 1.6.6 Let / be a multivalued linear operator on X. Suppose

that the condition
K > 0, R :
_
_
_
_
d
k
d
k
(I /)
1

n
_
_
_
_

K k!
( )
k+1
(1.6.15)
holds for all > and k = 0, 1, . . . . Then the Cauchy problem (IP) is
(n, )-well-posed on R
n+1
A
()D(/).
Proof If the estimate (1.6.15) holds for some K > 0 and R, then by
Theorem 1.2.1 / is the generator of an (n +1)-times integrated semigroup
V (t), t 0 with the property
|V (t +h) V (t)| Kh, t 0, h 0. (1.6.16)
Note that in this case the semigroup can be degenerate. By the denition
of the generator, for all y X we have
_

0

n+1
e
t
V (t)R
A
()ydt = R
A
()R
A
()y = R
A
()R
A
()y
=
_

0

n+1
e
t
R
A
()V (t)ydt.
The uniqueness of the Laplace transform implies
V (t)R
A
()y = R
A
()V (t)y, y X.
Let y = (I /)x, x D(/), then we have
V (t)x = R
A
()V (t)(I /)x,
2001 CRC Press LLC
2001 CRC Press LLC
and applying (I /):
(I /)V (t)x = (I /)R
A
()V (t)(I /)x.
From V (t)(I /)x (I /)R
A
()V (t)(I /)x we obtain
V (t)(I /)x (I /)V (t)x.
Hence for all x D(/), V (t)/x /V (t)x, and
_

0

n+2
e
t
t
n+1
(n + 1)!
xdt
= x = R
A
()x R
A
()/x
=
_

0

n+2
e
t
V (t)xdt
_

0

n+1
e
t
V (t)/xdt
=
_

0

n+2
e
t
V (t)xdt
_

0

n+2
e
t
_
t
0
V (s)/xdsdt.
From the uniqueness of the Laplace transform we have
t
n+1
(n + 1)!
x = V (t)x
_
t
0
V (s)/xds, x D(/),
hence
V

(t)x =
t
n
n!
x +V (t)/x,
(1.6.17)
V

(t)x
t
n
n!
x +/V (t)x, x D(/).
Using property (1.6.16) and the closedness of / we can extend the inclusion
in (1.6.17) to D(/). Now take x D(/
2
) (as /x D(/) ,= ), then from
(1.6.17) we have
/V

(t)x =
t
n
n!
/x +/V (t)/x, x D(/
2
),
V

(t)/x
t
n
n!
/x +V (t)/
2
x

t
n
n!
/x +/V (t)/x, x D(/
2
),
therefore V

(t)/x /V

(t)x for x D(/

2
), and
V

(t)x =
t
n1
(n 1)!
x +V

(t)/x,
(1.6.18)
V

(t)x
t
n1
(n 1)!
x +/V

(t)x, x D(/
2
).
2001 CRC Press LLC
2001 CRC Press LLC
Now we show that the equation from (1.6.18) holds for all x R
A
()D(/).
Let y D(/). By the closedness of /, there exists a sequence y
n
D(/)
such that y
n
y. Take x
n
= R
A
()y
n
, then x
n
R
A
()D(/) and
x
n
R
A
()y = x R
A
()D(/). Moreover
/x
n
= /R
A
()y
n
= (I /)R
A
()y
n
+R
A
()y
n
,
and the set (I /)R
A
()y
n
contains the sequence y
n
convergent to
y. Therefore /x
n
D(/) ,= , and y
n
+ R
A
()y
n
D(/) converges
to y +R
A
()y D(/). Hence
/x
n
/x and V

(t)/x
n
V

(t)/x,
as V

(t) is bounded (property (1.6.16)). Since

t
n1
(n1)!
x
n

t
n1
(n1)!
x and
V

(t) is closed, we have

V

(t)x =
t
n1
(n 1)!
x +V

(t)/x, x R
A
()D(/). (1.6.19)
From the inclusion in (1.6.18), the closedness of / and the boundedness of
V

(t), we obtain
V

(t)x
t
n1
(n 1)
x +/V

(t)x, x R
A
()D(/).
For x R
2
A
()D(/) there exists V

(t)/x, therefore, after dierentiating

(1.6.18), we have
V
(3)
(t)x =
t
n1
(n 2)!
x +V

(t)/x

t
n2
(n 2)!
x +/V

(t)x, x R
2
A
()D(/).
We continue the process and obtain
V
(n+2)
(t)x /V
(n+1)
(t)x, x R
n+1
A
()D(/),
V
(n+1)
(0)x = x.
Therefore for any x R
n+1
A
()D(/), V
(n+1)
()x is a solution of the Cauchy
problem (IP).
Now we show that the solution is unique. Let y() be another exponen-
tially bounded solution of (IP). Then we have
_

0
e
t
y(t)dt =
1

_
x +
_

0
e
t
y

(t)dt
_

_
x +/
_

0
e
t
y(t)dt
_
,
2001 CRC Press LLC
2001 CRC Press LLC
hence,
x (I /)
_

0
e
t
y(t)dt.
Applying R
A
() to this inclusion, we obtain
R
A
()x =
_

0
e
t
y(t)dt,
or
_

0

n+2
e
t
V (t)xdt =
_

0
e
t
V
(n+2)
(t)dt =
_

0
e
t
y(t)dt.
Therefore V
(n+2)
(t)x = y(t). We have shown that if there exists a k-
times integrated semigroup then any solution of (IP) coincides with the
k-th derivative of the integrated semigroup.
As in Theorem 1.2.4, the unique solution u() of (IP) satises the esti-
mate in Denition 1.6.7, thus the Cauchy problem (IP) is (n, )-well-posed
on R
n+1
A
()D(/).
Now we compare the obtained results with the results of Theorem 1.6.2
on the uniform well-posedness of the Cauchy problem (IP). Decomposition
(1.6.6) generalizes in the degenerate case the property of the generator of
a nondegenerate C
0
-semigroup to be densely dened. Only in this case we
have the projector
P = U(0) : X X
1
= D(/),
such that X
1
= PX, /0 = ker P. Due to the decomposition (1.6.6) the
MFPHY condition for (I

/)
1
on X
1
can be written as the condition
for (I /)
1
on X. In this subsection, without any decomposition of
X in Theorem 1.6.5, we obtained the MFPHY-type estimates for
(IA)
1

n
only on X
1
, and on X we obtain the estimates for
(IA)
1

n+1
. And note that
the estimates (1.6.15) guarantee (n, )-well-posedness of (IP) only on some
subset of D(/), not on the whole of it.
Recall the property discussed in Proposition 1.5.1: if an operator /
satises the MFPHY estimates, then
x D(/), lim

R
A
()x x.
The following proposition justies the generalization of the decomposition
(1.6.6), which we will use later in this section.
Proposition 1.6.5 Let / satisfy (1.6.8), then
x D
n+1
, lim

R
A
()x x.
2001 CRC Press LLC
2001 CRC Press LLC
Proof Let x D
n+1
, then there exists y X such that x = (
0
I
/)
(n+1)
y for some
0
(/) and
|(I /)
1
x x|
= |(I /)
1
[(
0
I /)
1
]
n+1
y [(
0
I /)
1
]
n+1
y|
= |(
0
)
1
(I /)
1
[(
0
I /)
1
]
n
y

0
(
0
)
1
[(
0
I /)
1
]
n+1
y|
=
= |(
0
)
(n+1)
(I /)
1
y

0
(
0
)
1
[(
0
I /)
1
]
n+1
y|
0 as .
It follows that ker R
A
() D
n+1
= 0.
Thus we have, that in general, the estimates (1.6.8) do not imply
/0 X
1
= 0 or /
k+1
0 X
k+1
= 0
for some k = 1, . . . , n, and consequently, they do not imply any decompo-
sition of X.
Now we assume the decomposition
X = X
n+1
ker R
n+1
A
= X
n+1
/
n+1
0, (1.6.20)
the generalization of decomposition (1.6.6) for the case of a degenerate n-
times integrated semigroup. Having the decomposition (1.6.20), we can
prove the necessary and sucient condition for the (n, )-well-posedness of
the Cauchy problem (IP).
Theorem 1.6.7 Let / be a multivalued linear operator on X with (/) ,=
. Suppose that the decomposition (1.6.20) holds for some n N and C
with Re > , R. Then the Cauchy problem (IP) is (n, )-well-posed
on D
n+1
if and only if the condition (1.6.15) is satised.
Proof Suppose the Cauchy problem (IP) is (n, )-well-posed on D
n+1
. To
show the estimates (1.6.15) we dene U
0
(t), t 0, by
U
0
(t)x :=
_
u(t), x [D
n+1
]
n
0, x /0
.
Since U
0
(t) are dened on [D
n+1
]
n
/0, we can extend R
A
()U
0
(t) to
[D
n
]
n1
and /
2
0, respectively, and therefore to the closure of D
n
in the
norm | |
R
n1. Hence, U
k
(t)x, k = 1, . . . n are dened for x D
n+1k
,
they are exponentially bounded in the norm | |
R
nk, and we can extend
them to [D
n+1k
]
nk
and /
k+1
0, respectively. In particular, the operators
2001 CRC Press LLC
2001 CRC Press LLC
U
n
(t) are dened and bounded on D(/) and on /
n+1
0, and hence on
X = X
n+1
/
n+1
0. As in the proof in Theorem 1.6.5, we have that
V (t) := U
n
(t), t 0 is an exponentially bounded n-times integrated
semigroup degenerate on /0. Its generator / is dened by
R
A
()x =
_

0

n
e
t
V (t)xdt.
Hence, the condition (1.6.15) holds.
Taking into account the decomposition (1.6.20), we now clarify the
structure of the constructed degenerate semigroup V on X by considering it
on X
n+1
and on /
n+1
0. By construction we have that all operator-functions
U
k
(), k 1, are equal to the primitives of U
0
() of the corresponding order,
which are extended from D
n+1
to X
n+1
= D
n+1
. They are equal to zero
on /0 = ker R
A
(), hence they are independent of on /0. Now we show
that U
k
(t)x are independent of on the sets /
k+1
0.
U
1
(t)x = R
A
()x, x /
2
0 = ker R
2
A
(),
U
2
(t)x = [t 1]R
2
A
()x +tR
A
()x, x /
3
0,
U
3
(t)x =
_

2
t
2
2
2t + 1
_
R
3
A
()x +t
_

t
2
1
_
R
2
A
()x
+
t
2
2
R
A
()x, x /
4
0,
. . .
U
n
(t)x =
n

i=1
a
i
(t, )R
i
A
()x, (1.6.21)
a
i
=
i1

k=o
(1)
k+i1
C
k
i1

k
t
k
k!
, x /
n+1
0.
The equality
d
d
U
1
(t)x = R

A
()x = R
2
A
()x = 0, x /
2
0,
implies that U
1
(t) is independent of . By induction one can prove
d
d
U
k
(t)x = 0, x /
k+1
0, k = 2, 3, . . . .
Suppose that the estimates (1.6.15) hold. Then by Theorem 1.6.6 the
Cauchy problem (IP) is (n, )-well-posed on R
n+1
A
()D(/) D
n+1
. By
the decomposition (1.6.20)
D
n+1
= R
n+1
A
()X = R
n+1
A
()X
n+1
.
2001 CRC Press LLC
2001 CRC Press LLC
On the other hand, we have
R
n+1
A
()X
n+1
= R
n+1
A
()R
n+1
A
()X R
n+1
A
()D(/),
and hence, D
n+1
R
n+1
A
()D(/). So R
n+1
A
()D(/) = D
n+1
, therefore,
the Cauchy problem (IP) is (n, )-well-posed on D
n+1
.
From this theorem we obtain conditions for the (n, )-well-posedness of
the degenerate Cauchy problems (DP) and (DP1).
Corollary 1.6.2 Consider the Cauchy problem (DP) with linear operators
A and B such that AB
1
is closed. Let

B
(A) :=
_
C

(B A)
1
B is bounded
_
,= ,
and suppose that the decomposition (1.6.20) holds. Then the Cauchy prob-
lem (DP) is (n, )-well-posed on D
n+1
if and only if the condition
K > 0, R :
_
_
_
_
d
k
d
k
R
A
()

n
_
_
_
_

K k!
( )
k+1
(1.6.22)
holds for all > and k = 0, 1, . . . .
Corollary 1.6.3 Consider the Cauchy problem (DP1) with linear operators
A and B such that B
1
A is closed. Let the set
_
C

B(B A)
1
is bounded
_
,= ,
and suppose that the decomposition (1.6.20) holds. Then for any x D
n+1
the Cauchy problem (DP1) has solutions v such that Bv is uniquely dened
and |Bv(t)| Ke
t
|x|
R
n if and only if the condition (1.6.22) holds.
1.7 Second order equations
Let X, Y be Banach spaces and A, B, Q be linear operators from X to Y .
We suppose that Q is bounded and ker Q ,= 0. Consider the abstract
Cauchy problems
u

(t) = Au

(t) +Bu(t), t 0, u(0) = x, u

(0) = y, (1.7.1)
and
Qu

(t) = Au

(t) +Bu(t), t 0, u(0) = x, u

(0) = y, (1.7.2)
In this section we use semigroup methods for studying the well-posedness
of these problems. First, we construct /, ^-functions, the solution op-
erators for problem (1.7.1), which generalize semigroups and cosine and
2001 CRC Press LLC
2001 CRC Press LLC
sine operator-functions. Second, we study (1.7.1) and (1.7.2) by reducing
them to a rst order Cauchy problem in a product space. Using the results
of Sections 1.2 and 1.6 on integrated semigroups we obtain necessary and
sucient conditions for well-posedness of (1.7.1) and (1.7.2) in terms of
resolvent operators
R
r
(
2
) := (
2
I AB)
1
and R
d
(
2
) := (Q
2
AB)
1
,
respectively.
The relation between operators A and B plays a very important role
in studying the well-posedness of second order problems. We show that
for the equivalence of the well-posedness of (1.7.1) and the existence of an
integrated semigroup for the Cauchy problem
v

(t) = v(t), t 0, v(0) = v

0
, (1.7.3)
where
=
_
0 I
B A
_
, v(t) =
_
u(t)
u

(t)
_
, v
0
=
_
x
y
_
,
operators A, B have to be biclosed. With the help of the theory of integrated
semigroups we prove that the MFPHY-type condition:
K > 0, 0 :
(1.7.4)
_
_
_
_
d
k
d
k
R
r
(
2
)
_
_
_
_
,
_
_
_
_
d
k
d
k
R
r
(
2
)
_
I A
_
_
_
_
_

K k!
(Re )
k+1
,
for all C with Re > and k = 0, 1, . . . ,
is necessary and sucient for well-posedness of (1.7.1) with biclosed op-
erators A, B. Using the /, ^-functions theory, we obtain the following
MFPHY-type well-posedness condition:
K > 0, 0 :
(1.7.5)
_
_
_
_
d
k
d
k
R
r
(
2
)
_
_
_
_
,
_
_
_
_
d
k
d
k
_
I A
_
R
r
(
2
)
_
_
_
_

K k!
(Re )
k+1
,
for all C with Re > and k = 0, 1, . . . ,
for problem (1.7.1) with commuting operators A, B. In Subsection 1.7.3,
using the theory of degenerate integrated semigroups, we prove MFPHY-
type necessary and sucient conditions for well-posedness of the degenerate
Cauchy problem (1.7.2).
2001 CRC Press LLC
2001 CRC Press LLC
1.7.1 /, ^-functions method
Consider the Cauchy problem
u

(t) = Au

(t) +Bu(t), t 0, u(0) = x, u

(0) = y, (1.7.1)
where A and B are closed linear operators in a Banach space X.
Denition 1.7.1 The Cauchy problem (1.7.1) is called -well-posed on
E
1
, E
2
if for any x E
1
, y E
2
there exists a unique exponentially bounded
solution:
K > 0, 0 : |u(t)| K e
t
_
|x| +|y|
_
, t 0.
Denition 1.7.2 Let A, B : X X be closed linear operators. A one-
parameter family of bounded commuting operators /(t), ^(t), t 0 is
called a family of /, ^-functions generated by operators A, B if
(M1) /(t +h) = /(t)/(h) +B^(t)^(h),
^(t +h) = /(t)^(h) +/(h)^(t) +A^(t)^(h), t, h 0;
(M2) ^(0) = 0, /(0) = I, ^

(0) = I, /

(0) = 0;
(M3) /(t) and ^(t) are strongly continuous in t 0;
(M4) K > 0, 0 : |/(t)| Ke
t
and |^(t)| Ke
t
, t 0.
The operators A, B are called the generators of the family of /, ^-
functions.
If /(t) and ^(t) commute with A and B on D(A) and D(B) respectively,
then it follows from (M1) that operators

(0)x := lim
h0
/(2h) 2/(h) +I
h
2
x,

(0)x := lim
h0
^(2h) 2^(h)
h
2
x,
dened for those x X where limits exist, are the generators of the family
of /, ^-functions:

(0) = B,

^

(0) = A. (1.7.6)
2001 CRC Press LLC
2001 CRC Press LLC
Denition 1.7.3 Operators A, B are called -closed if the operator A+B
is closed for any C with Re > . Operators A, B are called biclosed,
if for any sequences x
n
D(A), y
n
D(B) such that
x
n
x and y
n
y ,
we have
Ax
n
+By
n
z = x D(A), y D(B)
and Ax +By = z.
It is not dicult to see that if A, B are biclosed, then A, B are -closed,
and both A and B are closed. The following two examples illustrate that
if A and B are closed operators, they may be not -closed, and if A, B are
-closed, they may be not biclosed.
1. Consider closed operators A = d
2
/ds
2
+ d/ds, B = cd
2
/ds
2
on
C[a, b]. Let x
n
be such that
D(A) = D(B) x
n
(s) x(s), x

n
(s) y(s),
and x

n
(s) is not convergent in C[a, b]. Then (cA + B)x
n
cy, but x ,
D(A) D(B) = D(cA+B). Therefore, operators A, B are not -closed for
< c.
2. Consider operators A = d/ds, B = d
2
/ds
2
on C[a, b]. They are
-closed for any . Let sequences x
n
(s) D(A) and y
n
(s) =
_
s
0
x
n
(t)dt
be such that x
n
x but x

n
is not convergent in C[a, b]. Then A, B are
not biclosed.
All pairs of operators where one is closed and the other is bounded, and
pairs of closed operators with ranges in orthogonal subspaces, are examples
of biclosed operators.
Denote d
1
= D(AB), d
2
= D(A
2
)D(B). Using the /, ^-functions we
obtain the following necessary and sucient conditions for well-posedness
of (1.7.1).
Theorem 1.7.1 Let A, B be commuting closed linear operators with do-
mains such that d
1
= d
2
= X. Then the following statements are equiva-
lent:
(I) the Cauchy problem (1.7.1) is -well-posed on d
1
, d
2
;
(II) the operators A, B are the generators of a family of /, ^-functions,
and

^

(0) = A,

/

(0) = B. In this case, the solution of (1.7.1) has

the form
u() = /()x +^()y, x d
1
, y d
2
;
2001 CRC Press LLC
2001 CRC Press LLC
(III) for Re > there exists R
r
(
2
) and
R
r
(
2
)x =
_

0
e
t
^(t)xdt, (1.7.7)
(I A)R
r
(
2
)x =
_

0
e
t
/(t)xdt (1.7.8)
for all x X;
(IV) the condition (1.7.5) if fullled for R
r
(
2
):
K > 0, 0 :
_
_
_
_
d
k
d
k
R
r
(
2
)
_
_
_
_
,
_
_
_
_
d
k
d
k
_
I A
_
R
r
(
2
)
_
_
_
_

K k!
(Re )
k+1
for all C with Re > and k = 0, 1, . . . .
Proof (I) = (II). Bounded operators /(t) and ^(t) are dened as
extensions of the solution operators for initial values (x, 0) and (0, y), re-
spectively:
/

(t)x = A/

(t)x +B/(t)x, t 0, /(0)x = x, /

(0)x = 0,
^

(t)y = A^

(t)y +B^(t)y, t 0, ^(0)y = 0, ^

(0)y = y.
Commutativity of A and B and the uniqueness of a solution imply com-
mutativity of /(t) and ^(h) for any t, h 0, commutativity of operator-
functions /, ^ with operators A, B, property (M1), and equality (1.7.6).
Property (M2) follows from the initial conditions. Continuity and expo-
nential boundedness of a solution imply (M3) and (M4).
From (1.7.6) and the commutativity of /, ^-functions we obtain the
following relations for derivatives
u D(B), /

(t)u = ^(t)Bu = B^(t)u,

u D(A), ^

(t)u = /(t)u +^(t)Au

= /(t)u +A^(t)u,
u D(AB), /

(t)u = B^(t)u +AB^(t)u

= A/

(t)u +B/(t)u,
u D(A
2
) D(B), ^

(t)u = A^

(t)u +B^(t)u
for all t 0. These relations imply that
u() = /()x +^()y, x d
1
, y d
2
is a solution of the Cauchy problem (1.7.1). Since /, ^-functions are
exponentially bounded, the solution is stable.
2001 CRC Press LLC
2001 CRC Press LLC
(II) = (III). As in the proof of the closedness of a generator of a C
0
-
semigroup in Proposition 1.1.1, one can prove the following property of
generators of a family of /, ^-functions:
x
n
D(A) D(B), (1.7.9)
x
n
x D(B) & Px
n
y D(P) = x D(P) & Px = y,
where P := A+B =

^

(0) +

/

(0), Re > .
Note that property (1.7.9) is weaker than the -closedness and hence,
the biclosedness of A, B, but it allows us to conclude that
(
2
I AB)
_

0
e
t
^(t)xdt =
_

0
e
t
(
2
I AB)^(t)xdt
for all x D(A
2
) D(B). Integrating by parts and using the formulae for
the derivatives of /, ^-functions, we obtain
(
2
I AB)
_

0
e
t
^(t)xdt = x
for all x X = D(A
2
) D(B). Applying
_

0
e
t
^(t)dt to (
2
IAB)x
for x D(A
2
) D(B), we obtain
_

0
e
t
^(t)(
2
I AB)xdt = x
for all x D(A) D(B). Thus (1.7.7) is proved. The equality (1.7.8) can
be proved by similar arguments.
(III) = (IV) follows from (1.7.7), (1.7.8) and exponential estimates
(M4). For proof of (IV) = (I) see our book [128], where the detailed
proof of this theorem is given.
1.7.2 Integrated semigroups method
We consider the Cauchy problem (1.7.3):
v

(t) = v(t), t 0, v(0) = v

0
,
on the Banach space A := X X, equipped with the norm
|(x, y)|
X
= max
_
|x|
X
, |y|
X
_
.
Here
=
_
0 I
B A
_
, v(t) =
_
u(t)
u

(t)
_
, v
0
=
_
x
y
_
,
2001 CRC Press LLC
2001 CRC Press LLC
Suppose that is the generator of an exponentially bounded n-times
integrated semigroup V on A. We recall (see Proposition 1.2.1) that the
operator dened by
x = x
__

0

n
e
t
V (t)dt
_
1
x
is called the generator of V .
Let () be the resolvent set of and (A, B) be the set of all C
such that the operator R
r
(
2
) : X X is bounded.
Theorem 1.7.2 Let A, B : X X be biclosed linear operators such
that d
1
= d
2
= X and (A, B) ,= . Then the following statements are
equivalent:
(I) the operator =
_
0 I
B A
_
is the generator of a 1-time integrated
semigroup on A;
(II) the condition (1.7.4) is fullled:
K > 0, 0 :
_
_
_
_
d
k
d
k
R
r
(
2
)
_
_
_
_
,
_
_
_
_
d
k
d
k
R
r
(
2
)
_
I A
_
_
_
_
_

K k!
(Re )
k+1
for all C with Re > and k = 0, 1, . . . ;
(III) the Cauchy problem (1.7.1) is -well-posed on d
1
, d
2
.
Proof (I) = (II). The operator =
_
0 I
B A
_
is closed if and only if
A, B are biclosed. So is closed and densely dened: D() = A. Since
is the generator of a 1-time integrated semigroup, by Theorem 1.2.3, the
MFPHY-type condition holds for the resolvent of :
K > 0, 0 :
_
_
_
_
_
R

()

_
(k)
_
_
_
_

K k!
( )
k+1
(1.7.10)
for all > and k = 0, 1, . . . , where
R

() = (I )
1
=
_
R
r
(
2
)(I A) R
r
(
2
)
R
r
(
2
)B R
r
(
2
)
_
. (1.7.11)
Note that (1.7.11) implies that estimates (1.7.10) are equivalent to estimates
(1.7.4) for R
r
(
2
).
2001 CRC Press LLC
2001 CRC Press LLC
(II) =(III). We have that the condition (1.7.4) is equivalent to (1.7.10)
and the condition d
1
= d
2
= X is equivalent to the condition D() = A.
Therefore the operator is densely dened and (1.7.10) holds. Thus is
the generator of a 1-time integrated semigroup V , and the Cauchy problem
(1.7.3) is (1, )-well-posed. That is, for all
v
0
D(
2
) = D(AB)
_
D(A
2
) D(B)

there exists the unique solution v() = V

()v
0
with
|v(t)| = |(u(t), u

(t))| Ke
t
|v
0
|

.
Then w() = R

()v() is a solution of (1.7.3) with the initial value w(0) =

R

()v
0
and with the stability property
|w(t)| Ke
t
|R

()v
0
|

Ke
t
|v
0
|.
Applying (I )
1
to the equation (1.7.3) and integrating it, we obtain
the equality
_
t
0
v(s)ds = w(t) +w(0) +
_
t
0
w(s)ds. (1.7.12)
From (1.7.12) we have
_
_
_
_
_
t
0
v(s)ds
_
_
_
_
K
_
2 +[[t
_
e
t
|v
0
|,
hence
|u(t)| Ke

1
t
_
|x| +|y|
_
for some
1
> . Thus, for all x d
1
, y d
2
the Cauchy problem (1.7.1)
has the unique solution u() stable in X. It is equal to the rst coordinate
of v(), the unique solution of (1.7.3).
(III) = (I). If (A, B) ,= then () ,= and the -well-posedness
of the Cauchy problem (1.7.1) on d
1
, d
2
is equivalent to the (1, )-well-
posedness of (1.7.3). In particular,
_
t
0
v(s)ds is stable in A if and only if
u(t) is stable in X. Since (1.7.3) is (1, )-well-posed then is the generator
of a 1-time integrated semigroup (see Theorem 1.2.4).
Comparing the /, ^-functions method and the integrated semigroups
method, we note that by Theorem 1.7.1, if A, B are commuting operators
satisfying the density property and (1.7.5), then (1.7.1) is -well-posed
on d
1
, d
2
and A, B are biclosed. By Theorem 1.7.2, if A, B are biclosed
operators satisfying the density property and (1.7.4), then (1.7.1) is well-
posed on d
1
, d
2
. The condition (1.7.5) is the same as (1.7.4) if and only if
A and B commute.
2001 CRC Press LLC
2001 CRC Press LLC
1.7.3 The degenerate Cauchy problem
Let X, Y be Banach spaces. We consider the Cauchy problem
Qu

(t) = Au

(t) +Bu(t), t 0, u(0) = x, u

(0) = y, (1.7.2)
where A, B : X Y are biclosed linear operators and operator Q : X Y
is linear and bounded.
Let A = X X and = Y Y . Consider the closed linear operator
=
_
0 I
B A
_
: A
and the bounded linear operator
=
_
I 0
0 Q
_
: A .
As in the regular case it is not dicult to prove that the existence of a
unique solution
u() C
_
[0, ), D(B)
_
C
1
_
[0, ), D(A)
_
C
2
_
[0, ), X
_
of (1.7.2) is equivalent to the existence of a unique solution
v() =
_
u()
u

()
_
C
_
[0, ), D()
_
C
1
_
[0, ), A
_
of the degenerate Cauchy problem
v

(t) = v(t), t 0, v(0) =

_
x
y
_
. (1.7.13)
Using Theorem 1.6.5 and Corollary 1.6.4 about the connection between
the well-posedness of (1.7.13) and existence of a degenerate n-times in-
tegrated exponentially bounded semigroup V with the pair of generators
, :
1
d
() := ()
1
=
_

0

n
e
t
V (t)dt, A, Re > ,
and the MFPHY-type condition for 1
d
(), we obtain the following theorem
on the well-posedness of (1.7.2).
Theorem 1.7.3 Suppose operators A, B are biclosed and the operator
( )
1
is bounded for some . Assume that A admits the
decomposition:
A = T
n+1
ker 1
n+1
d
(1.7.14)
where T
n+1
:= 1
n+1
d
()A. Then the following statements are equivalent:
2001 CRC Press LLC
2001 CRC Press LLC
(I) the Cauchy problem (1.7.13) is (n, )-well-posed on T
n+1
;
(II) the MFPHY-type condition holds for 1
d
() = ()
1
:
K > 0, R :
_
_
_
_
_
1
d
()

n
_
(k)
_
_
_
_

K k!
( )
k+1
(1.7.15)
for all > and k = 0, 1, . . . ;
(III) the MFPHY-type condition holds for R
d
(
2
) := (
2
QAB)
1
:
K > 0, R :
_
_
_
_
_
_
R
d
(
2
)

n1
_
(k)
_
_
_
_
_

K k!
(Re )
k+1
,
_
_
_
_
_
_
R
d
(
2
)(QA)

n1
_
(k)
_
_
_
_
_

K k!
(Re )
k+1
or all C with Re > and k = 0, 1, . . . ;
(IV) the Cauchy problem (1.7.2) is (n 1, )-well-posed on the sets of
initial values E
1
and E
2
such that
_
x
y
_
T
n+1
.
Proof (I) (II) follows from Corollary 1.6.4.
(II) (III) follows from the equality
1
d
() = ()
1
=
_
R
d
(
2
)(QA) R
d
(
2
)
R
d
(
2
)B R
d
(
2
)
__
I 0
0 Q
_
.
(I) (IV). Unique solutions of (1.7.13) and (1.7.2), stable with respect to
corresponding initial data, are connected by the relation v(t) =
_
u(t)
u

(t)
_
.
Corollary 1.7.1 Suppose operators A, B are biclosed and the opera-
tor ( )
1
is bounded for some . Assume that A admits the
decomposition:
A = T
2
ker 1
2
d
.
Then the Cauchy problem (1.7.2) is well-posed on the sets of initial values
E
1
and E
2
such that
_
x
y
_
T
2
if and only if the MFPHY-type condition
holds for R
d
(
2
) := (
2
QAB)
1
:
2001 CRC Press LLC
2001 CRC Press LLC
K > 0, R :
_
_
_
_
_
R
d
(
2
)
_
(k)
_
_
_
_

K k!
(Re )
k+1
,
_
_
_
_
_
_
R
d
(
2
)
_
QA
_
_
(k)
_
_
_
_
_

K k!
(Re )
k+1
or all C with Re > and k = 0, 1, . . . .
2001 CRC Press LLC
2001 CRC Press LLC
Chapter 2
Abstract Distribution
Metho ds
2.1 The Cauchy problem
In the previous chapter we showed that under dierent conditions on the
resolvent of an operator A : D(A) X X, the Cauchy problem
u

(t) = Au(t), t [0, T), T , u(0) = x, (CP)

has a unique solution for x from dierent correctness classes stable with
respect to x in dierent norms. In this section, we consider the Cauchy
problem (CP) for any x X in the space of distributions. Our main aim
is to obtain necessary and sucient conditions for well-posedness in the
space of distributions in terms of the resolvent of A. We show that they
are the same as in the case when A is the generator of an integrated semi-
group. That is, only the Cauchy problem with a generator of an integrated
semigroup can be well-posed in the space of distributions.
In the rst subsection, we consider abstract distributions and distribu-
tion semigroups. In the second, we prove necessary and sucient conditions
for the well-posedness of (CP) in the space of distributions. As in the well-
posedness theorems in Chapter 1, it is done via intermediate results about
the existence of some semigroups. Here, they coincide with solution oper-
ator distributions and distribution semigroups. In the third subsection, we
consider the particular case of exponential distributions.
2.1.1 Abstract (vector-valued) distributions
Let X be a Banach space and T be the Schwartz space of innitely dier-
entiable functions : R R with compact supports and with the following
2001 CRC Press LLC
2001 CRC Press LLC
convergence:
n
in T if there exists a compact set K R such that
supp
n
K and
sup
tK
[
(i)
n
(t)
(i)
(t)[
n
0, i = 0, 1, . . . .
Let T
+
, T

, T
0
, c be subspaces of functions from T with supports bounded
from the left, from the right, with support in [0, ), and with any support,
respectively. Let o be the countably normed space of rapidly decreasing
functions with the system of norms
||
j,k
:= sup
0ik
sup
t
(1 +[t[)
j
[
(i)
(t)[. (2.1.1)
Denition 2.1.1 The space of distributions on X is the space of linear
continuous operators from T to X:
T

(X) := L(T, X), T

(X) := L(T

, X); ,
T

0
(X) is the subspace of distributions vanishing on (, 0), o

(X) :=
L(o, X) is the space of tempered distributions, c

(X) := L(c, X) is the

space of distributions with compact supports, c

0
(X) is the subspace of dis-
tributions from c

(X) vanishing on (, 0).

We denote by U() or U, ) or U(

t ), (

t )) the value of a distribution U

for a function from a space of test functions. For a locally integrable
function f : R X we write
f() = f, ) = f(

t ), (

t )) :=
_
R
f(t)(t)dt, f T

(R).
Denition 2.1.2 A sequence U
n
is called convergent to U in T

(X) if
|(U
n
U)()|
n
0 uniformly with respect to from a bounded set
/ T. A set / T is said to be bounded if for any
n
/ and

n
0 (
n
R), we have
n

n
0 in T.
The following proposition for abstract distributions is an analogue of the
well-known fact that every linear continuous operator on a normed space
is bounded.
Proposition 2.1.1 Let U T

(X) and K be a compact set from R. Then

there exist p 0, C > 0 such that for any T with supp K
|U()| C||
p
, where ||
p
:= sup
0jp
sup
tK
[
(j)
(t)[. (2.1.2)
The number p in (2.1.2) is called the local order of a distribution U.
Let T
p
(K) be the Banach space of p-times continuously dierentiable
functions with supports in a compact set K R and with the norm (2.1.2),
then any function from T
p
(K) can be approximated by an innitely dier-
entiable function.
2001 CRC Press LLC
2001 CRC Press LLC
Proposition 2.1.2 For any T
p
(K) and C > 0 there exists T
such that
supp K

=
_
t R

dist(t, K)
_
and | |
p
.
Thus, every distribution from T

(X) can be locally extended to a smooth

function. The following structure theorems are based on this fact.
Theorem 2.1.1 Let U T

(X) and be an open bounded set from R.

Then there exists a continuous function f : R X and m > 0 such that
U() = f
(m)
()
for all T with supp .
If U = 0 on (, a), then f(t) = 0 for t < a.
The number m in the structure theorems is equal to the local order of
U plus two.
For the space o

(X) the structure theorem is global. The primitive of

a distribution U is dened by U,
t
) for some
t
T
p
(K). Later on, we
use this construction of the m-th continuous primitive for constructing an
m-times integrated semigroup related to U.
Theorem 2.1.2 Let U o

(X). Then there exist m, r > 0 and a contin-

uous function f : R X such that
U() = f
(m)
()
for all o, and [f(t)[ =
|t|
O([t[
r
).
If U = 0 on (, a), then f(t) = 0 for t < a.
For distributions with a point support t
0
the structure theorems give
the following representations.
Proposition 2.1.3 Let U be a distribution with a point support t
0
, then
p > 0, U =

kp

(k)
(t t
0
).
Here (t t
0
) =
t
0
is the Dirac distribution (delta function) concentrated
at t
0
.
Now we dene a convolution of distributions. Let Z, X, Y be Banach
spaces such that the bilinear mapping (u, v) uv : Z X Y is dened.
If Z = X = Y = R, then this mapping is the usual multiplication. We
consider the particular case of Z = L(X, Y ) which is relevant to the study
of the Cauchy problem. Let U T

0
(Z), V T

0
(X) and c > 0. By
2001 CRC Press LLC
2001 CRC Press LLC
Theorem 2.1.1 there exist m, p 0 and continuous functions f : R Z
and g : R X vanishing for t < 0 such that
U() = f
(m)
(), V () = g
(p)
().
Then we dene
(U V ), ) := (f g)
(m+p)
, ) (2.1.3)
for any with supp (, c). This denition does not depend on
m, p, c, f, g. If U o

(L(X, Y )) and V o

(X), then U V o

(Y ).
Example 2.1.1 Let S T

0
(L(X, Y )) and x X, then
S ( x), ) = S, )x = I S, )x, (2.1.4)
S (
(k)
x), ) = S
(k)
, )x =
(k)
I S, )x, (2.1.5)
here
(k)
x, ) :=
(k)
, )x,
(k)
I, ) :=
(k)
, )I, and I = I
X
is
the identity operator on X.
It follows from Theorem 2.1.1 that for any T there exists a contin-
uous function f such that f(t) = 0 for t < 0, and m 0 such that
S, ) = (1)
m
_

0

(m)
(t)f(t)dt.
Let
(t) =
_
t, t 0
0, t < 0
,
then

= . Suppose supp (, c). By denition of the convolution

we have
S ( x), ) = (1)
m+2
_
c
0

(m+2)
(t)dt
_
t
0
(t s)f(s)ds
= (1)
m
_
c
0

(m)
(t)f(t)xdt = S, )x.
The second equality in (2.1.4) and the equalities (2.1.5) can be proved in
the similar way.
To dene the Laplace transform for distributions, one considers the
space o

(X) of exponential distributions of type . A distribution U be-

longs to o

(X) if e

t
U(

t) o

(X). The Laplace transform of U o

(X)
is the function

U() = LU(t) := U, e

t
) := e

t
U(

t ), (

t )e
()

t
), (2.1.6)
where
(t) =
_
1, t 0
0, t c < 0
2001 CRC Press LLC
2001 CRC Press LLC
and () C

(R). The following properties of Laplace transforms for

distributions are similar to the properties of classical Laplace transforms.
Proposition 2.1.4 Let U, V o

(X), then LU() is an analytic function

for Re > and
(LU)

() = L(

tU), LU

() = LU(),
L(U V )() = LU()LV (), Re > .
Now we consider the Cauchy problem in spaces of distributions. It is
not dicult to verify that if u is a classical solution of (CP), then the
corresponding distribution
U() := uH, ) =
_

0
(t)u(t)dt, T,
(here H() is the Heaviside function) is a solution for the equation
P U = x, (P)
where
P :=

I A T

0
(L(X
A
, X)),
A, ) := , )A
and
X
A
:=
_
D(A), |x|
A
:= |x| +|Ax|
_
.
Indeed, we have
_

0
(t)u

(t)dt = (0)x
_

0

(t)u(t)dt
= ( x)() +U

()
and
_

0
(t)u

(t)dt = A
_

0
(t)u(t)dt = AU().
Taking into account Example 2.1.1, we obtain
P U, ) = (

I) U, ) ( A) U, )
= U

, ) AU, ) = x, ).
Denition 2.1.3 A distribution U T

0
(X
A
) is called a solution of the
Cauchy problem (CP) in the sense of distributions if it is a solution of
(P).
2001 CRC Press LLC
2001 CRC Press LLC
Denition 2.1.4 The Cauchy problem (CP) is called well-posed in the
sense of distributions (or (P) is well-posed) if
(a) for any x X there exists a unique solution U T

0
(X
A
) of (P);
(b) for any sequence of initial values x
n
X such that x
n
0, the
corresponding sequence of solutions U
n
converges to 0 in T

0
(X
A
).
The Cauchy problem (CP) is called -well-posed in the sense of distri-
butions if
(a1) for any x X there exists a unique solution U o

(X
A
) of (P);
(b1) for any x
n
0 we have U
n
0 in o

(X
A
).
Similarly to the case of a uniformly well-posed Cauchy problem, the solu-
tion of a well-posed problem (P) can be constructed with the help of a
semigroup, namely a distribution semigroup .
Denition 2.1.5 We say that a distribution Q T

0
(L(X)) is a distribu-
tion semigroup if
(i) , T
0
, Q, ) = Q, )Q, ).
A distribution semigroup is said to be nondegenerate if
(ii) T
0
, Q()x = 0 =x = 0.
A distribution semigroup is said to be regular on
ranQ :=
_
Q()x

T
0
, x X
_
or simply regular if
(iii) for any y = Q()x with T
0
and x X, the distribution Qy is a
continuous function u() such that u(t) = 0 for t < 0 and u(0) = y.
A distribution semigroup is said to be with dense range if
(iv) ranQ = X.
A distribution semigroup Q is said to be an exponential distribution semi-
group of type if Q o

(L(X)).
The generator of distribution semigroup is dened by A := Q(

).
2001 CRC Press LLC
2001 CRC Press LLC
To justify the denition of the generator, for any distribution E with a
compact support in [0, ), we give the denition of the operator Q(E).
Let E c

and let
n
T
0
be any sequence convergent to (Dirac
-function). Then
D(Q(E)) := (2.1.7)
_
x X

Q(
n
)x x and y such that Q(E
n
)x y
_
,
and Q(E)x := y. It is not dicult to prove that this denition does not
depend on the choice of the sequence
n
.
Now we state the properties of distribution semigroups. In Theorem
2.1.4 we use these properties to show that every distribution semigroup
coincides with a solution operator distribution.
Proposition 2.1.5 Let E c

0
and Q be a distribution semigroup with
dense range. Then Q(E) has the following properties
(Q1) T
0
, x X,
Q()x D(Q(E)) and Q(E)Q()x = Q(E )x;
(Q2) D(Q(E)) = X;
(Q3) T
0
, x D(Q(E)),
Q(E)Q()x = Q()Q(E)x;
(Q4) There exists Q(E), the closure of Q(E), and
T
0
, x X,
Q(E)Q()x = Q(E)Q()x.
If Q is regular, then
(Q5) T,
Q(
+
) = Q(),
where
+
() c

0
is dened by

+
(t) =
_
(t), t 0
0, t < 0
.
Proof (Q1). Taking into account the fact that
n
T
0
and property
(i), we have
Q(
n
)Q()x = Q(
n
)x Q()x,
Q(E
n
)Q()x = Q(E
n
)x Q(E )x, x X.
2001 CRC Press LLC
2001 CRC Press LLC
Hence Q()x D(Q(E)) and Q(E)Q()x = Q(E )x. (Q1) and (iv)
imply (Q2).
(Q3). For x, y from (2.1.7) we have
Q()Q(E
n
)x Q()y = Q()Q(E)x,
Q()Q(E
n
)x = Q(E
n
)x Q(E )x = Q(E)Q()x.
(Q4). It is well-known that a linear operator A has a closure if and only if
x
n
D(A) : x
n
0,
Ax
n
y = y = 0.
Let x
n
D(Q(E)) be such that x
n
0 and Q(E)x
n
y, then
T
0
,
Q()Q(E)x
n
= Q(E )x
n
Q()y.
Since Q(E ) L(X, X), we obtain Q()y = 0. It follows from (ii) that
y = 0.
(Q5). Let T, then by (Q4), for
+
c

0
we have
Q(
+
)Q()x = Q(
+
)Q()x = Q(
+
)x
for all T
0
and x X. If we show that for any x X
Q(
+
)Q()x = Q()Q()x, (2.1.8)
then by (Q2) we have (Q5). By the structure theorem, in a general case
there exists m 0 and a continuous function f() C
_
R, L(X)
_
such that
f(t) = 0, t < 0 and
Q()x = (1)
m
_

0
f(t)
(m)
(t)xdt.
for all x X. Due to the regularity of Q, for any y ranQ there exists
f C
_
R, L(X)
_
such that
Q()y =
_

0
f(t)(t)ydt.
Note that in this case m = 0, and the proof is also valid if m = 1.
Hence for any y = Q()x, where x X, and any sequence
j
T
0
such that
j
, we have
Q(
+
)y = lim
j
Q(
+

j
)y
= lim
j
_

0
f(t)dt
_

0
()
j
(t )yd
=
_

0
f(t)(t)ydt.
Therefore (2.1.8) and (Q5) hold.
2001 CRC Press LLC
2001 CRC Press LLC
2.1.2 Well-posedness in the space of distributions
Theorem 2.1.3 Let A be a closed linear operator on a Banach space X.
Then
(W) the Cauchy problem (CP) is well-posed in the sense of distributions
if and only if
(S) there exists a solution operator distribution S T

0
(L(X, X
A
)) such
that
P S = I, (2.1.9)
S P = J, (2.1.10)
where I = I
X
, J = I
X
A
. In this case
U, ) = S, )x,
for all T and x X.
Proof (W) = (S). Consider the distribution S dened by
S()x = Sx() := U()
for all T, where U is the unique solution of (P). Then Sx T

0
(X
A
).
The well-posedness of the Cauchy problem implies that for every T
|x
n
| 0 = |S()x
n
|
A
= |U
n
()|
A
0,
therefore S() L(X, X
A
). We show that S T

0
(L(X, X
A
)), i.e.,

n
T,

n
0 =|S(
n
)|
L(X,X
A
)
0.
This implication follows from the estimate
m, C > 0:
|Sx,
n
)| C|
n
|
m
which holds on the set of all Sx, where x are from a bounded set M, and
for all sequences
n
such that
n
0. This estimate is the analogue
of Proposition 2.1.1 for bounded sets. Here m depends on M and on a
compact set K containing supp
n
.
Now, using the structure theorem, we show (2.1.9). Let T

and
supp (, a). There exist
g C
_
R, L(X
A
, X)
_
, f C
_
R, L(X, X
A
)
_
, v C
_
R, X
A
_
,
such that
P = g
(p)
, S = f
(m)
, Sx = v
(q)
2001 CRC Press LLC
2001 CRC Press LLC
on (, a) and
P S, )x = g f
(m+p)
, )x
= lim
n
(1)
m+p
1
n

k=
g
_
k
n
__
R
f
_
t
k
n
_
x
(m+p)
(t) dt
= lim
n
(1)
m+p
1
n

k=
g
_
k
n
__
R
f(t)
(m+p)
_
t +
k
n
_
xdt
= lim
n
(1)
p
1
n

k=
g
_
k
n
_
_
Sx,
(p)
_

t +
k
n
_
_
= (1)
p
_
R
g(s)Sx,
(p)
(

t +s))ds
= (1)
(p+q)
_
R
g(s)
_
R
v(t)
(p+q)
(t +s)dtds
= (1)
(p+q)
_
R

(p+q)
(t)
_
R
g(s)v(t s)dsdt
= P Sx, ) = P U, ) = I, )x.
Let us prove (2.1.10). Equality (2.1.9) implies the following equalities
(P S

)x = (P S)

x =

x, x X,
(P S)Ax = Ax, x D(A).
By denition of P we have
(P x) =

x Ax, x D(A).
Hence
P (S

x SAx x) = 0
and since the Cauchy problem is well-posed, we have
S

x SAx = x, x D(A).
Therefore,
(S P) = S

SA = J.
(S) = (W). Let S T

0
(L(X, X
A
)). For any T
0
and x X, we dene
U() := S()x = Sx(),
then U T

0
(X
A
) is a solution of (P):

I Sx, ) A Sx, ) = x, ).
2001 CRC Press LLC
2001 CRC Press LLC
The associativity of convolution implies the uniqueness of the solution
U = J U = S P U = S ( x) = Sx.
Let |x
n
| 0. Since for any T, S() L(X, X
A
), we have
|Sx
n
()|
A
|S()x
n
| +|S

()x
n
| +|(0)x
n
| 0.
We show that this convergence is uniform with respect to from any
bounded set / T. Since / is a bounded set in T, it follows that
a) there exists a compact set K such that supp K for any /;
b) for any i 0, there exists C
i
such that
/, sup
tR
[
(i)
(t)[ C
i
.
By Theorem 2.1.1 there exist m 0 and f C
_
R, L(X, X
A
)
_
such that
S = f
(m)
on K, therefore
/, |S()|
L(X,X
A
)
=
_
_
_
_
_
K
f(t)
(m)
(t) dt
_
_
_
_
L(X,X
A
)
C
m
_
_
_
_
_
K
f(t) dt
_
_
_
_
L(X,X
A
)
= C,
and
|S()x
n
|
A
C|x
n
|.
Thus we have shown that S is a solution operator distribution for the
Cauchy problem (P). Now we prove that every solution operator distribu-
tion coincides with a distribution semigroup generated by A.
Theorem 2.1.4 Let A be a closed densely dened linear operator on a
Banach space X. Then
(S) there exists a solution operator distribution S T

0
(L(X, X
A
)) satis-
fying (2.1.9) and (2.1.10)
if and only if
(D) A is the generator of a reqular distribution semigroup Q with dense
range.
In this case S = Q.
Proof (D) = (S). Let Q be a distribution semigroup with the generator
A. For any T and the corresponding function
+
introduced in (Q5)
of Proposition 2.1.5, we have
+
c

0
and

+
=

+
+(0).
2001 CRC Press LLC
2001 CRC Press LLC
Hence, by (Q1) and by the denition of the generator of distribution semi-
group, we write
x X, T
0
,
Q(

+
)x = Q(

+
)x +(0)Q()x = Q(

)Q(
+
)x
= AQ(
+
)x = AQ(
+
)Q()x
= Q(
+
)Q(

)x = Q(
+
)(A)Q()x.
From (Q5) we have
AQ()Q()x = Q(

)Q()x +(0)Q()x
= Q()AQ()x, (2.1.11)
or for y = Q()x
AQ()y Q(

)y = (0)y, (2.1.12)
Q()Ay Q(

)y = (0)y. (2.1.13)
Since the range of a distribution semigroup is dense in X and the operator
A is closed, we have Q()y D(A), and (2.1.12) holds for every y X.
Let |y
n
| 0 in (2.1.12), then
|AQ()y
n
| 0 =Q() L(X, X
A
).
If
n
0 (
n
T) in (2.1.12), then |Q(

n
)| 0 and |AQ(
n
)| 0.
Hence, since Q T

0
(L(X, X)), we have Q T

0
(L(X, X
A
)).
To complete the proof it remains to show (2.1.13) for y D(A). Let
x D(Q(

)) and
n
in (2.1.11), then Q(
n
)x x, Q(

n
)x
Ax, and we have
Q()Ax = Q(

)x +(0)x, x D(Q(

)) (2.1.14)
Since
x D(A), x
n
D(Q(

)) :
x
n
x and Ax
n
Ax,
(2.1.14) holds for x D(A). Thus, (2.1.9), (2.1.10) are valid for S = Q, i.e.
the implication (D) = (S) is proved.
(S) = (D). Let Q T

0
(L(X, X
A
)) be a solution of (2.1.9), (2.1.10).
For any F T

+
(X), U = Q F is a solution of the equation
P U = F. (2.1.15)
It follows from (2.1.10) that the solution is unique. Hence the
well-posedness of (2.1.15) is equivalent to the well-posedness of (P).
2001 CRC Press LLC
2001 CRC Press LLC
Now, using the fact that (2.1.15) has a unique solution, we prove prop-
erties (i) (iv) for Q.
(i). Let , T
0
. Dene

1
(t) := (t),
1
(t) := (t), t 0,
and let u(

t) be a solution of (2.1.15) with F =

1
x:
u

Au =
1
x. (2.1.16)
Then u() = (Q
1
)x is an innitely dierentiable function with values in
D(A). Let w(

t ), v(

t ) be solutions of (2.1.15) with F = (

1

1
) x and
F =
1
u(0), respectively. Then w(t), v(t) are also innitely dierentiable
in t. These functions have the following initial values
u(0) = Q()x, w(0) = Q( )x,
v(0) = Q()u(0) = Q()Q()x.
To prove (i) one needs to show v(0) = w(0). From (2.1.16) we have
A(u
1
) + (u
1
)

= (
1

1
) x,
hence w(t) = u
1
and w(0) = u().
Let H() be the Heaviside function, then H(t)u(t) T

0
(X
A
). Since A
does not depend on t, we have
A(Hu) + (Hu)

= H(t)(Au +u

) + u(0).
Since
T
0
, H(t)(Au +u

) = (H
1
) x = 0,
we obtain
A(Hu) + (Hu)

= u(0) (2.1.17)
and
A((Hu)
1
) + ((Hu)
1
)

=
1
u(0).
Therefore,
u = (Hu)
1
, v(0) =
_

0
u(t)(t) dt,
and
T
0
, v(0) = u() = w(0).
Thus
, T
0
, Q( ) = Q()Q().
2001 CRC Press LLC
2001 CRC Press LLC
(iii). Let Z be a solution of (2.1.15) with F = y, where y = Q()x,
T
0
and x X. Since u(0) = y, by (2.1.17) we have
Z(t) = Qy = H(t)u(t),
and Qy is a continuous function on (0, ).
(ii). Let x be such that
T
0
, Q()x = 0. (2.1.18)
We show that x = 0. Since Q T

0
(L(X, X
A
)), we have Qx = 0 on
(, 0). Furthermore, (2.1.18) implies that Qx = 0 on (0, ). Hence, by
Proposition 2.1.3
Qx =
k

i=1

(i)
x
i
, x
i
X
A
.
Since U = Qx is a solution of the equation AU +U

= x, we have

(k+1)
x
k
+
(k)
(x
k1
Ax
k
) +. . .
+

(x
1
Ax
2
) + (x Ax
1
) = 0,
and x
k
= . . . = x
1
= x = 0.
(iv). Let X

and X

A
be the dual spaces of X and X
A
. Dene Q

() :=
(Q())

, then
Q

() L(X

A
, X

), Q

0
(L(X

A
, X

)), A

L(X

, X

A
).
Since (2.1.12), (2.1.13) hold for Q, we have
Q

_
d
dt
A

_
= I
X

and
_
d
dt
A

_
Q

= I
X

A
.
Therefore Q

0
(L(X

, X

)) is a distribution semigroup which is non-

degenerate, as Q is nondegenerate. Hence if z

A
is such that
T
0
, x X, Q()x, z

) = x, Q

()z

) = 0,
then
Q

()z

= 0 = z

= 0 = Q()x = X.
To complete the proof we show that if A
1
is a generator of Q then A
1
= A.
From the well-posedness of (2.1.15) we have that the mapping F Q F
is an isomorphism from T

0
(X) to T

0
(X
A
). Since A
1
is a generator of Q,
2001 CRC Press LLC
2001 CRC Press LLC
the mapping F QF is an isomorphism from T

0
(X) to T

0
(X
A
1
). Hence
D(A) = D(A
1
).
Since
AQx +Q

x = Ix = A
1
Qx +Q

x,
we have
AQ()x = A
1
Q()x
for all T
0
and x X. Taking into account that Q()x = X, we
conclude A = A
1
.
Theorem 2.1.5 Let A be a closed linear operator on a Banach space X.
Then
(S) there exists a solution operator distribution S T

0
(L(X, X
A
)) satis-
fying (2.1.9) and (2.1.10)
if and only if
(R) there exist M > 0 and n 0 N such that
|R
A
()|
M[[
n
log(1 +[[)
for all from
=
_
C

Re >
n

log(1 +[[) +
1

log
C

_
,
where
(0, T), C > 0, 0 < < 1.
Proof (R) = (S). For any m 2n + 2 we dene the function
T
m
(t) :=
1
2i
_

m
e
t
R
A
() d,
where is the boundary of . On the function under the integral
sign grows not faster than [[
nt

+nm
/ log [[. Hence, this integral denes
a continuous function on (,
m
) with values in L(X, X
A
), where
m
=

n
(mn 1). In this case, by the abstract Cauchy theorem T
m
(t) = 0 for
t 0.
We now dene S := T
(m)
m
on (,
m
). Since for any T, supp
(,
m
), and
m

> m, T
(m

m)
m

(t), ) = T
m
(t), )
2001 CRC Press LLC
2001 CRC Press LLC
for all t (,
m
), we have
_
R

(m)
T
m
(t) dt =
_
R

(m

)
T
m
(t) dt,
and thus S is well-dened.
Let us show that S is a solution of (2.1.10). Let x X. For any T
with supp (, a) we can choose
m
= a, then
S P, )x = S

()x S()Ax
=
(1)
m+1
2i
_
a
0

(m+1)
(t)
_

m
e
t
R
A
()xddt

(1)
m+1
2i
_
a
0

(m)
(t)
_

m
e
t
R
A
()Axddt
=
(1)
m+1
2i
_
a
0

(m)
(t)
_

m
e
t
xddt.
By the residue theorem, for any t [0,
m
] we obtain

1
2i
_

m
e
t
d =
t
m1
(m1)!
.
This yields that
S P, )x = (1)
m
_
a
0

(m)
(t)
t
m1
(m1)!
dt x
= (0)x = x, ).
In the same way, one can show that S is a solution of (2.1.9).
(S) = (R). Let S T

0
(L(X, X
A
)) be a solution of (2.1.9), (2.1.10). By
Propositions 2.1.1, 2.1.2 we have
T > 0, > 0, p Z
+
, C > 0 : T, supp [1 , T +],
|S()| C||
p
,
and S can be extended to T
p
[1, T]. We denote the extension by the same
symbol. Using this extension we can construct a continuous primitive of
order p + 2 for S. Let t [0, T], and consider the function

t,p
(s) = (s)
p
(t s) T
p
[1, T],
2001 CRC Press LLC
2001 CRC Press LLC
where () C

(R) is dened by
(s) =
_
0, s 1
1, s 0
(2.1.19)
and

p
(t) =
_
t
p+1
(p+1)!
, t 0
0, t < 0
.
Dene
V (t) := S,
t,p
). (2.1.20)
We now show that V (t), 0 t < T is a local (p + 2)-times integrated
semigroup.
By Proposition 2.1.2 there is
n
T such that
supp
n
[1 , T +]
for all n N, and
|
n

t,p+1
|
p+1
0
for all t [0, T]. Hence
|
n

t,p+1
|
p
0 and |

t,p+1
|
p
0.
Consider (2.1.9), (2.1.10) with =
n
. Let n , then
S,

t,p+1
)x AS,
t,p+1
)x =
t,p+1
(0)x
for all x X and 0 t T. Furthermore
S,

t,p+1
)x S,
t,p+1
)Ax =
t,p+1
(0)x, (2.1.21)
for all x D(A). Since

t,p+1
(s) =

(s)
(t s)
p+2
(p + 2)!
+(s)
p
(t s)
+

(s)
(t s)
p+2
(p + 2)!
+

t,p
(s)
and
supp

(s)
(t s)
p+2
(p + 2)!
[1, 0],
for dened by (2.1.20) V (t) we have
t [0, T], V (t) = S,

t,p+1
) = S,
t,p
), (2.1.22)
2001 CRC Press LLC
2001 CRC Press LLC
and V (t) L(X). Since the mapping t
t,p
: [0, T] T
p
[1, T] is
continuous, the function V () is continuous in t [0, T]. Since supp
0,p

[1, 0] and suppS [0, ) it follows that V (0) = 0. Furthermore,
t (0, T), k = 1, 2, . . . , p + 1,

t+h,k
(s)
t,k
(s)
h

t,k1
(s) (2.1.23)
uniformly with respect to s [1, T]. Hence we have the convergence in
T
p
[1, T]. Therefore
d
dt
S(
t,p+1
) = lim
h0
S(
t+h,p+1
) S(
t,p+1
)
h
= lim
h0
S
_

t+h,p+1

t,p+1
h
_
= S(
t,p
).
Since S(
0,p+1
) = 0, we have
S(
t,p+1
) =
_
t
0
V (s) ds. (2.1.24)
From (2.1.21) (2.1.24) we obtain
V (t)x =
t
p+2
(p + 2)!
x +
_
t
0
V (s)Axds, t [0, T), x D(A), (2.1.25)
V (t)x =
t
p+2
(p + 2)!
x +A
_
t
0
V (s)xds, t [0, T), x X. (2.1.26)
As it was proved in Theorem 1.2.5 for A satisfying (2.1.25) (2.1.26), there
exists the resolvent of A and (R) holds for R
A
() with n = p + 2, where p
is the local order of S.
Remark 2.1.1 It follows from the obtained Theorems 2.1.32.1.5 and
Propositions 1.2.3, 1.2.4, that for a closed linear operator A condition
(W) and polynomial estimates (R) with some n are equivalent to k-well-
posedness of (CP), with some k.
2.1.3 Well-posedness in the space of
exponential distributions
Now we consider (CP) in the space of exponential distributions
o

(L(X, X
A
)). We show that the MFPHY-type condition:
M > 0, R :
_
_
_
_
d
k
d
k
_
(I A)
1

n
__
_
_
_
=
M k!
(Re )
k+1
(2.1.27)
2001 CRC Press LLC
2001 CRC Press LLC
for all C with Re > and all k = 0, 1, . . . ,
is necessary and sucient for the -well-posedness of (P) with a closed
densely dened operator A. Hence, by Theorem 1.2.1, (P) with such A is
-well-posed if and only if A is the generator of an exponentially bounded n-
times integrated semigroup with the same n as in (2.1.27). In the following
theorem we give the constructive proof of these connections.
Theorem 2.1.6 Let A be a linear closed densely dened operator on a
Banach space X. Then the following statements are equivalent:
(W
exp
) the Cauchy problem (CP) is -well-posed in the sence of distribu-
tions;
(S
exp
) there exists a solution operator distribution S o

(L(X, X
A
)) such
that
P S = I, S P = J, (2.1.28)
In this case
U, ) = S, )x
for all T and x X;
(D
exp
) A is the generator of an exponentially bounded distribution semi-
group Q with dense range. In this case S = Q;
(I
exp
) A is the generator of an exponentially bounded n-times integrated
semigroup V (t), t 0 for some n;
(R
exp
) the MFPHY-type condition (2.1.27) holds.
Proof Implications (W
exp
) (S
exp
) (D
exp
) follow from the corre-
sponding implications in Theorems 2.1.3 2.1.4. If a solution S of (2.1.28)
belongs to o

(L(X, X
A
)), then we have that Q = S o

(L(X, X
A
)) and
U = Qx o

(X
A
). The converse is also valid.
(I
exp
) = (S
exp
). Let n N and A be the generator of an exponentially
bounded n-times integrated semigroup V . Since V (t) is dened for all t 0,
S o

(L(X, X
A
)) can be dened as S = V
(n)
. Let
S() := (1)
n
_

0

(n)
(t)V (t) dt, T.
Taking into account the following equality for an n-times integrated semi-
group:
V

(t)x =
t
n1
(n 1)!
x +AV (t)x, t 0, x D(A),
2001 CRC Press LLC
2001 CRC Press LLC
we have
(1)
n+1
_

0

(n+1)
(t)V (t)xdt
= (1)
n
_

0

(n)
(t)
t
n1
(n 1)!
xdt + (1)
n
A
_

0

(n)
(t)V (t)xdt
for any T and x D(A). Since A is closed and all other operators
in this equality are bounded, it is valid for all x X. Integrating the rst
term in the right-hand side by parts, we obtain
AS()x = S

()x (0)x, x X. (2.1.29)

Hence S is a solution of the rst equation in (2.1.28). Using the commuta-
tivity of V (t) and A on D(A), we obtain the second one. From (2.1.29) we
have
|S()x|
A
|S()x| +|Q

()x| +[(0)[ |x|.

Therefore S() L(X, X
A
) for any , and S T

0
(L(X, X
A
)). Moreover,
exponential boundness of V implies
R : e
t
S o

(L(X, X
A
)),
hence S o

(L(X, X
A
)).
(S
exp
) = (R
exp
) = (I
exp
). Let S o

(L(X, X
A
)) be a (unique)
solution of (2.1.29). Applying the Laplace transform to (2.1.29), we have
(I A)LS() = I
and
LS()(I A) = J
for all C with Re > . From these equalities we have that the half-
plane Re > belongs to the resolvent set of operator A, and (I A)
1
=
LS(). Since e
t
S o

(X), by the analogue of Proposition 2.1.1 for

exponentially bounded distributions, we have
p 0, C > 0 : o, |(e
t
S)()| C||
p,p
, (2.1.30)
where
||
j,k
:= sup
0ik
sup
t
(1 +[t[)
j
[
(i)
(t)[.
Using the density of o in the space o
p
(R) of p-times continuously dif-
ferentiable functions with the norm | |
j,k
and the estimate (2.1.30),
we can extend S to o
p
(R). Let
p
and be dened by (2.1.19), then
(s)
p
(t s) o
p
(R). Introduce the function
V (t) = S
_
(s)
p
(t s)
_
:= (e
s
S)
_
e
s
(s)
p
(t s)
_
. (2.1.31)
2001 CRC Press LLC
2001 CRC Press LLC
It is not dicult to verify that the mapping t e
s
(s)
p
(t s) from R
to o
p
(R) is continuous, hence V () is a continuous function with values in
L(X, X
A
). Since suppS [0, +) and supp(e
s
(s)
p
(t s)) [1, t],
we obtain that V (t) = 0 for t 0. Moreover, (2.1.30) implies the following
estimate
|V (t)| = |(e
s
S)
_
e
s
(s)
p
(t s)
_
|
C|e
s
(s)
p
(t s)|
p,p
Me
t
(1 +[t[)
p+1
Me

t
for

> and t 0.
Let us show that S is n-th derivative of V for n = p +2. For any T
we have
V
p+2
, ) = (1)
p+2
_

0

(p+2)
(t)V (t)dt
= (1)
p+2
_

0

(p+2)
(t)S( s), ( s)
p
(t s))dt
= S( s), ( s)(1)
p+2
_

0

(p+2)
(t)
p
(t s)dt)
= S, ( s)( s)) = S, ).
Using properties of the Laplace transform we have
(I A)
1
=
p+2
_

0
e
t
V (t)dt, Re >

.
Hence, for n = p + 2 and all k = 0, 1, . . ., we obtain
_
_
_
_
d
k
d
k
_
(I A)
1

n
__
_
_
_
=
_
_
_
_
d
k
d
k
LV ()
_
_
_
_
= |L(t
k
V )()|

_

0
e
tRe
t
k
|V (t)|dt
M
_

0
t
k
e
(Re

)t
dt
=
M k!
(Re

)
k+1
, Re >

.
By Theorem 1.2.1, these estimates are the necessary and sucient con-
dition for a closed densely dened operator A to be the generator of an
exponentially bounded n-times integrated semigroup V .
2001 CRC Press LLC
2001 CRC Press LLC
2.2 The degenerate Cauchy problem
Let X, Y be Banach spaces. In this section we consider the degenerate
Cauchy problem
Bu

(t) = Au(t), t 0, u(0) = x, (DP)

with linear operators B, A : X Y such that A is closed and B is bounded
with ker B ,= 0. Suppose

B
(A) := (2.2.1)
_
C

(B A)
1
B =: R() is bounded in X
_
,= ,
Recall that the problem (DP) is equivalent to the Cauchy problem for the
inclusion
u

(t) /u(t), t 0, u(0) = x, (IP)

with / = B
1
A.
We prove the necessary and sucient conditions for the well-posedness
of (DP) and (IP) in spaces of distributions, generalizing the conditions ob-
tained in Section 2.1. Naturally, the semigroups connected with these prob-
lems are degenerate on X. The important fact is that an n-times integrated
semigroup with the generator / is degenerate on ker B = ker R() = /0,
but the corresponding solution operator distribution and distribution semi-
group are degenerate on ker R
n+1
:= ker R
n+1
() = /
n+1
0. We assume
that X admits the decomposition
X = X
n+1
ker R
n+1
, (2.2.2)
where
X
n+1
:= D
n+1
, D
n+1
:= R
n+1
()X,
which generalizes the decomposition X = X
1
/0 and the density condition
for a generator of a C
0
-semigroup.
In the rst subsection we describe the structure of ker R
n+1
. It is used
in the main theorem (second subsection) where we discuss the structure
of solution operator distributions and distribution semigroups on ker R
n+1
.
We prove that (DP) is well-posed in the space of distributions if and only
if

/ (the part of operator / in X
n+1
) is the generator of a distribution
semigroup, or equivalently,

/ is the generator of a local k-times integrated
semigroup for some k. In the third subsection, we consider (DP) in the space
of exponential distributions. The distinctive feature of the exponential case
is that the parameter k of the integrated semigroup coincides with the
parameter in the estimate for the resolvent of the generator and with the
parameter n in the decomposition (2.2.2).
2001 CRC Press LLC
2001 CRC Press LLC
2.2.1 A-associated vectors and degenerate
distribution semigroups
Denition 2.2.1 The sets
K
i
=
_
x X

y K
i1
: Ay = Bx, y ,= 0, i 1,
K
0
= ker B,
are called the sets of i-th A-associated with ker B vectors.
From this denition we have that for any
i
K
i
there exists a sequence
of A-associated vectors
0
,
1
, ...
i1
such that
k
K
k
and A
k
=
B
k+1
for k = 0, 1, . . . , i 1. This sequence is unique up to ker A.
Proposition 2.2.1 Consider linear operators B, A : X Y with

B
(A) ,= . Then
(I) ker B ker A = 0; K
i
ker A = , i 1; K
i
K
j
= , j ,= i;
(II) K
i
0 is a closed linear manifold in X for all i 0;
(III) let C be a region containing some sequence convergent to innity.
If there exists M > 0 such that
|R()|
M[
n
[
[f()[
for all and some f such that [f()[

, then
x D
n+1
, lim

(B A)
1
Bx = x,
D
n+1
ker B = 0, and K
n+1
= ;
(IV) any x ker R
n+1
can be represented in the form
x = x
1
+x
2
+. . . +x
n
, where x
i
K
i
.
If ker B is complemented in ker R
n+1
, then the decomposition
ker R
n+1
= ker B /
1
... /
n
,
holds, where /
i

=
_
K
i
0
_
/ ker B and /
i
X.
Proof (I). Condition (2.2.1) implies that ker A ker B = 0. For
i
and
for
B
(A) the following relations hold
(B A)
1
B
i
=
i1

i2
...
i1

0
,
2001 CRC Press LLC
2001 CRC Press LLC
_
(B A)
1
B

i
= (2.2.3)
(1)
j
_
C
0
j1

ij
+C
1
j

ij1
+. . . +C
ij
i

ij

,
j = 1, . . . , i.
In particular,
_
(B A)
1
B

i
= (1)
i

0
,
for any
i
K
i
,
0
ker B. To prove these relations we write
(B A)
1
B
i
= (B A)
1
(AB)
i1
(2.2.4)
=
i1
+(B A)
1
B
i1
=
i1

i2
. . .
i1

0
.
Let
i
K
i
ker A, then from (2.2.3) we have
A
0
= A(1)
i
_
(B A)
1
B

i
= B(1)
i
_
(B A)
1
B

i1
(B A)
1
A
i
= 0.
Since ker A ker B = 0, we have
0
= 0. Therefore
K
i
ker A = for any i 1.
Let j = 0. Take
i
K
i
and
i1
K
i1
such that A
i1
= B
i
. Since
K
i
ker A = , then A
i1
,= 0. Hence,
i
, ker B, and K
0
K
i
= .
Now let j > 0 and i > j. Consider any x K
j
K
i
. We have
_
(B A)
1
B

i
x ,= 0 since x K
i
, and
_
(B A)
1
B

i
x = 0 since
x K
j
. Hence, K
j
K
i
= .
(II). For any i 1, K
i
0 are obviously linear manifolds. We now
show that they are closed. Let
k
i
be a sequence from K
i
convergent to
f
i
, then
_
(B A)
1
B

k
i

_
(B A)
1
B

j
f
i
for all j = 1, . . . , i and
B
(A). In particular,

k
0
= (1)
i
_
(B A)
1
B

i
is convergent. Furthermore, (2.2.3) implies that all
k
j
, j = 1, . . . , i, are
convergent. Denote their limits by f
j
. Considering (2.2.4) for
k
0
,
k
1
, ...
k
i
and letting k , we obtain
(B A)
1
Bf
i
= (B A)
1
(AB)f
i1
,
i.e. Bf
i
= Af
i1
. Similarly, Bf
j
= Af
j1
for any j = 1, 2, . . . , i 1. Hence
f
i
K
i
.
2001 CRC Press LLC
2001 CRC Press LLC
(III). Let x D
n+1
, then there exists y X such that
x =
_
(
0
B A)
1
B

n+1
y
for some
0

B
(A), and
|(B A)
1
Bx x|
= |(B A)
1
B
_
(
0
B A)
1
B

n+1
y
_
(
0
B A)
1
B

n+1
y|
= |(
0
)
1
(B A)
1
B
_
(
0
B A)
1
B

n
x

0
(
0
)
1
_
(
0
B A)
1
B

n+1
y|
= |(
0
)
(n+1)
(B A)
1
By . . .

0
(
0
)
1
_
(
0
B A)
1
B

n+1
y|

0.
Hence, if x X
n+1
= D
n+1
ker B, then x = 0.
Consider x K
n+1
. Let x
0
ker B be the rst vector in the sequence
of A-associated vectors x
0
, x
1
, x
n1
, x, then |x
0
| = |R(
0
)
n
R()x|. Ap-
plying the resolvent identity, we obtain that x
0
= 0, therefore K
n+1
= .
(IV). Let x ker R
n+1
, then we have
_
(B A)
1
B

n+1
x = 0 and
either
j = 1, 2, . . . n,
_
(B A)
1
B

j
x = 0 x ker B
or
1 j n,
_
(B A)
1
B

j+1
x = 0 and
_
(B A)
1
B

j
x ,= 0.
If j = 1, we have y = (B A)
1
Bx ker B. Hence Bx = Ay and
x K
1
. If j = 2, then
y = (B A)
1
B(B A)
1
Bx ker B
=y
1
= (B A)
1
Bx K
1
,
and x + y
1
K
2
, and so on. Thus, for any x ker R
n+1
we have
x = x
1
+ x
2
+ . . . + x
n
, where x
i
K
i
, 1 i n. If ker B complemented
in ker R
n+1
(i.e., if there exists Y such that ker R
n+1
= Y ker B), then
x = x
0
+ x
1
+ x
2
+. . . + x
n
, x
i
/
i
, 0 i n.
We now prove by contradiction that this representation is unique. Suppose
that it is not unique, then we have 0 = z
1
+z
2
+. . . +z
n
, where z
i
/
i
, and
there exists z
j
,= 0 such that z
i
= 0 for all i > j. Applying
_
(
0
BA)
1
B

j
to this equality, we have for x
0
that the vector from ker B for which z
j
is
2001 CRC Press LLC
2001 CRC Press LLC
the j-th A-associated vector, is equal to 0. Hence the decomposition in (iv)
holds true.
Corollary 2.2.1 If the decomposition (2.2.2) holds, and ker B is comple-
mentable in ker R
n+1
, then we have the decomposition
X = X
n+1
ker B /
1
... /
n
=: A ker B, (2.2.5)
and K
n+1
= .
Taking into account the discussions in the previous section we can give
the following denition of a solution for (DP) in spaces of distributions.
Denition 2.2.2 An abstract distribution U T

0
(X
A,B
) is a solution of
(DP) in the sense of distributions if
T, (2.2.6)
BU,

) +AU, ) = , )Bx, x X,
or equivalently
P U = Bx, x X, (DP)
where
P :=

B A,

B, ) :=

, )B,
A, ) := , )A
and
X
A,B
=
_
x D
1
, |x|
R
= inf
y:R()y=x
|y|
_
.
A solution U is called degenerate on Z X, if for U corresponding to
x Z we have U() = 0 for any T
0
.
For the degenerate Cauchy problem, the set
D
1
:= R()X
=
_
x D(A)

there exists y such that By = Ax

_
is the generalization of D(/) in the nondegenerate case.
It follows from Denition 2.2.2 that U T

0
(X
A,B
) is a solution of
(DP) if and only if U T

0
(X
A
) is a solution of
T U x, x X, (IP)
where T :=

I /, / = B
1
A, and
X
A
:=
_
D(/), |x|
A
= |x| +|/x|
_
.
2001 CRC Press LLC
2001 CRC Press LLC
Here |/x| is the factor-norm on X//0.
If ker B is complemented in X:
X = A ker B = A /0,
then there exists

/, the single-valued branch of / (i.e.,

/ is the single-
valued operator with D(

/) = D(/)) and X
A,B
is homeomorphic to the
space
X

A
:= [D(

/)] =
_
D(

/), |x|

A
= |x| +|

/x|
_
.
If, moreover, the decomposition (2.2.2) holds, then

/ , the part of / in
X
n+1
, is the part of

/ in X
n+1
.
Denition 2.2.3 The problem (DP) is called well-posed in the sense of
distributions (or (DP) well-posed) if for any x X there exists a unique
solution of (DP), and U
j
0 for any sequence x
j
0. If U o

(X

A
),
then (DP) is called w-well-posed.
If / = B
1
A, then (IP) is well-posed if and only if (DP) is well-posed.
In this case U T

0
(X
A
).
As in the nondegenerate case, the solution of well-posed (DP) can be
constructed with the help of a semigroup of solution operators. In this case
it is a distribution semigroup (see Denition 2.1.5), which is degenerate on
ker R
n+1
.
Here we say that a distribution semigroup Q is degenerate on
Z = ker Q :=
_
x X

T
0
, Q()x = 0
_
,
if Z contains at least one non-zero element, and we say that Q is nonde-
generate if ker Q = 0.
For any E c

0
we dene an operator Q(E). Let E c

0
and let

n
T
0
be any sequence covergent to . Then for any x from
D(Q(E)) := (2.2.7)
_
x

Q(
n
)x x and y such that Q(E
n
)x y
_
,
we have Q(E)x := y.
Now we state the properties of degenerate distribution semigroups. In
the main theorem we use these properties to show that the distribution
semigroup connected with the well-posed (DP) is regular on X
n+1
and
degenerate on ker R
n+1
.
Proposition 2.2.2 Let Q be a distribution semigroup degenerate on Z and
let E c

0
. Then Q(E) has the following properties
2001 CRC Press LLC
2001 CRC Press LLC
(Q1) T
0
, x X,
Q()x D(Q(E)) and Q(E)Q()x = Q(E )x;
(Q2) D(Q(E)) Z = 0;
(Q3) T
0
, x D(Q(E)),
Q(E)Q()x = Q()Q(E)x;
(Q4) if X = A Z and ranQ A, then there exists a closure of Q(E)
and
T
0
, x X,
Q(E)Q()x = Q(E)Q()x;
(Q5) if Q is regular and ranQ A, then for any x ranQ and T,
we have
Q(
+
)x = Q()x,
where
+
() c

0
is dened by

+
(t) =
_
(t), t 0
0, t < 0
.
Proof In addition to the proof of the Proposition 2.1.3 for the nondegen-
erate case, we need to show only (Q2), (Q4).
(Q2). Consider x Z and a sequence
j
from (2.2.7), then Q(
j
)x =
0. Hence, for any E c

0
we have x , D(Q(E)), i.e.
D(Q(E)) Z = 0.
(Q4). A linear operator A has a closure if and only if
x
j
D(A) such that x
j
0, we have
Ax
j
y = y = 0.
Consider x
j
D(Q(E)) such that x
j
0 and Q(E)x
j
y, then
y ranQ A. By (Q1) we have
T
0
,
Q()Q(E)x
j
= Q(E )x
j
Q()y = 0 = y Z,
and y ranQ A. Since Z A = 0, we have y = 0.
Due to (Q4), the operator Q(

) is dened for any distribution semi-

group Q with ranQ A, where X = A Z. This operator is called the
generator of Q.
2001 CRC Press LLC
2001 CRC Press LLC
2.2.2 Well-posedness in the sense of distributions
In the following theorem, assuming the decomposition (2.2.5), we show the
connection between the well-posedness of (DP), the existence of a degen-
erate distribution semigroup, the existence of a degenerate local k-times
(k n + 1) integrated semigroup with the generator

/, polynomial es-
timates for R

A
() in some region containing a right semiaxis, and the
existence of a unique classical solution for local (CP) on D
k+1
.
We recall that the operator A
0
, where A
0
x := lim
t0
t
1
[V
(k)
(t)x x]
is dened for those x where the limit exists, is called the generator of
a k-times integrated semigroup V . A local k-times integrated semigroup
V (t), t [0, T) is called degenerate on Z if
t [0, T),
V (t)x = 0 = x Z,
and V is called nondegenerate if Z = 0.
Theorem 2.2.1 Let A, B : X Y be linear operators, A be closed and
invertible, B be bounded. Suppose that the decomposition (2.2.5) holds.
Then the following statements are equivalent:
(W) the problem (DP) is well-posed. Its solution belongs to T

0
(X

A
) for
x X
n+1
and is degenerate for x ker R
n+1
;
(S) there exists a solution operator distribution
S T

0
(L(X, X
A,B
)) T

0
(L(X
n+1
, X

A
))
such that
(P S)x = Bx, x X, (2.2.8)
and
(S

T)x = x, x D(/), (2.2.9)
where

T :=

I

/,

/ is the part of / in X
n+1
, and

/ is the
single-valued branch of /.
In this case, for any x X,
U, ) = S, )x
for all T. Furthermore, S is degenerate on ker R
n+1
;
(D) there exists a regular distribution semigroup Q degenerate on ker R
n+1
with ranQ dense in X
n+1
. Its generator is equal to

/. In this case
S = Q;
(I) for any T > 0, the operator

/ is the generator of a local k-times inte-
grated semigroup V (t), t [0, T) on X
n+1
, for some k = k(T) n;
2001 CRC Press LLC
2001 CRC Press LLC
(R) for the resolvent of

/, the polynomial estimate
|R

A
()|
M[[
l
log(1 +[[)
M
1
[[
l
(2.2.10)
holds for some M > 0, l 0, and any
=
_
C

Re >
l

log(1 +[[) +
1

log
C

_
,
where
C > 0, 0 < < 1, > 0;
(Wk) for any T > 0, the local Cauchy problem
u

(t) =

/u(t), 0 t < T, u(0) = x, (CP)
is k-well-posed on D
k+1
for some k = k(T) n.
Proof (W) = (S). Consider the distribution S dened by
S()x = Sx() := U(), T, x X,
where U is a unique solution of (DP). Then Sx T

0
(X
A,B
) = T

0
(X

A
)
and Sx T

0
(X

A
) for x X
n+1
.
The well-posedness of (DP) implies that for every T,
|x
n
| 0 =|S()x
n
|

A
= |U
n
()|

A
0.
Therefore S() L(X, X

A
). We now show that S T

0
(L(X, X

A
)), i.e.,
suppS [0, ) and

n
T,

n
0 =|S(
n
)|
L(X,X

A
)
0.
Consider the set
B =
_
Sx, |x| c
_
T

0
(X).
Since for any Sx
j
B and any
j
0, we have
j
Sx
j
= S(
j
x
j
) 0,
then B is bounded in T

0
(X). For any bounded set B and any
j
0 in T
p N, C > 0 : j, Sx B,
|Sx(
j
)| C|
j
|
p
,
where
||
p
= sup
k=0,1,...p, tsupp
|
(k)
(t)|.
2001 CRC Press LLC
2001 CRC Press LLC
Therefore, S(
j
)x 0 uniformly in x from a bounded set, or equivalently
|S(
j
)|
L(X,X

A
)
0. Verication of (2.2.8), as in the nondegenerate case
in Theorem 2.1.3, is based on the structure theorem. Let us prove (2.2.9).
Equality (2.2.8) implies the following equalities
(P S

)x = (P S)

x =

Bx, x X,
(P S)

/x = Ax, x D(

/) = D(/).
By the denition of P we have
(P x) =

Bx Ax, x D(/).
Hence
P (S

x S

/x x) = 0, x D(/),
and since the Cauchy problem is well-posed, we have
S

x S

/x = (S

T)x = x, x D(/).
(S) = (W). Let
S T

0
(L(X, X
A,B
)) T

0
(L(X
n+1
, X

A
)).
For any T and x X we dene U() := S()x = Sx(). Then
U T

0
(X
A,B
) is a solution of (DP):
BSx,

) ASx, ) = , )Bx, x X.
For x X
n+1
we have
Sx,

)

/Sx, ) = , )x,
and the associativity of convolution implies the uniqueness of the solution
on X
n+1
:
U = ( I) U = S

T U = S ( x) = Sx, x X
n+1
.
We now show that on ker R
n+1
the degenerate solution U is also unique.
Consider x ker R
n+1
. Since for any T
0
, S()x = 0, then we have
that Sx = 0 on (0, ). Since S T

0
(L(X, X
A
)), we have Sx = 0 on
(, 0). Thus, suppSx = 0. By Proposition 2.1.3 we have
Sx =
p

i=0

(i)
z
i
,
where p is the local order of the distribution Sx. Since Sx is a solution of
(2.2.8), we have

(p+1)
Bz
p
+
(p)
(Bz
p1
Az
p
) +. . .
+

(Bz
0
Az
1
) + (Bx Az
0
) = 0
2001 CRC Press LLC
2001 CRC Press LLC
and z
p
ker B, z
p1
K
1
, . . . z
0
K
p
, x K
p+1
. Hence, for x
K
i
(1 i p +1) and for the corresponding the sequence of A-associated
vectors x
0
, . . . x
i1
, we have
Sx =
i1

j=0

(j)
x
i1j
. (2.2.11)
Due to Corollary 2.2.1, K
n+1
= , and hence p+1 n. Since A is invertible,
x
k
in (2.2.11) are dened uniquely. In a similar way, any degenerate on
ker R
n+1
solution U has the same form and is equal to Sx. In particular,
for x K
0
= ker B, all x
k
in (2.2.11) are equal to zero and U = 0. Thus,
U = Sx for x ker R
n+1
and for x X
n+1
, hence U = Sx for all x X.
Let |x
j
| 0. Then S() L(X, X

A
) for all T, and
|Sx
j
()|

A
|S()x
j
| +|S

()x
j
| +|(0)x
j
| 0,
i.e. Sx
j
0 in T

0
(X

A
).
(S) = (D). Let
Q T

0
(L(X, X

A
)) T

0
(L(X
n+1
, X

A
))
be a solution of (2.2.8) and (2.2.9). For any F T

+
(X
n+1
), U = Q F is
a unique solution of the equation

T U = F. (2.2.12)
Using this fact, we now prove that Q satises properties (i) (ii) from
Denition 2.1.5, that Q is regular on X
n+1
, degenerate on ker R
n+1
, and
that ranQ is dense in X
n+1
.
Let , T
0
. Dene
1
(t) := (t),
1
(t) := (t), t 0. Let u(

t )
be a solution of (2.2.12) with F =
1
x, where x X
n+1
:

/u +u

=
1
x. (2.2.13)
Let w(

t ) and v(

t ) be solutions of (2.2.12) with F = (

1

1
) x and F =

1
u(0), respectively. Then u() = (Q
1
)x is an innitely dierentiable
function with values in D(

/). The same is valid for w and v. These
functions have the following initial values
u(0) = Q()x, w(0) = Q( )x,
v(0) = Q()u(0) = Q()Q()x, x X
n+1
.
To prove (i) one needs to show v(0) = w(0). From (2.2.13) we have

/(u
1
) + (u
1
)

= (
1

1
) x,
2001 CRC Press LLC
2001 CRC Press LLC
hence w(t) = u
1
and w(0) = u(). Let H() be the Heaviside function,
then H(t)u(t) T

0
(X

A
). Since

/ does not depend on t, we have

/(Hu) + (Hu)

= H(t)(

/u +u

) + u(0).
Since for any T
0
we have H(t)(

/u + u

) = (H
1
) x = 0, then we
obtain

/(Hu) + (Hu)

= u(0) (2.2.14)
and

/((Hu)
1
) + ((Hu)
1
)

=
1
u(0).
Therefore, v = (Hu)
1
, v(0) =
_

0
u(t)(t) dt, and for any T
0
we
have v(0) = u() = w(0). Thus
, T
0
,
Q( )x = Q()Q()x, x X
n+1
.
Note that Qx = Sx dened for x ker R
n+1
by formula (2.2.11), also
satises relation (i). Thus, we have constructed the distribution semigroup
Q degenerate on ker R
n+1
.
To show that Q is a regular distribution semigroup, we let z be a solution
of (2.2.12) with F = y, where y ranQ. Since u(0) = y, by (2.2.14) we
have
z(t) = Qy = H(t)u(t)
and Qy is a continuous function on (0, ).
We now show that ran Q is dense in X
n+1
. We have

/ L(X

A
, X
n+1
)
and Q() = S() L(X
n+1
, X

A
). Let X

n+1
and X

A
be the dual spaces
of X
n+1
and X

A
, then

/

L(X

n+1
, X

A
). Consider Q() on X
n+1
and
dene Q

() = (Q())

for T
0
. Then
Q

() L(X

A
, X

n+1
), Q

0
(L(X

A
, X

n+1
)).
Since (2.2.8), (2.2.9) hold for Q, we have
Q

I

/

_
= I
X

n+1
,
and
_

I

/

_
Q

= I
X

A
.
Therefore, Q

0
(L(X

A
, X

n+1
)) is a distribution semigroup, that is non-
degenerate on X

A
as Q is nondegenerate on X
n+1
. Hence, if z

A
is
such that
T
0
, x X
n+1
,
Q()x, z

) = x, Q

()z

) = 0,
2001 CRC Press LLC
2001 CRC Press LLC
then
Q

()z

= 0 = z

= 0 = Q()x = X
n+1
.
Thus, the set
_
Q()x

x X
n+1
_
is dense in X
n+1
. Since Q()x = 0 for x ker R
n+1
, we have ranQ() =
X
n+1
.
To complete the proof we note that

/ generates the nondegenerate
distribution semigroup Q which is densely dened on X
n+1
. Since the
domain of the generator of a degenerate distribution semigroup belongs to
the complement of the set where Q is degenerate,

/ is the generator of the
distribution semigroup Q constructed on X.
(D) = (S). Let Q be the distribution semigroup from (D). For T
and the function
+
c

0
introduced in (Q5) of Proposition 2.2.2, we have

+
=

+
+(0).
Hence, by (Q1) and by the denition of the generator of a distribution
semigroup we have
x X
n+1
, T
0
,
Q(

+
)x = Q(

+
)x +(0)Q()x
= Q(

)Q(
+
)x
=

/Q(
+
)x =

/Q(
+
)Q()x
= Q(
+
)Q(

)x = Q(
+
)(

/)Q()x.
Taking into account (Q5), we obtain

/Q()Q()x = Q(

)Q()x +(0)Q()x = Q()

/Q()x
or for y = Q()x

/Q()y Q(

)y = (0)y, (2.2.15)
Q()

/y Q(

)y = (0)y. (2.2.16)
Since the range of the distribution semigroup is dense in X
n+1
and the
semigroups generator

/ is closed, we have that Q()y D(

/) for every
y X
n+1
and (2.2.15) holds for every y X
n+1
. Let |y
j
| 0 in (2.2.15),
then
|

/Q()y
j
| 0 =Q() L(X, X

A
).
If
j
0 (
j
T) in (2.2.16), then |Q(

j
)| 0 and |/Q(
j
)| 0.
Hence, since Q T

0
(L(X, X)), we have
Q T

0
(L(X
n+1
, X

A
)) T

0
(L(X, X
A
)).
2001 CRC Press LLC
2001 CRC Press LLC
Thus, S = Q on X
n+1
, and for x X
n+1
, S satises (2.2.8). Furthermore,
S = Q dened by formula (2.2.11) satises (2.2.8) for x /
n+1
. Hence S
satises (2.2.8) for all x X.
Now we show (2.2.9), i.e. we verify (2.2.16) for y D(

/). Let x
D(Q(

)) and
j
. Then Q(
j
)x x and Q(

j
)x

/x. From
(2.2.16), for x D(Q(

)) we have
Q()

/x = Q(

)x +(0)x, T. (2.2.17)
Since
x D(

/), x
j
D(Q(

)):
x
j
x and

/x
j


/x,
then (2.2.17) holds for x D(

/). Thus, considering Sx = Qx on X
n+1
,
we have that equation (2.2.9) holds on D(

/). For S dened by (2.2.11),
equation (2.2.9) holds on D(/) ker R
n+1
. Since D(

/) = D(/) X
n+1
,
we have (2.2.9) on D(/).
(I) = (R). Let

/ be the generator of a local k-times integrated semi-
group V (t), 0 t < T, then for any (0, T) the operator-function
R(, ) =
_

0

k
e
t
V (t)dt (2.2.18)
is dened and bounded on X. Taking into account the following equality
for a local k-times integrated semigroup:
V (t)x =
_
t
k
k!
_
x +
_
t
0
V (s)

/xds, x D(

/), t (0, T),
we have
R(, )(I

/)x = (I G())x, x D(

/),
(I

/)R(, )x = (I G())x, x X
n+1
,
where G() is dened by the formula:
G()x =
k
e

V ()x +
k1

i=0
()
i
i!
e

x. (2.2.19)
It was proved in Theorem 1.2.5 that the estimates
|G()| < ,
_
_
(1 G())
1
_
_
<
1
1
hold for . Therefore, (R) holds for (I

/)
1
= R

A
() with l = k.
2001 CRC Press LLC
2001 CRC Press LLC
Now we prove (R) = (S). To construct S, we consider the operator-
function:
P
m
(t)x :=
1
2i
_

m
e
t
R

A
()xd, x X
n+1
. (2.2.20)
On the function under the integral sign grows not faster than
[[
lt

+lm
/ log [[. Hence, this integral denes a continuous function on
(,
m
) with values in L(X
n+1
, X

A
), where
m
=

l
(m l 1) and
m > l + 1. In this case, by the abstract Cauchy theorem P
m
(t) = 0 for
t 0. Now we dene the distribution S := P
(m)
m
on (,
m
). This de-
nition does not depend on m

> m : P
(m)
m
= P
(m

)
m

. Furthermore, for any

T with supp (, a),
m
= a can be chosen by increasing m. In
Theorem 2.1.5 it was shown that the dened operator distribution S satis-
es (2.2.8) and (2.2.9) on X
n+1
and X

A
, respectively. For x
0
ker B we
dene P
m
(t)x
0
= 0. For x
1
K
1
using formally (2.2.20) and the equality
Bx
1
= Ax
0
, by the abstract Cauchy formula we have
P
m
(t)x
1
=
1
2i
_

m
e
t
(B A)
1
Bx
1
d
=
1
2i
_

m
e
t
x
0
d =
t
m1
(m1)!
x
0
,
. . . (2.2.21)
P
m
(t)x
i
=
i1

k=0
t
m1k
(m1 k)!
x
i1k
, x
i
K
i
, i n.
Then Sx
i
:= P
(m)
m
x
i
on ker R
n+1
coincides with S dened by (2.2.11), and
satises (2.2.8) and (2.2.9) on ker R
n+1
and ker R
n+1
D(/), respectively.
Thus, S satises the equation (2.2.8) on X, and the equation (2.2.9) on
D(/).
(S) = (I). Let
S T

0
(L(X, X

A
)) T

0
(L(X
n+1
, X

A
))
be the solution of (2.2.8) (2.2.9) degenerate on ker R
n+1
. Then the dis-
tribution S can be extended to T
p
[1, T], where p is the order of S. We
denote this extension by the same symbol. Let t [0, T], consider the func-
tion
t,p
(s) = (s)
p
(t s) T
p
[1, T], where and are the same as
(2.1.19):
(s) =
_
0, s 1
1, s 0
and

p
(t) =
_
t
p+1
/(p + 1)! , t 0
0, t < 0
.
2001 CRC Press LLC
2001 CRC Press LLC
Dene V (t) = S,
t,p
). We have that the operator-function V () is strongly
continuous in t. We now show that V (t), t [0, T) is a (p + 2)-times
integrated semigroup. The function
t,p+1
can be approximated by
n
T
so that
|
t,p+1

n
|
p
0 and |

t,p+1

n
|
p
0.
Consider (2.2.8) and (2.2.9) with =
n
. Taking the limit, we have
BS,

t,p+1
)x AS,
t,p+1
)x =
t,p+1
(0)Bx
for x X, and
S,

t,p+1
)x S,
t,p+1
)

/x =
t,p+1
(0)x
for x D(/). As in the nondegenerate case, we have
BV (t)x =
t
p+2
(p + 2)!
Bx +A
_
t
0
V (s)xds (2.2.22)
for x X, and
V (t)x =
t
p+2
(p + 2)!
x +
_
t
0
V (s)

/xds (2.2.23)
for x D(/). By denition of the operators V (t) via S, we have that
V (t)x X
n+1
for x X
n+1
. Then equations (2.2.22) and (2.2.23) hold on
X
n+1
and D(

/), respectively. Hence, as it was proved in Theorem 2.1.6,
V is a nondegenerate (p +2)-times integrated semigroup on X
n+1
with the
generator

/.
(I) (Wk). Let

/ be the generator of a nondegenerate local k-
times (k n) integrated semigroup V (t), t [0, T) on X
n+1
. Since the
domain D(

/) of its generator is equal to D
n+1
and is dense in X
n+1
, then
by Proposition 1.2.4, for any x R
k+1
()X
n+1
there exists the unique
solution u(t) = V
(k)
(t)x, t [0, T) of the local Cauchy problem (CP). Due
to the decomposition (2.2.2), R
k+1
()X
n+1
is equal to R
k+1
()X = D
k+1
.
If for any x D
k+1
, u() is a unique solution of the local Cauchy problem
(CP), then for any x D
k+1
, u() is the unique solution of (DP). Since

B
(A) ,= , and hence (

/) ,= , then by Lemma 1.2.1 we have the stability
of the solution:
|u(t)| C|x|
R
k, where |x|
R
k := inf
y:R
k

A
()y=x
|y|.
As in Theorem 1.6.4 we dene operators U(t)x := u(t), t [0, T) for
x D
k+1
. We extend them to [D
k+1
]
k
, the closure of D
k+1
in the norm
| |
R
k. Then U
1
(t) are dened on [D
k
]
k1
, . . . , U
i
(t) on [T
k(i1)
]
ki
.
In particular V (t) := U
k
(t) are dened and bounded on D(

/) = X
n+1
.
2001 CRC Press LLC
2001 CRC Press LLC
By construction, V (t)x is continuous in t [0, T) for any x X
n+1
, and
satises equations (2.2.22) and (2.2.23) (with the semigroup parameter k)
for x X
n+1
and D(

/), respectively. It follows from Theorem 1.2.5 that V
is a local nondegenerate (k + 1)-times integrated semigroup on X
n+1
with
the generator

/, which is densely dened in X
n+1
.
Remark 2.2.1 We note that the condition that A is invertible implies the
uniqueness of the solution on ker R
n+1
. If ker A ,= 0, then x
i
in (2.2.11)
are dened up to ker A.
Summarizing the discussion of this subsection we note that assuming
that X admits the decomposition (2.2.5), we have that the condition (R) is
the criterion for the well-posedness of (DP). Here we used the technique
of integrated and distribution semigroups with a generator equal to the
single-valued restriction of / on X
n+1
.
Now, using the technique of degenerate semigroups with multivalued
generators and not considering the detailed structure of semigroups on K
i
,
we obtain the well-posedness criterion for (DP) and (IP) under a less
restrictive assumption.
We say that a closed linear multivalued operator / is the generator of
a degenerate local k-times integrated semigroup V (t), t [0, T) if
V (t)x +
t
k
k !
x =
_
t
0
V (s)/xds, x D(/) (2.2.24)
and
V (t)x +
t
k
k !
x /
_
t
0
V (s)xds, x X. (2.2.25)
Theorem 2.2.2 Let / be a linear multivalued operator with (/) ,= .
Suppose that the decomposition
X = X
n+1
/
n+1
0 (2.2.26)
holds for some n N, where X
n+1
:= D(/
n+1
). Then the following state-
ments are equivalent:
(J) the Cauchy problem (IP) is well-posed, and its solution U T

0
(X
A
)
is degenerate for x /
n+1
0;
() for the resolvent of /, the polynomial estimate (2.2.10):
|R
A
()|
M[[
l
log(1 +[[)
M
1
[[
l
holds for some M > 0, l 0, and any
=
_
C

Re >
l

log(1 +[[) +
1

log
C

_
,
2001 CRC Press LLC
2001 CRC Press LLC
where
C > 0, 0 < < 1, > 0;
(o ) there exists a solution operator distribution S T

0
(L(X, X
A
)) such
that
(T S)x x, x X,
(2.2.27)
(S T)x = x, x D(/),
where T :=

I /. In this case
U = Sx, x X
and S is degenerate on /
n+1
0;
(1 ) for any T > 0, the operator / is the generator of a local k-times
integrated semigroup V (t), t [0, T) for some k = k(T) n,
which is degenerate on /
n+1
0 ;
(J k) for any T > 0, the local Cauchy problem (IP) with operator / having
nonempty resolvent set is k-well-posed on D(/
k+1
) for some k =
k(T) n.
Pro of (J) (o). As in the nondegenerate case in Section 2.1, this
proof is based on the equality that connects U T

0
(X
A
), the solution of
(IP), with the solution operator distribution S T

0
(L(X, X
A
)):
U, ) = S, )x = Sx, ), T, x X.
(o) (1). To prove (1) = (o) we dene S := V
(k)
. Conversely, as
in Theorem 2.2.1, we dene
V (t)x := S,
t,p
)x,
for x X
n+1
. This operator-function V () satises (2.2.24) and (2.2.25)
with k = p+2 on X
n+1
and D(/), respectively. On /
n+1
0, V ()x is dened
by formula (1.6.21).
(Jk) =() =(1) =(Jk). Let the local (IP) be k-well-posed on
D(/
k+1
). Then one can verify the operator family
V (t) := U
k
(t), t [0, T)
that was constructed in Theorem 1.6.6, satises the inclusion (2.2.25) with
the semigroup parameter k, for x X. Hence, due to the assumption
2001 CRC Press LLC
2001 CRC Press LLC
(/) ,= , for operators R(, ) and G() dened by formulae (2.2.18) and
(2.2.19), we have
R
A
() := R(, )(I G)
1
and the condition () holds with l = k.
Suppose () is fullled. Then the operator-function V
p
() dened by
(2.2.20):
V
p
(t) :=
1
2i
_

p
e
t
R
A
()d,
is continuous in t (,
l
), there
l
:=

l
(p l 1) and p > l +1 . Using
the abstract Cauchy theorem we show that V
p
satises (2.2.24) with k = p
on D(/):
_
t
0
V
p
(s)(/I)xds
=
_
t
0
_
_
1
2i
_

p+1
e
s
R
A
()x
p
e
s
x
_
d
_
_
ds
= V
p
(t)x
t
p
p!
x, t [0,
l
), x D(/).
Next, applying /I to
_
t
0
V
p
(s)xds, x X, we obtain (2.2.25) with k = p.
Hence, V
p
(t), t [0,
l
) is a local p-times integrated semigroup on X
(p > l + 1).
By Theorem 1.6.6, the Cauchy problem (IP) is p-well-posed on R
p+1
X
1
,
which is equal to R
p+1
X since p n.
2.2.3 Well-posedness in the space of
exponential distributions
From Theorems 2.2.1 and 2.1.6 we obtain the following necessary and suf-
cient conditions for the well-posedness of (DP) in the space o

(X).
Theorem 2.2.3 Let A, B : X Y be linear operators, A be closed and
invertible, B be bounded. Suppose that the decomposition (2.2.5) holds for
some n N. Then the following statements are equivalent:
(W
exp
) (DP) is -well-posed. Its solution belongs to T

0
(X

A
) for x
X
n+1
and is degenerate for x ker R
n+1
;
(S
exp
) there exists a solution operator distribution
S o

(L(X, X

A
)) o

(L(X
n+1
, X

A
))
2001 CRC Press LLC
2001 CRC Press LLC
such that
(P S)x = Bx, x X,
(S

T)x = x, x D(/),
In this case,
U = Sx, x X
and S is degenerate on ker R
n+1
;
(D
exp
) there exists a regular exponentially bounded distribution semigroup
Q with ranQ dense in X
n+1
, and degenerate on ker R
n+1
. Its gener-
ator is equal to

/. In this case S = Q;
(I
exp
)

/ is the generator of an exponentially bounded n-times integrated

semigroup V (t), t 0 on X
n+1
;
(R
exp
) the estimates (2.1.28) hold for the resolvent of

/;
(E
exp
) for any x D
n+1
there exists a unique solution of local (CP).
2.3 Ultradistributions and new distributions
In Section 2.1 we studied the Cauchy problem
u

(t) = Au(t), t 0, u(0) = x, (CP)

where A : D(A) X X, in spaces of abstract distributions. Recall that
the polynomial estimates for the resolvent of A in a certain logarithmic
region of the complex plane is the criterion for the well-posedness of (CP)
in the sense of distributions. This region and the resolvent estimates are
connected with the order of the corresponding distribution semigroup and
with the order of the solution operator distribution. Now, we consider the
Cauchy problem (CP) with an operator A having the resolvent in a certain
region smaller than a logarithmic region from Section 2.1. We investigate
the well-posedness of such a (CP) in spaces of ultradistributions. The
spaces of ultradistributions are the dual of spaces of innitely dierentiable
functions with a locally convex topology. These spaces are more general
than the space of Schwartz distributions. We show that the existence of
a solution operator ultradistribution is equivalent to exponential estimates
for the resolvent of A in . As it was proven in Theorem 1.3.3 the latter
fact is equivalent to the existence of a unique solution of the convoluted
equation
v(t)

= Av(t) + (t)x, v(0) = 0, (2.3.1)

where function is dened by the resolvent of A.
2001 CRC Press LLC
2001 CRC Press LLC
Spaces of test functions for ultradistributions consist of innitely dier-
entiable functions, and they are dened in terms of estimates on derivatives
of these functions. These estimates depend on some numerical sequence
M
n
. The same sequence allows one to estimate coecients of dierential
operators of innite order, which are called the ultradierential operators.
The spaces of ultradistributions are invariant under ultradierential op-
erators. Some properties of ultradistributions are considered in the rst
subsection. In the second subsection, the well-posedness of the (CP) in
the spaces of ultradistributions is investigated. The limit case for the (CP)
well-posed in the spaces of ultradistributions is the case of the Cauchy prob-
lem with an operator A having no regular points in a right semiplane (e.g.
the reversed Cauchy problem for the heat equation). It is proved in the
third subsection that such a (CP) is well-posed in spaces of new distribu-
tions. Spaces of test elements x X in the construction of spaces of new
distributions are dened in terms of the behaviour of A
n
x for any n.
2.3.1 Abstract ultradistributions
Let M
n
be a sequence of positive real numbers, satisfying the following
conditions:
(M.1) M
2
n
M
n1
M
n+1
, n = 1, 2, . . . ;
(M.2) , R :
M
n

n
min
0sn
M
s
M
ns
, n = 0, 1, . . . ;
(M.3)

s=1
M
s1
M
s
.
Let K be a compact set in R and h > 0. We consider a space of functions
C

(R) with support K and with the estimates

|
(n)
|
C(K)
CM
n
h
n
, n N, (2.3.2)
for some constant C. The space T
M
n
,h
K
of such functions with the norm
||
M
n
,h,K
:= sup
nN
|
(n)
|
C(K)
M
n
h
n
is a Banach space.
Denition 2.3.1 Let R be an open set. The space
T
(M
n
)
() = ind lim
K
proj lim
h0
T
M
n
,h
K
2001 CRC Press LLC
2001 CRC Press LLC
with the topology of inductive limit is called the space of ultradierentiable
functions with supports in of Beurling type or of class (M
n
). The dual
space T
(M
n
)

() is the space of ultradistributions of class (M

n
). If = R,
then we simply write T
(M
n
)
and T
(M
n
)

.
For more details see [135] and [153], where it is also shown that spaces
of Schwartz distributions are subspaces of spaces of ultradistributions.
Property (M.1) is called the logarithmic convexity. Property (M.2) guar-
antees the invariance of T
(M
n
)
() under dierential operators of innite
order:
P(D) =

n=0
a
n
D
n
(2.3.3)
with coecients satisfying the estimates
[a
n
[
C L
n
M
n
for some L > 0 and some C > 0. Such operators P(D) are called the
ultradierential operators of class (M
n
). Property (M.3) is called the non-
quasi-analyticity. As in the case of distributions, multiplication by any
function with estimates (2.3.2) can be dened for ultradistributions. We
also note that any ultradierential operator P(D) is continuous in the space
of ultradistributions.
For every sequence M
n
we dene the associated function
M() := sup
n
ln
[[
n
M
0
M
n
,
which is closely related to Laplace transforms of ultradistributions.
For M
n
satisfying (M.1) each of the conditions
_

0
M()

2
d < ,

n=1
1
M
1/n
n
<
is equivalent to (M.3).
Examples of such sequences are the following Gevrey sequences: (n!)
s
,
n
ns
and (1+ns) for s > 1, where is the Euler function. For such M
n
the
corresponding ultradierential polynomial P(z) has the following estimates
e
(l|z|)
1/s
[P(z)[ e
(L|z|)
1/s
, z C,
for some l, L > 0. The typical example of the ultradierential polynomial
of class (n!)
s
is
P(z) :=

n=1
_
1 +
lz
n
s
_
,
2001 CRC Press LLC
2001 CRC Press LLC
it satises the above estimates with L = 2l. Owing to the Denjoy-Carleman-
Mandelbrojt theorem [135] there are suciently many functions in
T
(M
n
)

().
The structure theorem for ultradistributions states that
f T
(M
n
)

()
if and only if on every relatively compact open set G , f can be written
in the form
f =

n=0
D
n
f
n
with measures f
n
on G satisfying the estimate
|f
n
|
C

(G)

C L
n
M
n
, n N,
for some constants L and C. Let K be a compact subset of R. We remind
here that C(K) is the space of all continuous functions in K equipped with
the supremum norm
|u|
C(K)
= sup
tK
[u(t)[.
Then C

(K), the dual to C(K), is the space of all nite Borel measures
dened on K. For f C

(K) dene the total variation [f[ by

[f[(G) = sup

j
[f(G
j
)[,
where G is any f-measurable set and supremum is taken over all possible
decompositions of G into a disjoint union of nite number of f-measurable
sets G
j
. Then [f[ C

(K) and we can dene a norm in C

(K) by
|f| = [f[(K).
The space C

(K) equipped with this norm is a Banach space.

Denition 2.3.2 Let X be a Banach space. The space T
(M
n
)

0
(X) of ab-
stract ultradistributions of class (M
n
) is the space L(T
(M
n
)
, X) of bounded
linear operators with supports in [0, ), equipped with the topology of uni-
form convergence on bounded sets from T
(M
n
)
.
We now dene the convolution of abstract ultradistributions. Let X
1
,
X
2
and X
3
be Banach spaces. Suppose that multiplication
(x
1
, x
2
) x
1
x
2
: X
1
X
2
X
3
, x
i
X
i
2001 CRC Press LLC
2001 CRC Press LLC
is dened. Consider T T
(M
n
)

0
(X
1
) and S T
(M
n
)

0
(X
2
). Then their
convolution
(T, S) T S : T
(M
n
)

0
(X
1
) T
(M
n
)

0
(X
2
) T
(M
n
)

0
(X
3
),
is dened by
T S, ) := S,

T ),
where T
(M
n
)
and (

T )() := (T )(). This mapping is bilinear

and is bounded [135]. In the particular case when T, S T
(M
n
)

0
(R) and
x
i
X
i
, the convolution of ultradistributions is dened by
(T x
1
) (S x
2
) = (T S) (x
1
x
2
).
2.3.2 The Cauchy problem in spaces of
abstract ultradistributions
Consider the Cauchy problem
u

(t) = Au(t), t 0, u(0) = x, (CP)

where A : D(A) X X. Denote
Y = [D(A)] :=
_
D(A), |x|
A
= |x| +|Ax|
_
.
Recall that Y is a Banach space. Furthermore, the operator A : Y X
belongs to L(Y, X), the space of bounded linear operators from Y to X.
Introduce
P :=

I
Y
A, (2.3.4)
then
P T
(M
n
)
0

(L(Y, X)).
Thus, in the space of ultradistributions T
(M
n
)

0
(X), the Cauchy problem
(CP) can be written in the form
P U = x,
where x X, U T
(M
n
)
0

(Y ) and Y = [D(A)].
In general, Y does not have to be equal to [D(A)]. Let X and Y be
Banach spaces such that Y X and A L(Y, X). Consider the Cauchy
problem
P U = x, x X, (uP)
where U T
(M
n
)
0

(Y ) and P T
(M
n
)
0

(L(Y, X)) is dened by (2.3.4).

This problem is referred to as the Cauchy problem (CP) in the space of
ultradistributions T
(M
n
)

0
(X).
2001 CRC Press LLC
2001 CRC Press LLC
Denition 2.3.3 The Cauchy problem (CP) is said to be well-posed in the
sense of ultradistributions (or (uP) is well-posed) if for any x X, there
exists a unique solution U T
(M
n
)
0

(Y ) such that for any x

n
0 in X,
corresponding solutions of (uP) U
n
0 in T
(M
n
)
0

(Y ).
The next theorem generalizes Theorems 2.1.3 2.1.5.
Theorem 2.3.1 Let X and Y be Banach spaces, Y X and A L(Y, X).
Then the following statements are equivalent:
(I) the Cauchy problem (CP) is wellposed in the sense of
ultradistributions;
(II) there exists a solution operator ultradistribution E T
(M
n
)
0

(L(X, Y ))
satisfying
P E = I
X
, E P = I
Y
; (2.3.5)
(III) for any > 0, there exists region
:=
_
C

Re M() +
_
such that for any ,
|R()|
L(X,Y )
Ce
M()
. (2.3.6)
for some constant C > 0.
Proof (I) = (II). We have that for any x X there exists a unique
solution U T
(M
n
)
0

(Y ). Dene the operator E() by

E()x := U(), T
(M
n
)
, x X.
Then Ex T
(M
n
)
0

(Y ), where Ex() = E()x. It follows from the well-

posedness of the Cauchy problem that for any sequence of initial values
x
n
X such that x
n
0, the corresponding sequence |E()x
n
|
Y
=
|U
n
()|
Y
converges to 0 uniformly in from a bounded set. Therefore,
E() L(X, Y ).
We now show that E T
(M
n
)
0

(L(X, Y )), i.e., E is a continuous linear

operator from T
(M
n
)
to L(X, Y ):

j
T
(M
n
)
,

j
0 =|E(
j
)|
L(X,Y )
0,
and suppE [0, ). Consider the set
B =
_
Ex

|x|
X
C
_
T
(M
n
)
0

(Y ).
2001 CRC Press LLC
2001 CRC Press LLC
From the stability of the solution we have
sup
ExB
|E()x|
Y
= sup
x
X
C
|U()|
Y
< ,
therefore B is a bounded set. As in the scalar case, the set B is bounded
in T
(M
n
)
0

(Y ) if and only if for any relatively compact open set G

there exist constants C
1
, L such that for any element Ex B there exist
measures f
n
satisfying such that
Ex

G
=

n=0
D
n
f
n
and
|f
n
|
C

{G, Y }

C
1
L
n
M
n
,
where C

_
G, Y
_
is the space of functions continuous on G with values in
Y . Hence, for Ex B and any
j
0, we have
|E(
j
)x|
Y
=
_
_
_
_

n=0
D
n
f
n
,
j
)
_
_
_
_
Y
=
_
_
_
_

n=0
f
n
, (1)
n

(n)
j
)
_
_
_
_
Y

n=0
|f
n
|
C

{G, Y }
|
(n)
j
|
C(G)
C
1

n=0
L
n
h
n
|
(n)
j
|
C(G)
M
n
h
n
.
Since
sup
n
|
(n)
j
|
C(G)
M
n
h
n
= |
j
|
M
n
,h,G
,
we obtain
|E(
j
)x|
Y
C
1
|
j
|
M
n
,h,G

n=0
L
n
h
n
.
This estimate holds for any h > 0, in particular for h =
1
2L
we have
|E(
j
)x|
Y
2C
1
|
j
|
M
n
,
1
2L
,G
0 as
j
0.
Thus, for
j
0, E(
j
)x tends to zero in Y uniformly in x from a bounded
set in X. Hence E() 0 in the norm of the space L(X, Y ).
Now we show that E satises equations (2.3.5). We have
P E, )x = (

I
Y
A) E, )x
= E

AE, )x = (E

AE)x, ).
2001 CRC Press LLC
2001 CRC Press LLC
For E T
(M
n
)
0

(L(X, Y )) we have the equality E

x = (Ex)

, therefore
(Ex)

AEx, ) = U

AU, ) = P U, )
= x, ) = I
X
, )x.
The second equation:
(P E

)x = P (E

x) = P U

= U

I
X
U

A
= (P Ex)

x, x X,
(P E)Ax = Ax, x Y,
P ( x) =

x Ax, x Y.
For x Y we have the equality P (E

x EAx x) = 0. Hence,
E

EA = I
Y
, that is E P = I
Y
.
(II) = (I). For E T
(M
n
)
0

(L(X, Y )) we consider U := Ex, x X.

Then U belongs to T
(M
n
)
0

(Y ) and satises the equation:

P U = P Ex = x.
Let x
j
X be a sequence of initial data converging to zero. We now show
that the corresponding sequence U
j
= Ex
j
tends to zero in T
(M
n
)
0

(Y ), i.e.,
converges uniformly in from any bounded set / in T
(M
n
)
. Let
/ =
_
T
(M
n
)

supp K, ||
M
n
,h,K
< C
_
for some compact set K R. Since E() is a bounded linear operator, we
have
|U
j
()|
Y
= |Ex
j
()|
Y
|E()|
L(X,Y )
|x
j
|
X
.
By the structure theorem, for any E T
(M
n
)
0

(L(X, Y )) and any rel-

atively compact open set G R with G = K,there exist measures
f
n
C

_
K, L(X, Y )
_
such that for some constants L, C
1
> 0
E

G
=

n=0
D
n
f
n
,
where
|f
n
|
C

{K, L(X,Y )}
C
1
L
n
M
n
.
Using the boundedness of test functions we have
|E()|
L(X,Y )
2001 CRC Press LLC
2001 CRC Press LLC
=
_
_
_

n=0
D
n
f
n
, )
_
_
_
L(X,Y )
=
_
_
_

n=0
f
n
,
(n)
)
_
_
_
L(X,Y )

n=0
_
_
f
n
,
(n)
)
_
_
L(X,Y )

n=0
|f
n
|
C

{K, L(X,Y )}
|
(n)
|
C(K)
C
1

n=0
L
n
M
n
|
(n)
|
C(K)
C
1
C

n=0
L
n
h
n
.
Let C
1
C

n=0
L
n
h
n
= C
2
for h =
1
2L
, then
|U
j
()|
Y
|E()|
L(X,Y )
|x
j
|
X
C
2
|x
j
|
X
.
Hence |U
j
()|
Y
tends to zero uniformly with respect to /.
Now we show the uniqueness of the solution. Let P V = x. Then
E P V = E ( x) = Ex = U T
(M
n
)
0

(Y ).
Using the second equation in (2.3.5) we obtain
U = E P V = ( I
Y
) V = V T
(M
n
)
0

(Y ).
(II) = (III). For 0 < c < c

we consider the local unity function in

the space T
(M
n
)
:
(t) =
_
_
_
0, t < 1
1, 0 t c
0, t > c

.
For arbitrary C we denote

(t) = e
t
(t) and consider the rst
equation in (2.3.5) for this test function
(P E)(

) = E

) AE(

) = I
X

(0) = I
X
. (2.3.7)
From the linearity of E() we have
E

) = E([

(t)]

) = E(e
t
(t) +e
t

(t))
= E(

) E(e
t

(t)).
Hence, from (12.3.7) we obtain
(I A)E(

) = I
X
+E(e
t

(t)). (2.3.8)
To show the existence of a resolvent, we estimate the function e
t
(t) for
any test function T
(M
n
)
and then we estimate the operator-functions
E(e
t
(t)) and E(e
t

(t)). Let supp R

+
K = [a, b], then
D
p
(e
t
(t)) = e
t
p

j=0
C
p
j
()
pj

(j)
(t),
2001 CRC Press LLC
2001 CRC Press LLC
and
sup
tK
[D
p
(e
t
(t))[
e
aRe
p

j=0
sup
tK
[
(j)
(t)[
M
j
L
j
C
p
j
[[
pj
L
pj
M
j
L
p
||
M
n
,L,K
e
aRe
p

j=0
sup
pj
[[
pj
L
pj
M
pj
L
p
C
p
j
M
j
M
pj
.
From the equality e
M()
= sup
n
||
n
M
n
and the property of logarithmic con-
vexity: M
j
M
pj
M
p
, we have
sup
tK
[D
p
(e
t
(t))[ ||
M
n
,L,K
e
M(/L)aRe
L
p
p

j=0
C
p
j
M
j
M
pj
||
M
n
,L,K
e
M(/L)aRe
L
p
2
p
M
p
.
Hence for any T
(M
n
)
|e
t
(t)|
M
n
,2L,K
||
M
n
,L,K
e
M(/L)aRe
. (2.3.9)
From the structure theorem, for E T
(M
n
)
0

(L(X, Y )) and any relatively

compact set G [0, ) there exist f
n
C

_
G, L(X, Y )
_
and constants
, C > 0 such that
E

G
=

n=0
D
n
f
n
,
where
|f
n
|
C

{G, L(X,Y )}
C

n
M
n
and K = G.
Hence
_
_
_E(e
t
(t))
_
_
_
L(X,Y )
=
_
_
_

n=0
D
n
f
n
, e
t
(t))
_
_
_
L(X,Y )

n=0
|f
n
|
C

{G, L(X,Y )}
_
_
D
n
(e
t
(t))
_
_
C(K)
C

n=0

n
|D
n
e
t
(t)|
C(K)
L
n
M
n
L
n
2C|e
t

(t)|
M
n
,L,K

n=0
L
n

n
.
2001 CRC Press LLC
2001 CRC Press LLC
Since K = [0, c

] for , and K = [c, c

] for

, from (2.3.9) we obtain for

L =
1
2
:
|E(e
t

(t))|
L(X,Y )
2Ce
M(2)cRe
|

|
M
n
,
1
4
,K
,
|E(e
t
(t))|
L(X,Y )
2Ce
M(2)
||
M
n
,
1
4
,K
.
From (2.3.8) and the equality 2Ce
M(2)cRe
= < 1 we obtain that the
operator I A has the inverse on
:=
_
C

Re 1/c[M(2) ln(/2C)] =: M() +

_
,
and the resolvent estimate there:
|R
A
()|
2C
1
e
M()
||
M
n
,
1
2
,K
.
Thus
> 0 , , C > 0 such that
, |R

| Ce
M()
.
(III) = (II). For T
(M
n
)
we consider the operator
E() :=
1
2i
_

R
A
()() d,
where () =
_
R
e
t
(t) dt. From (2.3.6) we have
|R
A
()| Ce
1

(Re )
, . (2.3.10)
Let supp K = [a, b], then
[()[ =

(1)
n

n
_
K=[a,b]
e
(t)

(n)
(t) dt

sup
n
|
(n)
|
C
K
M
n
h
n

M
n
h
n
[[
n
_
K
e
(Re t)
dt
= ||
M
n
,h,K
M
n
h
n
[[
n
e
(bRe )
e
(aRe )
Re
, h > 0.
Since this inequality holds for any n, it is valid for
inf
n
M
n
h
n
[[
n
= e
M(/h)
,
2001 CRC Press LLC
2001 CRC Press LLC
therefore
[()[ C
1
||
M
n
,h,K
e
bRe M(/h)
. (2.3.11)
Since
M(/h) =
Re

(
1
h
)
on , from (2.3.10), (2.3.11) we have
|E()|
L(X,Y )

1
2
_

|R
A
()|
L(X,Y )
[()[[d[

1
2
_

CC
1
e
1/(Re )+bRe M(/h)
||
M
n
,h,K
[d[
=
_

C
2
e
(
1

+b
1
h
)Re
||
M
n
,h,K
[d[.
For any , , b we can take h such that
1

+b
1
h
< 0, hence, the operator
E() is dened. We now show that it satises (2.3.5). We have
P E, ) = (

I
Y
A) E, ) = E

AE, )
= I
Y
E,

) AE, ).
Integrating by parts and taking into account that R
A
() is the resolvent of
A we obtain
P E, ) =
1
2i
_

R
A
()
_
R
e
t

(t) dt d AE()
= I
Y
1
2i
_

_
K
e
t
(t) dt d,
supp K. It follows from (2.3.10) that we can continuously deform
contour into the imaginary axis
P E, ) = I
Y
1
2i
_
+i
i
_
K
e
t
(t) dt d
= I
Y
1
2i
_
+i
i
_
K
e
t

(t) dt d.
Using the inverse Laplace transform of the identity in the space of ultra-
distributions
L
1
(1), ) =
_
1
2i
_
+i
i
e
t

2
d,

_
= , ),
2001 CRC Press LLC
2001 CRC Press LLC
we have
P E, ) = I
Y
, ).
Corollary 2.3.1 By Theorem 1.3.3, each statement (I), (II) and (III) of
Theorem 2.3.1 is equivalent to the existence of -convoluted semigroup,
where () =
||
O(e
M()
).
2.3.3 The Cauchy problem in spaces of
new distributions
We continue the study of the generalized well-posedness of the Cauchy
problem (CP):
u

(t) = Au(t), t 0, u(0) = x. (CP)

In Section 2.1 we considered the (CP) with an operator A generating an
n-times integrated semigroup. Such a (CP) is well-posed in the space of
Schwartz distributions and can be referred to as slightly ill-posed problem.
In this subsection, we study the generalized well-posedness of essentially
ill-posed Cauchy problems, namely problems with operators A generating
C-regularized semigroups. We note that Cauchy problems that are well-
posed in the space of ultradistributions, i.e., problems with operators A
generating -convoluted semigroups, can be referred to as intermediate case
between slightly and essentially ill-posed Cauchy problems.
In this subsection, we use spaces of new distributions (or new general-
ized functions) introduced by V.K. Ivanov. The construction of spaces of
new distributions is based on the development of Sobolevs and Schwartzs
concepts: instead of dierentiation operators applied to test functions we
take an unbounded operator : X X giving a solution of an ill-posed
problem. The test functions X are constructed in terms of their Fourier
coecients in the case Hilbert space X, and as functions from D(
k
) or

k=1
D(
k
) in the general case.
In a Hilbert space H we consider a self-adjoint operator generating
an orthonormal basis of eigenvectors
k
corresponding to the eigenvalues
[
1
[ [
2
[ . We denote by H the class of such operators. The domain
D() of such an operator is dense in H, but generally does not coincide
with H. The same holds for D(
k
). We introduce the normed spaces
T
k
=
_
u D(
k
) H, |u|

k =
k

i=0
|
i
u|
H
_
, k = 1, 2, ...
and the countably normed space
T

=
_
u

k=1
D(
k
) H, |u|

k, k 1
_
.
2001 CRC Press LLC
2001 CRC Press LLC
The convergence of a sequence u
n
u in T

means that
k
u
n

k
u in
H for all k 1. Note that T

is not empty since it contains all eigenvectors

of operator .
Spaces of new distributions are dened as the dual spaces T

k
(T

).
The space T

is the generalization of the Schwartz space of test functions

c. The space T

is the generalization of the space c

of distributions with
compact support. Spaces T
k
generalize the Sobolev spaces. If is equal to
the dierentiation operator dened on H = L
2
(G), then T
k
coincides with
W
k,2
(G) H
k
(G), and T

k
coincides with H
k
(G). This is the reason that
we denote T

k
by T
k
.
Later on we use the sequences of values of functionals generated by
elements T
k
and f T
k
on eigenvectors
n
:
n
= ,
n
) = (,
n
)
and

f
n
= f,
n
). Here
n
and

f
n
are analogous to the Fourier coecients.
The inverse transforms have the form:
=

n=1

n

n
, f =

n=1

f
n

n
. (2.3.12)
It follows from the denition of T
k
that the series for T
k
is convergent
in T
k
. Since for any T
k
,
lim
m
_
m

n=1

f
n

n
,
_
= lim
m
_
m

n=1

f
n

n
,

n=1

n

n
_
= lim
m
m

n=1

f
n
, = lim
m
_
f,
m

n=1

n

n
,
_
= f, ),
the series for f T
k
is convergent in T
k
. This implies that spaces T
k
are dense in H, and H is dense in T
k
in the sense of weak topology.
Now we construct spaces of test functions and new distributions dened
in the form of Fourier series, via the behaviour of the coecients of the
series. Such spaces can coincide with those already introduced, T
k
and
T

, but both these approaches turn out to be useful for the investigation
of generalized well-posedness.
Let
k
with [
1
[ [
2
[ . . . , be some given numerical sequence, and
let
n
be an orthonormal system in H. We consider the space of test
functions:

:=
_
H

k 1,

n=1
[
n
[
2
[
n
[
2k
<
_
with the countable system of norms
||
J
k
=
_

n=1
[
n
[
2
[
n
[
2k
_
1/2
, (2.3.13)
2001 CRC Press LLC
2001 CRC Press LLC
and consider the corresponding space

of new distributions f
dened in the form of series (2.3.12) whose coecients are subject to the
conditions
k
0
1 :

n=1
[

f
n
[
2
[
n
[
2k
0
< .
We also dene the spaces
k
and
k
such that for =

n=1

n

n

k
and f =

n=1

f
n

n

k
the coecients
n
and

f
n
are subject to the
conditions

n=1
[
n
[
2
[
n
[
2k
< and

n=1
[

f
n
[
2
[
n
[
2k
< ,
and the norm in
k
is dened by (2.3.13). Then

k=1

k
,

_
k=1

k
.
If we take the system of eigenvectors of an operator H as the orthonor-
mal system
n
, and the sequence of eigenvalues
n
as the sequence
n
,
then the spaces

and
k
coincide with T

and T
k
, respectively. The
spaces
k
with
n
chosen in this way are used for studying the gener-
alized well-posedness of the problem of evaluating of unbounded operator
: u = f. If there exists
1
= , then this problem is equivalent to
the operator equation of the rst kind
u = f. (2.3.14)
If
n
is taken to be a sequence of the form F(
n
), where the
n
are the
eigenvalues of an operator A H and F is an analytic function, then
k
co-
incides with the space T
k
generated by the operator = F(A). Such spaces
are used in the study of generalized well-posedness of dierent boundary
value problems for dierential-operator equations: if the initial data are
taken in T
k
(
k
), k = 0, 1, . . . (in particular, in H = T
0
=
0
), then
the existence, the uniqueness, and the stability of a solution can be estab-
lished in the spaces T
m
(
m
) for m > k. If initial data are taken in
T

), then the well-posedness can be proved in the same spaces.

Let H
+
be the subclass of operators from H with eigenvalues
k
such
that
1

2
. We now investigate the generalized well-posedness of
the following problems:
1) the ill-posed Cauchy problem (CP) with an operator A H
+
and
= e
At
,
2) the problem of evaluating of unbounded operator written in the
form (2.3.14) with =
1
.
2001 CRC Press LLC
2001 CRC Press LLC
In the next chapter we show that these essentially ill-posed Cauchy
problems can be solved by dierential regularization methods. Writ-
ing them in the form (2.3.14) one can use the variational regularization
methods.
Theorem 2.3.2 Suppose that =
1
H. Then for any
f T
k
(T

), k = 1, 2, . . . , the solution of the equation

u, ) = u, ) = f, ), T
k
(2.3.15)
exists, is unique, and is stable in the space T
k1
(T

). For this solution

the following expansion holds:
u =

n=1

n

f
n

n
. (2.3.16)
Proof We show that for any f T
k
the formal solution f of (2.3.15)
belongs to T
k1
. According to the denition of the space T
k
, if T
k
,
then T
k1
. Since f, ) = f, ) for f T
k
, and the operator
maps T
k+1
to T
k
and T
k
to T
k1
for k = 0, 1, . . ., we have f T
(k+1)
.
As =
1
: T
k
T
k+1
, for the unbounded operator we have that
f =: u satises the equation
T
k
,
u, ) = u, ) = f, ), k = 0, 1, . . . .
We now show that the expansion (2.3.16) holds. According to (2.3.15),
a function u can be expanded in the series u =

n=1
u
n

n
in the space
T
k1
, where u
n
satises the equalities
u
n
= u,
n
) = f,
n
) = f,
n
) =
n
f,
n
) =
n

f
n
.
This implies (2.3.16). It remains to verify the stability of the solution
in the space T
k1
, which is equivalent to the continuity of the operator
: T
k
T
k1
. Let f
n
f in T
k
, i.e.
T
k
, lim
n
f
n
, ) = f, ).
Then for any T
k
,
u
n
, ) = f
n
, ) = f
n
, ) f, ) = f, ) = u, ),
i.e. u
n
u in T
k1
.
Consider A H
+
with eigenvalues
n
. Let T < . By
exp
k
we denote

k
with
n
= e

n
T
. We now establish the well-posedness of the local
Cauchy problem
u

(t) = Au(t), 0 t < T, u(0) = x, (LCP)

in these spaces of new distributions.
2001 CRC Press LLC
2001 CRC Press LLC
Theorem 2.3.3 Suppose that A H
+
. Then for any x
exp
k
(
exp

)
the solution of (LCP) exists, is unique, and is stable in the space
exp
k1
(
exp

):

exp
k+1
,
_
du(t)
dt
,
_
= Au(t), ), u(0), ) = x, ),
here
u(t), ) =
_

n=1
e

n
t
x
n

n
,
n
_
=

n=1
e

n
t
x
n

n
.
The operator A considered in Theorem 2.3.3 generates the local C-
regularized semigroup
S(t)x =

n=1
e

n
(tT)
x
n

n
, with Cx :=

n=1
e

n
T
x
n

n
,
and the space
exp
k
coincides with T
k
, where = C
1
. It follows from
the construction of
exp
k
and the denition of C-semigroups that Theorem
2.3.3 can be extended in terms of C-regularized semigroups in the following
way.
Theorem 2.3.4 Suppose that an operator A H
+
is the generator of a C-
regularized semigroup. Then for any x T
k
, where = C
1
, a solution
of (LCP) exists, is unique, and is stable in the space T
k1
. In the space
T

the (LCP) is well-posed in the Hadamard sense: for any x T

,
the solution exists, is unique, and is stable in T

.
The unbounded operators U(t) giving the weak solution of the (LCP), are
dened by
U(t)x := C
1
S(t)x, x X
They satisfy the equations
T
k
, k 1,
U(t +s)x, ) = U(t)U(s)x, ), U(0)x, ) = x, ),
i.e., they form a semigroup in the space of new distributions.
2001 CRC Press LLC
2001 CRC Press LLC
Chapter 3
Regularization Methods
In the rst two chapters we investigated the Cauchy problem
u

(t) = Au(t), t 0, u(0) = x, (CP)

where A : D(A) X X, which is not uniformly well-posed. The
technique of integrated and C-regularized semigroups presented in Chapter
1 allows one to construct a solution of (LCP) for initial values x from
various subsets of D(A) stable in X with respect to x in corresponding
graph-norms. The technique of distributions presented in Chapter 2 allows
the construction of a generalized solution for any x X stable in a space
of distributions.
In this chapter, we consider regularizing operators that allow the con-
struction of an approximate solution stable in X for concordant parame-
ters: - the error of initial data and the regularization parameter . In
Section 3.1, we study three main dierential regularization methods for
an ill-posed Cauchy problem: Lattes-Lions quasi-reversibility method, the
auxiliary boundary conditions (ABC) method, and the method of reduc-
tion to a Dirichlet problem (Carassos method). In Section 3.2 we study
the connections between dierent regularization methods and between reg-
ularization methods and the C-regularized semigroup method.
3.1 The ill-posed Cauchy problem
3.1.1 Quasi-reversibility method
Let X be a Banach space. Consider the local Cauchy problem
u

(t) = Au(t), 0 t < T, u(0) = x, (LCP)

2001 CRC Press LLC
2001 CRC Press LLC
with a densely dened linear operator A. Suppose that spectrum of A lies
in the region
G
1
=
_
C

[arg [ < <

4
_
,
and
M > 0, / G
1
,
|R
A
()|
M
1 +[[
. (3.1.1)
Denote the class of such operators by /
1
. In a general case, under the
assumption (3.1.1) the problem (LCP) is ill-posed: its solution does not
exist for every initial data from D(A) and it is not stable in any graph-
norm. We can say that such problems are essentially ill-posed, in contrast
to n-well-posed problems from the rst two chapters, which can be referred
to as slightly ill-posed problems. Essentially ill-posed problems can be
treated with the help of regularization methods.
Suppose, as it is customary in the theory of ill-posed problems, that
for some exactly given initial value x there exists a solution of (LCP) u()
(it is unique due to the estimate (3.1.1)) and the initial value x

is given
with an error : |x

x| . We show that a regularizing operator for

(LCP) with A /
1
can be constructed by the quasi-reversibility method.
The following denition of a regularizing operator for the ill-posed Cauchy
problem is given according to the general denition of regularizing operators
in the theory of ill-posed problems [129], [258].
Denition 3.1.1 An operator R
,t
: X C
_
[0, T], X
_
is called the regu-
larizing operator for the Cauchy problem (LCP) if the following conditions
are fullled:
1) operator R
,t
is dened for all x X. (Usually R
,t
supposed to be
bounded).
2) there exists a dependence = () (()
0
0) such that
|R
(),t
x

u(t)|
0
0.
Consider the Cauchy problem corresponding to the quasi-reversibility
method
u

,
(t) = (AA
2
)u
,
(t) =: A

u
,
(t), 0 < t T, (3.1.2)
u
,
(0) = x

.
We now show that there exist the unique generalized solution and a depen-
dence = () (()
0
0) such that for any x

X the generalized
2001 CRC Press LLC
2001 CRC Press LLC
solution of the Cauchy problem (3.1.2) converges to the solution of (LCP)
when 0:
|u
(),
(t) u(t)|
0
0, 0 < t T.
First we prove the well-posedness of the Cauchy problem (3.1.2) and then
nd its solution.
Theorem 3.1.1 Let A /
1
, then the Cauchy problem (3.1.2) is well-posed
on D(A
2
).
Proof Consider the identity
A

I =
_
A
1 +

1 4
2
I
__
A
1

1 4
2
I
_
=: (A
1
I)(A
2
I).
From the estimate (3.1.1), for R
A
(
j
), j = 1, 2, we obtain the corresponding
estimate for R
A

() at points =
j

2
j
, j = 1, 2,:
|R
A

()| (3.1.3)

|R
A
(
1
)||R
A
(
2
)|
1

M
2
(1 +[
1
[)
1
(1 +[
2
[)
1
=
1

M
2
_
1 +

1 +

1 4
2

_
1
_
1 +

1 4
2

_
1

M
2
_
1 +
[[

_
1
M
2
[[
1
.
If
1
and
2
are lying on the left of the curve
=
_
C

= t(1 i tan ), 0 t <

_
,
then the estimate (3.1.3) for R
A

() holds for lying on the right of the

curve

1
=
_
=
2
,

= t (1 tan
2
)t
2
+i(t tan 2t
2
tan )
_
.
In particular, for
Re > C() = max
0t<
_
t (1 tan
2
) t
2
_
=
1
4(1 tan
2
)
the following condition is fullled
2001 CRC Press LLC
2001 CRC Press LLC
L = M
2
> 0 :
, Re > C() =|R(, A
e
)| L[[
1
. (3.1.4)
As C() 0, for Re > C() we have [ C()[ < [[ and
|R
A

()| L[ C()[
1
.
Therefore, the operator A

generates an analytic C
0
-semigroup V
A

(), and
the Cauchy problem (3.1.2) is well-posed on D(A

) = D(A
2
).
According to Theorem 1.1.2, for x

D(A
2

) = D(A
4
) the solution of
(3.1.2) equal to V
A

can be written in the integral form:

u
,
(t) = v.p.
1
2i
_
+i
i
e
t
R
A

()x

d, (3.1.5)
t > 0, > C().
We substitude the resolvent R
A

() in the form of the Cauchy-Dunford

integral:
R
A

() =
1
2i
_

R
A
()
(
2
)
d, = G
1
into (3.1.5) with t = T. For x

D(A
4
) we obtain the solution of (3.1.2)
u
,
(T) =
1
2i
_

e
(
2
)T
R
A
()x

d. (3.1.6)
The integral (3.1.6), giving for x

D(A
4
) the classical solution of problem
(3.1.2) at t = T, converges for all x

X. Note that D(A

4
) = X. Ex-
tending (3.1.6) by continuity from D(A
4
) to X, we obtain the generalized
solution of problem (3.1.2) equal to V
A

(t)x

. We dene the operator R

,T
as the operator giving the generalized solution of (3.1.2) at t = T:
R
,T
x

:= V
A

(T)x

=
1
2i
_

e
(
2
)T
R
A
()x

d.
We now show that for R
,T
dened on the whole space X, there exists a
dependence
= (), with ()
0
0
such that
R
,T
x

0
u(T),
i.e. R
,T
is the regularizing operator for (LCP) at t = T.
Theorem 3.1.2 Let A /
1
. Suppose that for some x there exists a solu-
tion of the Cauchy problem (LCP). Then R
,T
is the regularizing operator
for (LCP) at t = T.
2001 CRC Press LLC
2001 CRC Press LLC
Proof Consider supremum of the dierence |R
,T
x

u(T)| for all x

from -neighbourhood of x:

Q
(A, u(T), , )
= = sup
xx

|R
,T
x

u(T)|
|u

(T) u(T)| + sup

xx

|R
,T
x

(T)|
=:
1
+
2
,
here u

(T) is the generalized solution of problem (3.1.2) with the initial

data x: u

(T) = R
,T
x = V
A

(T)x. Since V
A

(t) is a C
0
-semigroup, then
|V
A

(t)| Le
C()t
, t 0. (3.1.7)
Hence we obtain

2
= sup
xx

|V
A

(T)(x x

)| (3.1.8)
Le
C()T
= Le
T/4(1tan
2
)
.
Now we estimate
1
= |u

(T) u(T)|. First we show that u

(T) =
V
A
2(T)u(T). Due to condition (3.1.1) the Cauchy problem
w

(t) = Aw(t), w(0) = w

0
, (3.1.9)
is well-posed. Its solution at t = T for w = u(T) can be written in the form
of the contour integral:
x = V
A
(T)u(T) =
_

e
T
R
A
()u(T)d, (3.1.10)
where contour

is lying on the right of contour and for all

with suciently large absolute values, [arg()[ < /4. This choice of

guarantees the convergence of integral in (3.1.10). From equality (3.1.10)

and estimate (3.1.3) we have
u

(T) = V
A

(T)x
=
1
2i
_

e
(
2
)T
R
A
()xd
=
1
2i
_

e
(+
2
)T
R
A
()xd
=
1
2i
_

e
(+
2
)T
R
A
()

1
2i
_

e
T
R
A
()u(T)d
_
d.
2001 CRC Press LLC
2001 CRC Press LLC
Using the resolvent identity, the Cauchy formula, and the analyticity of
functions e
T
, e
(+
2
)T
in the region C G
1
, we obtain
u

(T) =
1
2i
_

e
(+
2
)T

1
2i
_

e
T
R
A
() R
A
()

u(T)d
_
d
=
1
2i
_

2
T
R
A
()u(T)d
=
1
2i
_

2
T
[R
A
() + 2R
A
2(
2
)]u(T)d
=
1
2i
_

2
T
R
A
2(
2
)u(T)2d
=
1
2i
_

2
e
T
R
A
2()u(T)d,
where

2
=
_
C

=
2
,
_
.
Therefore
u

(T) = V
A
2(T)u(T),
where V
A
2 is the analytic C
0
-semigroup generated by A
2
. The ana-
lyticity of V
A
2 may be proved similarly to the analyticity of the semi-
group V
A

. Hence, using the properties of integrals of a C

0
-semigroup, for
u(T) D(A
4
) we obtain

1
= |V
A
2(T)u(T) u(T)|
=
_
_
_
_
_
_
T
0
V
A
2(s)A
2
u(T)ds
_
_
_
_
_
L
1
, (3.1.11)
and
Le
T/4(1tan
2
)
+L
1
. (3.1.12)
It follows from (3.1.12) that we can choose = () so that 0 as
0. Hence, if we eliminate the additional condition u(T) D(A
4
), it
will be proved that R
,T
is the regularizing operator for problem (LCP)
at t = T. Note that, in general, the condition u(T) D(A
4
) used in the
estimate (3.1.11) is not fullled. If u(T) / D(A
4
), then due to the density
of D(A
4
) and the well-posedness of the Cauchy problem (3.1.9), there exists
w
0
D(A
4
) in a neighbourhood of u(T) such that for the solution of (3.1.9)
with w
0
= w
1
, the estimate |xw(T)| holds. Now if we consider (LCP)
with the initial data w(T), then
u(T) = w
0
D(A
4
), |w(T) x

| |w(T) x| +|x x

| 2,
2001 CRC Press LLC
2001 CRC Press LLC
and the order of the obtained estimate (3.1.12) is unchanged.
3.1.2 Auxiliary bounded conditions (ABC) method
Consider the ill-posed Cauchy problem (LCP) with a densely dened linear
operator A such that the spectrum of A lies in the half-strip
G
2
=
_
C

[Im[ < <

T +
, Re > , R, > 0
_
,
and suppose that for / G
2
the resolvent of A is bounded:
M > 0 : / G
2
, |R
A
()| M. (3.1.13)
Denote the class of such operators by /
2
. Suppose for some x there exists a
solution of (LCP) u(). We show that (LCP) with A /
2
can be regularized
by the ABC method: i.e. there exists a dependence = () (()
0
0)
such that the generalized solution of the boundary problem
u

,
(t) = A u
,
(t), 0 < t < T +, > 0, (3.1.14)
u
,
(0) + u
,
(T +) = x

, > 0,
converges to the solution of (LCP):
| u
,
(t) u(t)|
0
0, 0 t < T +.
In [128] we investigated the well-posedness of the boundary value problem
u

(t) = Au(t), 0 < t < T, u(0) u(T) = x, (3.1.15)

in a Banach space X. It was proved that if a densely dened operator A
satises the condition
K > 0, a region G C, and an integer k 1:
G, |R
A
()| K(1 +[[)
k
,
and / exp (TG), then there exists a unique solution of (3.1.15). The
assumptions of this theorem are fullled for A /
2
, x

D(A
2
), and any
positive . Therefore, the problem (3.1.14) for x

D(A
2
) has a unique
solution that can be written in the integral form:
u(t) =
1
2i
_

e
t
1 +e
(T+)
R
A
()x

d, = G
2
. (3.1.16)
2001 CRC Press LLC
2001 CRC Press LLC
The integral in (3.1.16) converges for all x

X and 0 t < T + .
Extending by continuity the equality (3.1.16) to the whole space X, we
dene a generalized solution of (3.1.14)

R
,T
x

:=
1
2i
_

e
T
1 +e
(T+)
R
A
()x

d, x

X,
and show that

R
,T
is a regularizing operator for (LCP) at t = T.
Theorem 3.1.3 Let A /
2
. Suppose that for some x there exists a solu-
tion of (LCP) for 0 t < T
1
, where T
1
> T. Then

R
,T
is the regularizing
operator for (LCP) at t = T.
Proof Consider |

R
,T
x

u(T)|. Let

A
(A, u(T), , )
= = sup
xx

R
,T
x

u(T)|
|

R
,T
x

u(T)| + sup
xx

R
,T
u
0

R
,T
x

|
=:
1
+
2
.
Here

R
,T
x = u

(T) is a solution of (3.1.14) at t = T with the boundary

condition u

(0) +u

(T +) = x, < T
1
T. We have

2
=
2
(T)
= sup
xx

_
_
_
_
1
2i
_

e
T
1 +e
(T+)
R
A
()(x x

)d
_
_
_
_

2
__

e
T
1 +e
(T+)

|R
A
()|[d[
+
_

[e
T
[|R
A
()|[d[
_

C
1
(T),
where

+
=
_

Re > 0
_
and

=
_

Re 0
_
.
To estimate
1
=
1
(T) we introduce the function v

() := u

() u().
Then v

() is a solution of problem (3.1.14) with the boundary condition

v

(0) +v

(T +) = u(T +).
2001 CRC Press LLC
2001 CRC Press LLC
As was shown in Theorem 3.1.2, we can assume without loss of generality
that u(T +) D(A
2
). Therefore,

1
(T) = |v

(T)| (3.1.17)
=
_
_
_
_
1
2i
_

e
T
1 +e
(T+)
R
A
()u(T +)d
_
_
_
_
.
We divide into three parts:

=
_

Re < 0
_
,

N
=
_

Re > N
_
,

0
= (

N
),
and obtain estimates of integral (3.1.17) along each part of :
_
_
_
_
_
1
2i
_

e
T
1 +e
(T+)
R
A
()u(T +)d
_
_
_
_
_
(3.1.18)

2
_

[e
T
[ |R
A
()| |u(T +)| [d[ C
2
,
_
_
_
_
1
2i
_

N
e
T
1 +e
(T+)
R
A
()u(T +)d
_
_
_
_
(3.1.19)

1
2
_

N
[e

[ |R
A
()| |u(T +)| [d[ C
3
e
N
.
Taking into account that
max
0<Re N

e
T
1 +e
(T+)

=
T
T
T+
()

T+
T +
,
for the integral along
0
we obtain
_
_
_
_
1
2i
_

0
e
T
1 +e
(T+)
R
A
()u(T +)d
_
_
_
_
(3.1.20)

1
2
_

0
T
T
T+
()

T+
T
|R
A
()| |u(T +)| [d[
C
4
N

T+
.
From (3.1.18) (3.1.20) we obtain the estimate for
1
(T):

1
(T) C
2
+C
3
e
N
+C
4
N

T+
.
2001 CRC Press LLC
2001 CRC Press LLC
Choosing N =
1
T+
ln
1

, we have

1
(T)
_
C
2

T
T+
+C
3
+C
4
1
T +
ln
1

T+
.
Finally, we obtain
(T)
_
C
2

T
T+
+C
3
+C
4
1
T +
ln
1

T+
+C
1
(T)

. (3.1.21)
Therefore, we can choose = () such that (T) 0 as 0, i.e.

R
,T
is the regularizing operator of (LCP) at t = T.
3.1.3 Carassos method
Consider the ill-posed Cauchy problem (LCP) with a densely dened linear
operator A such that the spectrum of A lies in a half-strip
G
3
=
_
C

[Im[ < b, Re > a

_
,
and suppose that for / G
3
the resolvent of A is bounded:
M > 0 : / G
3
, |R
A
()| M.
Denote the class of such operators by /
3
. Suppose that for some x there
exists a solution of (LCP) such that |u(T)| K for some K > 0, and the
initial value x

is given with an error . We show that (LCP) with A /

3
can be regularized by reduction to a Dirichlet problem.
Using the substitution w(t) = e
k(Tt)
u(t) for some k > 0 and dierenti-
ating both sides of the equation in (LCP), we obtain the Dirichlet problem
w

(t) = (AkI)
2
w(t), 0 < t < T, (3.1.22)
w(0) = e
kT
x, |w(T)| K.
We show that if w is the solution of (3.1.22) with boundary conditions
w(0) = e
kT
x

, w(T) = 0,
then w(t)e
k(Tt)
converges to u(t) as 0, where k = k() =
1
T
ln(K/).
This method of construction of regularizing operator R
k,t
by reduction to
the Dirichlet problem
w

k
(t) = (AkI)
2
w
k
(t), 0 < t < T, (3.1.23)
w
k
(0) = e
kT
x

, w
k
(T) = 0,
we call Carassos method. Here R
k,t
x

:= w
k
(t)e
k(Tt)
.
Using the technique of Cauchy-Dunford integrals we obtain a solution
of (3.1.23).
2001 CRC Press LLC
2001 CRC Press LLC
Theorem 3.1.4 Let A /
3
, then
J
k
(t) =
e
k(Tt)
2i
_

e
t
(1
Tt
T
e
2(Tt)
)
1 e
2T
R
A
() x

d, = G
3
,
is a twice continuously dierentiable solution of the Dirichlet problem
(3.1.23). Here = e
2kT
and G
3
(that is b and a) are chosen so that
k =
1
T
ln(K/) / G
3
.
Now we show that the approximate solution
u
,
(t) := e
k(Tt)
J
k
(t) (3.1.24)
=
1
2i
_

e
t
(1
Tt
T
e
2(Tt)
)
1 e
2T
R
A
()x

d
converges to the exact solution u(t) as 0. We write the integral in
(3.1.24) as a sum of two integrals:
J
k
(t)
1
=
1
2i
_

e
t
1 e
2T
R
A
()x

d,
J
k
(t)
2
=

Tt
T
2i
_

e
2(Tt)
1 e
2T
R
A
()x

d
and choose in a way that the lines Im =
i
T
m do not belong to it.
Therefore the denominator is not equal to zero as 0.
Comparing J
k
(t)
1
and J
k
(t)
2
with the solution obtained by the ABC
method, one can see that J
k
(t)
1
is a solution of the boundary value problem
v

() = Av(), 0 < < 2T, v(0) v(2T) = x

, (3.1.25)
at = t : J
k
(t)
1
= v(t). Further, J
k
(t)
2
is a solution of (3.1.25) at
= 2T t multiplied by a small parameter:
J
k
(t)
2
= v(2T t)
Tt
T
.
Thus, using the results for the ABC method we obtain convergence u

(t) to
u(t) as 0 and estimates for u(t) satisfying the condition |u(2T)| K
(or on the well-posedness class /=
_
u(t)

|u(2T)| K
_
):
|u

(t) u(t)| |J
k
(t)
1
u(t)| +|J
k
(t)
2
| C

, 0 < < 1,
where = e
2kT
= c
2
.
2001 CRC Press LLC
2001 CRC Press LLC
3.2 Regularization and regularized
semigroups
The general theory of ill-posed problems usually treats them in the form of
the equation of the rst kind
u = f, : U F, (3.2.1)
where, in general, operator
1
either does not exist or is not bounded.
Main regularization methods for (3.2.1) are variational: Ivanovs quasival-
ues method, the residual method, and Tikhonovs method. In Section 3.1
we studied regularization methods for the ill-posed Cauchy problem
u

(t) = Au(t), 0 t < T, u(0) = x, (LCP)

with operator A : D(A) X X that may have points of spectrum
anywhere in the half-plane Re > .
Those methods employ the dierential nature of this problem and allow
the construction of a regularized solution of (LCP) without reducing it to
the form (3.2.1).
In this section, we will compare the obtained regularization methods for
(LCP) by the error estimates. We discuss connections between dierential
methods and variational methods. Furthermore, we establish the connec-
tion between regularization methods for the ill-posed Cauchy problem and
the C-regularized semigroup method.
3.2.1 Comparison of the ABC and
the quasi-reversibility methods
The error estimates (3.1.12) for the regularizing operator R
,T
in the quasi-
reversibility method:

Q
Le
T/4(1tan
2
)
+L
1
,
and the error estimates (3.1.21) for the regularizing operator

R
,T
in the
ABC method:

A

_
C
2

T
T +
+C
3
+C
4
1
T +
ln
1

T +
+C
1
(T)

were obtained in Section 3.1 under the following assumption: operator A

belongs to class /
1
or /
2
, respectively, and there exists an exact solution
u(T) of (LCP) with the property
|A
2
u(T)| r or |u(T +)| r, r > 0,
2001 CRC Press LLC
2001 CRC Press LLC
respectively. For a comparison of these methods on the basis of the error
estimates, we introduce the following correctness classes:
/
1
=
_
u(T)

|A
2
u(T)| r
_
and /
2
=
_
u(T)

|u(T +)| r
_
,
and errors of a method on a class of operators / and on a correctness class
/:
(/, /, , ) = sup
u(T)M,AA
(A, u(T), , ).
We show that for the quasi-reversibility method there exists a class of op-
erators /, a correctness class /and a dependence between the error and
the regularization parameter , such that for

Q
(/, /, , ) = sup
u(T)M,AA

Q
(A, u(T), , ),
the exact by the order of logarithmic estimate holds. Here

Q
(A, u(T), , )
=
Q
= sup
xx

|R
,T
x

u(T)|
|R
,T
x u(T)| + sup
xx

|R
,T
x

R
,T
x|
=
1
Q
+
2
Q
.
For the error in the ABC method

A
(/, /, , ) = sup
u(T)M,AA

A
(A, u(T), , ),
the exact by the order of polynomial estimate holds, where

A
(A, u(T), , )
=
A
= sup
xx

R
,T
x

u(T)|
|u

(T) u(T)| + sup

xx

R
,T
x

(T)|
=
1
A
+
2
A
.
Denition 3.2.1 The error estimate (/, /, , ) (, ) is called ex-
act by the order of , if
C > 0, A /, u(T) /:
(A, u(T), , ) C(, ).
2001 CRC Press LLC
2001 CRC Press LLC
Theorem 3.2.1 For the quasi-reversibility method the exact by the order
of , error estimate

Q
(/
1
, /
1
, , ) Le
T/4(1tan
2
)
+L
1
(3.2.2)
and the exact by the order of error estimate

Q
_
/
1
, /
1
,
C
[ ln [
,
_
C[ ln [
1
hold. For the ABC method the corresponding estimates are

A
(/
2
, /
2
, , ) (3.2.3)
C
1
(T)

+
_
C
2

T
T+
+C
3
+C
4
1
T +
ln
1

T+
and
(/
2
, /
2
, , ) Cr

, <

T + 2
.
Proof Since constants L and L
1
in the estimate (3.1.12) do not depend
on A /
1
and the solution u(T) is from /
1
, the estimate (3.1.12) implies
(3.2.2). Similarly, (3.1.21) implies (3.2.3).
Now we show that (3.2.2) is exact by the order of , . From denitions
of (both
Q
and
A
) and
1
,
2
we have the inequalities

1
,
2

1
,
1
2

2
. (3.2.4)
Hence, it is sucient to prove that the estimates for
1
and
2
are exact
by the order of , . We show that if the spectrum of operator A /
1
contains suciently large (by the absolute value) points, then the error
estimate (3.1.7) for the quasi-reversibility method:

2
= sup
xx

|V
A

(T)(x x

)| Le
T/4(1tan
2
)
is exact by the order of , . An example of such an operator is
A =
d
2
dx
2
, D(A) =
_
u C[0, l]

xx
C[0, l], u(0) = u(l) = 0
_
,
Let
n
be the sequence of eigenvalues of operator A corresponding to
the eigenvectors
n
such that
lim
n
Re
n
= + and lim
n

Im
n
Re
n

< 1.
2001 CRC Press LLC
2001 CRC Press LLC
Then, for suciently large n, we have [arg
n
[ <

4
, the sequence
n
=

2
n
is the sequence of eigenvalues for operator A

= AA
2
, > 0,
and
V
A

(T)
n
= e

n
T

n
= e
(
n

2
n
)T

n
.
If we take (n) = 1/(2Re
n
), then
|V
A

(T)
n
| = [e
(
n
(n)
2
n
)T
[|
n
| = e
TRe (
n
(n)
2
n
)
|
n
|
= e
TRe
n
T(n)[(Re
n
)
2
(Im
n
)
2
]
|
n
|
= e
T
2
Re
n
+T(Im
n
)
2
|
n
| e
T
4(n)
|
n
|.
Therefore, the estimate for
2
is exact by the order of . Now consider the
estimate (3.1.10)

1
= |V
A
2(T)u(T) u(T)| =
_
_
_
_
_
_
T
0
V
A
2(s)A
2
u(T)ds
_
_
_
_
_
L
1
.
Since V
A
2(t) strongly converges to the identity operator as t 0, there
exist C > 0, A /
1
and u(T) with A
2
u(T) ,= 0 such that

1
=
_
_
_
_
_
_
T
0
V
A
2(s)A
2
u(T)ds
_
_
_
_
_
C. (3.2.5)
Hence, the estimate for
1
is exact by the order of . From (3.2.4) and the
obtained estimates for
1
,
2
with the dependence = C/[ ln [ we obtain
the exact by the order of estimate

Q
_
/
1
, /
1
,
C
[ ln [
,
_
C[ ln [
1
.
For the ABC method from the estimate (3.1.21) with the dependence =

T+
T+2
we obtain the polynomial estimate:
(/
2
, /
2
, , )
(C
1
+C
3
)

T+2
+C
2

T+
T+2
+
C
4
T + 2
[ ln [

T+2
C

, <

T + 2
.
Thus it is proved that the obtained estimates for the quasi-reversibility
and ABC methods hold on the class of operators /
1
and /
2
, respectively.
To compare these methods we obtain for the quasi-reversibility method an
error estimate on /
1
/
2
, /
1
/
2
.
2001 CRC Press LLC
2001 CRC Press LLC
Theorem 3.2.2 The error estimate (3.2.12) for the quasi-reversibility
method is exact by the order of , on the class of operators /
1
/
2
and
the correctness class /
1
/
2
= /
2
.
Proof In the proof of the estimate (3.1.2) (Theorem 3.2.1) for an operator
A /
1
we can take
n
R. Then A belongs to the class /
2
as well, and
the estimate of
2
is exact by the order of , on this class. Hence,
2
is exact by the order , on the class of operators /
1
/
2
. Since in /
2
there exists u(T) such that A
2
u(T) ,= 0, the proof that the estimate for
1
is exact by the order , is valid on the correctness class /
1
/
2
. The
equality /
1
/
2
= /
2
on the class of operators /
1
/
2
follows from
boundedness of operator A
2
V
A
() for > 0:
|A
2
u(T)| = |A
2
V
A
()u(T +)| r|A
2
V
A
()|,
where constant r is from the denition of /
2
. Thus, the estimate (3.2.12)
on /
1
/
2
, /
2
is exact by the order of , .
We note here that if A /
1
/
2
and u(T) /
2
it is preferable to use
the ABC method.
3.2.2 Dierential and variational
methods of regularization
The main variational methods for the regularization of the ill-posed problem
(3.2.1)
u = f, : U F,
are Ivanovs quasivalues method, Tikhonovs regularization method and the
residual method. As an approximate solution of (3.2.1) one takes a solution
u U of one of the following variational problems:
inf
_
|u f|

|u| r
_
(Ivanov

s quasivalues method)
inf
_
|u f| +|u|, > 0
_
(Tikhonov

s regularization)
inf
_
|u|

|u f| :
_
(residual method)
It is proved in [129] that if : U F, : U V are closed linear
operators, D() = U, and U, F are reexive spaces, V is a Banach space,
then these methods are equivalent. That is there exists a relation between
positive parameters , , and r such that solutions u(r), u() and u() of
these variational problems coincide.
Now, using the technique of Hilbert spaces, we establish that there
exists a connection between the regularization parameters such that the
2001 CRC Press LLC
2001 CRC Press LLC
ABC method for ill-posed (LCP) coincides with the quasivalues method
for the corresponding equation (3.2.1).
By H
+
we denote the class of self-adjoint operators whose eigenvectors

k
, corresponding to eigenvalues <
1

2
. . ., form an orthonor-
mal basis. If A H
+
, then A /
1
/
2
. Consider the ill-posed Cauchy
problem (LCP) with operator A H
+
. For (LCP) with such operators the
error estimates are obtained for both methods on correctness classes /
1
and /
2
. If there exists a solution of (LCP), then it can be written in the
form of the Fourier series
u(t) =

k=1
e

k
t
x
k

k
, x
k
= (x,
k
), 0 t T. (3.2.6)
The solution operators R
,T
and

R
,T
at t = T, corresponding to problems
obtained by the quasi-reversibility method (3.1.2):
u

,
(t) = (AA
2
)u
,
(t) =: A

u
,
(t), 0 < t T, > 0,
u
,
(0) = x

,
and by the ABC method (3.1.14):
u

,
(t) = A u
,
(t), 0 < t T +, > 0, > 0,
u
,
(0) + u
,
(T +) = x

,
also can be written in the form of series:
R
,T
x

= u
,
(T) =

k=1
e
(
k

2
k
)T
(x

)
k

k
, (x

)
k
= (x

,
k
),

R
,T
x

= u
,
(T) =

k=1
e

k
T
1 +e

k
(T+)
(x

)
k

k
. (3.2.7)
On the other hand, the assumption A H
+
guarantees that A generates
a C
0
-semigroup U
A
(t), t 0 and the Cauchy problem
v

(t) = Av(t), 0 t T, v(0) = x, (3.2.8)

is uniformly well-posed. Its solution is v() = U
A
() x. Problem (3.2.8) is
equivalent to the inverse Cauchy problem
v(t) = A v(t), 0 t < T, v(T) = x, v(t) = v(T t). (3.2.9)
Writing down the solution of (3.2.9) at t = T, with x = u(T), we reduce the
ill-posed Cauchy problem to the operator equation of the rst kind (3.2.1):
u :=

k=1
u
k
e

k
T

k
= f, (3.2.10)
2001 CRC Press LLC
2001 CRC Press LLC
where u = u(T), u
k
= (u(T),
k
), f = x = u(0) = v(0), and is a compact
linear operator on H. Regularization of (3.2.10) by the quasivalues method
leads us to the following optimization problem
inf
uM
|u f

|, f

= x

on the compact set

/:=
_
u =

k=1
u
k

k
H, u
k
= (u,
k
)

k=1

2
k
u
2
k
r
2
, lim
k

k
=
_
.
For
k
=
2
k
, set / coincides with the correctness class /
1
, for
k
=
e

k
(T+)
with /
2
. Using the Lagrange multipliers method:
inf
_

k=1
(u
k
e

k
T
(f

)
k
)
2
+
_
r
2

k=1

2
k
(u
T
k
)
2
__
,
(f

)
k
= (x

,
k
),
we obtain the solution of the optimization problem

k=1
u
k

k
=

k=1
e

k
T
1 +
2
k
e
2
k
T
(f

)
k

k
=: u
,
(T). (3.2.11)
Comparing (3.2.11) and (3.2.7) we see that if = and
k
= e

k
T
, = 3T,
then the quasivalues method coincides with the ABC method.
3.2.3 Regularizing operators and local C-regularized
semigroups
In this subsection, we establish a connection between the existence of reg-
ularizing operators for (LCP) and the existence of a C-regularized semi-
groups with the generator A and operator C = C

depending on the regu-

larization parameter .
Theorem 3.2.3 Suppose that operator A generates a C
0
-semigroup on a
Banach space X. Then the following statements are equivalent:
(I) A is the generator of a local C

-regularized semigroup S(t), 0 t < T

with operator C

0
I on D(A);
(II) for (LCP) there exists a bounded linear regularizing operator R
,t
, 0 <
t < T. It is invertible and commutes with A.
2001 CRC Press LLC
2001 CRC Press LLC
Proof (II) =(I). By U(t) we denote the (unbounded in general) solution
operator of (LCP), U(t)x := u(t). The assumption that A generates a
C
0
-semigroup U
A
() guarantees the uniform well-posedness of the Cauchy
problem (3.2.8) and the uniform well-posedness of the equivalent to (3.2.8)
inverse Cauchy problem (3.2.9). Since the problem (3.2.8) is well-posed,
there exists an inverse operator to U
A
(t) and we have
[U
A
(t)]
1
= U(t).
Commutativity of R
,t
with A implies commutativity of R
,t
with U
A
(s)
for s 0. We dene
S(t) := R
,
U
A
( t) = R
,
U(t)U
A
(t)U
A
( t)
= R
,
U(t)U
A
(), t < T,
and
C

= S(0) = R
,
U
A
().
We claim that S(t) satises relations (LC1) (LC2) from Denition 1.4.5.
The equality
S(t +s)C

= S(t)S(s), 0 t, s, t +s < T,
follows from the commutativity of R
,
with U
A
(t) and the semigroup
property of U
A
(t):
S(t +s)C

= R
,
U
A
( t s)R
,
U
A
()
= R
,
U
A
( t)R
,
U
A
( s)
= S(t)S(s).
The operator R
,
is dened on the whole space X and is bounded. Fur-
thermore, U
A
( t) is strongly continuous in t, where t 0. Therefore,
S(t)x = R
,
U
A
( t)x is continuous in t for 0 t < and for all x X.
Thus S(t), 0 t < T is a local C

-regularized semigroup.
Now we show that operator C

converges strongly to the identity oper-

ator as 0. For x D(A) we have
|C

x x| = |R
,
U
A
()x U()U
A
()x|
= |R
,
y U()y|, (3.2.12)
where y = U
A
()x is the solution of (3.2.8) at t = , corresponding to
the initial value x. Formula (3.2.12) and condition 2) from Denition 3.1.1
imply that |C

x x|
0
0 for x D(A).
Now, we establish that for the complete innitesimal generator G and
the generator Z of the constructed C

-regularized semigroup the inclusion

2001 CRC Press LLC
2001 CRC Press LLC
G A Z holds and implies A = Z. By denition of the complete
innitesimal generator, for x D(G) we have
Gx = lim
h0
h
1
_
C
1

S(h)x x

= lim
h0
h
1
[U(h)x x]
= U

(0)x = AU(0)x = Ax,

therefore, D(G) D(A) and A[
D(G)
= G. This and the closedness of A
imply G A. Now we show that D(A) D(Z) and that Z[
D(A)
= A. For
x D(A) we have
C

Ax = AC

x = AR
,
U
A
()x = R
,
AU
A
()x
= R
,
lim
t0
t
1
[U
A
( t) U
A
()] x
= lim
t0
t
1
[S(t)x C

x] = C

Zx,
i.e., x D(Z) and Ax = Zx for x D(A). Hence A Z. From the
denition of the generator and the complete innitesimal generator for a
C-regularized semigroup it follows that C
1
GCx = Zx for x D(Z). This
and the inclusion G A imply equalities
ACx = CZx, C
1
ACx = Zx, x D(Z).
Under the assumption (A) ,= (which holds for A, the generator of a
C
0
-semigroup, and hence for A ), it follows A = Z.
(I) = (II). Suppose that for any < T, A is the generator of a local
C

-regularized semigroup S(t), t [0, ) with C

strongly convergent to I
as 0. We show that R
,t
:= S(t), 0 < t < T, is a regularizing operator
for (LCP). By our construction, the linear operator R
,t
is dened on the
whole X, is bounded, and therefore is continuous. Suppose that for some
x D(A) there exists a solution of (LCP) u(). For xed t (0, T) we
consider the error
|R
,t
x u(t)| = |S(t)C
1

x u(t)|.
Since C
1

and S(t) commute on R(C

), we have
S(t)C
1

x = C
1

S(t)C

x.
Consider y = C

x C

D(A). Then by Theorem 1.4.6,

v(t) = R
,t
x = C
1

S(t)C

x = C
1

S(t)y
is the unique solution of the Cauchy problem with the generator of a C

-
regularized semigroup:
v

(t) = Av(t), 0 < t < T, v(0) = y.

2001 CRC Press LLC
2001 CRC Press LLC
On the other hand, C

u(t) is also a solution of (LCP) with the initial value

C

u(0) = C

x = y, therefore R
,t
x = S(t)x = C

u(t). Hence, if for an

initial value x there exists a solution u(), then
|R
,t
x u(t)| = |C

u(t) u(t)|
0
0, 0 t < T,
i.e., R
,t
is a regularizing operator for (LCP).
For A H
+
we now consider the regularizing operators obtained by the
modied quasi-reversibility and ABC methods:
R
p
,t
x =

k=1
e

k
t
p
k
T
x
k

k
, p > 1, (3.2.13)
and

R
p
,t
x =

k=1
e

k
t
1 +e

p
k
T
x
k

k
, p 1. (3.2.14)
They are C

-regularized semigroups, where

C

x =

k=1
e

p
k
T
x
k

k
, C

x =

k=1
1
1 +e

p
k
T
x
k

k
,
respectively. In [185] it is proved that the classes of regularizing op-
erators (3.2.13), (3.2.14) have error estimates similar to (3.2.2), (3.2.3),
respectively.
3.2.4 Regularization of slightly ill-posed problems
In this subsection, we discuss the regularization of the Cauchy problem
(CP) in the cases when A is the generator of an integrated semigroup or of
a -convoluted semigroup.
Let X be a Banach space. Consider the Cauchy problem
u

(t) = Au(t), t 0, u(0) = x, (CP)

with a densely dened linear operator A generating an exponentially
bounded integrated semigroup V (t), t 0. Suppose that for some ex-
actly given initial value x there exists a solution of (CP) u() and the initial
value x

is given with an error : |x

x| . If x D(A
2
), then by
Theorem 1.2.4 the problem (CP) is (1, )-well-posed and its solution
u(t) = V

(t)x = V (t)Ax +x, t 0, (3.2.15)

satises the estimate
|u(t)| K e
t
(|Ax| +|x|).
2001 CRC Press LLC
2001 CRC Press LLC
The solution operators V

(t) are dened and (in general) unbounded on

D(A). From the representation (3.2.15), it is clear that in order to regularize
V

(t), one has to neutralize the action of the unbounded operator A. An

obvious candidate for this role is the resolvent R
A
() : X D(A). We
recall that if A is the generator of an exponentially bounded integrated
semigroup satisfying property (1.2.6), then
R
A
()x

x, x D(A), (3.2.16)
(see Proposition 1.5.1). Now for t 0 and > a = max, 0 we
dene the operator
R1

,t
:= V

(t)R
A
()
on X. From the relation
AR
A
()x =
_
AI

R
A
()x = R
A
()x x, x X,
we have
R1

,t
= V (t)
_
R
A
() I
_
+R
A
().
Hence the operator R1

,t
is bounded on X for any t 0 and .
Theorem 3.2.4 Suppose that a densely dened operator A is the generator
of an exponentially bounded integrated semigroup V (t), t 0 satisfying
property (1.2.6). Then the operator
R
,t
=
1

_
V (t)AR
A
(1/) +R
A
(1/)
_
, (0, 1/], t > 0,
is the regularizing operator for (CP).
Proof Let u() be an exact solution of (CP) with the initial value x
D(A
2
). Consider its approximation u
,
(t) = R1

,t
x

, t > 0, x

X,
where = 1/. We now estimate the error
_
_
u
,
(t) u(t)
_
_
=
_
_
_V

(t)R
A
()x

(t)x V

(t)R
A
()x
_
_
_

_
_
_R1

,t
(x

x)
_
_
_ +
_
_
_V

(t)
_
R
A
() I

x
_
_
_.
Since |R
A
()| K and |V (t)| K e
t
, we have
_
_
_R1

,t
x
_
_
_ =
_
_
_
2
V (t)R
A
()x V (t)x +R
A
()x
_
_
_
2
K e
t
|x|
and
_
_
_V

(t)
_
R
A
() I

x
_
_
_ =
_
_
_V

(t)
_
R
A
() I

R
A
(
0
)y
_
_
_
=
_
_
_
_
V (t)AR
A
(
0
) +R
A
(
0
)
_
R
A
() I

y
_
_
_
K e
t
_
_
_
_
R
A
() I

y
_
_
_,
2001 CRC Press LLC
2001 CRC Press LLC
where y :=
0
x Ax D(A). Choosing = o(
/2
) with some (0, 1),
we obtain that _
_
_R1

,t
x

u(t)
_
_
_
0
0.
Hence R
,t
is the regularizing operator with the regularization parameter
= 1/ 0.
If A is the generator of an exponentially bounded n-integrated semi-
group V (t), t 0 satisfying property (1.2.6), then (3.2.16) holds for all
x D(A
n
). In this case the operator
R
,t
:= V
(n)
(t)
_
1

R
A
(1/)
_
n
, (0, 1/], t > 0,
is the regularizing operator for (CP).
We now consider the Cauchy problem (CP) with an operator A gen-
erating a -convoluted semigroup S

(t), t 0. By Denition 1.3.1, we

have that S

(t)Ax = AS

(t)x for x D(A), and for x X

S

(t)x = A
_
t
0
S

(s)xds + (t)x, (t) :=

_
t
0
(s)ds, t 0.
We recall that if for some x there exists a solution u() of (CP), then
v = u is a solution of the corresponding convoluted Cauchy problem:
v

(t) = Av(t) + (t)x, t 0, v(0) = 0. (3.2.17)

Suppose now that we are given v

(t) =
_
t
0
S

(s)x

, t 0, a solution of the
convoluted problem (3.2.17) with x = x

X. In order to nd u, we need
to regularize the following convolution equation of the rst kind:
u := u =
_
t
0
(t s)u(s)ds = v(t) (3.2.18)
given v

such that |v v

| K. Let be the region from Theorem 1.3.1.

Applying the Laplace transform to (3.2.18) we obtain
L(, ) L(u, ) = L(v, ), C,
and
u = L
1
L(v, )
L(, )
=
1
v.
Note that the operator
1
is unbounded since L(, ) 0 as (see
Theorem 1.3.1). Using Tikhonov-Arsenins regularization [258] we dene
the regularizing operator
R
,t
x := L
1
_
L(v, )f(, )
L(, )
, t
_
2001 CRC Press LLC
2001 CRC Press LLC
for all x X, (0,
0
] and t > 0. Here f(, )
0
1 uniformly in
and
f(, ) =

O(L(, )) , (0,
0
].
For example, one can take
f(, ) =
L(, )
(L
2
(, ) +)
1/2
.
Finally, we note that the same procedure can be used in the case where
A is the generator of a C-regularized semigroup. In this case one has to
nd a solution of the Cauchy problem
v

(t) = Av

(t), v

(0) = Cx

,
and then to regularize the operator equation of the rst kind: Cu = v with
given v

.
2001 CRC Press LLC
2001 CRC Press LLC
Bibliographic Remark
We would like to emphasize here that our book is not a survey in any
sense. Each of the three approaches (semigroups, distributions and reg-
ularizations) to treating the abstract Cauchy problem deserves a separate
volume in order to cover all aspects of the theory and the applications. Our
aim was to present a concept of some of the connections between these ap-
proaches. For this reason our bibliographic notes are very minimal, and we
indicate only the sources that were directly used in writing this book. The
Bibliography also includes publications that are closely related and can be
recommended for further reading.
Chapter 0 is inspired by R. Dautray and J.-L. Lions [51], R. Latt`es and
J.-L. Lions [145], A. Pazy [219], E. Zeidler [284], H. R. Thieme [256].
Chapter 1
Section 1.1 is based on H. O. Fattorini [84], S. G. Kren [141], T. Kato
[130], E. B. Davies and M. M. H. Pang [52], A. V. Balakrishnan [14]. We
also recommend books by E. Hille and R. S. Phillips [120], K. Yosida [281],
R. Dautray and J.-L. Lions [51], A. Pazy [219], J. A. Goldstein [106], K.
Engel and R. Nagel [73].
Section 1.2 is based on W. Arendt [5], F. Neubrander [209], H. R. Thieme
[256], H. Kellerman and M. Hieber [131], N. Tanaka and N. Okazawa [255],
E. B. Davies and M. M. H. Pang [52], I. Cioranescu [36].
Section 1.3 is based on I. Cioranescu and G. Lumer [39, 38], I. V. Mel-
nikova, U. A. Anufrieva and A. Filinkov [177].
Section 1.4 is based on E. B. Davies and M. M. H. Pang [52], N. Tanaka
and N. Okazawa [255], I. Miyadera and N. Tanaka [201].
Section 1.5 is based on I. V. Melnikova and M. A. Alshansky [174, 176].
Section 1.6 is based on A. Yagi [280], I. V. Melnikova and A. V. Glad-
chenko [187], I. V. Melnikova [170, 171, 173], W. Arendt, O. El-Mennaoui
and V. Keyantuo [9], H. R. Thieme [256], A. Favini and A. Yagi [102].
2001 CRC Press LLC
2001 CRC Press LLC
Section 1.7 is based on I. V. Melnikova [171], I. V. Melnikova and A.
Filinkov [186, 182].
Chapter 2
Section 2.1 is based on H. O. Fattorini [84], J.-L. Lions [152], J. Chazarain
[30], I. V. Melnikova and M. A. Alshansky [175], I. V. Melnikova [169, 172].
Section 2.2 is based on I. V. Melnikova [169], I. V. Melnikova, U. A.
Anufrieva and V. Yu. Ushkov [178].
Section 2.3 is based on H. Komatsu [135], J. Chazarain [30], E. Magenes
and J.-L. Lions [153], I. V. Melnikova, U. A. Anufrieva and A. Filinkov
[177], V. K. Ivanov and I. V. Melnikova [124], I. V. Melnikova [167], I. V.
Melnikova and A. Filinkov [185].
Chapter 3
Sections 3.1 and 3.2 are based on I. V. Melnikova [167, 165, 166], I. V.
Melnikova and A. Filinkov [185], R. Latt`es and J.-L. Lions [145], V. K.
Ivanov, I. V. Melnikova and A. Filinkov [128], V. K. Ivanov, V. V. Vasin
and V. P. Tanana [129], M. M. Lavrentev [146], I. V. Melnikova and S. V.
Bochkareva [180], A. Carasso [23].
2001 CRC Press LLC
2001 CRC Press LLC
Bibliography
[1] K. A. Ames, G. W. Clark, J. F. Epperson, and S. F. Oppenheimer. A
comparison of regularizations for an ill-posed problem. Math. Comp.,
67(224):14511471, 1998.
[2] K. A. Ames and J. F. Epperson. A kernel-based method for the ap-
proximate solution of backward parabolic problems. SIAM J. Numer.
Anal., 34(4):13571390, 1997.
[3] K. A. Ames and L. E. Payne. Asymptotic behavior for two regulariza-
tions of the Cauchy problem for the backward heat equation. Math.
Models Methods Appl. Sci., 8(1):187202, 1998.
[4] W. Arendt. Resolvent positive operators. Proc. London Math. Soc.
(3), 54(2):321349, 1987.
[5] W. Arendt. Vector-valued Laplace transforms and Cauchy problems.
Israel J. Math., 59(3):327352, 1987.
[6] W. Arendt. Sobolev imbeddings and integrated semigroups. In Semi-
group Theory and Evolution Equations (Delft, 1989), pages 2940.
Dekker, New York, 1991.
[7] W. Arendt and C. J. K. Batty. Almost periodic solutions of rst- and
second-order Cauchy problems. J. Dierential Equations, 137(2):363
383, 1997.
[8] W. Arendt and A. V. Bukhvalov. Integral representations of resol-
vents and semigroups. Forum Math., 6(1):111135, 1994.
[9] W. Arendt, O. El-Mennaoui, and V. Keyantuo. Local integrated
semigroups: evolution with jumps of regularity. J. Math. Anal. Appl.,
186(2):572595, 1994.
[10] W. Arendt and A. Favini. Integrated solutions to implicit dierential
equations. Rend. Sem. Mat. Univ. Politec. Torino, 51(4):315329
(1994), 1993. Partial dierential equations, I (Turin, 1993).
2001 CRC Press LLC
2001 CRC Press LLC
[11] W. Arendt and H. Kellermann. Integrated solutions of Volterra inte-
grodierential equations and applications. In Volterra Integrodier-
ential Equations in Banach Spaces and Applications (Trento, 1987),
pages 2151. Longman Sci. Tech., Harlow, 1989.
[12] W. Arendt, F. Neubrander, and U. Schlotterbeck. Interpolation of
semigroups and integrated semigroups. Semigroup Forum, 45(1):26
37, 1992.
[13] W. Arendt and A. Rhandi. Perturbation of positive semigroups. Arch.
Math. (Basel), 56(2):107119, 1991.
[14] A. V. Balakrishnan. Applied Functional Analysis. Springer-Verlag,
New York, second edition, 1981.
[15] J. Banasiak. Generation results for B-bounded semigroups. Ann.
Mat. Pura Appl. (4), 175:307326, 1998.
[16] V. Barbu and A. Favini. On some degenerate parabolic problems.
Ricerche Mat., 46(1):7786, 1997.
[17] V. Barbu and A. Favini. The analytic semigroup generated by a sec-
ond order degenerate dierential operator in C[0, 1]. In Proceedings of
the Third International Conference on Functional Analysis and Ap-
proximation Theory, Vol. I (Acquafredda di Maratea, 1996), number
52, Vol. I, pages 2342, 1998.
[18] V. Barbu, A. Favini, and S. Romanelli. Degenerate evolution equa-
tions and regularity of their associated semigroups. Funkcial. Ekvac.,
39(3):421448, 1996.
[19] B. Baumer, G. Lumer, and F. Neubrander. Convolution kernels and
generalized functions. In Generalized Functions, Operator Theory,
and Dynamical Systems (Brussels, 1997), pages 6878. Chapman &
Hall/CRC, Boca Raton, FL, 1999.
[20] B. Baumer and F. Neubrander. Laplace transform methods for evo-
lution equations. Confer. Sem. Mat. Univ. Bari, (258-260):2760,
1994. Swabian-Apulian Meeting on Operator Semigroups and Evolu-
tion Equations (Italian) (Ruvo di Puglia, 1994).
[21] R. Beals. On the abstract Cauchy problem. J. Functional Analysis,
10:281299, 1972.
[22] H. Bremermann. Distributions, Complex Variables, and Fourier
Transforms. Addison-Wesley Publishing Co., Inc., Reading, Mass.-
London, 1965.
2001 CRC Press LLC
2001 CRC Press LLC
[23] A. Carasso. The abstract backward beam equation. SIAM J. Math.
Anal., 2:193212, 1971.
[24] A. Carasso. The backward beam equation and the numerical com-
putation of dissipative equations backwards in time. In Improperly
Posed Boundary Value Problems (Conf., Univ. New Mexico, Albu-
querque, N.M., 1974), pages 124157. Res. Notes in Math., No. 1.
Pitman, London, 1975.
[25] A. Carasso. Error bounds in the nal value problem for the heat
equation. SIAM J. Math. Anal., 7(2):195199, 1976.
[26] R. W. Carroll and R. E. Showalter. Singular and Degenerate Cauchy
Problems. Academic Press [Harcourt Brace Jovanovich Publishers],
New York, 1976. Mathematics in Science and Engineering, Vol. 127.
[27] J. Chazarain. Probl`emes de Cauchy au sens des distributions vecto-
rielles et applications. C. R. Acad. Sci. Paris Ser. A-B, 266:A10A13,
1968.
[28] J. Chazarain. Probl`emes de Cauchy dans des espaces dultra-
distributions. C. R. Acad. Sci. Paris Ser. A-B, 266:A564A566, 1968.
[29] J. Chazarain. Probl`emes de Cauchy abstraits et applications. In Sym-
posia Mathematica, Vol. VII (Convegno sulle Problemi di Evoluzione,
INDAM, Rome, 1970), pages 129134. Academic Press, London,
1971.
[30] J. Chazarain. Probl`emes de Cauchy abstraits et applications `a
quelques probl`emes mixtes. J. Functional Analysis, 7:386446, 1971.
[31] I. Cioranescu. Operator-valued ultradistributions in the spectral the-
ory. Math. Ann., 223(1):112, 1976.
[32] I. Cioranescu. Abstract Beurling spaces of class (M
p
) and ultradis-
tribution semi-groups. Bull. Sci. Math. (2), 102(2):167192, 1978.
[33] I. Cioranescu. On the second order Cauchy problem associated with
a linear operator. J. Math. Anal. Appl., 154(1):238242, 1991.
[34] I. Cioranescu. A generation result for C-regularized semigroups. In
Semigroups of Linear and Nonlinear Operations and Applications
(Cura cao, 1992), pages 121128. Kluwer Acad. Publ., Dordrecht,
1993.
[35] I. Cioranescu. On a class of C-regularized semigroups. In Evolu-
tion Equations, Control Theory, and Biomathematics (Han sur Lesse,
1991), pages 4550. Dekker, New York, 1994.
2001 CRC Press LLC
2001 CRC Press LLC
[36] I. Cioranescu. Local convoluted semigroups. In Evolution Equations
(Baton Rouge, LA, 1992), pages 107122. Dekker, New York, 1995.
[37] I. Cioranescu and V. Keyantuo. C-semigroups: generation and ana-
lyticity. Integral Transform. Spec. Funct., 6(1-4):1525, 1998. Gener-
alized functionslinear and nonlinear problems (Novi Sad, 1996).
[38] I. Cioranescu and G. Lumer. Probl`emes devolution regularises par un
noyau general K(t). Formule de Duhamel, prolongements, theor`emes
de generation. C. R. Acad. Sci. Paris Ser. I Math., 319(12):1273
1278, 1994.
[39] I. Cioranescu and G. Lumer. On K(t)-convoluted semigroups. In
Recent Developments in Evolution Equations (Glasgow, 1994), pages
8693. Longman Sci. Tech., Harlow, 1995.
[40] G. W. Clark and S. F. Oppenheimer. Quasireversibility methods for
non-well-posed problems. Electron. J. Dierential Equations, pages
No. 08, approx. 9 pp. (electronic), 1994.
[41] Ph. Clement, O. Diekmann, M. Gyllenberg, H. J. A. M. Heijmans,
and H. R. Thieme. A Hille-Yosida theorem for a class of weakly

continuous semigroups. Semigroup Forum, 38(2):157178, 1989.

Semigroups and dierential operators (Oberwolfach, 1988).
[42] Ph. Clement and S. Guerre-Delabri`ere. On the regularity of abstract
Cauchy problems and boundary value problems. Atti Accad. Naz. Lin-
cei Cl. Sci. Fis. Mat. Natur. Rend. Lincei (9) Mat. Appl., 9(4):245
266 (1999), 1998.
[43] Ph. Clement, H. J. A. M. Heijmans, S. Angenent, C. J. van Duijn, and
B. de Pagter. One-Parameter Semigroups. North-Holland Publishing
Co., Amsterdam, 1987.
[44] Ph. Clement and C. A. Timmermans. On C
0
-semigroups generated by
dierential operators satisfying Ventcels boundary conditions. Ned-
erl. Akad. Wetensch. Indag. Math., 48(4):379387, 1986.
[45] W. L. Conradie and N. Sauer. Empathy, C-semigroups, and inte-
grated semigroups. In Evolution Equations (Baton Rouge, LA, 1992),
pages 123132. Dekker, New York, 1995.
[46] G. Da Prato. Semigruppi regolarizzabili. Ricerche Mat., 15:223248,
1966.
[47] G. Da Prato. Weak solutions for linear abstract dierential equations
in Banach spaces. Advances in Math., 5:181245 (1970), 1970.
2001 CRC Press LLC
2001 CRC Press LLC
[48] G. Da Prato and E. Giusti. Una caratterizzazione dei generatori di
funzioni coseno astratte. Boll. Un. Mat. Ital. (3), 22:357362, 1967.
[49] G. Da Prato and U. Mosco. Semigruppi distribuzioni analitici. Ann.
Scuola Norm. Sup. Pisa (3), 19:367396, 1965.
[50] G. Da Prato and E. Sinestrari. Dierential operators with nondense
domain. Ann. Scuola Norm. Sup. Pisa Cl. Sci. (4), 14(2):285344
(1988), 1987.
[51] R. Dautray and J.-L. Lions. Mathematical Analysis and Numeri-
cal Methods for Science and Technology. Vol. 16. Springer-Verlag,
Berlin, 19881993.
[52] E. B. Davies and M. M. H. Pang. The Cauchy problem and a gener-
alization of the Hille-Yosida theorem. Proc. London Math. Soc. (3),
55(1):181208, 1987.
[53] R. deLaubenfels. Polynomials of generators of integrated semigroups.
Proc. Amer. Math. Soc., 107(1):197204, 1989.
[54] R. deLaubenfels. Integrated semigroups and integrodierential equa-
tions. Math. Z., 204(4):501514, 1990.
[55] R. deLaubenfels. Integrated semigroups, C-semigroups and the ab-
stract Cauchy problem. Semigroup Forum, 41(1):8395, 1990.
[56] R. deLaubenfels. C-existence families. In Semigroup Theory and
Evolution Equations (Delft, 1989), pages 295309. Dekker, New York,
1991.
[57] R. deLaubenfels. Entire solutions of the abstract Cauchy problem.
Semigroup Forum, 42(1):83105, 1991.
[58] R. deLaubenfels. Existence and uniqueness families for the abstract
Cauchy problem. J. London Math. Soc. (2), (2):310338, 1991.
[59] R. deLaubenfels. Unbounded well-bounded operators, strongly con-
tinuous semigroups and the Laplace transform. Studia Math.,
103(2):143159, 1992.
[60] R. deLaubenfels. C-semigroups and strongly continuous semigroups.
Israel J. Math., 81(1-2):227255, 1993.
[61] R. deLaubenfels. C-semigroups and the Cauchy problem. J. Funct.
Anal., 111(1):4461, 1993.
[62] R. deLaubenfels. Matrices of operators and regularized semigroups.
Math. Z., 212(4):619629, 1993.
2001 CRC Press LLC
2001 CRC Press LLC
[63] R. deLaubenfels. Existence Families, Functional Calculi and Evolu-
tion Equations. Springer-Verlag, Berlin, 1994.
[64] R. deLaubenfels. Regularized functional calculi and evolution equa-
tions. In Evolution Equations (Baton Rouge, LA, 1992), pages 141
152. Dekker, New York, 1995.
[65] R. deLaubenfels. Functional calculi, semigroups of operators, and
Hille-Yosida operators. Houston J. Math., 22(4):787805, 1996.
[66] R. deLaubenfels, Z. Huang, S. Wang, and Y. Wang. Laplace trans-
forms of polynomially bounded vector-valued functions and semi-
groups of operators. Israel J. Math., 98:189207, 1997.
[67] R. deLaubenfels and M. Jazar. Functional calculi, regularized semi-
groups and integrated semigroups. Studia Math., 132(2):151172,
1999.
[68] R. deLaubenfels, G. Z. Sun, and S. W. Wang. Regularized semigroups,
existence families and the abstract Cauchy problem. Dierential In-
tegral Equations, 8(6):14771496, 1995.
[69] G. V. Demidenko and S. V. Uspenski. Uravneniya i Sistemyi Ner-
azreshennyie Otnositelno Starshei Proizvodnoi. Nauchnaya Kniga,
Novosibirsk, 1998.
[70] A. A. Dezin. Partial Dierential Equations. Springer-Verlag, Berlin,
1987.
[71] O. Diekmann, M. Gyllenberg, and H. R. Thieme. Semigroups and
renewal equations on dual Banach spaces with applications to popu-
lation dynamics. In Delay Dierential Equations and Dynamical Sys-
tems (Claremont, CA, 1990), pages 116129. Springer, Berlin, 1991.
[72] N. Dunford and J. T. Schwartz. Linear Operators. Parts IIII. John
Wiley & Sons Inc., New York, 1988.
[73] K. Engel and R. Nagel. One-Parameter Semigroups for Linear Evolu-
tion Equations. Springer-Verlag, New York, 2000. With contributions
by S. Brendle, M. Campiti, T. Hahn, G. Metafune, G. Nickel, D. Pal-
lara, C. Perazzoli, A. Rhandi, S. Romanelli and R. Schnaubelt.
[74] M. V. Falaleev. Generalized Functions. Irkutsk University, Irkutsk,
1996. Preprint.
[75] M. V. Falaleev. The Cauchy problem for degenerate integrodieren-
tial equations in Banach spaces. Vestnik Chelyabinsk. Univ. Ser. 3
Mat. Mekh., 2(5):126136, 1999.
2001 CRC Press LLC
2001 CRC Press LLC
[76] H. O. Fattorini. Ordinary dierential equations in linear topological
spaces. I. J. Dierential Equations, 5:72105, 1969.
[77] H. O. Fattorini. Ordinary dierential equations in linear topological
spaces. II. J. Dierential Equations, 6:5070, 1969.
[78] H. O. Fattorini. On a class of dierential equations for vector-valued
distributions. Pacic J. Math., 32:79104, 1970.
[79] H. O. Fattorini. A representation theorem for distribution semi-
groups. J. Functional Analysis, 6:8396, 1970.
[80] H. O. Fattorini. Uniformly bounded cosine functions in Hilbert space.
Indiana Univ. Math. J., 20:411425, 1970/1971.
[81] H. O. Fattorini. Some remarks on convolution equations for vector-
valued distributions. Pacic J. Math., 66(2):347371, 1976.
[82] H. O. Fattorini. Structure theorems for vector-valued ultradistribu-
tions. J. Funct. Anal., 39(3):381407, 1980.
[83] H. O. Fattorini. Some remarks on second order abstract Cauchy
problems. Funkcial. Ekvac., 24(3):331344, 1981.
[84] H. O. Fattorini. The Cauchy Problem. Addison-Wesley Publishing
Co., Reading, Mass., 1983.
[85] H. O. Fattorini. Second Order Linear Dierential Equations in Ba-
nach Spaces. North-Holland Publishing Co., Amsterdam, 1985.
[86] A. Favini. Laplace transform method for a class of degenerate evolu-
tion problems. Rend. Mat. (6), 12(3-4):511536 (1980), 1979.
[87] A. Favini. Degenerate and singular evolution equations in Banach
space. Math. Ann., 273(1):1744, 1985.
[88] A. Favini. An operational method for abstract degenerate evolution
equations of hyperbolic type. J. Funct. Anal., 76(2):432456, 1988.
[89] A. Favini and M. Fuhrman. Approximation results for semigroups
generated by multivalued linear operators and applications. Dier-
ential Integral Equations, 11(5):781805, 1998.
[90] A. Favini, G. R. Goldstein, Je.A. Goldstein, and S. Romanelli. C
0
-
semigroups generated by second order dierential operators with
general Wentzell boundary conditions. Proc. Amer. Math. Soc.,
128(7):19811989, 2000.
2001 CRC Press LLC
2001 CRC Press LLC
[91] A. Favini, J. A. Goldstein, and S. Romanelli. An analytic semigroup
associated to a degenerate evolution equation. In Stochastic Pro-
cesses and Functional Analysis (Riverside, CA, 1994), pages 85100.
Dekker, New York, 1997.
[92] A. Favini, J. A. Goldstein, and S. Romanelli. Analytic semigroups on
L
p
w
(0, 1) and on L
p
(0, 1) generated by some classes of second order
dierential operators. Taiwanese J. Math., 3(2):181210, 1999.
[93] A. Favini and E. Obrecht. Conditions for parabolicity of second or-
der abstract dierential equations. Dierential Integral Equations,
4(5):10051022, 1991.
[94] A. Favini and P. Plazzi. On some abstract degenerate problems of
parabolic type. I. The linear case. Nonlinear Anal., 12(10):10171027,
1988.
[95] A. Favini and P. Plazzi. On some abstract degenerate problems of
parabolic type. II. The nonlinear case. Nonlinear Anal., 13(1):2331,
1989.
[96] A. Favini and S. Romanelli. Degenerate second order operators as
generators of analytic semigroups on C[0, +] or on L
p

1/2
(0, +).
In Approximation and Optimization, Vol. II (Cluj-Napoca, 1996),
pages 93100. Transilvania, Cluj-Napoca, 1997.
[97] A. Favini and S. Romanelli. Analytic semigroups on C[0, 1] generated
by some classes of second order dierential operators. Semigroup
Forum, 56(3):362372, 1998.
[98] A. Favini and H. Tanabe. Linear parabolic dierential equations of
higher order in time. In Dierential Equations in Banach Spaces
(Bologna, 1991), pages 8592. Dekker, New York, 1993.
[99] A. Favini and A. Yagi. Space and time regularity for degenerate
evolution equations. J. Math. Soc. Japan, 44(2):331350, 1992.
[100] A. Favini and A. Yagi. Multivalued linear operators and degenerate
evolution equations. Ann. Mat. Pura Appl. (4), 163:353384, 1993.
[101] A. Favini and A. Yagi. Abstract second order dierential equations
with applications. Funkcial. Ekvac., 38(1):8199, 1995.
[102] A. Favini and A. Yagi. Degenerate Dierential Equations in Banach
Spaces. Marcel Dekker Inc., New York, 1999.
[103] C. Foias. Remarques sur les semi-groupes distributions doperateurs
normaux. Portugal. Math., 19:227242, 1960.
2001 CRC Press LLC
2001 CRC Press LLC
[104] D. Fujiwara. A characterisation of exponential distribution semi-
groups. J. Math. Soc. Japan, 18:267274, 1966.
[105] I. M. Gelfand and G. E. Shilov. Generalized Functions. Vol. 13.
Academic Press [Harcourt Brace Jovanovich Publishers], New York,
19641967 [1977].
[106] J. A. Goldstein. Semigroups of Linear Operators and Applications.
The Clarendon Press Oxford University Press, New York, 1985.
[107] J. A. Goldstein. A survey of semigroups of linear operators and ap-
plications. In Semigroups of Linear and Nonlinear Operations and
Applications (Curacao, 1992), pages 957. Kluwer Acad. Publ., Dor-
drecht, 1993.
[108] J. A. Goldstein, R. deLaubenfels, and J. T. Sandefur, Jr. Regularized
semigroups, iterated Cauchy problems and equipartition of energy.
Monatsh. Math., 115(1-2):4766, 1993.
[109] J. A. Goldstein and C. Y. Lin. An L
p
semigroup approach to degen-
erate parabolic boundary value problems. Ann. Mat. Pura Appl. (4),
159:211227, 1991.
[110] M. Hieber. Integrated semigroups and dierential operators on L
p
spaces. Math. Ann., 291(1):116, 1991.
[111] M. Hieber. Integrated semigroups and the Cauchy problem for sys-
tems in L
p
spaces. J. Math. Anal. Appl., 162(1):300308, 1991.
[112] M. Hieber. Laplace transforms and -times integrated semigroups.
Forum Math., 3(6):595612, 1991.
[113] M. Hieber. On strongly elliptic dierential operators on L
1
(R
n
).
In Semigroups of Linear and Nonlinear Operations and Applications
(Cura cao, 1992), pages 177183. Kluwer Acad. Publ., Dordrecht,
1993.
[114] M. Hieber. On the L
p
theory of systems of linear partial dierential
equations. In Evolution Equations, Control Theory, and Biomath-
ematics (Han sur Lesse, 1991), pages 275283. Dekker, New York,
1994.
[115] M. Hieber. Spectral theory and Cauchy problems on L
p
spaces. Math.
Z., 216(4):613628, 1994.
[116] M. Hieber. Spectral theory of positive semigroups generated by dif-
ferential operators. Arch. Math. (Basel), 63(4):333340, 1994.
2001 CRC Press LLC
2001 CRC Press LLC
[117] M. Hieber, A. Holderrieth, and F. Neubrander. Regularized semi-
groups and systems of linear partial dierential equations. Ann.
Scuola Norm. Sup. Pisa Cl. Sci. (4), 19(3):363379, 1992.
[118] M. Hieber and F. R abiger. A remark on the abstract Cauchy problem
on spaces of Holder continuous functions. Proc. Amer. Math. Soc.,
115(2):431434, 1992.
[119] E. Hille. Functional Analysis and Semi-Groups. American Mathe-
matical Society, New York, 1948. American Mathematical Society
Colloquium Publications, vol. 31.
[120] E. Hille and R. S. Phillips. Functional Analysis and Semi-Groups.
American Mathematical Society, Providence, R. I., 1974.
[121] L. H ormander. Estimates for translation invariant operators in L
p
spaces. Acta Math., 104:93140, 1960.
[122] V. K. Ivanov. Ill-posed problems and divergent processes. Uspekhi
Mat. Nauk, 40(4(244)):165166, 1985.
[123] V. K. Ivanov. Hadamard conditions for well-posedness in spaces of
generalized functions. Sibirsk. Mat. Zh., 28(6):5359, 218, 1987.
[124] V. K. Ivanov and I. V. Melnikova. A general scheme for the elimina-
tion of divergences of dierent types. Sibirsk. Mat. Zh., 29(6):6673,
1988.
[125] V. K. Ivanov and I. V. Melnikova. Construction of quasivalues for
weakly well-posed problems. Dokl. Akad. Nauk SSSR, 306(3):530
535, 1989.
[126] V. K. Ivanov and I. V. Melnikova. New generalized functions and
weak well-posedness of operator problems. Dokl. Akad. Nauk SSSR,
317(1):2226, 1991.
[127] V. K. Ivanov and I. V. Melnikova. New generalized functions and
weak well-posedness of operator problems. In Complex Analysis and
Generalized Functions (Varna, 1991), pages 141147. Publ. Bulgar.
Acad. Sci., Soa, 1993.
[128] V. K. Ivanov, I. V. Melnikova, and A. I. Filinkov. Operator-
Dierential Equations and Ill-Posed Problems. Fizmatlit Nauka,
Moscow, 1995.
[129] V. K. Ivanov, V. V. Vasin, and V. P. Tanana. Theory of Linear
Ill-Posed Problems and its Applications. Nauka, Moscow, 1978.
2001 CRC Press LLC
2001 CRC Press LLC
[130] T. Kato. Perturbation Theory for Linear Operators. Springer-Verlag,
Berlin, 1995. Reprint of the 1980 edition.
[131] H. Kellerman and M. Hieber. Integrated semigroups. J. Funct. Anal.,
84(1):160180, 1989.
[132] C. Knuckles and F. Neubrander. Remarks on the Cauchy problem for
multi-valued linear operators. In Partial Dierential Equations (Han-
sur-Lesse, 1993), pages 174187. Akademie-Verlag, Berlin, 1994.
[133] H. Komatsu. Semi-groups of operators in locally convex spaces. J.
Math. Soc. Japan, 16:230262, 1964.
[134] H. Komatsu. Ultradistributions, hyperfunctions and linear dierential
equations. In Colloque International CNRS sur les

Equations aux
Derivees Partielles Lineaires (Univ. Paris-Sud, Orsay, 1972), pages
252271. Asterisque, 2 et 3. Soc. Math. France, Paris, 1973.
[135] H. Komatsu. Ultradistributions. I. Structure theorems and a char-
acterization. J. Fac. Sci. Univ. Tokyo Sect. IA Math., 20:25105,
1973.
[136] H. Komatsu. Operational calculus, hyperfunctions and ultradistribu-
tions. In Algebraic Analysis, Vol. I, pages 357372. Academic Press,
Boston, MA, 1988.
[137] H. Komatsu. Operational calculus and semi-groups of operators. In
Functional Analysis and Related Topics, 1991 (Kyoto), pages 213
234. Springer, Berlin, 1993.
[138] T. K omura. Semigroups of operators in locally convex spaces. J.
Functional Analysis, 2:258296, 1968.
[139] R. Kravarusic, M. Mijatovic, and S. Pilipovic. Integrated semigroup
of unbounded linear operatorsCauchy problem. II. Novi Sad J.
Math., 28(1):107122, 1998.
[140] R. Kravarusic, M. Mijatovic, and S. Pilipovic. Integrated semigroups
of unbounded linear operators in Banach spaces. I. Bull. Cl. Sci.
Math. Nat. Sci. Math., (23):4562, 1998.
[141] S. G. Kren. Linear Dierential Equations in Banach Space. Amer-
ican Mathematical Society, Providence, R.I., 1971. Translations of
Mathematical Monographs, Vol. 29.
[142] S. G. Kren. Linear Equations in Banach Spaces. Birkh auser Boston,
Mass., 1982.
2001 CRC Press LLC
2001 CRC Press LLC
[143] S. G. Kren and K.I. Chernyishev. Singular-Pertubed Dierential
Equations in a Banach Space. Rossiskaya Akademiya Nauk. Sibirskoe
Otdelenie. Institut Matematiki im. S. L. Soboleva., Novosibirsk, 1979.
Preprint.
[144] S. G. Kren and M. I. Khazan. Dierential equations in a Banach
space. In Mathematical Analysis, Vol. 21, pages 130264. Akad. Nauk
SSSR Vsesoyuz. Inst. Nauchn. i Tekhn. Inform., Moscow, 1983.
[145] R. Latt`es and J.-L. Lions. The Method of Quasi-Reversibility. Appli-
cations to Partial Dierential Equations. American Elsevier Publish-
ing Co., Inc., New York, 1969.
[146] M. M. Lavrentev. Conditionally Well-Posed Problems for Dierential
Equations. Novosibirsk State University, Novosibirsk, 1973. Preprint.
[147] M. M. Lavrentev, V. G. Romanov, and S. P. Shishatsky. Ill-Posed
Problems of Mathematical Physics and Analysis. American Mathe-
matical Society, Providence, R.I., 1986.
[148] M. M. Lavrentev and L. Ya. Savelev. Linear Operators and Ill-Posed
Problems. Consultants Bureau, New York, 1995.
[149] Y. S. Lei and Q. Zheng. The application of C-semigroups to dier-
ential operators in L
p
(R
n
). J. Math. Anal. Appl., 188(3):809818,
1994.
[150] J. Liang and T.-J. Xiao. Integrated semigroups and higher order
abstract equations. J. Math. Anal. Appl., 222(1):110125, 1998.
[151] J. Liang and T.-J. Xiao. Wellposedness results for certain classes
of higher order abstract Cauchy problems connected with integrated
semigroups. Semigroup Forum, 56(1):84103, 1998.
[152] J.-L. Lions. Les semi groupes distributions. Portugaliae Mathematica,
19:141164, 1960.
[153] J.-L. Lions and E. Magenes. Non-Homogeneous Boundary Value
Problems and Applications. Vol. IIII. Springer-Verlag, New York,
19721973.
[154] Ju. I. Ljubic. The classical and local Laplace transform in the abstract
Cauchy problem. Uspehi Mat. Nauk, 21(3 (129)):351, 1966.
[155] G. Lumer. Examples and results concerning the behavior of general-
ized solutions, integrated semigroups, and dissipative evolution prob-
lems. In Semigroup Theory and Evolution Equations (Delft, 1989),
pages 347356. Dekker, New York, 1991.
2001 CRC Press LLC
2001 CRC Press LLC
[156] G. Lumer. Generalized evolution operators and (generalized) C-
semigroups. In Semigroup Theory and Evolution Equations (Delft,
1989), pages 337345. Dekker, New York, 1991.
[157] G. Lumer. Evolution equations. Solutions for irregular evolution prob-
lems via generalized solutions and generalized initial values. Appli-
cations to periodic shocks models. Ann. Univ. Sarav. Ser. Math.,
5(1):iiv and 1102, 1994.
[158] G. Lumer. Singular problems, generalized solutions, and stability
properties. In Partial Dierential Equations (Han-sur-Lesse, 1993),
pages 204216. Akademie-Verlag, Berlin, 1994.
[159] G. Lumer. Singular evolution problems, regularization, and appli-
cations to physics, engineering, and biology. In Linear Operators
(Warsaw, 1994), pages 205216. Polish Acad. Sci., Warsaw, 1997.
[160] G. Lumer and F. Neubrander. Asymptotic Laplace transforms and
evolution equations. In Evolution Equations, Feshbach Resonances,
Singular Hodge Theory, pages 3757. Wiley-VCH, Berlin, 1999.
[161] T. Matsumoto, S. Oharu, and H. R. Thieme. Nonlinear perturbations
of a class of integrated semigroups. Hiroshima Math. J., 26(3):433
473, 1996.
[162] I. V. Melnikova. The family M, N of operator functions and second-
order equations in Banach space. Izv. Vyssh. Uchebn. Zaved. Mat.,
2:4552, 87, 1985.
[163] I. V. Melnikova. A theorem of Miyadera-Feller-Phillips type for
a complete second-order equation in a Banach space. Izv. Vyssh.
Uchebn. Zaved. Mat., 4:3440, 86, 1985.
[164] I. V. Melnikova. Conditions for the solvability of abstract boundary
value problems. Izv. Vyssh. Uchebn. Zaved. Mat., 10:1724, 1990.
[165] I. V. Melnikova. Regularization methods for ill-posed boundary prob-
lems. In Ill-Posed Problems in Natural Sciences (Moscow, 1991),
pages 93103. VSP, Utrecht, 1992.
[166] I. V. Melnikova. Regularization of ill-posed dierential problems.
Sibirsk. Mat. Zh., 33(2):125134, 221, 1992.
[167] I. V. Melnikova. General theory of the ill-posed Cauchy problem. J.
Inverse Ill-Posed Probl., 3(2):149171, 1995.
2001 CRC Press LLC
2001 CRC Press LLC
[168] I. V. Melnikova. Properties of an abstract pseudoresolvent and
well-posedness of the degenerate Cauchy problem. In Generaliza-
tions of Complex Analysis and Their Applications in Physics (War-
saw/Rynia, 1994), pages 151157. Polish Acad. Sci., Warsaw, 1996.
[169] I. V. Melnikova. Properties of the Lions d-semigroups, and gener-
alized well-posedness of the Cauchy problem. Funktsional. Anal. i
Prilozhen., 31(3):2334, 95, 1997.
[170] I. V. Melnikova. The degenerate Cauchy problem in Banach spaces.
Izv. Ural. Gos. Univ. Mat. Mekh., (1):147160, 183, 1998.
[171] I. V. Melnikova. The method of integrated semigroups for Cauchy
problems in Banach spaces. Sibirsk. Mat. Zh., 40(1):119129, iii,
1999.
[172] I. V. Melnikova. Well-posedness of dierential-operator problems. I.
The Cauchy problem in spaces of distributions. J. Math. Sci. (New
York), 93(1):121, 1999. Functional analysis, 2.
[173] I. V. Melnikova. Abstract well-posed and ill-posed Cauchy problems
for inclusions. In Semigroups of Operators, pages 203212. Birkh auser
Boston, Mass., 2000.
[174] I. V. Melnikova and M. A. Alshansky. Well-posedness of the degener-
ate Cauchy problem in a Banach space. Dokl. Akad. Nauk, 336(1):17
20, 1994.
[175] I. V. Melnikova and M. A. Alshansky. Generalized well-posedness of
the Cauchy problem, and integrated semigroups. Dokl. Akad. Nauk,
343(4):448451, 1995.
[176] I. V. Melnikova and M. A. Alshansky. Well-posedness of the Cauchy
problem in a Banach space: regular and degenerate cases. J. Math.
Sci. (New York), 87(4):37323780, 1997. Analysis, 9.
[177] I. V. Melnikova, U. A. Anufrieva, and A. I. Filinkov. Laplace trans-
form of K-semigroups and well-posedness of Cauchy problems. Inte-
gral Transform. Spec. Funct., 9(1):3756, 2000.
[178] I. V. Melnikova, U. A. Anufrieva, and V. Yu. Ushkov. Degenerate
distribution semigroups and well-posedness of the Cauchy problem.
Integral Transform. Spec. Funct., 6(1-4):247256, 1998. Generalized
functionslinear and nonlinear problems (Novi Sad, 1996).
[179] I. V. Melnikova and S. V. Bochkareva. The Maslov property for the
quasireversibility method and the method of boundary value prob-
lems. In Investigations in the Theory of Approximations (Russian),
pages 4650, 121122. Ural. Gos. Univ., Sverdlovsk, 1988.
2001 CRC Press LLC
2001 CRC Press LLC
[180] I. V. Melnikova and S. V. Bochkareva. C-semigroups and the reg-
ularization of an ill-posed Cauchy problem. Dokl. Akad. Nauk,
329(3):270273, 1993.
[181] I. V. Melnikova and A. I. Filinkov. Classication and well-posedness
of the Cauchy problem for second-order equations in Banach space.
Dokl. Akad. Nauk SSSR, 276(5):10661071, 1984.
[182] I. V. Melnikova and A. I. Filinkov. A connection between the well-
posedness of the Cauchy problem for an equation and for a system in
a Banach space. Dokl. Akad. Nauk SSSR, 300(2):280284, 1988.
[183] I. V. Melnikova and A. I. Filinkov. Classication with respect to
boundary value problems of a complete second-order equation in a
Banach space. Izv. Vyssh. Uchebn. Zaved. Mat., 6:3945, 1990.
[184] I. V. Melnikova and A. I. Filinkov. An integrated family of M and
N functions. Dokl. Akad. Nauk, 326(1):3539, 1992.
[185] I. V. Melnikova and A. I. Filinkov. Integrated semigroups and
C-semigroups. Well - posedness and regularization of operator-
dierential problems. Uspekhi Mat. Nauk, 49(6(300)):111150, 1994.
[186] I. V. Melnikova and A. I. Filinkov. Well-posedness of dierential-
operator problems. II. The Cauchy problem for complete second-order
equations in Banach spaces. J. Math. Sci. (New York), 93(1):2241,
1999. Functional analysis, 2.
[187] I. V. Melnikova and A. V. Gladchenko. Well-posedness of the
Cauchy problem for inclusions in Banach spaces. Dokl. Akad. Nauk,
361(6):736739, 1998.
[188] M. Mijatovic and S. Pilipovic. Integrated semigroups, relations with
generators. Novi Sad J. Math., 27(2):6575, 1997.
[189] M. Mijatovic and S. Pilipovic. Integrated semigroups and distribution
semigroupsCauchy problem. Math. Montisnigri, 11:4365, 1999.
[190] M. Mijatovic, S. Pilipovic, and F. Vajzovic. -times integrated semi-
groups ( R
+
). J. Math. Anal. Appl., 210(2):790803, 1997.
[191] I. Miyadera. On one-parameter semi-group of operators. J. Math.
Tokyo, 1:2326, 1951.
[192] I. Miyadera. Generation of a strongly continuous semi-group opera-
tors. Tohoku Math. J. (2), 4:109114, 1952.
2001 CRC Press LLC
2001 CRC Press LLC
[193] I. Miyadera. On the generation of semi-groups of linear operators.
T ohoku Math. J. (2), 24:251261, 1972. Collection of articles dedi-
cated to Gen-ichir o Sunouchi on his sixtieth birthday.
[194] I. Miyadera. On the generators of exponentially bounded C-
semigroups. Proc. Japan Acad. Ser. A Math. Sci., 62(7):239242,
1986.
[195] I. Miyadera. A generalization of the Hille-Yosida theorem. Proc.
Japan Acad. Ser. A Math. Sci., 64(7):223226, 1988.
[196] I. Miyadera. C-semigroups, semigroups and n-times integrated semi-
groups. In Dierential Equations and Control Theory (Iasi, 1990),
pages 193207. Longman Sci. Tech., Harlow, 1991.
[197] I. Miyadera. C-semigroups and semigroups of linear operators. In Dif-
ferential Equations (Plovdiv, 1991), pages 133143. World Sci. Pub-
lishing, River Edge, NJ, 1992.
[198] I. Miyadera, S. Oharu, and N. Okazawa. Generation theorems of
semi-groups of linear operators. Publ. Res. Inst. Math. Sci., 8:509
555, 1972/73.
[199] I. Miyadera, M. Okubo, and N. Tanaka. -times integrated semi-
groups and abstract Cauchy problems. Mem. School Sci. Engrg.
Waseda Univ., (57):267289 (1994), 1993.
[200] I. Miyadera, M. Okubo, and N. Tanaka. On integrated semigroups
which are not exponentially bounded. Proc. Japan Acad. Ser. A Math.
Sci., 69(6):199204, 1993.
[201] I. Miyadera and N. Tanaka. Exponentially bounded C-semigroups
and generation of semigroups. J. Math. Anal. Appl., 143(2):358378,
1989.
[202] I. Miyadera and N. Tanaka. A remark on exponentially bounded C-
semigroups. Proc. Japan Acad. Ser. A Math. Sci., 66(2):3134, 1990.
[203] I. Miyadera and N. Tanaka. Generalization of the Hille-Yosida the-
orem. In Semigroup Theory and Evolution Equations (Delft, 1989),
pages 371381. Dekker, New York, 1991.
[204] V. A. Morozov. Methods for Solving Incorrectly Posed Problems.
Springer-Verlag, New York, 1984.
[205] F. Neubrander. Well-posedness of abstract Cauchy problems. Semi-
group Forum, 29(1-2):7585, 1984.
2001 CRC Press LLC
2001 CRC Press LLC
[206] F. Neubrander. Laplace transform and asymptotic behavior of
strongly continuous semigroups. Houston J. Math., (4):549561, 1986.
[207] F. Neubrander. Well-posedness of higher order abstract Cauchy prob-
lems. Trans. Amer. Math. Soc., 295(1):257290, 1986.
[208] F. Neubrander. On the relation between the semigroup and its in-
nitesimal generator. Proc. Amer. Math. Soc., 100(1):104108, 1987.
[209] F. Neubrander. Integrated semigroups and their applications to the
abstract Cauchy problem. Pacic J. Math., 135(1):111155, 1988.
[210] F. Neubrander. Integrated semigroups and their application to com-
plete second order Cauchy problems. Semigroup Forum, 38(2):233
251, 1989. Semigroups and dierential operators (Oberwolfach, 1988).
[211] F. Neubrander. The Laplace-Stieltjes transform in Banach spaces and
abstract Cauchy problems. In Evolution Equations, Control Theory,
and Biomathematics (Han sur Lesse, 1991), pages 417431. Dekker,
New York, 1994.
[212] F. Neubrander and A. Rhandi. Degenerate abstract cauchy problems.
In Seminar notes in Functional Analalysis and Partial Dierential
Equations, pages 112. Louisiana State University, 1992/1993.
[213] S.

Oharu. Semigroups of linear operators in a Banach space. Publ.
Res. Inst. Math. Sci., 7:205260, 1971/72.
[214] N. Okazawa. A generation theorem for semigroups of growth order
. T ohoku Math. J. (2), 26:3951, 1974.
[215] N. Okazawa. A remark on innitesimal generators of C-semigroups.
SUT J. Math., 25(2):123127, 1989.
[216] S.

Ouchi. Semi-groups of operators in locally convex spaces. J. Math.
Soc. Japan, 25:265276, 1973.
[217] M. M. H. Pang. Resolvent estimates for Schr odinger operators in
L
p
(R
N
) and the theory of exponentially bounded C-semigroups.
Semigroup Forum, 41(1):97114, 1990.
[218] E. Pap and S. Pilipovic. Semigroups of operators on the space of
generalized functions exp/

. J. Math. Anal. Appl., 126(2):501515,

1987.
[219] A. Pazy. Semigroups of Linear Operators and Applications to Partial
Dierential Equations. Springer-Verlag, New York, 1983.
2001 CRC Press LLC
2001 CRC Press LLC
[220] A. Pazy. Semigroups of operators in Banach spaces. In Equadi 82
(W urzburg, 1982), pages 508524. Springer, Berlin, 1983.
[221] S. Pilipovic. Generalization of Zemanian spaces of generalized func-
tions which have orthonormal series expansions. SIAM J. Math.
Anal., 17(2):477484, 1986.
[222] A. G. Rutkas. The Cauchy problem for the equation Ax

(t)+Bx(t) =
f(t). Dierentsialnye Uravnenija, 11(11):19962010, 2108, 1975.
[223] J. T. Sandefur. Higher order abstract Cauchy problems. J. Math.
Anal. Appl., 60(3):728742, 1977.
[224] J. T. Sandefur, Jr. Convergence of solutions of a second-order Cauchy
problem. J. Math. Anal. Appl., 100(2):470477, 1984.
[225] N. Sanekata. A note on the abstract Cauchy problem in a Banach
space. Proc. Japan Acad., 49:510513, 1973.
[226] N. Sanekata. Some remarks on the abstract Cauchy problem. Publ.
Res. Inst. Math. Sci., 11(1):5165, 1975/76.
[227] L. Schwartz. Theorie des distributions ` a valeurs vectorielles. I. Ann.
Inst. Fourier, Grenoble, 7:1141, 1957.
[228] L. Schwartz. Theorie des distributions ` a valeurs vectorielles. II. Ann.
Inst. Fourier. Grenoble, 8:1209, 1958.
[229] L. Schwartz. Some applications of the theory of distributions. In
Lectures on Modern Mathematics, Vol. I, pages 2358. Wiley, New
York, 1963.
[230] L. Schwartz. Mathematics for the Physical Sciences. Hermann, Paris,
1966.
[231] M. Shimizu and I. Miyadera. Perturbation theory for cosine families
on Banach spaces. Tokyo J. Math., 1(2):333343, 1978.
[232] R. E. Showalter. Quasi-reversibility of rst and second order parabolic
evolution equations. In Improperly Posed Boundary Value Problems
(Conf., Univ. New Mexico, Albuquerque, N.M., 1974), pages 7684.
Res. Notes in Math., No. 1. Pitman, London, 1975.
[233] R. E. Showalter. Hilbert Space Methods for Partial Dierential Equa-
tions. Pitman, London, 1977. Monographs and Studies in Mathemat-
ics, Vol. 1.
2001 CRC Press LLC
2001 CRC Press LLC
[234] N. A. Sidorov and E. B. Blagodatskaya. Dierential equations with
a Fredholm operator in the main dierential expression. Dokl. Akad.
Nauk SSSR, 319(5):10871090, 1991.
[235] N. A. Sidorov and M. V. Falaleev. Generalized solutions of degenerate
dierential and integral equations in Banach spaces. In The Method
of Lyapunov Functions in the Analysis of the Dynamics of Systems
(Irkutsk, 1985) (Russian), pages 308318, 328. Nauka Sibirsk. Ot-
del., Novosibirsk, 1987.
[236] Yu. T. Silchenko and P. E. Sobolevski. Solvability of the Cauchy
problem for an evolution equation in a Banach space with a non-
densely given operator coecient which generates a semigroup with
a singularity. Sibirsk. Mat. Zh., 27(4):93104, 214, 1986.
[237] P. E. Sobolevski. Semigroups of growth . Dokl. Akad. Nauk SSSR,
196:535537, 1971.
[238] P. E. Sobolevski and Yu. T. Silchenko. Semigroups of bounded linear
operators. In Mathematics Today 93, No. 8 (Ukrainian), pages 34
63. Vishcha Shkola, Kiev, 1993.
[239] M. Sova. Cosine operator functions. Rozprawy Mat., 49:47, 1966.
[240] M. Sova. Semigroups and cosine functions of normal operators in
Hilbert spaces.

Casopis Pest. Mat., 93:437458, 480, 1968.
[241] M. Sova. Linear dierential equations in Banach spaces. Rozpravy

Ceskoslovenske Akad. Ved

Rada Mat. Prrod. Ved, 85(sesit 6):82,
1975.
[242] M. Sova. Abstract Cauchy problem. In Equadi IV (Proc. Czechoslo-
vak Conf. Dierential Equations and their Applications, Prague,
1977), pages 384396. Springer, Berlin, 1979.
[243] M. Sova. Concerning the characterization of generators of distribution
semigroups.

Casopis Pest. Mat., 105(4):329340, 409, 1980.
[244] M. Sova. On a fundamental theorem of the Laplace transform theory.

Casopis Pest. Mat., 106(3):231242, 316, 1981.

[245] G. A. Sviridyuk. On the general theory of operator semigroups. Us-
pekhi Mat. Nauk, 49(4(298)):4774, 1994.
[246] T. Takenaka and N. Okazawa. A Phillips-Miyadera type perturba-
tion theorem for cosine functions of operators. T ohoku Math. J. (2),
30(1):107115, 1978.
2001 CRC Press LLC
2001 CRC Press LLC
[247] T. Takenaka and N. Okazawa. Abstract Cauchy problems for second
order linear dierential equations in a Banach space. Hiroshima Math.
J., 17(3):591612, 1987.
[248] T. Takenaka and N. Okazawa. Wellposedness of abstract Cauchy
problems for second order dierential equations. Israel J. Math.,
69(3):257288, 1990.
[249] N. Tanaka. On the exponentially bounded C-semigroups. Tokyo J.
Math., 10(1):107117, 1987.
[250] N. Tanaka. Holomorphic C-semigroups and holomorphic semigroups.
Semigroup Forum, 38(2):253261, 1989. Semigroups and dierential
operators (Oberwolfach, 1988).
[251] N. Tanaka. On perturbation theory for exponentially bounded C-
semigroups. Semigroup Forum, 41(2):215236, 1990.
[252] N. Tanaka and I. Miyadera. Some remarks on C-semigroups and inte-
grated semigroups. Proc. Japan Acad. Ser. A Math. Sci., 63(5):139
142, 1987.
[253] N. Tanaka and I. Miyadera. Exponentially bounded C-semigroups
and integrated semigroups. Tokyo J. Math., (1):99115, 1989.
[254] N. Tanaka and I. Miyadera. C-semigroups and the abstract Cauchy
problem. J. Math. Anal. Appl., 170(1):196206, 1992.
[255] N. Tanaka and N. Okazawa. Local C-semigroups and local integrated
semigroups. Proc. London Math. Soc. (3), 61(1):6390, 1990.
[256] H. R. Thieme. Integrated semigroups and integrated solutions to
abstract Cauchy problems. J. Math. Anal. Appl., 152(2):416447,
1990.
[257] H. R. Thieme. Positive perturbations of dual and integrated semi-
groups. Adv. Math. Sci. Appl., 6(2):445507, 1996.
[258] A. N. Tikhonov and V. Y. Arsenin. Solutions of Ill-Posed Problems.
V. H. Winston & Sons, Washington, D.C.: John Wiley & Sons, New
York, 1977.
[259] C. C. Travis and G. F. Webb. Compactness, regularity, and uniform
continuity properties of strongly continuous cosine families. Houston
J. Math., 3(4):555567, 1977.
[260] C. C. Travis and G. F. Webb. Cosine families and abstract nonlinear
second order dierential equations. Acta Math. Acad. Sci. Hungar.,
32(1-2):7596, 1978.
2001 CRC Press LLC
2001 CRC Press LLC
[261] C. C. Travis and G. F. Webb. Second order dierential equations
in Banach space. In Nonlinear Equations in Abstract Spaces (Proc.
Internat. Sympos., Univ. Texas, Arlington, Tex., 1977), pages 331
361. Academic Press, New York, 1978.
[262] C. C. Travis and G. F. Webb. An abstract second-order semilin-
ear Volterra integro-dierential equation. SIAM J. Math. Anal.,
10(2):412424, 1979.
[263] C. C. Travis and G. F. Webb. Perturbation of strongly continuous
cosine family generators. Colloq. Math., 45(2):277285, 1981.
[264] T. Ushijima. Some properties of regular distribution semi-groups.
Proc. Japan Acad., 45:224227, 1969.
[265] T. Ushijima. On the strong continuity of distribution semi-groups. J.
Fac. Sci. Univ. Tokyo Sect. I, 17:363372, 1970.
[266] T. Ushijima. On the abstract Cauchy problems and semi-groups of
linear operators in locally convex spaces. Sci. Papers College Gen.
Ed. Univ. Tokyo, 21:93122, 1971.
[267] T. Ushijima. On the generation and smoothness of semi-groups of
linear operators. J. Fac. Sci. Univ. Tokyo Sect. IA Math., 19:65127,
1972.
[268] T. Ushijima. Approximation theory for semi-groups of linear opera-
tors and its application to approximation of wave equations. Japan.
J. Math. (N.S.), 1(1):185224, 1975/76.
[269] V. V. Vasilev, S. G. Kren, and S. I. Piskarev. Operator semigroups,
cosine operator functions, and linear dierential equations. In Math-
ematical Analysis, Vol. 28 (Russian), pages 87202, 204. Akad. Nauk
SSSR Vsesoyuz. Inst. Nauchn. i Tekhn. Inform., Moscow, 1990. Trans-
lated in J. Soviet Math. 54 (1991), no. 4, 10421129.
[270] D.V. Widder. An Introduction to Laplace Transform Theory. Aca-
demic Press, New York and London, 1971.
[271] T.-J. Xiao and J. Liang. Integrated semigroups, cosine families and
higher order abstract Cauchy problems. In Functional Analysis in
China, pages 351365. Kluwer Acad. Publ., Dordrecht, 1996.
[272] T.-J. Xiao and J. Liang. Widder-Arendt theorem and integrated
semigroups in locally convex space. Sci. China Ser. A, 39(11):1121
1130, 1996.
2001 CRC Press LLC
2001 CRC Press LLC
[273] T.-J. Xiao and J. Liang. Laplace transforms and integrated, regular-
ized semigroups in locally convex spaces. J. Funct. Anal., 148(2):448
479, 1997.
[274] T.-J. Xiao and J. Liang. Semigroups arising from elastic systems with
dissipation. Comput. Math. Appl., 33(10):19, 1997.
[275] T.-J. Xiao and J. Liang. The Cauchy Problem for Higher-Order Ab-
stract Dierential Equations. Springer-Verlag, Berlin, 1998.
[276] T.-J. Xiao and J. Liang. Dierential operators and C-wellposedness
of complete second order abstract Cauchy problems. Pacic J. Math.,
186(1):167200, 1998.
[277] T.-J. Xiao and J. Liang. Approximations of Laplace transforms and
integrated semigroups. J. Funct. Anal., 172(1):202220, 2000.
[278] T.-J. Xio and J. Liang. On complete second order linear dierential
equations in Banach spaces. Pacic J. Math., 142(1):175195, 1990.
[279] A. Yagi. Abstract quasilinear evolution equations of parabolic type
in Banach spaces. Boll. Un. Mat. Ital. B (7), 5(2):341368, 1991.
[280] A. Yagi. Generation theorem of semigroup for multivalued linear
operators. Osaka J. Math., 28(2):385410, 1991.
[281] K. Yosida. Functional Analysis. Springer-Verlag, Berlin, 1995.
Reprint of the sixth (1980) edition.
[282] S. Zaidman. Well-posed Cauchy problem and related semigroups of
operators for the equation Bu

(t) = Au(t), t 0, in Banach spaces.

Libertas Math., 12:147159, 1992.
[283] S. Zaidman. Functional Analysis and Dierential Equations in Ab-
stract Spaces. Chapman & Hall/CRC, Boca Raton, FL, 1999.
[284] E. Zeidler. Applied Functional Analysis. Springer-Verlag, New York,
1995. Applications to mathematical physics.
[285] A. H. Zemanian. Generalized Integral Transformations. Dover Pub-
lications Inc., New York, second edition, 1987.
[286] J. Zhang and Q. Zheng. On -times integrated cosine functions. Math.
Japon., 50(3):401408, 1999.
[287] Q. Zheng. Solvability and well-posedness of higher-order abstract
Cauchy problems. Sci. China Ser. A, 34(2):147156, 1991.
2001 CRC Press LLC
2001 CRC Press LLC
[288] Q. Zheng. Applications of integrated semigroups to higher order ab-
stract Cauchy problems. Systems Sci. Math. Sci., 5(4):316327, 1992.
[289] Q. Zheng. A Hille-Yosida theorem for the higher-order abstract
Cauchy problem. Bull. London Math. Soc., 24(6):531539, 1992.
[290] Q. Zheng. Higher order abstract Cauchy problems with N-closedness.
Math. Japon., 38(3):531539, 1993.
[291] Q. Zheng. Perturbations and approximations of integrated semi-
groups. Acta Math. Sinica (N.S.), 9(3):252260, 1993.
[292] Q. Zheng. Strongly continuous M,N-families of bounded operators.
Integral Equations Operator Theory, 19(1):105119, 1994.
[293] Q. Zheng. Integrated cosine functions. Internat. J. Math. Math. Sci.,
19(3):575580, 1996.
[294] Q. Zheng and Y. S. Lei. Exponentially bounded C-semigroup and
integrated semigroup with nondensely dened generators. I. Approx-
imation. Acta Math. Sci. (English Ed.), 13(3):251260, 1993.
[295] Q. Zheng and Y. S. Lei. Exponentially bounded C-semigroups and
integrated semigroups with non-densely dened generators. III. Ana-
lyticity. Acta Math. Sci. (English Ed.), 14(1):107119, 1994.
[296] Q. Zheng and Y. Li. Abstract parabolic systems and regularized
semigroups. Pacic J. Math., 182(1):183199, 1998.
[297] S. Zubova and K.

Cernysov. A linear dierential equation with a Fred-
holm operator acting on the derivative. Dierentsialnye Uravnenija
i Primenen.Trudy Sem. Processy Optimal. Upravlenija I Sektsija,
(Vyp. 14):2139, 107, 1976.
2001 CRC Press LLC
2001 CRC Press LLC
Glossary of Notation
Generalities
:= equality by denition
= logical implication
for all
there exist
the empty set
the end of proof
v.p. Cauchy principal value (of an integral)
Sets, functions and numbers
_
x E [ P
_
the subset of E whose elements possess property P
E F the cartesian product of sets E and F
E F the set E is contained in F with continuous injection
I
E
the identity mapping in E
f[

the restriction of function f : E F to the

subset of E
a
k
the sequence a
1
, . . . , a
k
, . . .
E the closure of E
E the boundary of E
N, R, C the sets of natural numbers, real numbers and
complex numbers
R
N
the set of all real N-tupels (r
1
, . . . , r
n
)
2
E
the set of all subsets of E
f(x) =
xa
O(g(x)) [f(x)[ const [g(x)[ in a neighborhood of a
o(x) a function satisfying [o(x)/x[ 0 as x 0
Re z the real part of z C
Imz the imaginary part of z C
arg z the argument of z C
[[x]] the integer part of x R
2001 CRC Press LLC
2001 CRC Press LLC
or (t) the Dirac distribution

a
or (t a) the Dirac distribution concentrated at a
H() the Heaviside function
f g the convolution of f and g
suppf the support of f
Operators and spaces
X Y the direct sum of vector spaces X and Y
X/Y the quotient space of X by Y if Y X
|x|
X
or |x| norm of x X
I
X
the identity operator in X
D(A) the domain of an operator A : X Y
ranA the range (image) of A
ker A the kernel of A
(A) the resolvent set of A

B
(A) the B-resolvent set of A
(A) the spectrum of A
A
1
the inverse of A
A

the adjoint of A
A the closure of A
|A| the norm of A
A[
F
the restriction of A to the subspace F X
R
A
() the resolvent of A:
R
A
() := (I A)
1
for (A)
[D(A)] the Banach space
_
D(A), |x|
A
= |x| +|Ax|
_
[ D(A
n
) ] the Banach space
_
D(A
n
), |x|
A
n = |x| +|Ax| +. . . +|A
n
x|
_
L(X, Y ) the set of all continuous (or bounded) linear mappings
of X with values in Y . If X and Y are Banach spaces,
then so is L(X, Y ) equipped with the norm:
|A|
L(X,Y )
= sup
x
X
|Ax|
2001 CRC Press LLC
2001 CRC Press LLC
L(X) L(X) := L(X, X)
X

the dual of X, i.e. X

= L(X, C)
x, y) the bracket of duality between x X and y X

l
2
the space of sequences x
k
C
with

k=1
[x
k
[
2
<
Spaces of functions and distributions
Let be an open set in R
N
and K be a compact set in R
N
.
C(K) the space of functions continuous on K
with |f|
C(K)
= sup
tK
[f(t)[
C
0
(R
N
) the space of continuous functions on R
vanishing at innity
C
k
(K) the space of k-times continuously dierentiable
functions equipped with the norm:
|f|
C
k
(K)
=

k
|j|=0
|D
j
f|
C(K)
,
where
D
j
=

j
1
+...+j
N
x
j
1
1
...x
j
N
N
, [j[ = j
1
+. . . +j
N
T() the space of Schwartz test functions
T

() the space of distributions on , i.e.

T

() = L(T(), C)
L
p
() the space of classes of measurable functions
on equipped with the norm:
|f|
L
p =
_
_

[f(x)[
p
dx
_
1/p
, 1 p <
H
k
() Sobolev space:
:=
_
f L
2
()

f L
2
(), : [[ k
_
,
where
= (
1
, . . . ,
N
), [[ =
1
+ +
N
, D

f =

||
f
x

2001 CRC Press LLC

2001 CRC Press LLC
H
k
0
() the closure of T() in H
k
()
H
k
() the dual space of H
k
0
()
W
k,p
() :=
_
f L
p
()

f L
p
(), : [[ k
_
C
_
E, X the space of functions continuous on E,
with values in a Banach space X
C
k
_
E, X the space of k-times continuously dierentiable functions
on E, with values in a Banach space X
L
2
_
E, X the space of square integrable functions on E,
with values in a Banach space X
2001 CRC Press LLC
2001 CRC Press LLC