Cazenave & Haraux - An Introduction to Semilinear Evolution
-
Upload
juan-montealegre -
Category
Documents
-
view
219 -
download
0
Transcript of Cazenave & Haraux - An Introduction to Semilinear Evolution
-
8/18/2019 Cazenave & Haraux - An Introduction to Semilinear Evolution
1/198
n Introduction to
Sem ilinear Evolution E quations
n z 1 N I V E R S I D A D C O M P L U T E N S E
Revised edition
. ^ f ?
I I I I I ^ ^ ^ ^ I I I I ^ I I ^ ^ ^ I ^ I I I I ^ I I I I I ^ I I ^ I I I
3 9 45
Thierry C azenave
CN R S and Univ ersity of Paris V I France
n
lain H araux
CN R S and Univ ersity of Paris V I France
Translated by
Yvan Martel
Univ ersity of Cergy-Pontoise France
k
0R 5/ ??0
CL RENDON PRESS • OXFORD
99
-
8/18/2019 Cazenave & Haraux - An Introduction to Semilinear Evolution
2/198
Oxford University Press, Great Clarendon Street, Oxford 012 6DP
Oxford New Y ork
A thens A uckland B angkok B ogota B uenos A ires Calcutta
Cape Town Chennai Dares Salaam Delhi Florence HongKong Istanbul
Karachi KualaLumpur Madrid Melbourne Mexico City Mumbai
Nairobi Paris
Sao
Paolo Singapore Taipei Tokyo Toronto Warsaw
and associated com panies in
Berlin Ibadan
Ox ford is a trade m ark o f O xf ord University Press
Published in the U nited States
by O xf ord University Press Inc., N ew Y ork
Introduction aux problem es dev olution sem i-lineaires
© E dition M arketing SA , 1990
First published by Ellipses
Translation © Ox ford University Press, 1998
A ide par le m inistere f rancais charge de la culture
A ll rights reserv ed. N o part of thispublication may be reproduced,
stored in a retrieval syste m , or transm itted, in any f orm or by any m eans,
without the prior permission in writing of Oxford University Press.
W ithin the UK, ex ceptions are allow ed in respect of any f air dealing for the
purpose of research or private study, or criticism or review, as permitted
under the Copy right, Designs and Patents A ct, 1988, or in the case of
reprographic reproduction in accordance w ith the term s of licences
issued by the Copy right Licensing A gency . Enquiries concerning
reproduction outside tho se terms and in other countries should be sent to
the Rights Department, Oxford University Press,
at the address abov e.
This book is sold subject to the condition that it shall not, by w ay
of trade o r otherwise, be lent, re-sold, hired out, or otherw ise circulated
w ithout the publisher's prior consent in any f orm of binding or cover
other than that in w hich it is published and w ithout a similar condition
including this condition being impo sed on the subseq uent purchaser.
A catalogue record for this book is available from the B ritish L ibrary
Library of Congress Cataloging in Publication Data
(Data available)
ISBNO 19 850277 X(Hbk)
Typeset by Y van Martel
Printed in Great B ritain by
Bookcraft (Bath) Ltd,
M idsomer Norton, Av on
-
8/18/2019 Cazenave & Haraux - An Introduction to Semilinear Evolution
3/198
Preface
This book is an expa nded version of a post-gradua te course taught for severa l
yea rs at the La boratoire d'Ana lyse Nume rique of the Universite Pierre e t Marie
Curie in Pa ris. The purp ose of this course w as to give a self-contained pre sen-
tation of some rece nt results concerning the fundam ental properties of solutions
of semilinear e volution partial differe ntial equations, with special em phasis on
the asym ptotic be haviour of the solutions.
We begin with a brief description of the abstract theory of sem ilinear evolu-
tion equa tions, in orde r to provide the re ade r with a sufficient bac kground. In
par ticula r, we rec all the ba sic results of vector integra tion (Chapter 1) and lin-
ear semigroup theory
i n
Bana ch spaces (Chapters 2 and 3). Chapter 4 c oncerns
the local e xistence, uniqueness, and re gularity of solutions of abstract sem ilinear
problems.
In Nature, many propagat ion phenome na are described by evolut ion equa-
tions or evolution systems wh ich ma y include non-linea r intera ction or self-
intera ct ion term s. In Cha pters 5, 6, and 7, we a pply some gene ral me thods
to the following thre e problem s.
(1) The he at equa tion
ut
=Au 01
which models the therma l energy transfer in a hom ogeneous me dium, is the
simplest exam ple of a di f fusion equ ation. This equa tion, as we ll as the sel f -
interac tion problem
Ut
= Au
+f u,
0.2)
can be considered on the entire space
RN
or on various doma ins S1 (bounded
o r
not) of RN . In the case in which
c i
# RN,
we ne ed to speci fy a boundary
condition on I' = 852. It ca n be, for exam ple, a hom ogeneous D irichlet condition
u=0on, 03
o r a
homogeneous Neu ma nn condition
au _
an
0 our,
0.4)
-
8/18/2019 Cazenave & Haraux - An Introduction to Semilinear Evolution
4/198
vi
reface
Chapter 5 studies in detail the properties of the solutions of (0.2)—(0.3) when Sl is
bounde d. In this problem, the m aximu m principle plays an important role. This
is the rea son for studying equa tion (0.2) in the spa ce of continuous functions.
Vector-valued generalizations of the form
aui
c
czAui
+ fi(ul,
... ,
uk),
= 1 , ... , k,
0.5)
ca lled re ac tion— diffusion system s, often arise in chem istry and biology. One of
the m ain tools in the study of these system s (and in pa rt icular o f their non-
nega tive solutions) is the ma ximum principle, which gives
a priori
estimates
in
L
0°
(5l) k
for the trajectories. We thus develop Co methods rather than L
2
me thods, which are ea sier but less suitable in this fram ework.
(2 )
The wa ve equa tion (also ca lled the Klein— Gordon equa tion)
t
A u — m u ,
0.6)
with m > 0, mode ls the propa gation of different kinds of wave s (for exam ple light
wave s) in homogene ous media. Non-linea r m odels of conservative type ar ise in
quantum m echa nics, where as variants of the form
Utt
= Au —
f (u,
Ut)
0.7)
appe ar in the study of vibrat ing system s with or without dam ping, and with
or without forcing terms. Other perturbations of the wave equation arise in
electronics (the telegraph e qua tion, semi-conductors, etc.).
The ba sic me thod for studying
(0.6)
with suitable bounda ry c onditions (for
exa mple (0.3)) consists of introducing the a ssociated isome try group in the e n-
ergy space Hl
x L
2
. Loca l existenc e a nd uniqueness of solutions is established in
this spac e. Howe ver, in genera l, the solutions are differe ntiable only in the sense
of the larger spac e
L 2 x H-1.
These local questions are considered in Chapter
6.
(3 )
The S chrodinger e quation
iUt
= Au,
0.8)
possesses a c ombination of the pr operties described in (1) and (2). Pr imarily a
simplified mode l for some problem s of optics, this equa tion also arises in qua n-
tum field theory, possibly coupled w ith the Klein— Gordon e qua tion. Various
non-linear pe rturbations of (0.7) have appe are d rec ently in the study of lase r
beam s when the cha racter istics of the medium depend upon the tempera ture;
for example, focusing phenomena in some solids (where the medium c an brea k
down if the temperature reaches a critical point) and contrastingly, defocusing
-
8/18/2019 Cazenave & Haraux - An Introduction to Semilinear Evolution
5/198
P r e f a ce
ii
from the source.
close examination of sharp properties of solutions of the
non-linear Schrodinger equation is delicate, since this problem has a mixed or
degenerate nature (neither parabolic nor hyperbolic).
n Chapter 7, which is
devoted to Schrodinger's equation, it becomes clear that even the local theory
requires very elaborate techniques.
The choice of these three problems as model examples is somewhat arbitrary.
This selection was motivated by the limited experience of the authors, as well as
by the desire to present the easiest models (in particular, semilinear models) for
a first approach to the theory of evolution equations. We do not address several
other equally worthy problems, such as transport equations, vibrating plates,
and fundam ental equations of fluid mechan ics (such as Bo ltzmann's equation, the
Navier—Stokes equation, etc.). Such complicated systems require many specific
methods which could not be covered or even approached in a work of this kind.
Chapters 8, 9, and 10 are devoted to some techniques and results concerning
the global behaviour of solutions of semilinear evolution problems as the time
variable converges to infinity. In Chapter 8, we establish that, for several kinds
of evolution equations, the solutions either blow up in finite time in the original
space or they are uniformly bounded in this space for all t >_ 0.
his is the
case for the heat equation and the Klein—Gordon equation with attractive non-
linearity,
as
well as for non-autonomous problems with dissipation.
o such
alternative is presently known for Schrodinger's equation. Chapter 9 is devoted
to some basic notions of the theory of dynamical systems and its application to
models (1) and (2) in an open, bounded domain of R
N
.
We restrict ourselves
to the basic properties, and we give an extensive bibliography for the interested
reader. In Chapter 10, we study the asymptotic stability of equilibria. We also
discuss the connection between stability and positivity in the case of the heat
equation.
Finally, in the notes at the end of each chapter there are various bibliograph-
ical comments which provide the reader with a larger overview of the theories
discussed. Moreover, the limited character of the examples studied is compen-
sated for by a rather detailed bibliography that refers to similar works.
e
hope that this bibliography will serve our goal of a sufficient yet comprehensible
introduction to the available theory of evolution problems. At the time of pub-
lication, new results will have made some parts of this book obsolete. However,
we think that the methods presented are, and will continue to be for some years,
an indispensable basis for anyone wanting a global view of evolution problems.
Pa r is
. C.
1998
. H .
U
-
8/18/2019 Cazenave & Haraux - An Introduction to Semilinear Evolution
6/198
Contents
Notation .
.
.
.
. . . . .
iii
1. Preliminary results . . . . . . .. . . . . . . . . . . . . . . . 1
1.1.
Some abstrac t tools
.
.
1.2.
The e xponential of a linea r continuous operator
.
1.3.
Sobolev spaces
.
.
1.4. Vector-valued functions
.
1.4.1.
easurable functions
.
.
1.4.2. ntegrable functions
..
.
1.4.3.
he spaces
LP(I,X)
.
1.4.4.
ec tor-valued distributions
.
0
1.4.5.
he spaces W 1, P(I,X)
.
3
2. m-
dissipative operators
. . . . . . . . . . . . . . . . . . .
8
2.1.
Unbounded operators in Ba nach spaces
.
.
8
2.2.
De finition a nd m ain prope rties of m-dissipative ope rators . .
9
2.3.
Extrapolation
.
1
2.4.
Unbounde d operators in Hilbert spac es .
2
2.5.
Complex Hilbert space s
.
. 5
2.6.
Exam ples in the theory of pa rtial differential equations
.
6
2.6:1.
he La plac ian in an open subset of R
N:
2 theory
.
.
6
2.6.2.
he La plac ian in an open subset of R
N
:
o theory . .
7
2.6.3.
he wa ve operator (or the Klein— Gordon operator)
in Ha (1l)
x
L
2 (1l)
.
9
2.6.4.
he wa ve operator (or the Klein— Gordon operator)
in L
2
(1) x H
—
'(Il)
.
0
2.6.5.
he Schrodinger operator
.
.
3. The Hille—Yosida—Phillips Theorem and applications . . . . 33
3
The sem igroup gene rated by an m -dissipative operator
.
3
3.2.
Two important special cases
.
.
5
-
8/18/2019 Cazenave & Haraux - An Introduction to Semilinear Evolution
7/198
x Contents
1 The heat equation. . . . . . . . . . . . . . . . .
2
3.5.2. The wave equation (or the Klein—Gordon equation) . . . 47
3.5.3. The Schrodinger equation . . . . . . . . . . . . . .
7
3.5.4. The Schrodinger equation in Rr` . . . . . . . . . . .
8
4.
Inhomogeneous equations and abstract semilinear
problems .
.
.
. . . . . . .
. . .
.
.
0
4.1. Inhomogeneous equations . . . . . . . . . . . . . . . . .
0
4.2. Gronwall's lemma . . . . . . . . . . . . . . . . . . . .
4
4.3. Semilinear problems . . . . . . . . . . . . . . . . . . .
5
4.3.1. A result of local existence . . . . . . . . . . . . . .
6
4.3.2. Continuous dependence on initial data . . . . . . . .
9
4.3.3. Regularity
. . . . . . . . . . . . . . . . . . .
0
4.4. Isometry groups . . . . . . . . . . . . . . . . . . . . .
1
5.
The
heat equation . . . . . . . . . . . . . . . . . . . . .
2
5.1.
Preliminaries
.
2
5.2.
Local existence
.
.
4
5.3.
Global existenc e
.
5
5.4.
Blow-up in finite time
.
.
2
5.5.
Applica t ion to a m odel case
.
.
6
6.
The Klein—Gordon equation . . . . . . . . . . . . . . . .
8
6.1.
Preliminaries
.
8
6.1.1.
n abstrac t result
.
8
6.1.2.
unctionals on Ho (S2)
.
9
6.2.
Local e xistence .
.
2
6.3.
Global existence
.
4
6.4.
Blow-up in finite time
.
.
7
6.5.
Applica t ion to a m odel case
.
.
9
7.
The Schrodinger equation
. . . . . . . . . . . . . . . . .
1
7.1. Preliminaries
. . . . . . . . . . . . . . . . . . . . .
1
7.2. A gene ral result
. . . . . . . . . . . . . . . . . .
2
7.3. The linear Schrodinger equation in RN . . . . . . . . . . .
5
7.4. The non-linea r Sc hrodinger equa tion in R
N:
local existence . . 100
7.4.1. Some estimates . . . . . . . . . . . . . . . . . . 101
7.4.2. Proof of Theorem 7.4.1 . . . . . . . . . . . . . . . 106
7.5. The non-linea r Sc hrodinger equa tion in R
N
:
global existence . . 112
-
8/18/2019 Cazenave & Haraux - An Introduction to Semilinear Evolution
8/198
Contents xi
7.6.
The non-linea r Sc hrodinger e quation in R
N
:
low up
I
n finite time
.
14
7.7.
A rem ark c oncer ning behaviour at infinity
.
.
20
7.8.
Applica tion to a model ca se
.
2 1
Bounds on global solutions
.
24
8.1.
The he at equa t ion
.
24
8.1.1.
singular G ronwa ll's lemm a: a pplica tion to the he at
equation
.
25
8.1.2.
niform estima tes
.
29
8.2.
The Klein—G ordon equation
.
30
8.3.
The non-autonomous heat equation
.
.
34
8.3.1.
he Ca uchy problem for the non-autonomous heat
equation
.
34
8.3.2.
priori
estimates
.
3 5
8.4.
The dissipative non-autonomous Klein— Gordon e quation
.
37
The
invariance principle and some applications
.
4 2
9.1.
Abstract dynamica l system s
.
.
4 2
9.2.
Liapunov functions and the invarianc e principle
.
4 3
9.3.
A dynam ical system associated with a semilinear evolution
equation
.
4 5
9.4.
Application to the non-linea r hea t equation
.
46
9.5.
Application to a dissipative Klein— Gordon equa tion
. .
49
. Stability
of
stationary solutions
.
54
10.1. Definitions and simple exam ples
.
54
10.2. A simple genera l result
10.3. Exponentially stable system s govern ed by PD E
.
5 6
.
5 8
10.4. S tability a nd positivity
.
64
10.4.1. The one-dimensional case
.
.
6 5
10.4.2. The m ultidimensional ca se
.
67
.
6 9
.
.
8 5
-
8/18/2019 Cazenave & Haraux - An Introduction to Semilinear Evolution
9/198
the space of linear, continuous mappings from
X to
Y
the space of linear, continuous mappings from
X to
X
the topological dual of the vector space X
the Banach space
(D(A)
,
( I
I I D (A) )
with II
u I ID ( A ) = I I
uiI + IIAuII,
when A is a linear operator with a closed graph
the space of C°° (real-valued or complex valued) functions with
compact support in S2
= C°°(^) = D( l)
the space of continuous functions with compact support in S2
the space of functions of C(S2) w hich are zero on 011
the space of distributions on 11
the space of measurable functions on 11 such that
I u I P is integrable
(1 < p< oo)
= (f
n
u ) , for u E
Lp(1)
the space of measurable functions u on SZ such that there exists C
such that
I u(x ) I < C f or
almost every x E 1 1
= Inf{C > 0, Iu(x)I < C almost everywhere}, for u E L(1)
the conjugate exponent of
p,
i.e. p' = p/(p — 1)
for 1 < p < oo
0
kI
=
= (ai, ... , aN),
IaI = E a^
_ { f E
LP (St),
Da f E L P (S2)
for all
aE N' such that
I al < m }
_ .Ijcj
-
8/18/2019 Cazenave & Haraux - An Introduction to Semilinear Evolution
10/198
xiv Notation
H
Wo 2
(fl)
D(I,
X)
he space of C°° functions with compact support from
I
to
X
u
'
u
t
=
du/dt, for u
E D'(I, X)
C,(I,X)
the space of continuous functions with compact support from
I to
X
Cb(I, X)
the space of continuous and bounded functions from
I
to
X
Cb,
w
(I,X)
the space of uniformly continuous and bounded functions from
I
to X
LP(I,X)
he space of measurable functions u on
I
with values in
X
and such
that
I I u I I P
is integrable (1 < p < oo)
I I U I I L P
( I i IuIP)1
/P,
for u
E LP(I,X)
L(I,X
he space of measurable functions
u
on
I
such that there exists C
such that
I I u ( x ) I I
< C for almost every x E I
I I u I I L ° °
Inf{C
> 0,
Iu(x)I
< C almost everywhere}, for
u
E LO°(I, X)
W
1 'P(I,X) = {u
E LP(I,X), u' E LP(I,X),
in the sense of
D'(I,X)}
I I u I I w
i,P
I IU I I L P +
I I n ' I I L P
for u
E W1,P(I,X)
-
8/18/2019 Cazenave & Haraux - An Introduction to Semilinear Evolution
11/198
u
1
Preliminary results
1.1. Some abstract tools
We r eca ll here some classical theorem s of functional ana lysis that are nece ssary
for the study of semilinear evolution equations. The proofs can be found in
Brezis [2].
Theorem 1.1.1. (The Banach Fixed Point Theorem)
Let
(E,
d)
be a
c o m -
pletemetric
space and let f
: E –+E
be a mapping
such
that there
exists
k E [0, 1)
satisfying
d(f (x), f (y))
<
kd(x,
y) for all (x,
y) E E x
E. Then there
exists a unique point
10
E
E
such that f (xo) = xp.
Theorem 1.1.2.
(The Closed Graph Theorem)
Let X and Y
be Banach
spaces
and let A: X — Y be a linear mapping. Then A E
L(X,Y) if and only
if the graph of A
is a c losed subspace
of X x Y.
Remark 1.1.3. We recall that the graph of
A is G(A) = {(
x, y) E
X x Y; y =
Ax}.
Theorem 1.1.4. (The Lax—Milgram Theorem)
Let H be a Hilbert space
and let
a : H x H – IR
be a
bilinear functional.
Assume that there e xist two
constants C <
oo,
a
> 0 such that:
(i ) Ia(u,v)I
ClIull IMI
for all (u, v)
E
H x H (continuity);
(ii)
a(u, u)
>
a I I u I I 2
for all u
E
H (coerc iveness) .
Then,
f o r every
f
E
H* (the
dual space of H), there exists a unique
u
E
H such
that
a(u, v) = (f, v) for ally E H.
1.2. The expon ential of a linear continuous operator
Let
X
be a Ba nach space a nd let
A
E
C(X).
Definition 1.2.1. We denote by e A
the sum of the series E
n
An.
n> 0
It is clear tha t the ser ies is norm c onver gent in
C(X)
an d that Ile
A ll <
e
l I A 1 I
.
-
8/18/2019 Cazenave & Haraux - An Introduction to Semilinear Evolution
12/198
2
reliminary resu lts
In ad dition, for f ixed
A,
the func tion
t
'-4
e tA belongs to C°° (R, £(X))
and we
have
dt
e
tA = e1AA = Ae
IA
for a ll t
E
R. Fina lly, we have the following classica l result.
Proposition 1.2.2.
Let A
E
£(X).
For a ll
T> 0 and all x
E
X,
there
exists
a unique solution u E
C
1
([0,
T I,
X) of the following problem:
u (t) = Au(t), for alit
E [0,T];
u(0) = x.
This
solution is
given by u(t) = e
tA
x, for all t E [0, T].
Proof. It is clea r that e
tA
x
is a solution there fore, we ne ed only show unique-
ness. Let v be anothe r solution a nd let z(t)
= e-tAV(t). We have
z
'(
t ) = e
-tA
(Av(t)) - A(e
-tA
V(t)) = 0.
Therefore,
z(t) __ z(0) = x;
and so v(t) = ex.
1 3 Sobolev spaces
We refer to Adams [1] for the proofs of the results given below. Consider an
open subset S2 of R
N
. A distribution
T
E
D'(S2)
is said to belong to LP(11)
(1
-
8/18/2019 Cazenave & Haraux - An Introduction to Semilinear Evolution
13/198
Sobolev spaces 3
Then Hm(1l) is a Hilbert space with the scalar produc t
u )Hm
=
D
a
uD
a
v dx
IIm
If S2 is bounded, there e xists a c onstant C(S2) such that
u I I L 2 < _ C ( c l )I IV U I IL 2 ,
for all u E Ho
(S2) (this is Poincar e's inequality). It may be more convenient to
equip Ho (1l)
with the following sca lar produc t
(u, v) =
in
Vu • Vv dx,
which defines an e quivalent norm to
n the c losed space HH(SZ). The
following two resu lts ar e e ssential in the theory of pa rtial differe ntial equations.
Theorem
1 3 1
If S2
is open and
has a Lipschitz continuous
boundary, then:
(i) if 1 < p < N, then W
1
'P(IZ)
q(fl),
for every q
E [p,p *],
where p* _
Npl (N — p);
(ii)
if p = N,
then W 1 'P(S2) y L
9
(1l), for every q
E [p, oo);
(iii) if p> N, then W r'r(cl) - - - >
L(c) fl C°' 1 1
(52),
where
a = (p — N)/p.
Theorem 1.3.2.
In
addition, if 11
is bounded, embeddings (ii) and (iii)
of
Theorem 1.3.1
are com pact . Embedding (
i) is
compact for q
E [p,p*).
Remark
1.3.3.
The c onclusions of Theorem s 1.3.1 a nd 1.3.2 re ma in valid
without any smoothness assum ption on
5 2 ,
i f one replac es
W
1
'P(1l)
by WW'P(c)
We also reca ll the following result (see F riedman [1], Theorem 9 .3, p. 24).
Theorem 1.3.4.
Let q,
r
be such that 1
< q,
r
< oo,
and let j, m be integers,
0 < j < m.
Leta
E
[j/m,1]
(a _ 0), and
let p
be
given by
p
1/
Then there
exists C(q, r,
j, m, a, n)
such that
a
DauIIL1
C
I D a u I I L -
I u I I L 9 a ,
IcI=7
k I = ' m
-
8/18/2019 Cazenave & Haraux - An Introduction to Semilinear Evolution
14/198
4 Preliminary results
Fina lly, we r ec all the following com position rule (see M ar cus a nd Mizel [11).
Proposition 1.3.5.
Let F
: R - IR be a Lipsch itz continuous func tion, and
let 1
-
8/18/2019 Cazenave & Haraux - An Introduction to Semilinear Evolution
15/198
Vector-valued functions 5
I \En .
Let k(n) be such tha t II
fn,k(n)
fnII < 1/n on
I \E
n
and let g
n
= fk(n).
Take F = E U (f u E
n
) then IFI = 0. Let
t E I \ F.
We have
f, (t) —* f (t);
m>O n>m
on the other ha nd, for
n
large enough,
t E I \E
n
.
It follows that IIg
n
— fn
I
f (t)
and so f
is masurabe.
Remark 1.4.4. If
f : I -- X
and
cp : I
-4 R are mea surable, then
cp f : I --> X
is measurable.
Remark 1.4.5. If
(xn)n>o
is a family of elements of
X
and if (w
n ) >
0
is
a family of measurable subsets of
I
uch that
w i
n
w^
_ 0 for
i j,
hen
i
>o
x1
s measurable.
Proposition 1.4.6. (Pettis' Theorem)
Consider f : I —i X. Then f is
mea surable i f
and
only i f the fo l lowing two co nditions
are satisfied:
(i )
f is weakly
me asura ble (i.e.
for every x' E X*, the function t H (x', f (t))
is measurable);
(ii)
th ere
exists a set
N C I of measure 0
such that
f (I \ N) is sep a ra ble.
Proof.
First, since
f
is mea surable, it is clear that
f
is weakly measurable.
Now let (fn)n>o
C C
c
(I,X)
be a sequence such that
fn -i f on
I\N
as n --> oo,
where INI = 0. It is clear that
f
n
(I \ N)
is separa ble, and then so is
f (I \ N).
Conversely, we ma y assume that
f (I)
is separa ble, so that
X
is sepa rable
(by possibly re placing
X
by the smallest closed subspace of
X
containing
f (I)).
We nee d the following lem ma (see Yosida [1], p. 132).
Lemma 1.4.7.
Let X be
a separable Banach space,
let X* be its dual, and
let S*
be the unit
ball of X. There exists a sequence (x)
>
of
f S* such
that,
for every x' E S*, there exists
a subsequence (x'
nk )k> o
of (x)> withith
xnk
(x)—>x'(x) for all x c X.
Proof.
Let
(xn)n>o
be dense in
X.
For a ll
n> 0,
define Fn
: S* -* 2
2
(n), by
Fn(x
) = (x
(x1)
...,x
'
1xn)),
for a ll
x' E X*. Since
t
2
(n)
is separ able, there exists a seque nce (xn
k )k>o of
S *
such that F
n
((xn
k
)k>
o
) is dense in F
n
(S*).
In particular , for all
x'
E X*, there
exists xn E S* such that
x'
( x.7 ) —
x
lk(,. (x
7 ) I
-
8/18/2019 Cazenave & Haraux - An Introduction to Semilinear Evolution
16/198
6 Preliminary
results
for 1
oo, for all
j
E N. Since
(xn )
n
>o is dense in
X ,
we deduce easily that
x n
k(n)
(x) —* x (x)
as
n --f
oo, for
all x E X. The result follows.
End
of the
proof of Proposition 1.4.6.
Let
x E X.
Then
t H
f(t) — xli
is
measurable. Indeed, for all a > 0,
{ t ,
IIf(t)—x11
-
8/18/2019 Cazenave & Haraux - An Introduction to Semilinear Evolution
17/198
Vector-valued functions 7
1.4.2.
ntegrable functions
Definition 1.4.10. A measurable function
f : I -> X
is integra ble if there
exists a sequenc e ( f
n
)
n
>o C C
c
(I, X) such tha t
f
Ilfn(t)
— f ( t ) I I d t - - , 0
asn ->oo.
Remark 1.4.11. II
I
f
- f
II is non-negative and measura ble, and so
f
IIfn
- f I
makes sense.
Proposition 1.4.12.
Let f : I - X
be integrable.
Ther e
exists
x
E X such
that
if a sequence (f
n
)
n
> o
C C,(I,
X) satisfies f
I
IIfn
- f
II —4
0, as n -
then
one
has
f
fn —* x as n -* oo.
Proof..
We have
f fn—^f
P
) < f
IIfn—fII+jIIf—ftI.
Therefore,
f
fn
is a Cauc hy sequence that converges to some element
x E X.
Consider a nother sequenc e
( 9 n
)
n >o
that satisfies
f
II9n
- f
II — > 0 as
n --> oo.
We have
I I J7
9 n
—
x l
I 9 n— A +LII fn— fA I + I
X.
Therefore,
fl g n , x as
n - .
oo.
Definition 1.4.13 The element
x
constructed a bove is denoted
by f
f,
or
f
I f.
If
I = (a, b),
it is also denote d
by fa f.
As for rea l-valued func tions, it is
convenient to set
l
a
f= - f
a
b f .
Proposition 1.4.14. (Bochner's Theorem)
Let f : I -4 X be measurable.
Then f is integrable i f and only
if 111 11 is integrable.
In addition,
we have
If
_ f I I . .
Proof.
Assume that
f
is integrable and c onsider a seque nce (fn)n>o C C^(I,
X)
such that
f
If
— I
ll --> 0. We have II
f
1I < IlfniI + Ilfn — f
1 I ; a n d s o I I
f
I 1 i s
i n t e g r a b l e .
-
8/18/2019 Cazenave & Haraux - An Introduction to Semilinear Evolution
18/198
8 Preliminary results
Conversely, suppose that
I I
f
I
is integrable. Let g
n
E
C,(I,
R) be a sequence
such that g
n — J ^f 1I in
L'(I)
and such that g, o
C C,(I, X)
be a sequence such that f
n
—* f
almost
everywhere. Finally, let
un
f' I n +
n fn
We have (u
n
I1
o be
a sequence
of
integrable functions
I --> X,
let g
: I
— > IR be an
integrable
function,
and
let f : I — X. If
for a l l n E
N, 1
f
n
II < g, almost
everywhere on I ,
f(t)—f(t) as n —> oo, for almost all t E I,
then
f is integrable and f1f =
lim
f
I f
n
.
n o0
1.4.3.
he spaces
LP(I,X)
Definition 1.4.16.
et
p E
[1,J. One denotes
by LP(I,X)
the set of
(equivalenc e classes of ) me asura ble functions
f : I —* X
such that
t -->
I ff
( t ) I I
belongs to
LP(I).
For
f E LP(I, X),
one defines
If (IL
P = NsSUPtEIII
A0111
^I f ( t )jjP dt) ,
f
p <
oo;
ifp=
cc.
Proposition 1.4.17.
(LT'(I,X)j
I • ^^LP) is a B ana ch space.
If p <
oo,
then
D(I,X)
is dense in
LP(I,X).
Proof.
The proof is similar to that of the real-valued case (in particular, the
den sity of
D(I, X)
is obtained by truncation and convolution).
Remark 1.4.18. Let
f E LP(I,X)
and let g
E LP (I,X*).
Then
t
g(t),f(t))x•,x
-
8/18/2019 Cazenave & Haraux - An Introduction to Semilinear Evolution
19/198
Vector-valued functions 9
is integrable and
J
I (g(t), f (t))X*,X I
I f I I L p I I J I I L p
The following result is related to the preceding remark. The proof is much
more difficult than that for real-valued functions.
Theorem 1.4.19.
I f 1 <
p < oo
and if X is reflexive or if X*
is separable,
then
(LP(I,X))*
Lp (I, X*). In addition, if 1
o be a bounded sequence of
LP(I,X)
and let f : I —; X
be such that
f(t) — f (t)
weakly in
X as n -->
for almost alit El.
Then
f E LP(I,X),
and I
f
L P
< lim of 1lfn11LN .
Pro o f .
y Corollary 1.4.9,
f
is measurable. We define g,,, and g by
g(t)
= inf
fk(t)I)
k>n
g(t) = lim
g(t)
g(t)
= l im inf I f , , , ( t )
I I
lmost everywhere.
I
-
8/18/2019 Cazenave & Haraux - An Introduction to Semilinear Evolution
20/198
10
Preliminary results
Since
g
n
(t)
-
8/18/2019 Cazenave & Haraux - An Introduction to Semilinear Evolution
21/198
Vector-valued functions
11
t}h
Th(t)
IIPdt <
f Jt
1f(s)IIIdsdt
h
Ilf(t)IIIdtds<
f
I l f ( t ) I I
P
d t .
E C(LP(R,X))
and that IlThjI i C D(R, X)
be a sequence such that
fn --> f
of -> ac in
L P
(R,
X) and on IR \ E, w ith IE I = 0 (such a sequence exists by
roposition 1.4.17). Let t E R \ E. W e have
I A h f ( t ) I I < I I A h ( f( t )
— f^(t))II + I J A h f n ( t) I I
1
/ ^
t + h
I 1 1 ( t ) — f(t)II + h J
If(s) — fn(s)IJ d s +
I IA h fn ( t ) [I
t
n large enough, one has II f (t) - f,(t)II < E/4.
On the other hand, since II f O -
fn(•)II
E L o^(R ), by the theory of Lebesgue
ints (see D unford and Schwartz [1], p. 217, Theorem 8 ) w e know that
1
If(s)
— f(s)IId s— >
I l f ( t )
n ( t ) I I ,
t
E R, as
h -> 0.
Therefore, for almost all t,
n
being fixed so that
f (t) - f(t)II
< E /4, and if
h
is small enough, we have
1 f
t}h
J l f(s)
- fn(s)II
ds 1
I I A hf I I LP(R,X)
< _ 2 1 1 f
— fn I I LP(R,X) + I [A hf^.I l LP(R,X) '
a ny n,
it is well known that
I I A h f
fl I I
r
P
R
x)
-' 0 as
h -*0;
it follow s that
A
h
f
I I L
„
(i,x) -+ 0, wh ich comp letes the proof.
Let f E Li ;(I, X) be such
that f = 0 in D'(I, X). Then
almost everywhere.
-
8/18/2019 Cazenave & Haraux - An Introduction to Semilinear Evolution
22/198
12 Preliminary
results
Proof.
First, we rem ark that if J
is
a bounded subinterval
of I,
we have
f,
f
0. Indeed , let
(n)n>1
C D(I), V,,
< 1, an d
c o n -->
1j almost everywhere. W e
have
f
f =
lim
f
fw =
lim(f,
ca) = 0.
J
Then fix a bounde d subinterva l
J C I
and consider
f
E
L i ( I R ,
X),
defined by
{
J(t)
(t) = f (t),
if t
E J,
J(t)=0
ift¢J.
It follows that
T h f
= 0 for all
h > 0. By
Proposition 1.4.29, w e obtain
f = 0
a lmost everywhere . There fore ,
f
= 0 almost everywhere on J. Since J is
arbitrary, we have
f
= 0 almost everywhere.
Corollary 1.4.31.
L et g E
L (I,
X), to E I a nd
let
f E C(I, X) be given by
1(t)
= f t.
9(s) ds.
Then:
(i )
f` = gin
D'(I, X);
(ii)
f is
differentiable a lmost everywhere
and f' = g
almost
everywhere.
Proof.
Rea soning as be fore, we m ay re strict ourselves to the case I = IR and
gEL
1
(R,X). We
have
Thg(t)
=
f(t+h)
-
(t)
h
By P roposition 1.4.29 ,
we
deduce (ii).
Now
consider
c p
E D(R). W e have
(f',
w) _ -(f, co ) = -
f
f
(t)w'(t) d t.
Furthermore,
p(
t
+ h) — P(t)
p'(t)
uniformly on R, a s
h — . 0 .
Therefore,
` ,
—
l im
(t)
c(t + h) — co(t)
h
= —
li
o
f
co(t)
f (t(t)
_
o
J R
T hgco = (g,
c o ) ,
by P roposition 1.4.29 ; henc e (i).
-
8/18/2019 Cazenave & Haraux - An Introduction to Semilinear Evolution
23/198
Vector-valued functions 13
Proposition 1.4.32.
Let T E D'(I, X)
be such that
T' = 0. Then there exists
z,o E
X
such that T = xo, i.e.
(T, ^) = xof,
I
for all tpED(I).
Proof. Let 0
E D(I) be such that
f6 =
1, and let xo =
(T, 6).
' Let (a, b) be
the support of 0
and let to E I, to < a. Now consider cp E D(I). We define
D(I)by
)
(V(s)
- 6(s) f
da) ds,
t o
for alltEI. We have
\ I
).
H e n c e
0
= (T,') _ (T,) — xo
J
.
It follows tha t
(T,) _ xo
J
;̂
I
hence the result.
1.4.5.
he spaces W' P(I, X)
Definition 1.4.33. Let 1
-
8/18/2019 Cazenave & Haraux - An Introduction to Semilinear Evolution
24/198
14
Preliminary results
(iv)
f is absolutely continuous, differentiable almost everywhere, and f' belongs
to LP(I,X);
(v )
f is weakly absolutely continuous,
weakly
differentiable
almost
everywhere,
and f' belongs to LP(I, X).
Proof.
(i) .(ii)` Let t
o
E I . For a ny
t E
I, set
w ( t ) _ . f ( t ) — . f ( to ) — L
t
. f ( s )
ds.
We have w E
C(I,X)
and
w'
= 0 in
D'(I, X)
by Corollary 1.4.31. There -
fore, there exists X
o
E X
such that
w
= xo in
D'(I„X)
(Proposition 1.4.32).
Since w (0) = 0, it follows from Corollary 1.4.30 that w = 0 a lmost everywher e;
henc e (ii).
(ii)(iii) is immediate.
( ii i)^(iv). By possibly modifying f on a se t of mea sure 0, we m ay a ssume
that
1(t) = xo + f
g(s)
d s ,
t o
for all t E I, and we apply Corollary 1 .4.31.
(iv)(v) is immediate.
(v)=(i). Let g be the weak derivative of
f.
Let t
o
E I,
and set
t
(t) = f(t) — f(to) - f
g(s) ds,
t
o
for a ll
t E I.
By C orollary 1.4.31, c p
is differentiable almost everywher e and its
derivative is 0 almost everywhere .
Let x' E X* and let zb be defined
by ib(t) = (x', p(t)). Ali
is absolutely
continuous, differentiable almost everywher e a nd
,b'(t)
= 0 a lmost everywhere.
Since 0(to) = 0, we obtain
0(t) - 0.
Since
x'
is arbitrary, we conclude that
co(t) - 0.
Hence, (i) follows from Corollary 1.4.31.
Corollary1436
Let 1
< p < oo.
Then W I"P(I, X )
b, v, (I, X ).
Proof.
We have
I l f(t) - f(s) I1 s f t
1 1 f (o ,
) I I d a ,
s
I
or a ll
s, t E I.
Hence we have uniform continuity.
Fu rtherm ore, if we se t h(.) = IIf (
)II, we have
h(t) — h(s)I : Ilf(t)
— f(s)II .
-
8/18/2019 Cazenave & Haraux - An Introduction to Semilinear Evolution
25/198
Vector-valued functions
15
It follow s that
^ t
I h(t) – h(s)I <
J
I If ( ^ ) II d a m ,
3
for a ll s, t E I. Ther efore , h is absolutely continuous a nd we have Ih' < II f'II E
L 5
(I)
almost everywhere . We obtain
h E W
1
P(I) ^-+ L°O(I),
which completes
the proof.
Corollary 1.4.37.
III is bounded, then C°° (7, X) is dense in W 1' ' (I, X ).
Proof.
Let
f E W ''r'(I,X)
an d let (g n ),,>1 C
D(I,X)
be such that
g, ---> f' in
LP(I,X).
Let to el, and set
fn(t) = .f (to)
+ f
t .
t g n ( s ) d s .
It is now ea sy to verify that
fn
E C
°O
(I,X)
and that
f,, —f
in
W
1
'T'(I,X),
a s
T1 -4
00.
Corollary 1.4.38.
Let 1
< P < oo.
Then W
1
'P(I,X )` ----> C°'
a
(I,X), with
a = ( p – 1
) / p
Proof.
By Holder's inequa lity, we ha ve
p
IIf(t+h) -f (t)jj
-
8/18/2019 Cazenave & Haraux - An Introduction to Semilinear Evolution
26/198
16
Preliminary
results
Proof.
It is clear that this condition is necessary (for example, take
p() _
Converse ly, since X is complete, we ea sily verify that we ca n modify f in a set
of me asure 0 so that (1.1) holds for all t -r E I. In particular,
f
is continuous and
we m ay consider only the case in which
I = R. Since
f
is continuous, it is clea r
that f (R) is sepa rable. By possibly considering the sma llest closed subspa ce of
X,
we re strict ourselves to the ca se in which
X
is ref lexive and separa ble, and
so X''
is sepa ra ble. For all
h > 0,
se t
fh(t) = f(t + h) - f(t)
h
If
p =
oc, it is clear that
fh
is bounded in
LP(IE€,X).
If
p <
oe, by Holder's
inequality, we h ave
t+h
l ifh(t)l ip < T I
s)I'ds.
Integrating on R, we find that
r
rt+h
r r5
]
fh(t) 1
we have (fh(t), x')
--> n(t), as h -* 0.
Let F be the complement of the set
of Lebesgue's points of (we know Fl = 0 ). By (1.1), for all t
E H \ F,
we have
fh(t)M < K(t)
< oo, if Ih is small enough. We claim that for all
t E hR \ (E U F), there exists w(t)
e X such that
fh(t) — w(t)
as
h -> 0 .
Indeed,
^ l fh(t)
( is bounded, and since X
is reflexive, there exists a sequence
h„ — 0 and
w(t) E
X, such that fh„ (t) — w(t)
weakly in
X as n -- x.
In particular, we have (w(t), x
t
n) = V'(t),
for all
n
E N. Since the se-
quence (x)>i is de nse in
X*, w(t) does not depe nd on the sequence
h
n
and so fh(t) — w(t).
By Proposition 1.4.24, we have
w E LP(R,X),
and
(W ( LP ( I.X
) < 'p(
LP(I,&)
. By Theorem 1.4.35(v) and Corollary 1.4.31, we have
f
E W"P(R. X) and w
-
8/18/2019 Cazenave & Haraux - An Introduction to Semilinear Evolution
27/198
Vector-valued functions 17
From this, we imm ediately deduce the fol lowing result
Corollary 1.4.41. Assume that
X is reflexive. Let f : I —^ X be bounded,
Lipschitz continuous
with Lipschitz
constant
L. Then, f E W(I,X)
and
< L .
orollary 1.4.42. Assume that
X is reflexive
and that 1 < p < co.
Let
(fn
),
a
>o be a bounded sequenc e in
W"r(I, X) and let f : I —^ X
be such that
fn
(t) — f (t)
as n
— > oc,
for almost
every t E I. Then f E
W
i
^P(I, X), and
M.f'M Lr(r, .x)
< liiminf (f,'ALF
,
(1,X).
oo.
y P roposit ion 1.4.24, we have
f E LP(I, X).
Let, E be a set of measure
0 such that fn (t) — f (t)
as n
for all
t E I \ E. For all
t, r C I \ E,
we
h a v e
) J (
t
) - f() < 1
im
of
))fn(t) - fn(
T ) 1 1
< u
rninf
f J i f(s)^I ds.
1 . 3 )
t
Consider
)f).
y?n
is bounded in
LP(I)
and so there exists a subsequence
(k )k>1 and
p E LP(I)
such that :p nk — ^ p
weakly in LP(I) (weak-* if p
= oc)
as k — oo, and litn inf ^^ ^p,^
k ^1 L n — lim inf
In particular, we have
—
--oo
J
4
(s) ds —>
J p(s) ds,
as
k
--^ oo,
1.4)
for all t, r E I,
and
((^F))Lv < liminf))fn(ILr,.
1.5)
n—+oo
It follows from (1.3) and (1.4) that
§f t)—f T)11
-
8/18/2019 Cazenave & Haraux - An Introduction to Semilinear Evolution
28/198
2
m-dissipative operators
Throughout this chapter, X is a Banach space, endowed with the norm
I I • I I
2 1 Unbounded operators in Banach spaces
Definition 2.1.1.
A linear unbounded operator in
X
is a pair
(D, A),
where
D
is a l inear subspa ce of
X
and
A
is a l inear ma pping
D — X.
We say that A
is bounded if there exists c > 0 suc h that
IIAu
I I < _
c,
for all
u
E
{x E
D.
I I r I I
< 1}. Otherwise,
A
is not bounded
Remark 2.1.2. Note that a linear unbounded operator can be either bounded
or not hounded. This somewha t strange terminology is in gene ral use and should
not lead to misunder standing in our a pplica tions.
Remark 2.1.3. If
A
is bounded,
A
is the restriction to
D of
an operator
A
EY, X),
where
Y
is a closed l inea r subspac e of
X,
such that
D
C
Y. If A
is not bounded, there exists no operator A
E
G(Y,X)
with Y
closed in
X and
D C Y,
such that
AID = A.
Definition 2.1.4. Let
(D, A)
be a linear operator in
X.
The graph
G(A) of
A
and the range
R(A) of
A are defined by
G(A)={(u,f)EXxX;uED and f=Au},
R(A) = A(D).
G(A)
is a linear subspac e
of X x X,
and
R(A)
is a l inear subspac e
of X
Remark 2.1.5. In this chapter, a linear unbounded operator is just called an
opera tor where there is no risk of confusion. As usual, we de note the pair
(D,
A)
by
A
with
D(A) = D,
me aning the doma in of
A is D.
Note, however , that when
one de fines an ope rator, it is absolutely nece ssary to define its doma in.
Remark 2.1.6.
When
D(A) = X,
it follows from Theorem 1.1.2 that
A E
G(X) if
and only
if G(A)
is closed in
X
More generally, for not bounded
opera tors, it is very use ful to know w hether or not the grap h is closed.
-
8/18/2019 Cazenave & Haraux - An Introduction to Semilinear Evolution
29/198
Definition and main properties of m-dissipative operators 19
2. Definition and main properties of m
dissipative operators
I ^ u - A A u l l > I lu l l ,
ion 2.2.2. An operator
A in
X
is m-dissipative if
i ) A is dissipative;
for all A > 0 and all
f
E X, there exists u E
D(A)
such that u — A A u = f .
emark 2.2.3. If
A
is in-dissipative in
X,
i t is clea r, from De finitions 2.2.1
f
E X and all A > 0, there exists a unique solution
u of
tion
u — AAu = f.
In add ition, one ha s I ull < I
f
1 I .
A
be an m-dissipative operator in X
and A > 0. For
f
E X, we denote by
Jaj or by
(I — AA)
-f
he solution a of the equa tion
emark 2.2.5. By Remark 2.2.3, one has Ja
E L(X) and IJaj I
c
( x ) < 1.
roposition 2.2.6. Let A
be a dissipa tive
operator in X. The following prop-
are
equivalent.
i )
A is m-dissipative in X;
there
exists Ao > 0 such tha t for
all f E X, there e xists a
solution u E D(A)
ofu—A0Au= f.
It is clear that (i)^(ii). Let us show that (ii)^(i). Let A > 0. Note
equation
u — AAu = f is equivalent to
U-A O Au=A°f+(1- A
)u.
A is dissipative an d
R(I — \o A) = X,
the operator J a
0
= (I — A0A)
-1 can
in Definition 2.2.4. This operator is a contraction on X. Next,
ced ing equa tion is also equivalent to
u=J50 L0
o
f+(1_ )u)
.
0
, this last equation is u =
F(u),
where F
is Lipschitz continuous
with a Lipschitz constant
k =
I(A — A0)/AI < 1. Applying Theo-
u
of
u — AAu = f,
for all A E (A0/2,).
-
8/18/2019 Cazenave & Haraux - An Introduction to Semilinear Evolution
30/198
2
m-dissipative operators
Iterating this argument
n
time s, ther e e xists a solution for all A
E (2-̀Ao, oo),
n > 1.
Since
n is
arbitra ry there exists a solution for all A > 0.
Proposition 2.2.7.
If A
is m-
dissipative, then
G(A) is
closed in X.
Proof.
Since J 1
E
£(X),
G(J1) is closed. It follows that
G(I - A )
is closed,
and so G(A) is closed.
Corollary
2.2.8.
Let A
be an
m-
dissipative operator. For every u
E
D(A),
let
I IL I ID(A)
= I I u l I +
IIAull.
Then (D(A),
I I
- I I
D(A))
is a Ba nach space , and
A E
12(D(A ), X ).
Rem ark 2 .2.9 . In wha t fol lows, and in part icular in Cha pters 3 and 4,
D ( A )
mea ns the B anac h space (D(A),
I I -
IID(A))•
Proposition 2.2.10.
If A
is
m-dissipative,
then li
o I I
Jau - ull = 0 for all
uED(A).
Proof.
We have IIJ) -
I I I
0, we denote
by A ) ,
the operator defined by
Aa=AJ)=
Ja-I
We have Aa
E G(X)
and IIAa IIc(x) 5 2/A.
Proposition 2.2.12.
If A is m-dissipative
and
if D(A) = X,
then
A
),u ---> A u
as A 0 for all u
E
D(A).
Proof.
Let
u E D(A).
By Proposition 2.2.10, one has
JA A u-Au— *0
asAJ0.
On the other ha nd, it follows ea sily from D efinition 2.2.11 that
Aau = J)Au.
Thus,
IIAau - AuII = IIJ)Au -
AuII --30
as A j 0;
hence the result.
-
8/18/2019 Cazenave & Haraux - An Introduction to Semilinear Evolution
31/198
Extrapolation
21
2.3. Extrapolation
In this section, we show tha t, given an m -dissipative opera tor
A on X
with dense
doma in, one c an extend it to an m-dissipative opera tor
A
on a larger space
X.
This result will be ver y useful for char ac terizing the we ak solutions in Chapter s 3
and 4.
Proposition 2.3.1.
Let A
be an
m-dissipative operator in X with
dense do-
main.
There
exists
a Banach space X,
and an
m-dissipative
operator A
in X ,
s u c h t h a t
(i)
X '—
ith
dense em bedding;
(ii)
for
all u E X,
the norm of u in X is equal to
liJlulI;
(iii)
D(A) = X,
with
equivalent
norms;
(iv)
Au = Au, for all u E D(A).
In addition, X
and
A satisfying (i)-(iv) are unique,
up to
isomorphism.
Proof.
For
u E
X, we define IMu IIt = I^J luII. It is clea r that III - III is a norm on
X.
Let X
be the com plet ion of
X
for the norm .
X
is unique, up to an
isomorphism, and X '— X, with dense em bedding. On the other han d, observe
t h a t
J1A u = Jlu — u, du E D(A ).
Thus,
II IA u I II < - IM u I II + h u h < - 2 1 1 u 1 1 , V u E D ( A ) .
He nce, A can be extended to an operator
A E £(X, Y).
We de fine the linea r
operator
A on X
by
D(4)=X ,
A u=A u, VuED(A ).
It is clea r tha t A satisfies (iii) and (iv). Now, let us show tha t A is dissipa tive.
Take X> 0. Let
u E D(A)
and let
v = J
lu.
One has
v — .XA v = Ji(u —
,Au)
Since A is dissipative, it follows tha t
I I I u —)
AuIII = liv — AAvII > IIvli = IIIuIIl
By continuity of A, we de duce that
-
8/18/2019 Cazenave & Haraux - An Introduction to Semilinear Evolution
32/198
m-dissipative operators
and so
A is
dissipative. Fina lly, let
f
E
X, and
(f>o C
X, with f ,, , - - - > f
in
X
as n ,
oo. Set u
n =
J1
f,.
Since (fn ) n,>
o is a Ca uchy sequence in
X,
(un,)n
>o
is also a Ca uchy sequence in
X;
and so there exists u
E
X,
such that u
n p
u
inXasn --*oo. We have
f
= u
n
— Au
n
=
un
— Au
n
Since
A
E
£(X,Y);
it follows that
f
= u — Au = u — Au.
Hence
A is m-
dissipative. The u niqueness of
A
follows from the uniqueness of
A.
Corollary
2.3.2.
Ifs E
X is such that
Ax
E
X,
then x
E
D(A) and Ax = Ax.
Proof.
Let
f = x — Ax
E
X.
Since
A is m-
dissipative, there exists
y E
D(A)
such that y — Ay =
f.
By P roposition 2.3.1(iii), we ha ve
(x — y) — A(x — y) = 0,
and since
A is
dissipative, w e obtain
x=y
2.4. Unboun ded operators in Hilbert spaces
Throughout this section, we a ssume tha t X is a Hilbert space , and we de note by
(•, •) its scala r produc t. If
A is
a linear operator in
X
with dense doma in, then
G(A*) = {(v, go)
E
X x X; (cp, u) = (v, f)
for a ll
(u, f)
E
G(A)},
defines a l inear operator
A*
(the a djoint of
A).
The doma in of
A* is
D(A*) = {v
E
X, 3C < o o , ((Au,
v)I <
u
E
D(A)},
and A*
satisfies
(A*v, u) = (v, Au), `d
E
D(A),
Indeed, the linea r ma pping u
H
v, Au),
defined on
D(A)
for a ll
v
E
D(A*),
can
be extended to a unique linea r ma pping
c p
E
X' X,
denoted by
cp
= A*v.
It
is clea r that
G(A*) is
systematically closed.
Fina lly, it follows e as ily tha t if
B
E
£(X),
then
(A + B)* = A* + B
*.
1
Proposition 2.4.1.
(R(A)) = {v E
D(A*); A*v = 0}.
Proof.
One has
v
E
(R(A))
v,Au) = 0,Vu
E
D(A) 0,v)
E
G(A*).
This last property is equivalen t to
v
E
D(A*)
and A*v = 0;
henc e the re sult.
Proposition 2.4.2.
A is dissipa tive in X
if and
only if (Au, u) < 0, for
a ll
-
8/18/2019 Cazenave & Haraux - An Introduction to Semilinear Evolution
33/198
Unbounded operators in Hilbert spaces 23
Proof.
If A is dissipative, one has
— 2A(A u, u)
+ A 2
I I A u 1 I
2
= I Iu
— AAuII
2 — I I U I I 2
> 0, VA > 0, Vu
E
D(A).
Dividing by A an d letting A j 0, we obtain
(Au, u) < 0, for all
u
E
D(A).
Conve rsely, if the last prope rty is satisfied, then for a ll A > 0 a nd u
E
D(A)
we
have
I I u
— AAuII
2
= I I u I I
2
— 2A(Au, u) + A
2
I I A u I I 2
> I Iu l l
2
and then A is dissipative.
Corollary 2.4.3. If
A is m-dissipative in X,
then D(A) is dense in X.
Proof. Let
z E
(D(A))
l
and let u = Jjz
E
D(A).
We have
0 = (z, u) = (u —
Au, u).
Hence,
I I u I I 2 = ( A u ,
u) < 0.
It follows that u = z = 0;
and so D (A) is dense in
X.
Corollary
2.4.4. If A is
m-dissipative in X, then
J,u -- u as A j 0,
for all u
E
X and
AAu --* Au as A j 0,
for all u
E
D(A).
Proof.
We apply Corollary 2.2.3 and Propositions 2.2.10 and 2.2.12.
Theorem
2.4.5. Let A be a linear dissipative operator
in
X
with dense do-
main.
Then A
is
m-dissipative
if and
only if A*
is dissipative and G(A) is
closed.
Proof.
If
A
is m-dissipative, then
G(A)
is closed, by P roposition 2.2.7. Let us
show that
A *
is dissipa tive. Let v
E
D(A*).
We have
(A*v, Jay) _ (v, AJav) = (v, AAv)
_ (v,J v—v)=
{(v,Jav)—
I I v I I 2 }
-
8/18/2019 Cazenave & Haraux - An Introduction to Semilinear Evolution
34/198
24
m -dissipative operators
Since
(A*v, Jav) -->(A*v, v)
as A J 0, it follows that A* is dissipative.
Conversely, since
A
is dissipative a nd
G(A)
is closed, it is clear that
R(I — A)
is closed in X. On the other ha nd, by Proposition 2.4.1, one ha s
(R(I — A)) = {v
E
D(A*); v — A*v = 0} = {0},
since A*
is dissipative. Therefore
R(I — A) = X,
and
A
is m-
dissipative, by
Proposition 2 .2.6.
Definition 2.4.6.
Let
A
be a l inear opera tor in
X
with dense dom ain. We
say that
A
is self-adjoint (resp ec tively ske w-adjoint) if
A*
= A (respectively
A* = —A).
Remark 2.4.7. The equality A* = +A has to be taken in the sense of opera-
tors. It me ans that
D(A) = D(A*)
and
A*u = ±Au,
for a ll
u
E
D(A).
Corollary 2.4.8.
If A
is a self-adjoint
operator in X,
and if
A < 0 (i.e.
(Au, u)
-
8/18/2019 Cazenave & Haraux - An Introduction to Semilinear Evolution
35/198
Complex Hilbert spaces 25
Corollary 2.4.11.
Let A be a linear operator in X with
dense domain.
Then
and
—A are m-dissipa t ive
if an d only if A is sk ew-a djoint.
P r o o f .
pplying Corollary 2.4.9 , it suff ices to show that if A a nd — A are
m-
dissipative, then A is skew -adjoint. Applying Pr oposition 2.4.2 to A
and —A,
we obtain
(Au, u) = 0, u E D(A).
For a ll
u, v E D(A),
we obtain
(Au, v) + (Av, u) = (A(u + v), u + v) — (Au, u) — (Av, v) = 0.
*) .
)here fore G(—A) C G ( A
It rema ins to show that
G(AC G(—A). Consider
(u, f) E G(A*) and let
g = u — A*u = u — f.
Since — A
is m-dissipative, there
exists
v E
D(A) such that g = v + Av, and since G(—A) C G(A*),
we have
v E D(A*)
and
f = v — A*v.
Hence (v — u) — A*(v — u)
= 0 a nd since
—A*
is
dissipative (Theore m 2.4.5), we obtain u = v.
Therefore,
(u, f) E G(A*); and so
A = —A*.
I
2.5.
omplex Hilbert spaces
In this sec t ion, we assum e that X is a com plex Hilbert space . Reca ll that by
definition X is a c omplex H ilbert spac e provided tha t ther e e xists a c ontinuous
R-bilinear mapping
b: X x X — > C satisfying the following properties:
b(iu, v) = ib(u, v),
(u,
v) E X x X;
1
b(v, u) = b(u, v),
(u, v) E
X x X;
b(u, u) =
^ l u l l
2 ,
u E X.
In that ca se
(u, v) = R e(b(u,_
v)) def ines a (rea l) sca lar produc t on X. Equipped
with this scalar product, X is a rea l Hilbert space . In wha t follows, we c onsider
X as a
real Hilbert space.
Let
A
be a linear opera tor on the real Hilbert space
X. I f A is C-linea r, we
ca n define iA as a l inear opera tor on the real Hilbert space X.
Proposition 2.5.1.
ssume that
D(A)
is dense a nd that A is C-linea r . Then
A*
is
C-linear, and (iA)* = —iA*.
P r o o f .
et
v E D(A), f = A*v
and let z E C . For al l
u E D(A),
we have
I
(zf, u) = (f, u) = (v, A(u)) = (v, zAu) = (zv, Au).
There fore zv
E D(A*)
and zf = A(zv).
Hence
A* is C-linear. In addition, I
-
8/18/2019 Cazenave & Haraux - An Introduction to Semilinear Evolution
36/198
6
m-dissipative operators
for a ll
(v,
f) E
G(A*)
and a ll
u
E
D(A);
and so
G(—iA*) C G ((iA)*).
Applying
this result to iA, we obtain
G(—i(iA)*) C
G ((i •
iA)*) = G(—A*).
It follows tha t
G ((iA)*)
C
G(—iA*),
and so G ((iA)*) = G(—iA*).
Corollary
2.5.2.
If
A is
self-adjoint, then
iA is
skew-adjoint.
Proof. (iA)* = —iA* = —iA.
2.6. Exam ples in the theory of partial differential equations
.6.1.
The
Laplacian in an open subset of
RN: L
2
theory
Let Sl be any open subset of
R
N
, and let
Y = L 2
(S2). We c an consider either rea l-
valued functions or complex-valued functions, but in both cases, Y is considered
as a rea l Hilbert space (see §2.5). We define the linea r opera tor
B
in
Y by
{
D(B) = {u
E Ho(S2);
Au
E L
2
( c l ) } ;
Bu = Au,
d u
E
D(B).
Proposition
2.6.1.
B is
m-
dissipative
with dense doma in.
More prec ise ly, B
is se l f -a djo int a nd B < 0.
We need the following lemma .
Lemma
2.6.2.
W e
have
s
f
vAudx=—
I
n
Vu•Vvdx.
2.1)
for alluED(B) andvEHo(5l).
oo.
2.1) is satisfied by
v
E D(l).
The lem ma follows by density, since both
term s of (2.1) are continuous in v on
Ho (S2).
Proof of Proposition 2.6.1.
First,
D(S2) C
D(B), and so
D(B) is
dense in
Y.
Let
u
E
D(B).
Applying (2.1) with
v
= u, we obtain
(Bu, u) < 0,
so that
B is
dissipative (Pr oposition 2 .4.2). The bilinear continuous ma pping
b(u, v)
(uv
+ Vu • Vv)dx
-
8/18/2019 Cazenave & Haraux - An Introduction to Semilinear Evolution
37/198
Examples in the theory of partial differential equations 27
is coe rcive in
Ho(f ).
It fol lows from The orem 1.1.4 that, for all
f
E L
2 (Sl),
there e xists u E H o (Sl) such that
J
(uv+Du•Vv)dx =
J
fvdx,
by E H o(Q).
We obtain
u — Du= f,
in the sense of distributions. Since , in addition
u E
Ho (S2), we obtain u E
D(B)
and
u — Bu = 1.
Therefore
B
is m-dissipative. Fina lly, for a ll u, v E
D(B),
we
have, by (2.1),
(Bu, v) = (u, Bv).
Therefore G(B) C G(B*),
an d by C orollar y 2.4.10, it follows that
B is self-
adjoint.
Remark 2.6.3. If Sl has a bounded boundary of class C
2 , then D(B) =
H2 (1)
f l Ho(c l), with equiva lent norms (see B rezis [2], Theorem IX.25, p. 187,
or Fr iedma n [1], Theore m 17.2, p. 67).
2.6.2. The Laplacian in an open subset of RN: Co theory
Let Sl be a bounded open subset of IR
N
, and let Z = L°°(Sl). We define the
l inea r opera tor C in Z by
D(C)
= {u E Ho (cl) n Z, Au E Z},
Cu = Au, b'u E D(C).
Proposition 2.6.4.
C is m-dissipative in Z.
Proof.
First, let us show that C is dissipative. Let ,A > 0, f E Z, and let
M = I f ^I L ^ .
Let u
E H o (f l) be a solution of
u
—
AAu= f,
in D'(S2). In particula r, this equa tion is satisfied in L
2
(S2), and we ha ve
(u— M)— A A (u— M)=f— M,
in L
2
(S1). On the other ha nd,
v = (u — M)+
E Ho(1l), with Vv = 1
{1uI>M}Du
(Corollary 1.3.6). Applying Lem ma 2.6.2, we obtain
J
v
2
dx+w
DuI2dx=
J(f— M)vdx_
-
8/18/2019 Cazenave & Haraux - An Introduction to Semilinear Evolution
38/198
8
m-dissipative operators
Therefore
f v 2
dx < 0, and so
v = 0.
We conclude that
u < M
a.e. on
Q. Similarly, we show that
u >_ — M
a.e. on Q. Hence u
E
L°°(f ), and
U
L°°
< I I f 1 I
L —
• It follows tha t C is dissipative. Now let
f
E
L°°(1) C L
2 ( c
).
By §2.6.1, there e xists u
E Ho(Sl),
with Du E
L
2
(1l), a solution of u — Au
= f ,
in
L
2
(S2). We a lready know that u
E
L(1l),
so that
u
E
D(C),
and
u—Cu
= f .
Thereore Cis mdsspaive.
Lemma 2.6.5.
If S2 has a L ipschitz continuous
boundary,
then
D(C) C C0(l) = {u
E
C(S2); u
an =
0}.
Proof.
The p roof is dif f icult, an d uses the notion of a ba rrier func tion (see
Gilbarg and Trudinger [1], Theorem 8.30, p. 206).
Remark 2.6.6.
It follows from Le mm a 2.6.5 that in gene ral the dom ain of C
is not dense in Z. The fa ct that the doma in is dense will turn out to be very
important (see C hapter 3). This is the rea son why we a re led to consider another
example.
We now set X = C
o
( l) , an d we def ine the opera tor
A as follows:
(D(A)=
{uEXnH.()),AuEX} ,
S
I A u = Au, Vu E
D(A).
Proposition 2.6.7. Assume that S2 has a Lipschitz
continuous boundary.
Then A is m-dissipative,
with dense domain.
Proof. D(Sl)
is dense in
X, and
D(SZ) C D(A );
and so
D(A) is
dense in
X .
On
the other hand,
X
is equipped with the norm of
L°°(S2),
and so
X — +
Z and
G(A) C G(C).
Since C is dissipative,
A is
also dissipative. Now let f
E
X y
L°° (f ). Since C is
m -
dissipative, there exists u E
D(C),
such that u — Au
= 1.
By Lemm a 2.6.5, we ha ve
u
E
X ,
and so Au
E
X. Therefore, u
E
D(A)
and
u — A u
= f.
Hence A is m-
dsspave
Rem ark 2.6.8. In the three exa mples of §2.6.1 a nd §2.6.2, note that the sa me
formula (the Laplac ian), corresponds to severa l operators that enjoy dif ferent
properties (since they are defined in dif fere nt domains). In pa rticular, the e x-
pression the opera tor A has a mea ning only if we spe cify the spac e in which this
-
8/18/2019 Cazenave & Haraux - An Introduction to Semilinear Evolution
39/198
Examples in the theory of partial differential equations 29
The wave operator (or
the Klein—Gordon operator) in Hp
(S2) x
2 (Q)
any open subset o f RN, and let X
= Ho (fl) x L
2
(1). We deal either
tions or with complex-valued func tions, but in both cases
dere d as a rea l Hilbert space (see §2.5). Let
A = inf
{11VU1IL2,u
E
Ho(S2), I I u I I
L2 = 1}.
2.2)
( In the c ase in which 5l is bounded, we rec all that A is the f irst eigenva lue of
— ,L in
Ho (S2),
and that A> 0). Let
m> —A.
Then
X
ca n be equipped with
((u, v), (w, z)) _
/ (Du • Vv + m aw + vz) dx.
t defines a norm on X which is equivalent to the usual norm.
f ine the linea r opera tor
A in X
by
D(A) = {(u,v)
E
X, Au
E
L
2
(S2),v
E Ho(S2)};
A(u, v) = (v, Au — mu), V(u, v)
E
D(A).
2.6.9.
A is skew-adjo int ,
an d in
particular A
and —A are m-
with dense do ma ins.
D(1) x V(l)
C
D(A)
and so
D(A) is
dense in X.
On the other hand,
l l
((u, v), (w, z))
E
D(A)z,
and by (2.1), we have
(A(u, v), (w, z))
= J
(Ov . Ow + mvw + (Du — mu)z) dx
= —
J
(vu
•Vz+muz+ (Aw — mw)v) dx
_ —((u, v), A(w, z))•
2.3)
(u, v) _
(w, z), it follows tha t
(A(u, v), (u, v)) = 0.
A is
dissipative (Proposition 2.4.2). Now let
(f,
g) E
X.
The e quation
is equivalent to the following system:
r2u—iu
=f+g;
2.4)
Slv=u— f.
2.5)
u E
H
o
(Sl)
of (2.4), satisfying L u
E
2
(1). Next, we solve (2.5) and we obtain v
E
Ho
(52). Therefore
(u, v)
E
D(A)
(u, v) — A(u, v) = ( f , g), so
that
A is
m-dissipative. Similarly, we show
—A is
m-dissipative. By (2.3), we have
G(A)
C G(—A*).
Corollary 2.4.11
A is
skewadon.
1
1
-
8/18/2019 Cazenave & Haraux - An Introduction to Semilinear Evolution
40/198
30
m-dissipative operators
2.6.4. The wave operator (or the Klein-Gordon operator) in L
2
(Sl) x
H(2)
Let S2 and m be as in §2.6.3. We recall that Ho (Q) ' L
2
(Q) '-+ (Ho (Q))' _
-1(Q) with dense embeddings. We equip Ho (1) with the scalar product de-
fined in §2.6.3. Theorem 1.1.4 shows that
H
-
(r) = {u E
D (),
5
Ho (ŝ ), o^
u
— m
= u in D'(sf)},
2.6)
and that we can equip H
-
(Q) with the scalar product
(u,v)-i =
J (v
0
.V +
Y = L2
(Q)
x H
-
'(St).
We deal either with real-valued functions or with
complex-valued functions, but in both cases X is considered as a real Hilbert
space (see X2.5). We define the linear operator
B
in
Y by
{
D(B) = Ho(l) x L
2
( Q ) ^
B(u, v) = (v, Au - inn) E Y,
u, v) E
D(B).
Proposition 2.6.10.
B
is
skew-adjoint.
In particular,
B and -B are m-
dissipative with dense domains.
Proof.
D(S2 ) x D(S2)
C D(B)
and so
D(B)
is den se in
Y.
Let
((u, v), (w, z)) E
D(B)2,
and consider cp„ and
z
defined by (2.6). Since v, z E L
2
(t1), we have
E L
2
(Q). Applying (2.1), we obtain
), (w, Z))L2
X
H_ =
f
vwdx+(Au
-
mu,
z ) - i
=
J
vw dx +
J
(Du • V(p
2
+ mu c)
z
) dx
=
J
vw dx -
f
u(A
z
- m) dx
=
J
vw dx -
J
uz dx.
Similarly, we have
((u,v),B(w,z))
t
, 2 XH
-1 =
J
zudx -
J
wvdx.
Therefore,
2.7)
-
8/18/2019 Cazenave & Haraux - An Introduction to Semilinear Evolution
41/198
Examples in the theory of partial differential equations 31
Applying (2.7) with
(u, v) = (w , z),
it follows that
(B(u, u), (u, v)) = 0.
Thus, B
is dissipative (Proposition 2.4.2). Now let
(f, g) E Y.
The equation
(u, v) — B(u, v) = (f, g)
is equivalent to the system (2.4)—(2.5) of §2.6.3. By
Theorem 1.1.4 (see the proof of Proposition 2.6.1), there exists a solution u E
H o (f2) of (2.4). Next, we solve (2.5) and w e obta in
v
E L
2
(fl). Therefore
(u, v) E
D(B)
and
(u, v) — B(u, v) = (f, g);
henc e B is m-dissipative. Similarly, we show
that —B
is m-dissipative. By (2.7), we have
G(B) C G(—B`). Corollary 2.4.11
proves tha Bis skewadoint.
Proposition 2.6.11.
We use the same notation as in §2.6.4.
Then Y
and
B
are the extensions of X and A given by Proposition 2.3.1.
Proof.
Properties (i), (iii), and (iv) are clearly satisfied. We need only show (ii),
i.e.
IIU)r
§(I — A)—'UMMx, VU E X.
Let U
e X and
V E D(A)
be such that U = (I — A)V.
We show that §(I
A)Vy Vf^x. Indeed, since
B
is skew -adjoint, we ha ve
II(I - A)V
I I Y
= ((I - B)V, (I - B)V)' -
II^IIY
+ II BV(Y.
Let
V = (u, v).
We have
J J B V 1 1 2 =
w 1 1
§§AU -
mu)
- §V L2 + IIuI1
2 , = § § V I 1
2 .
henc e the result.
2.6.5.
he Schrodinger operator
Let f be any open subset of Rh', and let Y = L 2
(52,C).
Y is considered as a
real Hilbert space (see §2.5). We define the linear operator B
in Y
by
D(B) = {u E
Ho (1l. C), L u E Y};
{ By = i^u, Vu E D(B).
In what follows, we w rite L
2
(4) and H) instead of L
2
(cl, C) and Ho (St, C).
Proposition 2.6.12.
B is skew-adjoint, and in
particular B
and -B are m-
dissipative with dense domains.
-
8/18/2019 Cazenave & Haraux - An Introduction to Semilinear Evolution
42/198
32 m-dissipative operators
Re ma rk 2.6.13. As in §2.6.1, if 11 ha s a bounde d boundar y of class C
2 , then
D(B) = H
2
(S2) fl
Ho(fl),
with equivalent norms.
We now set
X
= H-
(1, C) and, given u
E X,
we denote by cp
v
, E Ho (52, C)
the solution of
-^cp
v, +
u
in
X.
We equip X
with the scalar produc t
(u, v)-i = (^P .., ^Pv)H1
=Re J
x,
for u, v E
X.
We de fine the linea r opera tor
C in X by
D C) = Ho
1
l);
Cu=Lu,
VuED(C).
Proposition 2.6.14.
C is self-a djoint < 0.
Proof.
We have D(Sl, C) C
D(C)
so that
D(C)
is dense in X.
Furthermore,
for a ll u, v E
D(C),
(Cu,v)_i = (Cu —u,v)_1 +(u,v)_1 = (u,cp„)Hi +(u,V)_1
=
-
u,v)L2 + (u,v)_1.
2.8)
Ta king u = v, it follows that
(Cu,u)-1= —IIkIIL2 + IIUIIH-1 < 0,
and so C is dissipative. Theore m 1.1.4 proves tha t C is m-dissipative. By (2.8),
we have
Cu,v)_1 = u,Cv)_1,
for all u,v E
D C).
It follows that
G C)
C G(C*), and so C is self-adjoint
(Corollary 2.4.10).
Finally, consider the ope rator
A in X
given by
J D(A) = Ho (l);
Au=i/u, VuED(A).
Applying Proposition 2.6.14 and Corollary 2.5.2, we obtain the following result.
Corollary 2.6.15.
A
is skew-adjoint, an d in pa rticular
A and -A are m-
dissipa tive with de nse
domains.
Notes.
For more information about §2.6, see Brezis [2), Courant and Hilbert [1],
-
8/18/2019 Cazenave & Haraux - An Introduction to Semilinear Evolution
43/198
-
8/18/2019 Cazenave & Haraux - An Introduction to Semilinear Evolution
44/198
34 The Hille–Yosida–Phillips Theorem and applications
In particular,
Ilua t II
Ixil,
3.8)
for a l l A > 0 and a l l t > 0.
Step 2.
Assume that
x E D(A).
I t is c lear by construction that AA a nd A.
comm ute, for a ll A, p > 0. In particular, for all s, t > 0, we ha ve
ds
{Ta(st)T, (t – st)} = tTA(st)T
N
,(t – st)(A A – A
µ
) .
It follows that
Ilua t)
–u
µ (t)II = IIT. (t)x–T
µ (t)xII
i
<
s
{T(st)T(t – st)x}
dsl
< tIlA ax – A
µxII.
We deduc e (Proposition 2.2.12) that uA is a Ca uchy seque nce in
C([0,T],X),
for all
T > 0.
Let
u E
C([0, oo),
X) be its limit.
Step 3. Set
u(t) = T(t)x.
By (3.8), we ha ve
(IT(t)xII
Ixil,
for a ll t > 0 , x E
D(A); and so T(t)
ca n be extended to a unique operator
T(t) E
£(X)
satisfying IIT(t)II < 1, for all
t >_ 0.
Take x E X, and (x
n
,)>o C D(A),
such that
x,,, — x
as
n
–^ oo. We have
IITA (t)x –T (t)xII < IITA (t)x –T A (t)x
f
hj + IITA (t)x
n
–T(t)xIj
+ IIT(t)xn
–
(t)x I
<
2
11x. – xjI
+ IITa(t)xn
–
T(t)x,,Il ;
and so TA (t)x --> T ( t)x
as A j 0 uniformly on [0, T] for all
T > 0. Proper-
ties (3.1) and (3.2) follow. To show (3.3), it suffice s to rem ark that
TA(t)TA(s) _
TA (t + s), and so
IIT(t)T(s)x – T(t
+ s)xlI <
IIT(t)T(s)x – T(t)TA (s)xjI
+ (IT(t)TA (s)x – TA (t)TA
(s)xjj
+ IITA(t + s)x
– T(t + s)xI I.
It follows tha t
IIT(t)T(s)x – T(t +s)xII —;0
as A J 0.
Step 4. Returning to the case in which u E
D(A), set va(t) = AATA(t)x =
TA(t)Aax = u'(t). We have
-
8/18/2019 Cazenave & Haraux - An Introduction to Semilinear Evolution
45/198
Two important special cases
35
Hence,
v, --> T (t)Ax
as A .. 0, uniform ly on [0, T] for all
T > 0.
Taking
ua (t) = x +
J O
t
va (s) ds,
and letting .\
1 0, it follows tha t
f o
t
u(t) = x +
T(s)Ax dx.
Thus u
E
C l
([0, oo), X), an d
u (t) = T(t)Ax,
3.9)
for a ll
t > 0.
Finally, we ha ve
v(t) = A(JATA(t)x),
and
IJATT(t)x —T(t)xH) < IITA,(t)x —T(t)xjj + IIJaT(t)x — T(t)xjj.
Therefore, (J
a
T
a
(t)x, A(J
a T
A
(t)x)) --3(T(t)x,T(t)Ax) in
X x X a s A 1
0. Since
G(A)
is closed, it follows that
T(t)x
E
D(A) for all
t > 0, and
AT(t)x = T(t)Ax,
hence (3.7). We conclude that
u
E
C([0, oo), D(A)).
P utting together (3.7)
and (3.9 ), we obta in (3.5).
Step 5.
Uniquene ss of the solution of (3.4)— (3.6). Let
u be a solution, and let
T
> 0. Set
v(t) = T(r — t)u(t),
for
t
E
[0,r]. We have v
E
C([0,t],D(A))
nC
l
( [
0,t],X),
and
v'(t) = —AT(r — t)u(t) +T(rr — t)u'(t) = T(r — t)[u (t) — Au(t)) = 0,
for a ll
t
E
[0, r ]. H enc e, v(r ) = v(0), and so
u(r) =
T(T)x.
r >_ 0 be ing arbitrary,
the proof is complete.
3 .2. Two im portant special cases
W e a ssume in this section that X is a rea l Hilbert spac e. The following result
sharpens the conclusions of Theorem 3.1.1.
Theorem 3.2.1.
Assume that
A
is self-adjoint < 0.
Let x
E
X, and let
u(t) = T(t)x. Then
u is
the unique solution
of the fo l lowing problem:
u E
C([0, oo),
X) n
C((0, oo),
D(A))
n C
1 ((0, oo), X);
3.10)
u'(t) = Au(t), Vt > 0;
3.11)
u(0) = x.
-
8/18/2019 Cazenave & Haraux - An Introduction to Semilinear Evolution
46/198
36 The Hille-Yosida-Phillips Theorem and applications
In addition, we have
I I
Au(t)
I I
< - 1
1 x I I ;
3.13)
- (Au(t), u(t)) <
2t
I I x I I 2 •
3.14)
Finally,
Au(t)
1 1 2 <
2t
1
(Ax, x),
3.15)
if x
E
D(A).
Proof.
We easily verify that A. is self-adjoint < 0, for all A > 0. If
u(t) =
T.\(t)x,
the func tions J jua(t)II and IIua(t)II are non-incre asing with respe ct to t.
In addition, we ha ve
d
ua(t)11 2
= 2(Aaua(t),ua(t)),
3.16)
dt(A,\u),(t),ua(t)) =
2
(Aaua(t),ua(t))
= 2IIua(t)II•
3.17)
Fr om (3 .17), it follows that
-(Axua(t), ua(t))
is non-incre asing with respec t to
t. Integra ting (3.16) betwee n 0 a nd t > 0, it follows that
-
t(Aaua(t),ua(t))
< - f A),ua(s),ua(s))ds <
2 I x 1 1 2 .
3.18)
0
Integra ting (3.17), we obtain
2tIIua(t)II
-
8/18/2019 Cazenave & Haraux - An Introduction to Semilinear Evolution
47/198
Two important special cases 37
addition (by (3.14)), for a ll t > 0, IIAAJ au a(t)I ) is bounde d a s .1 j 0. The refore
u(t) E D(A), for a ll t > 0, with
A u(t) = 11i
a
A,\J.,ua(t),
in X weak. (3.10), (3.11), and (3.12) now follow from Theorem 3.1.1, and
(3.13), (3.14), and (3.15) ar e obta ined by pa ssing to the limit in (3.20), (3.18),
and (3.19).
It remains to show the uniqueness of
u. To do this, take t
> 0 and 0 0,
which completes the proof.
Rem ark 3.2.2. Theorem 3.2.1 m ea ns that T(t ) has a smoothing ef fect on the