Sie sind auf Seite 1von 198

An Introduction to

Semilinear Evolution Equations


nz 1NIVERSIDAD COMPLUTENSE
Revised edition ^.^,f ,? I I) I I ^ ^^^ I I I^ I ) ^^^I ^II I ^I I I^I I^I I
' 5311910045
Thierry Cazenave
CNRS and University of Paris VI, France

and

Alain Haraux
CNRS and University of Paris VI, France

Translated by
Yvan Martel
University of Cergy-Pontoise, France

k 7


0R . 5/.??0 r

CLARENDON PRESS OXFORD


1998
u

I
Oxford University Press, Great Clarendon Street, Oxford 012 6DP
Oxford New York
Athens Auckland Bangkok Bogota Buenos Aires Calcutta

1
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 companies in
Berlin Ibadan

Oxford is a trade mark of Oxford University Press

Published in the United States


by Oxford University Press Inc., New York

Introduction aux problemes devolution semi-lineaires


Edition Marketing SA, 1990
First published by Ellipses

I
Translation Oxford University Press, 1998

Aide par le ministere francais charge de la culture

All rights reserved. No part of thispublication may be reproduced,


stored in a retrieval system, or transmitted, in any form or by any means,
without the prior permission in writing of Oxford University Press.

t
Within the UK, exceptions are allowed in respect of any fair dealing for the
purpose of research or private study, or criticism or review, as permitted
under the Copyright, Designs and Patents Act, 1988, or in the case of
reprographic reproduction in accordance with the terms of licences
issued by the Copyright Licensing Agency. Enquiries concerning
reproduction outside those terms and in other countries should be sent to
the Rights Department, Oxford University Press,
at the address above.

' This book is sold subject to the condition that it shall not, by way
of trade or otherwise, be lent, re-sold, hired out, or otherwise circulated
without the publisher's prior consent in any form of binding or cover

I
other than that in which it is published and without a similar condition
including this condition being imposed on the subsequent purchaser.

A catalogue record for this book is available from the British Library

Library of Congress Cataloging in Publication Data


(Data available)

ISBNO 19 850277 X(Hbk)

I Typeset by Yvan Martel

Printed in Great Britain by


Bookcraft (Bath) Ltd,
Midsomer Norton, Avon

1
Preface

This book is an expanded version of a post-graduate course taught for several


years at the Laboratoire d'Analyse Numerique of the Universite Pierre et Marie
Curie in Paris. The purpose of this course was to give a self-contained presen-
tation of some recent results concerning the fundamental properties of solutions
of semilinear evolution partial differential equations, with special emphasis on
the asymptotic behaviour of the solutions.
We begin with a brief description of the abstract theory of semilinear evolu-
tion equations, in order to provide the reader with a sufficient background. In
particular, we recall the basic results of vector integration (Chapter 1) and lin-
ear semigroup theory in Banach spaces (Chapters 2 and 3). Chapter 4 concerns
the local existence, uniqueness, and regularity of solutions of abstract semilinear
problems.
In Nature, many propagation phenomena are described by evolution equa-
tions or evolution systems which may include non-linear interaction or self-
interaction terms. In Chapters 5, 6, and 7, we apply some general methods
to the following three problems.
(1) The heat equation
ut = Au, (0.1)

which models the thermal energy transfer in a homogeneous medium, is the


simplest example of a diffusion equation. This equation, as well as the self-
interaction problem
Ut = Au + f(u), (0.2)

can be considered on the entire space RN or on various domains S1 (bounded


or not) of R N . In the case in which ci # RN, we need to specify a boundary
condition on I' = 852. It can be, for example, a homogeneous Dirichlet condition

u=0 onr, (0.3)

or a homogeneous Neumann condition


au _
an 0 our, (0.4)
vi Preface

Chapter 5 studies in detail the properties of the solutions of (0.2)(0.3) when Sl is


bounded. In this problem, the maximum principle plays an important role. This
is the reason for studying equation (0.2) in the space of continuous functions.
Vector-valued generalizations of the form

aui
= czAui + fi(ul, ... , uk), i = 1, ... , k, (0.5)
ac
called reactiondiffusion systems, often arise in chemistry and biology. One of
the main tools in the study of these systems (and in particular of their non-
negative solutions) is the maximum principle, which gives a priori estimates
in L 0 (5l) k for the trajectories. We thus develop Co methods rather than L 2
methods, which are easier but less suitable in this framework.
(2) The wave equation (also called the KleinGordon equation)

Utt = Au mu, (0.6)

with m > 0, models the propagation of different kinds of waves (for example light
waves) in homogeneous media. Non-linear models of conservative type arise in
quantum mechanics, whereas variants of the form

Utt = Au f (u, Ut) (0.7)

appear in the study of vibrating systems with or without damping, and with
or without forcing terms. Other perturbations of the wave equation arise in
electronics (the telegraph equation, semi-conductors, etc.).
The basic method for studying (0.6) with suitable boundary conditions (for
example (0.3)) consists of introducing the associated isometry group in the en-
ergy space H l x L 2 . Local existence and uniqueness of solutions is established in
this space. However, in general, the solutions are differentiable only in the sense
of the larger space L 2 x H -1 . These local questions are considered in Chapter 6.
(3) The Schrodinger equation

iUt = Au, (0.8)

possesses a combination of the properties described in (1) and (2). Primarily a


simplified model for some problems of optics, this equation also arises in quan-
tum field theory, possibly coupled with the KleinGordon equation. Various
non-linear perturbations of (0.7) have appeared recently in the study of laser
beams when the characteristics of the medium depend upon the temperature;
for example, focusing phenomena in some solids (where the medium can break
down if the temperature reaches a critical point) and contrastingly, defocusing
in a gas medium which weakens the transmitted signal according to the distance
Preface vii

from the source. A 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). In 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 fundamental equations of fluid mechanics (such as Boltzmann's equation, the
NavierStokes 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. This is the
case for the heat equation and the KleinGordon equation with attractive non-
linearity, as well as for non-autonomous problems with dissipation. No 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. We
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.
Paris T. C.
1998 A. H.

U
Contents

Notation . . . . . . . . . . . . . . . . . . . . . . . . . . xiii
1. Preliminary results . . . . . . .. . . . . . . . . . . . . . . . 1

1.1. Some abstract tools . . . . . . . . . . . . . . . . . . . 1


1.2. The exponential of a linear continuous operator . . . . . . . . 1
1.3. Sobolev spaces . . . . . . . . . . . . . . . . . . . . . 2
1.4. Vector-valued functions . . . . . . . . . . . . . . . . . . . 4
1.4.1. Measurable functions . . . . . . . . . . . . . . . . . 4
1.4.2. Integrable functions ... . . . . . . . . . . . . . . . 7
1.4.3. The spaces LP(I,X) . . . . . . . . . . . . . . . . . 8
1.4.4. Vector-valued distributions . . . . . . . . . . . . . 10
1.4.5. The spaces W 1, P(I,X) . . . . . . . . . . . . . . . 13

2. m-dissipative operators . . . . . . . . . . . . . . . . . . . 18

2.1. Unbounded operators in Banach spaces . . . . . . . . . . . 18


2.2. Definition and main properties of m-dissipative operators . . . 19
2.3. Extrapolation . . . . . . . . . . . . . . . . . . . . . . 21
2.4. Unbounded operators in Hilbert spaces . . . . . . . . . . . 22
2.5. Complex Hilbert spaces . . . . . . . . . . . . . . . . . . 25
2.6. Examples in the theory of partial differential equations . . . . 26
2.6:1. The Laplacian in an open subset of R N L 2 theory : . . . 26
2.6.2. The Laplacian in an open subset of R N Co theory : . . . 27
2.6.3. The wave operator (or the KleinGordon operator)
in Ha (1l) x L 2 (1l) . . . . . . . . . . . . . . . . . 29
2.6.4. The wave operator (or the KleinGordon operator)
in L 2 (1) x H '(Il) . . . . . . . . . . . . . . . . 30
2.6.5. The Schrodinger operator . . . . . . . . . . . . . . 31

3. The HilleYosidaPhillips Theorem and applications . . . . 33

3.1. The semigroup generated by an m-dissipative operator . . . . 33


3.2. Two important special cases . . . . . . . . . . . . . . 35
3.3. Extrapolation and weak solutions . . . . . . . . . . . . . 38
3.4. Contraction semigroups and their generators . . . . . . . . . 39
3.5. Examples in the theory of partial differential equations . . . . 42

x Contents

3.5.1. The heat equation . . . . . . . . . . . . . . . . . 42


3.5.2. The wave equation (or the KleinGordon equation) . . . 47
3.5.3. The Schrodinger equation . . . . . . . . . . . . . . 47
3.5.4. The Schrodinger equation in Rr` . . . . . . . . . . . 48

4. Inhomogeneous equations and abstract semilinear


problems . . . . . . . . . . . . . . . . . . . . . . . . . . 50
4.1. Inhomogeneous equations . . . . . . . . . . . . . . . . . 50
4.2. Gronwall's lemma . . . . . . . . . . . . . . . . . . . . 54
4.3. Semilinear problems . . . . . . . . . . . . . . . . . . . 55
4.3.1. A result of local existence . . . . . . . . . . . . . . 56
4.3.2. Continuous dependence on initial data . . . . . . . . 59
4.3.3. Regularity . . . . . . . . . . . . . . . . . . . . 60
4.4. Isometry groups . . . . . . . . . . . . . . . . . . . . . 61

5. The heat equation . . . . . . . . . . . . . . . . . . . . . 62


5.1. Preliminaries . . . . . . . . . . . . . . . . . . . . . . 62
5.2. Local existence . . . . . . . . . . . . . . . . . . . . . . 64
5.3. Global existence . . . . . . . . . . . . . . . . . . . . . 65
5.4. Blow-up in finite time . . . . . . . . . . . . . . . . . . . 72
5.5. Application to a model case . . . . . . . . . . . . . . . . 76

6. The KleinGordon equation . . . . . . . . . . . . . . . . 78


6.1. Preliminaries . . . . . . . . . . . . . . . . . . . . . . 78
6.1.1. An abstract result . . . . . . . . . . . . . . . . . 78
6.1.2. Functionals on Ho (S2) . . . . . . . . . . . . . . . 79
6.2. Local existence . . . . . . . . . . . . . . . . . . . . . . 82
6.3. Global existence . . . . . . . . . . . . . . . . . . . . . 84
6.4. Blow-up in finite time . . . . . . . . . . . . . . . . . . . 87
6.5. Application to a model case . . . . . . . . . . . . . . . . 89

7. The Schrodinger equation . . . . . . . . . . . . . . . . . 91


7.1. Preliminaries . . . . . . . . . . . . . . . . . . . . . . 91
7.2. A general result . . . . . . . . . . . . . . . . . . . 92
7.3. The linear Schrodinger equation in RN . . . . . . . . . . . 95
7.4. The non-linear Schrodinger equation 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-linear Schrodinger equation in R N global existence . . 112 :
Contents xi I
7.6. The non-linear Schrodinger equation in R N blow up
in finite time . . . . . . . . . . . . .
:

. . . . . . . . . 114
I
7.7. A remark concerning behaviour at infinity . . . . . . . . . . 120
7.8. Application to a model case . . . . . . . . . . . . . . . . 121

8. Bounds on global solutions . . . . . . . . . . . . . . . . . 124


8.1. The heat equation . . . . . . . . . . . . . . . . . . . . 124
8.1.1. A singular Gronwall's lemma: application to the heat
equation . . . . . . . . . . . . . . . . . . . . . 125
8.1.2. Uniform estimates . . . . . . . . . . . . . . . . . 129
8.2. The KleinGordon equation . . . . . . . . . . . . . . . . 130
8.3. The non-autonomous heat equation . . . . . . . . . . . . . 134
8.3.1. The Cauchy problem for the non-autonomous heat
equation . . . . . . . . . . . . . . . . . . . . . 134
8.3.2. A priori estimates . . . . . . . . . . . . . . . . . 135
8.4. The dissipative non-autonomous KleinGordon equation . . . 137

9. The invariance principle and some applications . . . . . . 142


9.1. Abstract dynamical systems . . . . . . . . . . . . . . . . 142
9.2. Liapunov functions and the invariance principle . . . . . . . 143
9.3. A dynamical system associated with a semilinear evolution
equation . . . . . . . . . . . . . . . . . . . . . . . . 145
9.4. Application to the non-linear heat equation . . . . . . . . . 146
9.5. Application to a dissipative KleinGordon equation . . . . . . 149

10. Stability of stationary solutions . . . . . . . . . . . . . . 154


10.1. Definitions and simple examples . . . . . . . . . . . . . . 154
10.2. A simple general result . . . . . . . . . . . . . . . . . . 156
10.3. Exponentially stable systems governed by PDE . . . . . . . 158
10.4. Stability and positivity . . . . . . . . . . . . . . . . . . 164
10.4.1. The one-dimensional case . . . . . . . . . . . . . . 165
10.4.2. The multidimensional case . . . . . . . . . . . . . 167

Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . 169

Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 185
Notation

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 II D(A)) with II uIID(A) = II 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) which are zero on 011
the space of distributions on 11
the space of measurable functions on 11 such that I uI P is integrable
(1 < p< oo)

= (fn 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 for almost every x E 11
= 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 N
= a = (ai, ... , aN), IaI = E a^

_ { f E LP (St), Da f E LP (S2) for all a E N' such that I al < m}

_ .Ijcj<m II D"uII LP for u E Wm , r(SZ)


the closure of D(fl) with respect to the norm II Ilwm , p
= Wm,2(fl )
)
2 1/2
= (EIaI< m (IID - uIIL 2 ) for u E H m(1l)
xiv Notation

H) = Wo ' 2 (fl)
D(I, X) the space of C functions with compact support from I to X
u ' = ut = 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) the space of measurable functions u on I with values in X and such
that IIuIIP is integrable (1 < p < oo)

= (Ii IuIP)1 /P, for u E LP(I,X)


IIUIILP
L(I,X) the space of measurable functions u on I such that there exists C
such that IIu(x)II < C for almost every x E I

IIuIIL = 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)}

IIuIIw i, P = IIUIILP + IIn' IILP for u E W1,P(I,X)


u
1
Preliminary results

1.1. Some abstract tools


1
We recall here some classical theorems of functional analysis that are necessary
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 com-
plete metric 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 closed 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 LaxMilgram Theorem) Let H be a Hilbert space


and let a : H x H IR be a bilinear functional. Assume that there exist 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) >aIIuII 2 for all u E H (coerciveness).
Then, for 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.

I
1.2. The exponential of a linear continuous operator

Let X be a Banach space and 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 that the series is norm convergent in C(X) and that Ile A ll < e lIA1I
Furthermore, it is well known that if A and B commute, then e A+B = eAeB.
. I
2 Preliminary results

In addition, for fixed A, the function t '-4 e tA belongs to C (R, (X)) and we
have
tA = e1AA = Ae IA
dt e
for all t E R. Finally, we have the following classical result.

Proposition 1.2.2. Let A E (X). For all T> 0 and all x E X, there exists
a unique solution u E C 1 ([0, TI, 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 clear that e tA x is a solution therefore, we need only show unique-


ness. Let v be another solution and let z(t) = e -tA V(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 <p < oo) if there exists a function f E LP(S2) such that

(T,) f (x)cp(x) dx,


= fn
for all E D(52). In that case, it is well known that f is unique. Let m E N and
let p E {1, cc]. Define

Wm'() = { f E Lp(1l), D& f E LP(Sl) for all a E N tm such that Ial < m }.
Wm , P(cl) is a Banach space when equipped with the norm defined by

IIfIlwm,p = IID`fIILp,

for all f E W'r(Il). For all m,p as above, we denote by Wo '(S2) the closure
of D(1l) in W'P(f ). If p = 2, one sets Wt ,2 (f) = H-(l), Wo ' 2 (1) = Ho (1l)
and one equips Ht(SZ) with the following equivalent norm:

11A _ IIDc'UIIL2
JaI<m
Sobolev spaces 3

Then Hm(1l) is a Hilbert space with the scalar product

(u, v)Hm = f D a uD a v dx
IIm

If S2 is bounded, there exists a constant C(S2) such that

uIIL2 <_ C(cl)IIVUIIL2,


for all u E Ho (S2) (this is Poincare's inequality). It may be more convenient to
equip Ho (1l) with the following scalar product

(u, v) = Vu Vv dx,
in

which defines an equivalent norm to on the closed space HH(SZ). The


following two results are essential in the theory of partial differential 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) Lq(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 Wr'r(cl) ---> L(c) fl C' 11 (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 compact. Embedding (i) is compact for q E [p,p*).

Remark 1.3.3. The conclusions of Theorems 1.3.1 and 1.3.2 remain valid
without any smoothness assumption on 52, if one replaces W 1 'P(1l) by WW'P(c)

We also recall the following result (see Friedman [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 <1 if m j N/r is an integer >_ 0), and let p
be given by

p \1 m / q
Then there exists C(q, r, j, m, a, n) such that
a
IIDauIIL1 <C IIDauIIL- IIuIIL9a,
IcI=7 IkI='m

for every u E D(II81`')


4 Preliminary results

Finally, we recall the following composition rule (see Marcus and Mizel [11).

Proposition 1.3.5. Let F : R - IR be a Lipschitz continuous function, and


let 1 <p < co. Then, for all u E W l "r(52), the function F(u) belongs to W"PQ).
Moreover, if N is the set of points where F is not differentiable (INI = 0), then

F'(u)Vu, if u V N;
VF(u) _
0, if u E N;

almost everywhere in Q.

In particular, we have the following result.

Corollary 1.3.6. Let 1 < p < oo. For all u E W 1 'P(Sl), we have u+, u - , Jul E
W 1 'P(S2). Moreover,
V(u+) _Vu, if u > 0;
{0, ifu<0;

almost everywhere.

1.4. Vector-valued functions

We present here some results on vector integration and vector-valued distribu-


tions that will be useful throughout this book. We consider a Banach space X
and an open interval I C R.

1.4.1. Measurable functions

Definition 1.4.1. A function f : I ---> X is measurable if there exists a set


E C I of measure 0 and a sequence (f n ) n,>o C Cc (I, X) such that f(t) > f (t)
asn^oo, for alit EI\E.

Remark 1.4.2. If f : I > X is measurable, then Iii f 11 : I --# R is measurable.

Proposition 1.4.3. Let (ff )>o be a sequence of measurable functions I > X


and let f : I --+ X be such that ff (t) > f (t) as n > oo, for almost all t E I.
Then f is measurable.

Proof. f,,, > f on I \E with JEl = 0. Let (fn,k)k>o be a sequence of continuous


functions with compact supports such that fn ,k > f, almost everywhere as
k > oo. By applying Egorov's theorem to the sequence li fn ,k fn iI, we obtain
the existence of E^,, C I with IEj < 2', such that fn ,k -# f on
Vector-valued functions 5

I \En . Let k(n) be such that II fn,k(n) fnII < 1/n on I \En and let g n = fk(n).
Take F = E U (f u En ) then IFI = 0. Let t E I \ F. We have f, (t) * f (t);
m>O n>m
on the other hand, for n large enough, t E I \E n . It follows that IIg n fn I <1/n.
Therefore, g(t) > f (t) and so f is measurable.

Remark 1.4.4. If f : I -- X and cp : I -4 R are measurable, 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 such that w i n w^ _ 0 for i j, then
i n>o x1 is measurable.

Proposition 1.4.6. (Pettis' Theorem) Consider f : I i X. Then f is


measurable if and only if the following two conditions are satisfied:
(i) f is weakly measurable (i.e. for every x' E X*, the function t H (x', f (t))
is measurable);
(ii) there exists a set N C I of measure 0 such that f (I \ N) is separable.

Proof. First, since f is measurable, it is clear that f is weakly measurable.


Now let (fn)n>o C Cc (I,X) be a sequence such that fn -i f on I\N as n --> oo,
where INI = 0. It is clear that fn (I \ N) is separable, and then so is f (I \ N).
Conversely, we may assume that f (I) is separable, so that X is separable
(by possibly replacing X by the smallest closed subspace of X containing f (I)).
We need the following lemma (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 of S* such
that, for every x' E S*, there exists a subsequence (x'nk )k> o of (x)> withwith
xnk (x)>x'(x) for all x c X.

Proof. Let (xn)n>o be dense in X. For all n> 0, define Fn : S* -* 2 2 (n), by

Fn(x ) = (x (x1) ...,x 1xn)),


' ' ' 1
for all x' E X*. Since t 2 (n) is separable, there exists a sequence (xn k )k>o of S*
such that Fn ((xn k )k> o ) is dense in Fn (S*). In particular, for all x' E X*, there
exists xn E S* such that

x' ( x .7) x lk(,. ( x 7)I n


6 Preliminary results

for 1 <j < n. It follows that x'nk(.) (xi) --^ x'(xi) as n --> oo, for all j E N. Since
(x n ) n >o is dense in X, we deduce easily that xn 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 1 f(t) xli is


measurable. Indeed, for all a > 0,
{t, IIf(t)x11 <a}= f {t,Ix(f(t)-x)I <a} ;
Ilx'll<1
and it follows from Lemma 1.4.7 that
{t, 111(t) x11 < a} = (1 {t, Ixn'(.f (t) x)I < a}.
n>O
Since the set on the right-hand side is clearly measurable, t F-4 11 1(t) xli is
measurable. Now consider n >_ 0. The set f (I) (which is separable) can be cov-
ered by a countable union of balls B(x) of centres xj and radius 1/n. Consider
fn : I X, defined by
fn=Y"x,j1(;^
where wo = {t; f (t) E B(xo)} and wj = {t; f (t) E B(x 3 )} \ U B(xi). By
O<i<j
Remark 1.4.5, fn is measurable, and it is clear by construction that ii fn (t)
f (t)!1 < 1/n for all t E I. Therefore, it follows from Proposition 1.4.3 that f is
measurable.

Corollary 1.4.8. Let f : I + X be weakly continuous (i.e. if t o --> t, then


f (t n ) f (t) weakly in X). Then f is measurable.

Proof. By Proposition 1.4.6, it is sufficient to prove that f (I) is separable. Let


A be the convex hull of f (I n Q) and let A be the weak closure of A. We have
f (I) c A. On the other hand, A is also the strong closure of A and so A is
separable. D

Corollary 1.4.9. Let (fn ) n >o be a sequence of measurable functions I --p X


and let f: I --> X. If, for almost every t E I, fn (t) f (t) as n > oo, then f is
measurable.

Proof. Let x' E X*. Since (x', f(t)) ---* (x', f (t)) almost everywhere, the
function t f--* (x', f (t)) is measurable, and so f is weakly measurable.
On the other hand, for every n E N, there exists a set E n of measure 0 such
that fn (I \ En ) is separable. Consider the set E = U E, IEI = 0. Let A
n>1
be the convex hull of U f,,(I \ E) and let A be the weak closure of A. We
n>1 _
have f (I \ E) C A. Furthermore, A is also the strong closure of A and so A is
separable. It follows that f is measurable. El
Vector-valued functions 7

1.4.2. Integrable functions


Definition 1.4.10. A measurable function f : I -> X is integrable if there
exists a sequence (fn ) n >o C Cc (I, X) such that

f Ilfn(t) f(t)II dt --, 0


asn ->oo.

Remark 1.4.11. III f - f II is non-negative and measurable, and so fI IIfn - f I


makes sense.

Proposition 1.4.12. Let f : I - X be integrable. There exists x E X such


that if a sequence (fn ) n > o C C,(I, X) satisfies fI IIfn - f II 4 0, as n - then
one has fI fn * x as n -* oo.

Proof.. We have

f fn^fP ) < f IIfnfII+jIIfftI.


Therefore, fl fn is a Cauchy sequence that converges to some element x E X.
Consider another sequence (9 n ) n >o that satisfies f II9n - f II > 0 as n --> oo.
We have

xl X.
II J7 9n II9nA+LIIfnfAI +I

Therefore, fl gn , x as n -. oo.

Definition 1.4.13 The element x constructed above is denoted by f f, or


fI f. If I = (a, b), it is also denoted by fa f. As for real-valued functions, it is
convenient to set lb a f= - f ab f.

Proposition 1.4.14. (Bochner's Theorem) Let f : I -4 X be measurable.


Then f is integrable if and only if 111 11 is integrable. In addition, we have

If _ fII. .

Proof. Assume that f is integrable and consider a sequence (fn)n>o C C^(I, X)


such that f If I ll --> 0. We have II f 1I < IlfniI + Ilfn f 1I; and so II f I1 is
integrable.

8 Preliminary results

Conversely, suppose that II f I is integrable. Let gn E C,(I, R) be a sequence


such that gn J^f 1I in L'(I) and such that g, <_ g almost everywhere, for
some g E L 1 (I). Let (fn ) n >o C C,(I, X) be a sequence such that fn * f almost
everywhere. Finally, let

un I f' I(n+ n fn
We have (u n I1 <_ g E L'(I) and u n f almost everywhere. Therefore, we have
fl Ilan fII 0 and so f is integrable. Finally, it follows from Fatou's lemma
that

YI f I < lim J u, < lim J IIun i1 < f


n-.cx n_00

which completes the proof. LI

Corollary 1.4.15. (The Dominated Convergence Theorem) Let (fn ), >o be


a sequence of integrable functions I --> X, let g : I > IR be an integrable
function, and let f : I X. If

for all n E N, 1 fn 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 fI fn .


n o0

1.4.3. The spaces LP(I,X)

Definition 1.4.16. Let p E [1,J. One denotes by LP(I,X) the set of


(equivalence classes of) measurable functions f : I * X such that t --> I ff (t)II
belongs to LP(I). For f E LP(I, X), one defines

if p < oo;
^I f (t)jjP dt) ,
If (IL
P = NsSUPtEIII A0111 if p= cc.

Proposition 1.4.17. (LT'(I,X)j I ^^LP) is a Banach 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
density 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
Vector-valued functions 9

is integrable and I

J I (g(t), f (t))X*,X I (If IILpIIJIILp'


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. If 1 < p < oo and if X is reflexive or if X* is separable,


then (LP(I,X))* Lp (I, X*). In addition, if 1 <p < oo and if X is reflexive,
then LP(I, X) is reflexive.

Proof. See Dinculeanu [1] (p. 252, Ch. 13, Cor. 1 of Thm 8).

Remark 1.4.20. If I is bounded and if 1 < q <p < oo, then LP(I,X)
L9 (I, X).

Definition 1.4.21. Let 1 < p < oo. We denote by L , (I, X) the set of
measurable functions f : I * X such that for every compact interval J C I,
fj,^ E LP(J,X).
Proposition 1.4.22.
1
Let X and Y be Banach spaces and let A E L(X,Y).
if f E LP(I,X), then Af E LP(I,Y) and IlA.fIlLp C HAIjc(x,y)Ilf1ILP. If.f E
L'(I,X), then A(fi f) = f1 Af
I
Proof. First, assume that p < oo. The result is well known for f E D(I, X),
and the general case follows from a density argument (Proposition 1.4.17). If
p = oc, it is clear that Af is measurable, and that, for almost all t E I,

IIAf(t)II (jAIlL(x ,Y)Af(t)II C tAUL(X,Y)MfjjL l 0

hence the result. 1

Corollary 1.4.23. If X yY and if f E L 1 (I,X), then f E L'(I,Y) and the


integrals of f in the sense of X and Y coincide.

Proposition 1.4.24. Let 1 < p < oo. Let (f),,,>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 II LP < lim of 1lfn11LN.

Proof. By Corollary 1.4.9, f is measurable. We define g,,, and g by


g(t) = inf Ifk(t)I)
k>n
g(t) = lim g(t)

g(t) = l im in
nf I f,,, (t) II almost everywhere.
I
10 Preliminary results

Since gn (t) <_ f.(t) 1 almost everywhere, it follows from the monotone con-
vergence theorem that g is integrable and that j glILP = lim 11gn,jILP. By weak
lower semicontinuity of the norm, we have

IIfIILT(I.a) < .p,] L1 = u 119r.JLP <liin inf ^IIfn.^ILv,(I,X);

hence the result. El

1.4.4. Vector-valued distributions

Definition 1.4.25. We denote by D'((I, X)) the space L(D(I), X). It is called
the space of X-valued distributions on I.

Remark 1.4.26. For the definition of D(I) and its topology, see Schwartz [1].

Remark 1.4.27. Let f E Li, C (I, X). Then

(Tf,) = ff

for p E D(I), defines a distribution Tf E D'(I,X). We will sometimes denote


by f the distribution Tf.

Definition 1.4.28. Let T E D'(I, X). AVe define the distributional derivative
ofT,T'ED'(I,X),by
(T', v) _ (T,'),

for cp E D(I).

Proposition 1.4.29. Let 1 < p < oo and let f E L'(R, X). Let

1 /' t+h
Th.f ( t ) f ( s ) d s,
T1/t

for t E R and h 0. Then Ta f E LP(R, X) f Cb(R, X) and Th f f in L"" (R, X)


and almost everywhere as h i 0.

Proof. By the dominated convergence theorem, it is clear that Th f E C(R, X).


Furthermore, by Holder's inequality, we have
1 t+h
jITaf(t)II P < f JIf(s)jj''ds.
Vector-valued functions 11

Hence
1 t}h
j IIThf(t) IIPdt < f Jt 11f(s)IIIdsdt

f Ilf(t)II P dt.
h
Ilf(t)IIIdtds<
It follows that Th E C(LP(R,X)) and that IlThjI <_ 1. Let Ah = Th - I. One
has I Ah II L(LP) < 2. Let (fn ),I> i C D(R, X) be a sequence such that fn --> f
of -> ac in LP (R, X) and on IR \ E, with IEI = 0 (such a sequence exists by
Proposition 1.4.17). Let t E R \ E. We have

IAhf(t)II < IIAh(f(t) f^(t))II + IJAhfn(t)II


1 / t+h^

I11(t) f(t)II + h J IIf(s) fn(s)IJ ds + IIAhfn(t)[I


t
Let : > 0. For 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
)Dints (see Dunford and Schwartz [1], p. 217, Theorem 8) we know that
1
IIf(s) f(s)IIds> Ilf(t) fn(t)II,
j
for almost all 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

J
1 f t}h
Ilf(s) - fn(s)II ds <e/2.
t

Finally-, since ff E D(R, X), for h small enough, we find that

IIAhfn(t)IILP < E/4.

Therefore, for almost all t and for h sufficiently small, we have IIAhf(t)IILP < E._
Taking E = 1/n, we obtain IIAhf(t)IILP ;< 1/n if h is small enough, for all
t E JR \ En , where the measure of E n is 0. It follows that Thf --4 f as h -3 0, for
all t e R \ U E n , i.e. almost everywhere. Furthermore, we have
n>1

II Ahf II LP(R,X) <_ 211f fn II LP(R,X) + I [Ahf^.Il LP(R,X)'


Given any n, it is well known that II Ah ffl II r P ( R , x ) -' 0 as h -*0; it follows that
II A h f IIL (i,x) -+ 0, which completes the proof.

Corollary 1.4.30. Let f E Li ;(I, X) be such that f = 0 in D'(I, X). Then


f = 0 almost everywhere.

12 Preliminary results

Proof. First, we remark that if J is a bounded subinterval of I, we have


f, f
0. Indeed, let (n)n>1 C D(I), V,, < 1, and con --> 1j almost everywhere. We

f
have

f
J
f = lim fw = lim(f, ca) = 0.
Then fix a bounded subinterval J C I and consider f E L i (IR, X), defined by
J(t)
(t) = f (t), if t E J,
{
J(t)=0, iftJ.

It follows that Th f = 0 for all h > 0. By Proposition 1.4.29, we obtain f = 0


almost everywhere. Therefore, f = 0 almost everywhere on J. Since J is
arbitrary, we have f = 0 almost everywhere.

Corollary 1.4.31. Let g E L (I, X), to E I and let f E C(I, X) be given by


= ft. 9(s) ds. Then:
1(t)
(i) f` = gin D'(I, X);
(ii) f is differentiable almost everywhere and f' = g almost everywhere.

Proof. Reasoning as before, we may restrict ourselves to the case I = IR and


gEL 1 (R,X). We have

Thg(t) = f(t+h) - f(t)


h
By Proposition 1.4.29, we deduce (ii).
Now consider cp E D(R). We have

(f', w) _ -(f, co') = - f f (t)w'(t) dt.


Furthermore,

p( t + h) P(t) cp'(t)
uniformly on R, as h . 0.

Therefore,

`, _ lim. f (t) c(t + h) co(t)


h
o
= li o co(t) f (t(t)
f _ i
JR T hgco = (g, co),
by Proposition 1.4.29; hence (i).
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
(t) = (V(s) - 6(s) f ) da) ds,
to I

for alltEI. We have

\ I ).

Hence /'
0 = (T,') _ (T,) xo
Jr .

It follows that
(T,) _ xo J ^;
I
hence the result.

1.4.5. The spaces W' P(I, X)


Definition 1.4.33. Let 1 <p < oo. We denote by W1 P(I,X) the space of ,

(equivalence classes of) functions f e LP(I,X) such that f' E IY(I,X), in the
sflnse of D'(I, X). For f E W' P(I, X), we s et ILf Ilwl,p = II fJI LP + If IILp-
,

Proposition 1.4.34. (W 1 P(I,X),II (I w f,P) is a Banach space.

Proof. The proof is similar to that of the real case.

Theorem 1.4.35. Let 1 <p < oo and let f E LP(I,X). Then the following
properties are equivalent:
(I) f E W1,P(I X);,.'.
(ii) there exists g E LP(I, X) such that for almost all to, t E I, we have f (t) =;
f(to)+f o g(s)ds;
iii) there exists g E LP(I,X), xo E X, and to E I such that we have f (t) =
xo + fro g(s) ds, for almost all t E I;
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 any t E I, set

w(t) _ .f (t) .f (to) Lt .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'(IX) (Proposition 1.4.32).
Since w(0) = 0, it follows from Corollary 1.4.30 that w = 0 almost everywhere;
hence (ii).
(ii)(iii) is immediate.
(iii)^(iv). By possibly modifying f on a set of measure 0, we may assume
that
1(t) = xo + f g(s) ds,
to
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,
to

for all t E I. By Corollary 1.4.31, cp is differentiable almost everywhere 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 everywhere and ,b'(t) = 0 almost 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.

w Corollary 1.4.36. Let 1 < p < oo. Then W I"P(I, X) Cb, v, (I, X ).

I Proof. We have
Ilf(t) - f(s)I1 sf s
t 11f'(o )II da,
,

for all s, t E I. Hence we have uniform continuity.


Furthermore, if we set h(.) = IIf( . )II, we have
I

!h(t) h(s)I : Ilf(t) f(s)II.

1''
Vector-valued functions 15

It follows that /^ t
I h(t) h(s)I <
J3 IIf'(^)II dam,
for all s, t E I. Therefore, h is absolutely continuous and we have Ih' < II f'II E
which completes
L 5 (I) almost everywhere. We obtain h E W 1 'P(I) ^-+ LO(I),
the proof.

Corollary 1.4.37. III is bounded, then C (7, X) is dense in W 1' ' (I, X ).
be such that g, ---> f' in
Proof. Let f E W''r'(I,X) and let (g n ),,>1 C D(I,X)
LP(I,X). Let to el, and set

fn(t) = .f (to) + f tgn(s) ds.


t.

in W 1 'T'(I,X), as
It is now easy to verify that fn E C O (I,X) and that f,, f
T1 - 4 00.

a (I,X), with
Corollary 1.4.38. Let 1 < P < oo. Then W 1 'P(I,X)` ----> C'
a = (p 1 )/p
Proof. By Holder's inequality, we have
t+h 1p
IIf(t+h) -f(t)jj <h
( j
,
jjf'(s)jjPds {< ha ^If'fILr;
/
hence the result. 0
< p < oo and let f E
Corollary 1.4.39. Assume that I = (a, b). Let 1
W1'P(I,X). Then, for all CE I, we have

f( .+h) -f( .) f^
h
in LP((a,c),X), as h 1 0.
and we apply Propo-
Proof. We extend f on lib by a function f E W r 'P (R, X)
sition 1.4.28 and Theorem 1.4.35. 11

p < oo and let f E


Theorem 1.4.40. Suppose that X is reflexive. Let 1 '<
if and only if there exists cp E Lr(I,IR) such
LP(I,X). Then f E W l 'P(I,X)
that T
(1.1)
IIf(T) - f(t)1I < t f p(s)ds
for almost all t,T E I. In that case, we have II f'II LP(r,x) II F^ILn(I,n2)
16 Preliminary results

Proof. It is clear that this condition is necessary (for example, take p() _

Conversely, since X is complete, we easily verify that we can modify f in a set


of measure 0 so that (1.1) holds for all t -r E I. In particular, f is continuous and
we may consider only the case in which I = R. Since f is continuous, it is clear
that f (R) is separable. By possibly considering the smallest closed subspace of
X, we restrict ourselves to the case in which X is reflexive and separable, and
so X'' is separable. For all h > 0, set

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 have
t+h
lifh(t)lip < T I (s)I'ds.

Integrating on R, we find that


1r rt+h 1r r5
fh(t) <-J J ((s)I r dsdt= -J j w(t)V'dtds
] R h t h a-h (1.2)
_ f(s)ds.

Therefore fh is bounded in LP(R, X). Let (xn ) be dense in X*. For all n, the
function yhn (t) = (x; , f (t)) satisfies

^ n( T ) n ( t ) < xn ( Jt p(s) ds .

In particular, is absolutely continuous, and so differentiable on H \ E n , with


=0. Let E= U E. We have = 0, and for all t e H\E and nEN,
n>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 tn ) = V'(t), for all n E N. Since the se-
quence (x)>i is dense in X*, w(t) does not depend 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 f' = w.
Vector-valued functions 17

From this, we immediately deduce the following 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.
I
Corollary 1.4.42. Assume that X is reflexive and that 1 < p < co. Let
(fn ),a >o be a bounded sequence 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). ,

Proof. By Proposition 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
I
have

)J ( t ) - f() < 1 im of ))fn(t) - fn( T )11 < u rninf f Ji 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 Ln lim inf L In particular, we have
koc n--oo

t J
4 k (s) ds > J p(s) ds, as k --^ oo,
t
(1.4)

for all t, r E I, and


liminf))fn(ILr,.
((^F))Lv < n+oo (1.5)

It follows from (1.3) and (1.4) that

f(t)f(T)11 <_i p(s)ds, "

for all t, r E I; therefore f E W l 'P(I, X) (Theorem 1.4.40). We complete the


proof by applying (1.5).

Notes. For 1.1 and 1.2, consult Brezis [2], Dunford and Schwartz [2], and
Yosida [1]. For 1.3, see Adams [1], Bergh and Lofstrom [1], Brezis [2], and
Gilbarg and Trudinger [1], and for 1.4, see Dinculeanu [1], Dunford and
Schwartz [1], and the appendix of Brezis [1].
2
m-dissipative operators

Throughout this chapter, X is a Banach space, endowed with the norm II II


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 linear subspace of X and A is a linear mapping D X. We say that A
is bounded if there exists c > 0 such that

IIAu II <_ c,
for all u E {x E D. IIrII < 1}. Otherwise, A is not bounded
Remark 2.1.2. Note that a linear unbounded operator can be either bounded
or not hounded. This somewhat strange terminology is in general use and should
not lead to misunderstanding in our applications.

Remark 2.1.3. If A is bounded, A is the restriction to D of an operator


A E (Y, X), where Y is a closed linear subspace 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 subspace of X x X, and R(A) is a linear subspace of X

Remark 2.1.5. In this chapter, a linear unbounded operator is just called an


operator where there is no risk of confusion. As usual, we denote the pair (D, A)
by A with D(A) = D, meaning the domain of A is D. Note, however, that when
one defines an operator, it is absolutely necessary to define its domain.

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
operators, it is very useful to know whether or not the graph is closed.
Definition and main properties of m-dissipative operators 19

2.2. Definition and main properties of m dissipative operators


-

Definition 2.2.1. An operator A in X is dissipative if

I^u - AAull > Ilull,


for all u e D(A) and all A > 0.

Definition 2.2.2. An operator A in X is m-dissipative if

(i) A is dissipative;

(ii) for all A > 0 and all f E X, there exists u E D(A) such that u AAu = f.

Remark 2.2.3. If A is in-dissipative in X, it is clear, from Definitions 2.2.1


and 2.2.2, that for all f E X and all A > 0, there exists a unique solution u of
the equation u AAu = f. In addition, one has I ull < I f 1I .

Definition 2.2.4. Let A be an m-dissipative operator in X and A > 0. For


all f E X, we denote by Jaj or by (I AA) 'f the solution a of the equation
-

u AAu= f.

Remark 2.2.5. By Remark 2.2.3, one has Ja E L(X) and IJajI c ( x ) < 1.

Proposition 2.2.6. Let A be a dissipative operator in X. The following prop-


erties are equivalent.

(i) A is m-dissipative in X;

(ii) there exists Ao > 0 such that for all f E X, there exists a solution u E D(A)
ofuA0Au= f.

Proof. It is clear that (i)^(ii). Let us show that (ii)^(i). Let A > 0. Note
that the equation u AAu = f is equivalent to

U-A O Au= Af+(1- A )u.

Since A is dissipative and R(I \ o A) = X, the operator Ja 0 = (I A0A) -1 can


be defined as in Definition 2.2.4. This operator is a contraction on X. Next,
note that the preceding equation is also equivalent to

Ao
u=J50 L0 f+(1_ )u)
( .

If 2A > A 0 , this last equation is u = F(u), where F is Lipschitz continuous


x--X, with a Lipschitz constant k = I(A A0)/AI < 1. Applying Theo-
rem 1.1.1, there exists a solution u of u AAu = f, for all A E (A0/2,).
20 m-dissipative operators

Iterating this argument n times, there exists a solution for all A E (2 "`Ao, oo),
-

n > 1. Since n is arbitrary 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 J1 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 IILIID(A) = IIulI + IIAull. Then (D(A), II - II D(A)) is a Banach space, and
A E 12(D(A), X).
Remark 2.2.9. In what follows, and in particular in Chapters 3 and 4, D(A)
means the Banach space (D(A), II - IID(A))

Proposition 2.2.10. If A is m-dissipative, then li o II Jau - ull = 0 for all

uED(A).

Proof. We have IIJ) - III <2, and by density we need only consider the case
u E D(A). We have
Jau-u=Ja(u-(I-AA)u);
and so lIJ\ u ulI <IIu (IAA)uII= )IIAuII--*0,asA10.

Definition 2.2.11. Let A be an m-dissipative operator. For A > 0, we denote


by A), the operator defined by
Ja-I
Aa=AJ)=

We have Aa E G(X) and IIAaIIc(x) 5 2/A.

Proposition 2.2.12. If A is m-dissipative and if D(A) = X, then A ),u ---> Au


as A 0 for all u E D(A).

Proof. Let u E D(A). By Proposition 2.2.10, one has


JAAu-Au*0 asAJ0.
On the other hand, it follows easily from Definition 2.2.11 that

Aau = J)Au.
Thus,
IIAau - AuII = IIJ)Au - AuII --30 as A j 0;
hence the result.
Extrapolation 21

2.3. Extrapolation

In this section, we show that, given an m-dissipative operator A on X with dense


domain, one can extend it to an m-dissipative operator A on a larger space X.
This result will be very useful for characterizing the weak solutions in Chapters 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,
such that
(i) X ' with dense embedding;
(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 IMuIIt = I^JluII. It is clear that III - III is a norm on
X. Let X be the completion of X for the norm . X is unique, up to an
isomorphism, and X ' X, with dense embedding. On the other hand, observe
that
J1Au = Jlu u, du E D(A).
Thus,
IIIAuIII <- IMuIII + huh <- 211u11, Vu E D(A).
Hence, A can be extended to an operator A E (X, Y). We define the linear
operator A on X by
D(4)=X,
Au=Au, VuED(A).
It is clear that A satisfies (iii) and (iv). Now, let us show that A is dissipative.
Take X> 0. Let u E D(A) and let v = Jl u. One has

v .XAv = Ji(u ,Au)

Since A is dissipative, it follows that

IIIu AuIII = liv AAvII > IIvli = IIIuIIl


)

By continuity of A, we deduce that

^IIv AAuJIi ? lhIUJI1, `du E X;


22 m-dissipative operators

and so A is dissipative. Finally, 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 Cauchy sequence in X, (un,)n >o
is also a Cauchy sequence in X; and so there exists u E X, such that u n --p u
inXasn --*oo. We have

f = u n Aun = un Au n .
..

Since A E (X,Y); it follows that f = u Au = u Au. Hence A is m-


dissipative. The uniqueness of A follows from the uniqueness of A. c

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 Proposition 2.3.1(iii), we have (x y) A(x y) = 0,
and since A is dissipative, we obtain x = y.

2.4. Unbounded operators in Hilbert spaces

Throughout this section, we assume that X is a Hilbert space, and we denote by


(, ) its scalar product. If A is a linear operator in X with dense domain, then

G(A*) = {(v, go) E X x X; (cp, u) = (v, f) for all (u, f) E G(A)},

defines a linear operator A* (the adjoint of A). The domain of A* is

D(A*) = {v E X, 3C < oo, ((Au, v)I < du E D(A)},

and A* satisfies
(A*v, u) = (v, Au), `d E D(A),

Indeed, the linear mapping u H (v, Au), defined on D(A) for all v E D(A*), can
be extended to a unique linear mapping cp E X' X, denoted by cp = A*v. It
is clear that G(A*) is systematically closed.
Finally, it follows easily that 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 equivalent to v E D(A*) and A*v = 0; hence the result.

Proposition 2.4.2. A is dissipative in X if and only if (Au, u) < 0, for all


u E D(A).
Unbounded operators in Hilbert spaces 23

Proof. If A is dissipative, one has

2A(Au, u) + A 2 IIAu1I 2 = IIu AAuII 2 IIUII 2 > 0, VA > 0, Vu E D(A).

Dividing by A and letting A j 0, we obtain

(Au, u) < 0, for all u E D(A).

Conversely, if the last property is satisfied, then for all A > 0 and u E D(A) we
have
IIu AAuII 2 = IIuII 2 2A(Au, u) + A 2 IIAuII 2 > IIull 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,
IIuII2 = ( Au, 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 Proposition 2.2.7. Let us


show that A* is dissipative. Let v E D(A*). We have

(A*v, Jay) _ (v, AJav) = (v, AAv)


_ (v,JA vv)= {(v,Jav) IIvII2} <0.
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 and G(A) is closed, it is clear that R(I A)
is closed in X. On the other hand, by Proposition 2.4.1, one has

(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. 0

Definition 2.4.6. Let A be a linear operator in X with dense domain. We


say that A is self-adjoint (respectively skew-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 means that D(A) = D(A*) and A*u = Au, for all 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) <0, for all u E D(A)), then A is m- dissipative.

Proof. By Proposition 2.4.2, A is dissipative. Since A* = A, A* is dissipative.


Finally, G(A*) is closed, so that G(A) is closed. We finish the proof by applying
Theorem 2.4.5. 0

Corollary 2.4.9. If A is a skew-adjoint operator in X, then A and A are


m- dissipative.

Proof. Let u E D(A). One has (Au, u) = (u, A*u) = (u, Au). Hence
(Au, u) = 0. It follows from Proposition 2.4.2, that A and A are dissipative.
We conclude as in Corollary 2.4.8. El

Corollary 2.4.10. Let A be a linear operator in X with dense domain, such


that G(A) C G(A*) and A < 0. Then A is m- dissipative if and only if A is
self-adjoint.

Proof. Applying Corollary 2.4.8, we need only show that if A is m- dissipative


then A is self-adjoint. Let (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 g = vA*v. Therefore (vu)A*(vu) = 0 and
since A* is dissipative (Theorem 2.4.5), we obtain u = v. Thus, (u, f) E G(A);
and soA=A*. El
Complex Hilbert spaces 25

Corollary 2.4.11. Let A be a linear operator in X with dense domain. Then


A and A are m-dissipative if and only if A is skew-adjoint. I

Proof. Applying Corollary 2.4.9, it suffices to show that if A and A are m-


dissipative, then A is skew-adjoint. Applying Proposition 2.4.2 to A and A,
we obtain
(Au, u) = 0, Vu E D(A).
For all u, v E D(A), we obtain

(Au, v) + (Av, u) = (A(u + v), u + v) (Au, u) (Av, v) = 0.

Therefore G(A) C G(A *). It remains 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 and since A* is
dissipative (Theorem 2.4.5), we obtain u = v. Therefore, (u, f) E G(A*); and so
A = A*.

I
2.5. Complex Hilbert spaces

In this section, we assume that X is a complex Hilbert space. Recall that by


definition X is a complex Hilbert space provided that there exists a continuous
R-bilinear mapping b: X x X > C satisfying the following properties:

b(iu, v) = ib(u, v), V(u, v) E X x X; 1


b(v, u) = b(u, v), V(u, v) E X x X;
b(u, u) = ^lull 2 , Vu E X.

In that case (u, v) = Re(b(u,_ v)) defines a (real) scalar product on X. Equipped
with this scalar product, X is a real Hilbert space. In what follows, we consider
X as a real Hilbert space.
Let A be a linear operator on the real Hilbert space X. If A is C-linear, we
can define iA as a linear operator on the real Hilbert space X.

Proposition 2.5.1. Assume that D(A) is dense and that A is C-linear. Then
A* is C-linear, and (iA)* = iA*.

Proof. Let v E D(A), f = A*v and let z E C. For all u E D(A), we have I

(zf, u) = (f, u) = (v, A(u)) = (v, zAu) = (zv, Au).

Therefore zv E D(A*) and zf = A(zv). Hence A* is C-linear. In addition, I

(if, u) = (v, A(iu)) = (v, iAu),


aw
1 26 m-dissipative operators

for all (v, f) E G(A*) and all 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 that 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. Examples in the theory of partial differential equations

1 2.6.1. 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 can consider either real-
valued functions or complex-valued functions, but in both cases, 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(S2); Au E L 2 (cl)};

{
Bu = Au, du E D(B).

Proposition 2.6.1. B is m- dissipative with dense domain. More precisely, B


is self-adjoint and B < 0.

We need the following lemma.

Lemma 2.6.2. We have

f vAudx=
Js z n I
VuVvdx. (2.1)

i for alluED(B) andvEHo(5l).

I Proof. (2.1) is satisfied by v E D(l). The lemma follows by density, since both
terms 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 (Proposition 2.4.2). The bilinear continuous mapping

b(u, v) (uv + Vu Vv)dx


=J
Examples in the theory of partial differential equations 27

is coercive in Ho(f ). It follows from Theorem 1.1.4 that, for all f E L 2 (Sl),
there exists u E Ho (Sl) such that

J (uv+DuVv)dx = J fvdx, by E Ho(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. Finally, for all u, v E D(B), we
have, by (2.1),
(Bu, v) = (u, Bv).

Therefore G(B) C G(B*), and by Corollary 2.4.10, it follows that B is self-


adjoint. o

Remark 2.6.3. If Sl has a bounded boundary of class C 2 , then D(B) =


H 2 (1) fl Ho(cl), with equivalent norms (see Brezis [2], Theorem IX.25, p. 187,
or Friedman [1], Theorem 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
linear operator 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 Ho (f l) be a solution of

u AAu= f,

in D'(S2). In particular, this equation is satisfied in L 2 (S2), and we have

(uM)AA(uM)=fM,

in L 2 (S1). On the other hand, v = (u M)+ E Ho(1l), with Vv = 1 {1uI>M}Du


(Corollary 1.3.6). Applying Lemma 2.6.2, we obtain

Jv 2 dx+w
foul>M)
^DuI2dx= J(fM)vdx_<0.
28 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
11 U 11 L < II f 1I L It follows that C is dissipative. Now let f E L(1) C L 2 (c ).
By 2.6.1, there exists u E Ho(Sl), with Du E L 2 (1l), a solution of u Au = f,
in L 2 (S2). We already know that u E L(1l), so that u E D(C), and uCu = f.
Therefore C is m-dissipative.

Lemma 2.6.5. If S2 has a Lipschitz continuous boundary, then

D(C) C C0(l) = {u E C(S2); u 1 an = 0}.

Proof. The proof is difficult, and uses the notion of a barrier function (see
Gilbarg and Trudinger [1], Theorem 8.30, p. 206).

Remark 2.6.6. It follows from Lemma 2.6.5 that in general the domain of C
is not dense in Z. The fact that the domain is dense will turn out to be very
important (see Chapter 3). This is the reason why we are led to consider another
example.
We now set X = C o (l), and we define the operator A as follows:

(D(A)= {uEXnH.()),AuEX },
SI Au = 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 Lemma 2.6.5, we have u E X, and so Au E X. Therefore, u E D(A) and
u Au = f. Hence A is m- dissipative.

Remark 2.6.8. In the three examples of 2.6.1 and 2.6.2, note that the same
formula (the Laplacian), corresponds to several operators that enjoy different
properties (since they are defined in different domains). In particular, the ex-
pression the operator A has a meaning only if we specify the space in which this
operator applies and its domain.
Examples in the theory of partial differential equations 29

2.6.3. The wave operator (or the KleinGordon operator) in Hp (S2) x


L 2 (Q)
Let S2 be any open subset of RN, and let X = Ho (fl) x L 2 (1). 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 2.5). Let
A = inf {11VU1IL2,u E Ho(S2), IIuII L 2 = 1}. (2.2)

(In the case in which 5l is bounded, we recall that A is the first eigenvalue of
,L in Ho (S2), and that A> 0). Let m> A. Then X can be equipped with
the scalar product

((u, v), (w, z)) _ / (Du Vv + maw + vz) dx. 1


This scalar product defines a norm on X which is equivalent to the usual norm.
We define the linear operator A in X by
D(A) = {(u,v) E X, Au E L 2 (S2),v E Ho(S2)};

1
A(u, v) = (v, Au mu), V(u, v) E D(A).

Proposition 2.6.9. A is skew-adjoint, and in particular A and A are m-


dissipative with dense domains.

Proof. D(1) x V(l) C D(A) and so D(A) is dense in X. On the other hand,
for all ((u, v), (w, z)) E D(A)z, and by (2.1), we have

(A(u, v), (w, z)) (Ov . Ow + mvw + (Du mu)z) dx


=J
J
= (vu Vz+muz+ (Aw mw)v) dx
_ ((u, v), A(w, z)) (2.3)
Applying (2.3) with (u, v) _ (w, z), it follows that

(A(u, v), (u, v)) = 0.


Hence A is dissipative (Proposition 2.4.2). Now let (f, g) E X. The equation
(u, v) A(u, v) _ (1,9) is equivalent to the following system:

r2uiu =f+g; (2.4)


Slv=u f. (2.5)
o
By Proposition 2.6.1, there exists a solution u E H (Sl) of (2.4), satisfying L u E
L 2 (1). Next, we solve (2.5) and we obtain v E Ho (52). Therefore (u, v) E D(A)
and (u, v) A(u, v) = ( f, g), so that A is m-dissipative. Similarly, we show
that A is m-dissipative. By (2.3), we have G(A) C G(A*). Corollary 2.4.11
proves that A is skew-adjoint.
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))' _
H-1(Q) with dense embeddings. We equip Ho (1) with the scalar product de-
I fined in 2.6.3. Theorem 1.1.4 shows that

H '(r) = {u E D'(), 5
- E Ho (s^), 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 = L 2 (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 dense 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

I
v), (w, Z))L2 X H_ =
f vwdx+ (Au - mu, z) - i
= J vw dx + J (Du V(p + muc) ) dx
2 z

= J vw dx - f u(A - m) dxz

=
J vw dx - J uz dx.
Similarly, we have

I ((u,v),B(w,z)) t,2 XH -1 =
J zudx - J wvdx.
Therefore,

(B(u,v),(w,x))L xH-1= - ((u,v),B(w,z))L2xx-1. (2.7)


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
Ho (f2) of (2.4). Next, we solve (2.5) and we obtain v E L 2 (fl). Therefore (u, v) E
D(B) and (u, v) B(u, v) = (f, g); hence 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 that B is skew-adjoint.

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 have

II(I - A)V IIY = ((I - B)V, (I - B)V)' - II^IIY + II BV(Y.


Let V = (u, v). We have

JJBV 11 2 = ^w11 + AU - mu) 1 - V L2 + IIuI1 2 , = VI1 2 .


hence the result. 0

2.6.5. The Schrodinger operator

Let f be any open subset of R h', 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 write 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.

Proof. The result follows from Proposition 2.6.1 and Corollary 2.5.2.
32 m-dissipative operators

Remark 2.6.13. As in 2.6.1, if 11 has a bounded boundary 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 product

(u, v)-i = (^P.., ^Pv)H1 =Re J dx,



for u, v E X. We define the linear operator C in X by

D(C) = Ho ( 1l);
Cu=Lu, VuED(C).

Proposition 2.6.14. C is self-adjoint < 0.

Proof. We have D(Sl, C) C D(C) so that D(C) is dense in X. Furthermore,


for all 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)

Taking u = v, it follows that

(Cu,u)-1= IIkIIL2 + IIUIIH-1 < 0,


and so C is dissipative. Theorem 1.1.4 proves that 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). 0

Finally, consider the operator 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, and in particular A and -A are m-


dissipative with dense domains.

Notes. For more information about 2.6, see Brezis [2), Courant and Hilbert [1],
as well as Gilbarg and Trudinger [1].
El
3 1
The HilleYosidaPhillips Theorem and
applications
1
3.1. The semigroup generated by an m-dissipative operator I
Let X be a Banach space and let A be an m-dissipative operator in X, with
dense domain. For A > 0, we consider the operators Ja and A A defined in 2.2,
and we set TA(t) = eIA,, for t > 0.
,

Theorem 3.1.1. For all x E X, the sequence u(t) = T(t)x converges uni-
formly on bounded intervals of [0, T] to a function u E C([0, 00),X), as A J 0.
We set T(t)x = u(t), for all x E X and t > 0. Then

T(t) E (X) and I T(t)II < 1, Vt > 0; (3.1)


T(0) = I; (3.2)
(3.3)

t
T(t + s) = T(t)T(s), bs, t > 0.

In addition, for all x E D(A), u(t) = T(t)x is the unique solution of the problem

u E C([0, oo), D(A)) f1 C'([O, oo), X); (3.4)


u'(t)= Au(t), Vt > 0; (3.5)
u(0) = x. (3.6)

Finally,
T(t)Ax = AT(t)x. (3.7) I
forallxED(A) andt>0.

Proof. We proceed in five steps.


Step 1. By Definition 2.2.11, for all t > 0 and all A > 0, we have

T. (t) = e *J-1 e zr = e I e *JA ;

and so,
(ITa(t)11 S e zetII Jall <1
34 The HilleYosidaPhillips Theorem and applications

In particular,
Ilua(t)II IIxil, (3.8)
for all A > 0 and all t > 0.

Step 2. Assume that x E D(A). It is clear by construction that AA and A.


commute, for all A, p > 0. In particular, for all s, t > 0, we have

{Ta(st)T, (t st)} = tTA(st)TN,(t st)(AA A ).


ds

It follows that

Ilua(t) u (t)II = IIT. (t)xT (t)xII


i
< ds {T(st)T(t st)x} dsl < tIlAax A xII.

We deduce (Proposition 2.2.12) that uA is a Cauchy sequence 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 have

(IT(t)xII IIxil,
for all t > 0, x E D(A); and so T(t) can 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 TA(t)x f hj + IITA(t)x n T(t)xIj


+ IIT(t)xn T(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 suffices to remark 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)xII.

It follows that 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

llva(t) T(t)AxII < T(t)Ax TA(t)AxII + IIAAx AxII.


Two important special cases 35

Hence, v, --> T (t)Ax as A .. 0, uniformly on [0, T] for all T > 0. Taking

ua (t) = x +
JO t va (s) ds,

and letting .\ 1 0, it follows that


t
u(t) = x +
fo T(s)Ax dx.

Thus u E C l ([0, oo), X), and

u (t) = T(t)Ax, (3.9)

for all t > 0. Finally, we have 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, (Ja Ta (t)x, A(Ja TA (t)x)) --3(T(t)x,T(t)Ax) in X x X as 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)). Putting together (3.7)
and (3.9), we obtain (3.5).
Step 5. Uniqueness 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 all t E [0, r]. Hence, v(r) = v(0), and so u(r) = T(T)x. r >_ 0 being arbitrary,
the proof is complete. 0

3.2. Two important special cases

We assume in this section that X is a real Hilbert space. 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 following 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. (3.12)
36 The Hille-Yosida-Phillips Theorem and applications

In addition, we have

II Au(t) II < - 1 1 xII; (3.13)

- (Au(t), u(t)) < 2t IIxII2 (3.14)

Finally,
1
Au(t) 112 < (Ax, x), (3.15)
2t
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 functions Jjua(t)II and IIua(t)II are non-increasing with respect to t.
In addition, we have

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)

From (3.17), it follows that -(Axua(t), ua(t)) is non-increasing with respect to


t. Integrating (3.16) between 0 and t > 0, it follows that

- t(Aaua(t),ua(t)) < - f A),ua(s),ua(s))ds < 2(Ix112. (3.18)


0

Integrating (3.17), we obtain

2tIIua(t)II <2f1jua(s)jI ds = -(Aax,x) + (Aaua(t),ua(t)) < -(Aax,x).


(3.19)
and
sUu'(s)II ds = 2] s a (AauA(s), ua(s)) ds
d
t 2 llua(t)II. < 2] t

Hence, with (3.18),


5 -f 0
t
(Aaua(s),ua(s))

2t2IIua(t)112 < IIxII 2 . ( 3.20)


On the other hand, the vector subspace G(A) is closed in X x X strong, and
so is closed in X x X weak. Furthermore, u(t) = Aaua(t) = AAJaua(t),
and Jaua(t) 'u(t) as A 1 0 (see Step 4 of the proof of Theorem 3.1.1). In
Two important special cases 37

addition (by (3.14)), for all t > 0, IIAAJaua(t)I) is bounded as .1 j 0. Therefore


u(t) E D(A), for all t > 0, with
Au(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) are obtained by passing 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 <e < t.
It follows from (3.10) and Theorem 3.1.1 that u(t) = T(t - e)u(e), and so

lIu(t) - T(t)xI) < IIu(e) - xli + ljT(e)x - xli *0 as a ^, 0.

Therefore u(t) = T(t)x, for all t > 0, which completes the proof.

Remark 3.2.2. Theorem 3.2.1 means that T(t) has a smoothing effect on the
initial data. Indeed, even if x D(A), we have T(t)x E D(A), for all t > 0. This
is in contrast with the isometry groups generated by skew-adjoint operators.

Theorem 3.2.3. Assume that A is a skew-adjoint operator. Then (T(t))t>o


can be extended to a one-parameter group T(t) : R -> (X) such that

T(t)x E C(R,X), Vx E X; (3.21)


(IT(t)xll = ^lxii, dx E X,t E R; (3.22)
T(0) = I; (3.23)
T(s + t) = T(s)T(t), `ds, t E R. (3.24)
In addition, for all x E D(A), u(t) = T(t)x satisfies u E C(R, D(A)) fl C' (R, X )
and
u'(t) = Au(t), (3.25)
foralltER.

Proof. We denote by (T+(t)) t >o and (T (t)) t >o the semigroups corresponding
-

to A and -A. We set


( T(t), ift>0;
T(t)={

We easily verify (3.21), (3.22), (3.23), and (3.24) for x E D(A), and then for

a
x E X by density. Finally,

-t) (0) =
d u(t) (0= Ax -
) d+ .
dt dt (0)
(3.25) follows from the last identity and Theorem 3.1.1.
38 The HilleYosidaPhillips Theorem and applications

Remark 3.2.4. It is clear that if x V D(A), then u(t) g D(A) for all t E R.
Remark 3.2.5. The conclusions of Theorem 3.2.3 may be satisfied without
assuming that A is skew-adjoint. Indeed, it suffices (and the proof would the
same) that A and A are m-dissipative (and X may be any Banach space).

Corollary 3.2.6. Following the notation of Theorem 3.2.3, we have

T(t)* = T(t),

for all t E R.

Proof. Let x, y E D(A). We have

dt (T(t)x,T(t)y) = (AT(t)x, T(t)y) + (T(t)x, AT(t)y) = 0.

Therefore
(x, y) = (T(t)x, T(t)y),
for all t E JR and all x, y E D(A). Taking y = T(t)z, we have (x,T(t)z) _
(T(t)x, z), for all x, z E D(A); hence the result, by density.

3.3. Extrapolation and weak solutions

We know (Theorem 3.1.1) that if x E D(A) then T(t)x is the solution of (3.4)
(3.6). If X is a Hilbert space and A is a self-adjoint operator, then T(t)x is still
the solution of (3.10)(3.12), even for x E X. However, in general, if x D(A),
T(t)x is not differentiable in X and then it cannot satisfy (3.11). We will see that
the results of 2.3 allow us to identify T(t)x. We follow the notation introduced
in 2.3, and we denote by (T(t)) t >o and (T(t)) 9 >o the semigroups corresponding
to A and A. We begin with the following result.

Lemma 3.3.1. For all x E X and all t > 0, we have T(t)x = T(t)x.

Proof. The result is clear for x E D(A). The general case follows from an usual
density and continuity argument.

Corollary 3.3.2. Let x E X. Then u(t) = T(t)x is the unique solution of

u E C([o, oo), x) n C l ([0, oo), X );


u (t) = Au(t), Vt > 0;
u(0) = x.

Proof. We apply Lemma 3.3.1 and Theorem 3.1.1.


Contraction semigroups and their generators 39
9j

3.4. Contraction semigroups and their generators

Definition 3.4.1. A one-parameter family (T(t)) t > o C (X) is a contraction


semigroup in X provided that
(i) iiT(t)ii < 1 for all t > 0;
(ii) T(0) = I;
(iii) T(t + s) = T(t)T(s) for all s, t > 0;

(iv) for all x E X, the function t i-- T(t)x belongs to C([0, cc), X).

Definition 3.4.2. The generator of (T(t)) t >o is the linear operator L defined
by
( T(t)x - x 1
D(L)=S` xEX; h as a limit inX ash jO J7,
h v'^

and
T(t)x - x
Lx=1h mm h

for all x E D(L).

The following proposition justifies the introduction of m-dissipative operators


in Chapter 2.

Proposition 3.4.3. Let (T(t)) t >o be a contraction semigroup in X and let L


be its generator. Then L is m-dissipative and D(L) is dense in X.

Proof. We proceed in three steps.


Step 1. L is dissipative. For all x E D(L), A > 0, and h> 0, we have
T(h)x - x
II x - A 11 ? 11 (1 + h) x,, - h IIT(h)xil ? JJxJJ;
hence the result, letting h j. 0.
Step 2. L is m-dissipative. We define the operator J by

Jx = -t T(t)x dt,
f0 00 e
for all x E X. It is clear that J E (X), with liJlJ <_ 1. For x E X and h > 0,
we have
Th h f
I Jx = I
J e -t (T (t + h)x - T(t)x) dt
( ) -

0
=1 f ^
1 '
J e -t T(t)xdt
f
h
h Jh 0
eh h
- eh - 1 / e -t T (t)x dt - h e-tT(t)xdt
h ..11o
40 The Hille-Yosida-Phillips Theorem and applications

Letting h . 0, we obtain

h
m T(h -I Jx=Jx-x;
urn

and so Jx E D(L), with LJx = Jx - x, i.e. Jx - LJx = x.

Step 3. For all x E X and t > 0, we set

= 11 r t T(s)x ds.
xt
J0

It is clear that x t ----> x as t j 0. To show that D(L) is dense, it suffices to prove


that x t E D(L), for all t > 0. Now we have, for all h > 0,

t
t T (hh - I xt =
fh
+hT(s)xds
0 -J t T(s)x ds

f
- 1 tr
+h T(s)x ds -1 h T(s)x ds.
it

As h 1 0, the term on the right-hand side converges to T(t)x - x, and so x t E


D(L) with tLx t = T(t)x - x.

Theorem 3.4.4. (The Hille-Yosida-Phillips Theorem) A linear operator A


is the generator of a contraction semigroup in X if and only if A is m- dissipative
with dense domain.

Proof. If A is the generator of a contraction semigroup in X, Proposition 3.4.3


shows that A is m- dissipative with dense domain.
Conversely, assume that A is m,- dissipative with dense domain, and let
(T(t)) t >o be the semigroup corresponding to A given by Theorem 3.1.1. Then,
(T(t)) t >o is clearly a contraction semigroup. Denote its generator by L and let
us show that L = A.
For all x E D(A) and h > 0, we have (Theorem 3.1.1)

T(h)x = x + T(s)Ax ds,


j
and so x E D(L) with Lu = Au. Consequently, G(A) C G(L).
Finally, let y E D(L). Since A is m- dissipative, there exists x E D(A) such
that x - Ax = y - Ly; and since G(A) C G(L), we have (x - y) - L(x - y) = 0.
L being dissipative, we have x = y, and so G(L) C G(A). It follows that A = L,
which completes the proof.
Contraction semigroups and their generators 41 1
Finally, the following result shows the uniqueness of the semigroup generated
by an m-dissipative operator with dense domain. I
Proposition 3.4.5. Let A be an m-dissipative operator with dense domain.
Assume that A is the generator of a contraction semigroup (S(t)) t >o. Then
(S(t)) t >o is the semigroup corresponding to A given by Theorem 3.1.1.

Proof. Let (T(t)) t > o be the semigroup corresponding to A given by Theo-


rem 3.1.1. Let x E D(A), and u(t) = S(t)x. For all t > 0 and h > 0, we
have
u(t + ) u(t) _ S(h) I
u(t) = S(t)1 x > S(t)Ax as h j 0.

We deduce that S(t)x E D(A), for all t > 0, and that

AS(t)x = S(t)Ax = du (t),

for all t >_ 0. Thus u E C([0, oo), D(A)) f1 C'([O, oo), X) and u'(t) = An(t), for
t > 0. Therefore, by Theorem 3.1.1, we have S(t)x = T(t)x; hence the result,
by density.

The following definition is related to Theorem 3.2.3.

Definition 3.4.6. A one-parameter family (T(t))tE R of linear operators is said


to be an isometry group in X provided that
(i) T(t)xJ = Ijxl! for all x E X and all t E R;
(ii) T(0) = I;

I
(iii) T(t + s) = T(t)T(s) for all s, t E R;
(iv) for all x E X the function t T(t)x belongs to C(R,X).

Further to Theorem 3.2.3 and Remark 3.2.5, we have the following result.

Proposition 3.4.7. Let A be an m-dissipative operator with dense domain,


and let (T(t)) t>o be the contraction semigroup generated by A. Then (T(t))t>o
is the restriction to R + of an isometry group if and only if A is m-dissipative.

Proof. It is clear by Theorem 3.2.3 and Remark 3.2.5 that the condition A is
m-dissipative is sufficient. Assume that (T(t)) t >o is the restriction to R + of an
isometry group (T(t)) tE R, and set U(t) = T(t), for t > 0. Then (U(t)) t >o is
a contraction semigroup. Let B be its generator. For all h > 0 and x E X, we
have
U(h) I = T(h) I T(h) I
x=U(h) x.
We deduce immediately that B = A; hence the result.
42 The HilleYosidaPhillips Theorem and applications

3.5. Examples in the theory of partial differential equations

3.5.1. The heat equation


We use the notation of 2.6.1, and we denote by (S(t)) t >o the semigroup gener-
ated by B in Y.

Lemma 3.5.1. The embedding D(B) * Ho (1) is continuous.

Proof. Let u E D(B). In particular, we have u E Ho(S2) and, by Lemma 2.6.2,

IIuIIH1 = IIUIIL2 + IIVUIIL2 = IIujIL2 - (Bu, u).


Thus,
IIUIIHI < 2IIuIID(B)

for all u E D(B). C

Applying Proposition 2.6.1 and Theorem 3.2.1, we obtain the following


proposition.

Proposition 3.5.2. Let cp E L 2 (1l) and let u(t) = S(t)cp for t > 0. Then u is
the unique solution of the problem

u E C([0, oo), L 2 (1l)) n C l ((0, oo), L 2 (Sl)), Lu E C((0, oo), L 2 (S2)); (3.26)
u (t) = Du(t), Vt > 0; (3.27)

u(0) = V. (3.28)
In addition, we have

u E C((0, oo), Ho (1)); (3.29)

IIoUIIL2 <_ 71 II^2IIL2, Vt > 0 ; (3.30)


1
IIVuIIL2 <_ JJVIIL2, Vt > 0. (3.31)

Assuming more regularity on cp, the solution u is also more regular

Proposition 3.5.3. In Proposition 3.5.2, assume further that cp E Ho (fl).


Then u E C([0, oo), Ho (St)) and

IIIUIIL2 <_ IIV^PIIL2, (3.32)

for all t > 0. In addition, if A E L 2 (Q), then u E C 1 ([0,00),L 2 (1)), Du E


C([0, oo), L 2 (1)) and (3.27) is satisfied for t = 0.
Examples in the theory of partial differential equations 43

Proof. Assume first that cp E Ho (St) and AW E L Z (Sl). Then the result is a
straightforward consequence of Theorem 3.1.1 and (3.15).
By density, (3.32) is verified for all cp E Ho (S2). Now we need only show that
u E C([0, oo), Ho (52)), i.e. (by (3.26)), that u(t) --> cp in Ho (S2), as t J. 0. We
use the notation introduced in the proof of Theorem 3.2.1. Passing to the limit
in (3.19) as A J 0, it follows that

f0
1 Ilou(s)IIi2 ds < 2IIVwIIi2;

Iiou(t)IIL2 - IIo^IILz =21 t II u(s)IIL2 ds.


fo
Thus Ilou(t)IIL2 p IIV IIL2, and IIu(t)IIx1; ^I IIE1 as t 10. On the other
hand, we know that u(t) > cp in L 2 as t j 0; hence the result.

Remark 3.5.4. If ci has a bounded boundary of class C 2 , then (see Re-


mark 2.6.3) D(B) = H 2 (cl) n Ho (S2). Therefore, if cp E H 2 (S2) n Ho (S2), then we
have u E C([0, oo), H 2 (c)).

Proposition 3.5.5. Let A be defined by (2.2). Then

at, (3.33)
IIS(t)IIc(L2) 5 e -

for all t > 0.

Proof. Let cp E D(B), and let f (t) = (eAt1IS(t)cpll) 2 , for t > 0. We have

e -2At f (t) = 2 f u(t) 2 + 2 I u(t)u'(t)

= 2A f u(t) 2 + 2 f u(t)Au(t)

= 2f u(t) 2 -2
J IDu(t)I 2 <0.

Thus IIS(t)II < e -At II^PII for all t >_ 0 and all cp E D(B). The general result
follows by density.

We now assume that S2 has a bounded Lipschitz continuous boundary and


we follow the notation of 2.6.2. Let (T(t)) t > o be the semigroup generated by
A in X. We have X Y, and G(A) C G(B). We easily deduce the following
result.
44 The Hille-Yosida-Phillips Theorem and applications

Lemma 3.5.6. For all cp E X and all t > 0, we have T(t)cp = S(t)cp.

Consequently, for all cp E X, u(t) = T(t) cp satisfies the conclusions of Propo-


sition 3.5.2. In addition, the following estimates hold.

Proposition 3.5.7. Let 1 < q < p < oo. Then



I[ S ( t )[I L P < (47rt)- N! 'I'II L ,
(3.34)

for allt>0 and all cpEX.

The proof requires the following two results.


z
Lemma 3.5.8. Fort > 0, we define K(t) E S(RN ) by K(t)x = (4irt) - e- 4i .
Let 0 E CC (R N ) and let v(t) = K(t) * 0. Then v E C([0, oo), Cb(R N )) n
C((0, oo), Cb (R N )) and, for all 1 < p < oo, we have v E C([0, oc), LP(R N )) f1
C 00 ((0, oo), LP(R N )). In addition:

(i) vt =0v for allt>0;


(ii) v(0) = ^,b;
(iii) [v(t)IILP < (47rt) - a (9 P)JkbIIL9, for 1 < q < p < oo and for all t > 0.

Proof. Regularity and properties (i) and (ii) follow from easy calculations.
Property (iii) is a consequence of Young's inequality*, since

I K ( t )IIL P =p ^(47rt) 2( 1 P) < (4t) *@ P),


),

for 1<p<ooandforallt>0.

Lemma 3.5.9. Let cp E Y, cp > 0 a.e. on Q. Then, for all t > 0, we have
S(t)cp > 0 a.e. on Q.

Proof. By density, we may assume that cp E D(B). We set u(t) = S(t), and
we consider u E C([0, oo), Ho (S2)). By Proposition 3.5.2, we have, for all t > 0,
-

d +() 2 = - utu = -f U Au = f Vu - Vu = _f IVU - [ 2 < 0.


J
- -

From this, we deduce that fn(u)2 < 0, for all t > 0, and so u > 0.

* Recall Young's inequality: 11 f * 9II LP C 11f IIL91191[Lr, with 1 < p, q, r < oo,
and 1/p= 1/q+1/r- 1

Examples in the theory of partial differential equations 45 1

Proof of Proposition 3.5.7. By density, we may assume that cp E D(A). Let


= Icpl. Invoking Lemma 3.5.9, for all t > 0, we have

-S(t)( < S(t)cp < S(t)(,

almost everywhere on Sl; and so

I jS(t)^PI )Lp -< II S(t)CII Ln. (3.35)

We define E C(R N ) by

( on S2;

{
0 onR N \S2.

Then we set v(t) = K(t) * , and

u(t) = u(t)^ Q - S(t)(. (3.36) 1

We have u E C((0, oo), C(S2)) n C((0, oo), H l (S2)) n C 1 ((0, oo), L 2 (S2)) and Au E
C((0, oo), L 2 (St)). Furthermore, u(t) = v(t) > 0 on 3; u t = Du for t > 0; and
u(0) = 0. Thus,

dt f (u
- )2 =-
Juut - _ -
f u Au = f Du Vu
z
-

z
- _-
J IVu - I2 _<0,

and so u(t) < 0, for all t > 0. We then deduce from (3.35) and (3.36) that

II S(t) FPII LP C II v(t)II Lp I


We conclude by applying Lemma 3.5.8.

Corollary 3.5.10. Let A > 0 be given by (2.2), and let M = e \ l g l 2,N / (4 ">
Then
-at ,
II S(t)II L(x) <- Me (3.37)
for allt>0.

Proof. LetcpEXandletT>0. For0<t<T,wehave

II S(t)'PII L < IHIL= < e -At e^ n IIIPIIL.


For T < t, it follows from (3.34) that

II S(t)^PII L- << (47rt) 4 IIS(t - T)cpII L 2 < (47rt)- 4 e-AteATII^GIIL2


<I52I2(4^rt)-4e-at e"1 PIIL-.

The result follows by taking T = IS21 2 /N/(47r).


46 The Hille-Yosida-Phillips Theorem and applications

Remark. We cannot take M = 1 in (3.37). More precisely, we have

d {IIS(t)II'C(x)}t-o = 0 ;

and so, if IIS(t)IIc(x) < M'e - t with > 0, we have M' > 1. Indeed, let
cp E D(l) be such that cp - 1 in a neighbourhood of xo E Il and IIVIIx = 1, and
let u = S(t)cp. We see that u E C([0, cc) x S2). Thus we have u t (0) - 1 in a
neighbourhood of X. Consequently, for all e > 0 and for x in a neighbourhood
of x0, we have u(t, x) > 1- Et, for t small enough, and so in particular II u(t) II x >
1 - Et, for t small enough; hence the result.

Concerning L" inequalities, note that applying (3.37) and (3.34), we verify
easily that for all 1 < p < oo, there exists a constant Mp such that

II S(t)II e(LP) <- Mpe-\t, (3.37')

for all t > 0. Once more, we cannot take Mp = 1. Actually, one can see that,
for p > 2, one has Oat
II S(t)Il c(LP) <e,,
for all t > 0, and that this inequality is optimum in the following sense:

a t
{IIS(t)II^cLp )}t=o = - p2'

Indeed, for all cp E D(l), letting u(t) = S(t), and multiplying by lulp -2 u, we
obtain the equation satisfied by u. Applying (3.33), we obtain

o= 1 d Ivu2 I>
1d
j lu(t)IP + 4 f lu(t)IP + 2
4 f Iu(t)IP;
p dt l^ p p dt st p st

the inequality follows. To show optimality, it suffices to verify that, for all E > 0,
there exists 0 E D(1l) such that

fa IvV)2 I < (A+E)


J IV)IP.
To see this, we consider the first eigenfunction cp l of -A in Ho (S2) (see Brezis [21).
For e> 0, let 0, : [0, oo) -+ [0, oo) be such that 6 e - 0 in a neighbourhood of 0,
OE(x) < 1 and x - 8 E (x) <_ e for all x >_ 0. Set 2/' _ (9( 1 ))2/P . We verify that
Ali, E D(f1). Furthermore, Iv 12 , < IV^pil and, last,

V) >
Jn 1pi as E j 0.
Examples in the theory of partial differential equations 47

Consequently,

11 2 =A f cp 2 =Alim
J M 2 < fn Iv n EiJn

3.5.2. The wave equation (or the Klein-Gordon equation)


We use the notation introduced in 2.6.3 and 2.6.4, and we denote by (T(t))tER
the isometry group generated by A in X, and by (S(t)) tE R the isometry group
generated by B in Y.

Proposition 3.5.11. Let (cp, V) E X and let u(t) be the first component of
T(t)(p, z/'). Then u is the unique solution of the following problem:

UE C(R, Ho (fl)) n C 1 (R, L 2 (1)) n C 2 (R, H -1 ( 1l) ); (3.38)


utt-Du+mu=0, for alltER; (3.39)
u(0) = cp, u t (0) = z/i. (3.40)

In addition,

J {Jvu(t)1 2 + mju(t)J 2 + u(t)2} = Jst {IvcpI 2 + mIcpI 2 + ,0 2 } , ( 3.41)

for alit E R. Finally, if(cp,0) E D(A), wehaveu E C'(R,Ho(St))nC 2 (R,L 2 (1l))


and Du E C(R, L 2 (1)).

Proof. Let u E D'(R, H -1 (S2)) and set U = (u, u t ). Then U E C(R, D(B)) fl
C'(R, Y) if and only if U E C(R, Ho (52)) fl C(R, L 2 (SZ)) fl C 2 (R, H -1 (1)). Fur-
thermore, in that case, (3.39)-(3.40) is equivalent to the equation

U'(t) = BU(t),

for all t E R. The result then follows from Propositions 2.6.9, 2.6.10, and 2.6.11,
Theorem 3.2.3, and Corollary 3.3.2. Note that (3.41) is equivalent to (3.22).

3.5.3. The Schrodinger equation

We use the notation introduced in 2.6.5, and we denote by (S(t)) tE 1 and


(T(t)) tE R the isometry groups generated by B and A. We have G(B) c G(A),
and from this it is easy to deduce the following result.

Lemma 3.5.12. For all cp E Y, we have S(t)co = T(t)cp, for all t E R.

Then we have the following.


48 The Hille-Yosida-Phillips Theorem and applications

Proposition 3.5.13. Let cp E H0'(1) and let u(t) = T(t)cp. Then u is the
unique solution of the problem

u E C(R, Ho (l)) n C l (R, H-1(l)); (3.42)


ju t +,Lu = 0, for all t E R; (3.43)

u(0) = W. (3.44)

In addition,

f Iu(t) 2 =
f IcpI , for all t E R,
2 (3.45)

J IV I 2 ,
i n IVu(t)I 2 = for all t E R. (3.46)

Finally, if Lcp E L Z (cl), then u E C 1 (R, L 2 (S2)) and Au E C(R, L 2 (fl)).

Proof. We use Theorem 3.2.3, Corollary 3.3.2, and Lemma 3.5.12. (3.45) is
equivalent to (3.22). On the other hand, invoking (3.7) and (3.22), we obtain

IIAu(t)IIx = IIA(PIIx, (3.47)

for all t E R. But, for all V E Ho (S2), we have

(3.48)
II AvII x = II(Av - v) + vIIi = f IVvI 2 - JM 2 + IIvIIX

By (3.22), IIu(t)IIx = Ik IIx for all t E R; and so (3.46) follows by putting


together (3.47), (3.48), and (3.45). 0

3.5.4. The Schrodinger equation in R n'


We use the notation of 3.5.3, and we assume that Sl = 1[8N. We can state
estimates in the spirit of (3.34).

Proposition 3.5.14. For all p E [2, oo] and t 54 0. Then T(t) can be extended
to an operator belonging to (LT (RN), L(R")). In addition, we have

IIT(t)II,C(Lp ,LP) C (3.49)


(4^rItI)N(2-p)'

for all 0.

Proof. Let cp E S(IRN) and let u(t) E C 00 (IR,S(R N )) be defined by

Fu(t)(^) = e-'IE12t.F'G(^), (3.50)


Examples in the theory of partial differential equations 49

forall^ER' andtER. We have

idt P a(t)(6) - I6I 2 Fu(t)(C) =0 in R N ,

for all t E R; and so


iut + 0u = 0 in R N ,

for all t E R. Therefore we have (Proposition 3.5.13) u(t) = T(t), for all t E R.
Now we know that

f 2
.-1{e-;1e12t}(x) =
N e t := K(t)x,
(47rt) 2

for all x E R N and t # 0. It follows from (3.50) that u(t) = K(t) * cp for all
t 0. We deduce that

1
T(t) W II L- < II(P11t1,
(47rt) 2

for all t # 0 and cp E S(R N ). Thus, one can extend T(t) to an operator
of L(L l (R' ), L(1R ')) such that T(t)1I,c(L1, L c) < (4irItj) - . Furthermore,
T(t) E G(L 2 (R N ),L 2 (1R N )), with T(t)IIc(L2,L2) = 1. The general case follows
from the Riesz interpolation Theorem (see Dunford and Schwartz [1], p. 525, or
Bergh and Lofstrom [1], p. 2, Theorem 1.1.1).

Notes. Theorem 3.2.1 can be generalized to the case of the generators of an-
alytic semigroups; see Goldstein [1], Haraux [3], Pazy [1]. One can build semi-
groups for some classes of operators, rn-dissipative operators (non-linear), and
maximal monotone operators. These two classes coincide in Hilbert spaces. See
Brezis [1], Crandall and Liggett [1], Crandall and Pazy [1], and Haraux [1, 2].
Id

Inhomogeneous equations and abstract


semilinear problems

Throughout this chapter, we assume that X is a Banach space and that A


is an m-dissipative operator with dense domain. We denote by (T(t)) t >o the
contraction semigroup generated by A.

4.1. Inhomogeneous equations

Let T> 0. Given x E X and f: [0,T] --> X, our aim is to solve the problem
u E C([0,T],D(A)) nC'([O,T],X); (4.1)
u'(t) = Au(t) + f (t), `dt E [0, T]; (4.2)
t. u(0)=x. (4.3)
As in the case of ordinary differential equations, we have the following result
(the variation of parameters formula, or Duhamel's formula).

Lemma 4.1.1. Let x E D(A) and let f E C([0, T}, X). We consider a solution
uEC([0,T],D(A))nC'([0,T],X) of problem (4.1)(4.3). Then, we have

u(t) =T(t)x +
J0 T(t s)f(s)ds,
t
(4.4)

for all t E [0, T] .

Proof. Let t E (0,T]. Set


w(s) = T(t s)u(s),
for s E [0, t]. Lets E [0, t] and h E (0, t s]. We have
w(s +h(
) w(s)
7t s h) {
u(s + u(s) T(h) I u(s) 1
T(t s) {u'(s)Au(s)} = T(t s)f(s)
as h . 0. Since T(t ) f() E C([O,t],X), we deduce that w E C 1 ([0,t),X) and
that
w'(s) = T(t s) f (s), (4.5)

for all s E [0,t). Integrating (4.5) between 0 and rr < t, and letting T T t, we
obtain (4.4). 0
Inhomogeneous equations 51

Corollary 4.1.2. For all x E D(A) and f E C([O,T],X), problem (4.1)-(4.3)


has at most one solution.

Remark 4.1.3. For all x E X and all f E C([O,T],X), formula (4.4) defines
a function u E C([O, T], X). Now we are looking for sufficient conditions for u
given by (4.4) to be the solution of (4.1)-(4.3).

Remark 4.1.4. It is clear that if u is a solution of (4.1)-(4.3), then x E D(A).


However, this condition is not sufficient. Indeed, assume that (T(t)) tE R is an
isometry group, and let y E X \ D(A). Then (see Remark 3.2.4), T(t)y D(A),
for all t E R. Take f(t) = T(t)y, and x = 0 E D(A). It follows easily that (4.4)
gives u(t) = tT(t)y D(A), for t # 0.

Lemma 4.1.5. For all x E X and f E L 1 ((0,T),X), formula (4.4) defines a


function u E C([0,T],X). In addition, we have

IIUIIC([O,T],X) <IIxiI +IIfIILl(O,T,X)

Proof. The result is clear if f E C([0,T],X), and follows by density in the


general case. 0

Proposition 4.1.6. Let x E D(A) and let f E C([0,T],X). Assume that at


least one of the following conditions is satisfied:

(i) f E L'((0,T),D(A));

(ii) f E W 1,1 ((0,T),X).

Then u given by (4.4) is the solution of (4.1)-(4.3).

Proof. We proceed in four steps. Set

v(t) = J t T(t - s)f (s) ds t T(s)f (t - s) ds,


O = fo
for t E [0,T].

Step 1. We have v E C 1 ([0,T),X). Indeed, if f E L 1 ((0,T),D(A)), for t E


[0, t) and h E [0, t- s], write

v(t + h) - v(t) - f t T(h) - I


h T(t-s) h
1 l+h
f(s)ds + h T
(t+h-s)f(s)ds,
f
52 Inhomogeneous equations and abstract semilinear problems

and let h J 0. Note that


T(h)-I ,
Af
h f
in L 1 ((0,T),X), as h . 0, and apply Lemma 4.1.5. It follows that

d+ v
t T(t - s)A f (s) ds + f(t),
dt (t) =J
for all t E [0, T). If f E W 1,1 ((0, T), X), for t E [0, T) and h E [0, T - t], we

-h
write

v(t + h) - v(t) = O T(s) f (t + h - f (t - s)


ds+ T(h) f h
T (t - s)f (S) ds,

and we let h J, 0. Note that (Corollary 1.4.39)

.f(t+h.) ,f(t.) f'(t.), as h J 0


h
in L 1 ((0, t), X) and apply Lemma 4.1.5. It follows that

d+ v
dt (t) = fo t T(s) f'(t - s) ds + T(t)f(0),

for all t E [0,T). In both cases, we have d+v/dt E C([0,T),X); and so v E


C 1 ([O,T),X)
Step 2. Similarly, we show that (d v/dt) (T) makes sense and is equal to
-

i imv'(t); and so v E C'([O,T],X).

Step 3. Let t E [0, T) and let h E [0, T - t]. We have


11t
T (h) -I
h
v(t) = I t T (t + h - s) f (s) ds - f t T (t - s) f (s) ds

- v(t+h)-v(t) 1
-- [ 7(t + h- s) f (s) ds.
h h J t

Letting h 1 0, we deduce v(t) E D(A), and Av(t) = v'(t) - f(t). This is still true
for t = T, since the graph of A is closed. It follows that v E C([O,T], D(A)),
and that v satisfies (4.2).
Step 4. We have u(t) = T(t)x+v(t) E C([0,T], D(A))f C'([O, T], X), and (4.1)
follows. Furthermore,

u'(t) = AT(t)x + Av(t) + f(t) = Au(t) + f(t),

for all t E [0, T]. Hence, we have (4.2), and (4.3) is immediate.
Inhomogeneous equations 53

Corollary 4.1.7. Let x E X, f E C([O,T],X) and let u be given by (4.4).


Then using the notation of 2.3 and 3.3, u is the unique solution of the problem

u E C([O,T],X) nC'([0,T],X);
u'(t) = Au(t) + f (t), Vt E [0, T];
u(0) = x.

Proof. We apply Lemma 3.3.1 and Proposition 4.1.6.

Corollary 4.1.8. Let x E X, f E C([0,T],X) and let u be given by (4.4).


Assume that at least one of the following conditions is satisfied:
(i) u E C([0,T],D(A));
(ii) u E C 1 ([0,T],X).
Then u is the solution of (4.1)(4.3).

Proof. Assume that (i) holds. By Corollary 4.1.7, we have u' E C([0, T], X);
and so u E C i ([0, T], X ), hence the result.
Now assume that (ii) holds. By Corollary 4.1.7, we have Au E C([0,T],X);
and so (Corollary 2.3.2) u E C([0,T],D(A)); hence the result.

Throughout 4.1, we have supposed that f E C([0,T],X). But in order to


give a sense to (4.4), it suffices that f E L 1 ((0,T),X) (Lemma 4.1.5). In this
case, we have the following result.

Proposition 4.1.9. Let x E X, f E L 1 ((0,T),X) and let u E L 1 ((0,T),X).


Assume further that u E L 1 ((0,T),D(A)) or that u E W 1 1 ((0,T),X). Then u
"

verifies (4.4) if and only if u

u E L 1 ((0,T),D(A)) nW'" i ((O,T),X);


u'(t) = Au(t) + f (t), for almost every t E [0,T];
u(0) = x.

Proof. First note that if u E W 1 1 ((0,T),X) then u E C([0,T],X) (Corol-


'

lary 1.4.36) and so the condition u(0) = x makes sense. Let us first show that
the assumptions of the theorem are sufficient to have (4.4). To see this, we argue
as in Lemma 4.1.1. We consider t E (0, T] and we set w(s) = T(t s)u(s), for
almost every s E (0, t). For all h E (0, t), and for almost every s E (0, t h), we
have
w(s + h) w(s) u(s + h) u(s) T(h) I
_ T(t s h) (s)}
h
54 Inhomogeneous equations and abstract semilinear problems

It follows that w is absolutely continuous from [0, T] to Y. Moreover, the right-


hand member converges for almost every s E (0, t) to T(t s)u'(s) Au(s) =
T(t s) f (s), as h . 0 (Theorem 1.4.35). Therefore w is right differentiable
almost everywhere on (0, t) and

s
d (s) = T(t s) f (s)
d

Similarly, we show that w is left differentiable almost everywhere on (0, t) and


that
v (s) = T(t s)f(s).

Consequently (Theorem 1.4.35), w E W 1 1 ((0,T),Y) and w'(s) = T(t s)f(s)


'

almost everywhere. We deduce (4.4).


Conversely, assume that u satisfies (4.4). Let (fn ) n >o be a sequence of
C([0,T], X) such that fn if in L 1 ((O,T), X) as n + oo, and let u n be the
corresponding solutions of (4.4). Invoking Corollary 4.1.7, we have

I u(t) = Au(t) + fn(t),

for all t E [0,T]; and so

u(t) = x + J t (Aun(s) + fn(s)) ds


0

for all t E [0,T]. Letting n ti oo, we obtain (Lemma 4.1.5)

u(t) = x +
JO t (Au(s) + f(s))
ds,

for all t E [0,T]. It follows that u E W ' ((0,T),Y) and that


1 1

u'(t) = Au(t) + f(t),

for almost every t E [0,T]. If u E W 1 1 ((0,T),X), we have Au E L 1 ((O,T),X);


'

and so u E L 1 ((0,T),D(A)) (Corollary 2.3.2). If u E L 1 ((O,T),D(A)), we have


u' E L' ((O, T), X ); and so u E W 1 1 ((0, T ), X ). This completes the proof.
"

4.2. Gronwall's lemma

In this section, we give a result which is essential in the study of semilinear prob-
lems; not only for showing uniqueness of solutions but also for finding bounds
on the solutions.
Semilinear problems 55

Lemma 4.2.1. (Gronwall's lemma) Let T > 0, A E L l (0, T), A >_ 0 a.e. and
C1, C2 > 0. Let cp E L' (0, T), cp > 0 a.e., be such that A E L'(0, T) and

cp(t) < C1 + C2
J0
t A(s)cp(s) ds,

for almost everyt E (0,T). Then we have

rt
cp(t) < Cl exp (C2
Jo
A(s)ds )

for almost every t E (0,T).

Proof. We set
ft
(t) = C l + C2
J A(s)cp(s) ds.
0

i) is differentiable almost everywhere (since it is absolutely continuous), and we


have
z/^'(t) < C2 A(t)cp(t) < C 2 A(t)O(t),

for almost every t E (0, T). Consequently,

(
dt{ ip(t) exp (C2 f ll
t A(s) ds I 1 < 0,
l I ))J
and so
t

J
< Cl exp (C2 A(s) ds)
O

hence the result, since cp < 0.

Remark 4.2.2. In particular, if Cl = 0, we have cp = 0 a.e.

4.3. Semilinear problems

Definition 4.3.1. A function F : X > X is Lipschitz continuous on bounded


subsets of X provided that for all M > 0, there exists a constant L(M) such
that
IIF(y)-F(x)II <_L(M)IIy -xII, ex,yEBM,
where BM is the ball of center 0 and of radius M.

Throughout s4.3, F: X 4 X is a Lipschitz continuous function on bounded


subsets of X. We denote by L(M) the Lipschitz constant of F in BM for M > 0.
In particular, L(M) is a non-decreasing function. of M.
56 Inhomogeneous equations and abstract semilinear problems

Given x E X, we look for T> 0 and a solution u of the following problem:

u E C([0,T],D(A)) nC'([O,T],X); (4.6)


u'(t) = Au(t) + F(u(t)), Vt E [0, T]; (4.7)
u(0) = x. (4.8)

We also consider a weak form of the preceding problem. Indeed, by Lemma 4.1.1,
any solution u of (4.6)(4.8) is also solution of the following problem:

T(t s)F(u(s)) ds, Vt E [0, T].


u(t) = T(t)x +
JO t
(4.9)

Finally, note that, for all u E C([O,T],X), (4.9) is equivalent (following the
notation of Corollary 4.1.7) to the problem

u E C([0,T],X) nC'([O,T],X);
u'(t) = Au(t) + F(u(t)), Vt E [0, T];
u(0) = x.

4.3.1. A result of local existence

We begin with a uniqueness result.

Lemma 4.3.2. Let T > 0, x E X, and let u, v E C([0, T], X) be two solutions
to problem (4.9). Then u = v.

Proof. We set M = sup max{IIu(t)II, llv(t)ll}. We have


tE [O,T]

t II F(u(s)) F(v(s)) II ds < L(M)


II u(t) v(t)II < I0 fo t I u(s) v(s) II ds,
and we conclude using Remark 4.2.2.

Set
1
TM = 2L(2M + IIF( 0 )II) + 2 > 0,
for M > 0. We can state a first result of local existence.

Proposition 4.3.3. Let M > 0 and let x E X be such that < M. IIxII Then
there exists a unique solution u E C([O,TM], X) of (4.9) with T = TM.
Semilinear problems 57

Proof. Lemma 4.3.2 proves uniqueness. Let x E X and let M > IIxII. We let
K = 2M + IIF(0)II and

E = { u E C([O,TM],X); [In(t)I[ < K,`dt E [0,TM]},


and we equip E with the distance generated by the norm of C([O,TM], X), Let

d(u, v) = max IIu(t) v(t) II,


tE[0 ,TM I

for u, v E E. Since C([0, TM], X) is a Banach space, (E, d) is a complete metric


space. For all u E E, we define (D u E C([O,TM],X) by

= T(t)x +
f
T(t s)F(u(s)) ds,

for all t E [0, TM]. Note that for s E [0, TM], we have F(u(s)) = F(0)+(F(u(s))
F(0)); and so

IIF(0)II
IIF(u(s))II < IIF(0)II + KL(K) < M +TM

It follows that

0)II
II(t)It < IIxII +JO t IIF(u(s))Ilds <M+t M TM( <K.
Consequently, we have F : E --4 E. Furthermore, for all u, v E E, we have

II(t) (t)II < L(K)


fo IIv(s) u(s)II ds < TML(K)d(u, v) < 1 d(u, v).
Therefore, 1 is a contraction in E with Lipschitz constant 1/2, and so 1) has a
fixed point (Theorem 1.1.1) u E E, which satisfies the requirements of Proposi-
tion 4.3.3.

Theorem 4.3.4. There exists a function T : X (0, oo] with the following
properties: for all x E X, there exists u E C([0, T(x)), X) such that for all
0< T < T(x), u is the unique solution of (4.9) in C([O,T], X). In addition,

2 L(IIF( 0 )II + 2 11u(t)II) ? T(x) t 2, (4.10)

for all t E [0, T(x)). In particular, we have the following alternatives:

(i) T(x) = oo;


58 In homogeneous equations and abstract semilinear problems

(ii) T(x) < oo and lim Iu(t) ll = oo.


ttT(x)

Remark 4.3.5. If property (i) holds, we say that the solution u is global. On
the other hand, if (ii) holds, we say that u blows up in finite time. In other
words, the alternatives (i)-(ii) mean that the global existence of the solution
u is equivalent to the existence of an a priori estimate of IIu(t)II on [0,T(x)).
In applications, we establish such a priori estimates by standard methods in
the theory of partial differential equations (multipliers, comparison principles,
and maximum principles), as well as by various techniques involving differen-
tial or integral inequalities more specific to evolution equations (first integrals,
Liapunov functions, and variants of Gronwall's lemma).

Proof of Theorem 4.3.4. It is clear that (4.10) implies that if T(x) < oo, then
IIu(t)II-->ooastIT(x). Let x E X. We set

T(x) = sup{T > 0; 3u E C([0,T],X) solution of (4.9)}.

By Proposition 4.3.3, we know that T(z) > 0. On the other hand, the uniqueness
(Lemma 4.3.2) allows us to build a maximal solution u E C({0, T(x)), X) of (4.9).
It remains to show (4.10). Inequality (4.10) being immediate if T(x) = oo, we
may assume that T(x) < oo. We argue by contradiction, assuming that there
exists t E [0,T(x)) such that (4.10) does not hold. We then have

T(x) - t < TM,

with M = 1Iu(t)j^. Let v E C([O,TM],X) be the solution, given by Proposi-


tion 4.3.3, of
+
v(s) = T(s)u(t)
10 ,
T(s - a)F(v(c)) do,

for all SE [0, TM] . We then define w E C([0, t + TM], X) by

u(s), ifs E [0, t];


w(s) =
v(s - t), if s E [t, t + TM].

We verify easily that w is a solution of (4.9) with T = t + TM, which contradicts


the definition of T(x), since t + TM > T(x).

Remark 4.3.6. It may very well happen that, for the same equation, T(x) <
oo for some initial data, and T(x) = oo for others. For example, choose X = R,
A = 0, and F(u) = u 3 - u. This choice corresponds to the ordinary differential
equation u' = u 3 - u. If (xi <_ 1 we have T(x) = oo, and if xi > 1 we have
T(x) < oo. In the last case, (4.10) gives 12iu(t)1 2 > (T(x) - t) ' - 4. This
-
Semilinear problems 59

estimate describes the blow-up phenomenon sharply, since the solutions actually
blow up as (T(x) - t) I'. -

4.3.2. Continuous dependence on initial data

Proposition 4.3.7. Following the notation of Theorem 4.3.4, we have the


following properties:
(i) T : X -* (0, oo] is lower semicontinuous;
(ii) if x n --* x and if T < T(x), then u n -3 u in C([0, T], X), where u n and u
are the solutions of (4.9) corresponding to the initial data x n and x.

Proof. Let x E X, and let u be the solution of (4.9) given by Theorem 4.3.4.
Let 0 <T < T(x). It suffices to show that if (x n ) n > o C X is a sequence such
that x n *x as n -^ oo, then for n sufficiently large T(x) > T and u n --+ u in
C([0,T],X). To see this, set

M = 2 sup IIu(t)1I,
tE[O,TJ

and
Tn = sup{t E [0,T(x n )) ; IIun(s)II < 2M,Vs E [0,t]}.

For n large enough, we have (IxII < M; and so Tn > TM > 0. For all t < T,
t<Ti,,wehave

Ilu(t) u n(t)II (Ix xn ll + L(2M)


fo t IIu(S) un(S)IIds;
and it follows from Lemma 4.2.1 that

II u(t) u(t)ii < IIx x n lieTL( 2 M), (4.11)

for t < T, t <_ Tn . In particular, we deduce from (4.11) that for n large enough
we have
II un(t)II C Al,
for t _< min{T, rrn }; and so Tn > min{T, rrn }, i.e. Tn > T. We then have
T(x) > T. Applying (4.11) again, we see that u n --> u in C([0,T],X). This
completes the proof.

Remark 4.3.8. Actually, T may be discontinuous. For example, choose X =


1R 2 , A = 0, and F(u, v) = (vu 2 , -2). For x = (1, 2) we have T(x) = 1 and the
corresponding solution is ((1 - t) -2 , 2(1 - t)). For x E = ((1 + E) -1 , 2) we have
T(x e ) = oo and the corresponding solution is ((e + (1 - t) 2 )',2(1 - t)).
60 Inhomogeneous equations and abstract semilinear problems

4.3.3. Regularity
In some cases, it is possible to give a more precise result on the regularity of
solutions of (4.9). In particular, we have the following.

Proposition 4.3.9. Assume that X is reflexive. Let T > 0, x E X, and let


u E C([0,T],X) be a solution of problem (4.9). Then, if x E D(A), u is the
solution of problem (4.6)(4.8).

Proof. Let h> 0 and let t E [0,T h]. It is easy to see that

u(t + h) u(t) = T(h)x x+ J t T(s){F(u(t + h s) F(u(t s))} ds

f
0

+ h T(t + s)F(u(s)) ds.

Hence,

IIu(t + h) u(t)II c IIT(h)x x1j+h sup IIF(u(s))II


sE[O,T]

+ L(M)
J 0
t llu(s + h) u(s) I ds

Frthermore, we have

h
T(h)x x = J T(s)Axds;
O

and so IIT(h)x xII < hilAxil. Applying Lemma 4.2.1, we obtain

u(t + h) u(t)II < Ch,

for 0 < t <t + h < T. Consequently, u: [0, Ti 4 X is Lipschitz continuous, and


it follows that F(u) : [0, T] ---> X is also Lipschitz continuous. We conclude by
applying Corollary 1.4.41 and Proposition 4.1.6.

Remark 4.3.10. If X is not reflexive, the conclusion of Proposition 4.3.9 may


fail, as shown by the following example. Choose X = Co(R) x Co(IR), where
Co(IR) is the space of functions of C(R) which dies away to 0 as x ^ foo,
equipped with the norm L "(][8). We define the operator A by

D(A) = {( u, v) E X f1 C' (R 2 ); (u', v') E X};


A(u, v) = (u', v'), `d(u, v) E D(A).
{
Isometry groups 61

A is m-dissipative with dense domain, and generates the semigroup (T(t)) t >o
given by
I
T(t)(u, v) = (u(t + ), v(t + )),
for t > 0, x E R. Next, consider the Lipschitz continuous function F : X --+ X
given by

t
F(u, v) = (v,0), V(u, v) E X.
For all (x, y) E X, the corresponding solution (u, v) of (4.9) is given by

(u, v)(t) = (x(t +) + t y +(t + ), y(t + ))

Taking (x, y) E D(A) such that y(0) = 0 and y'(0) 54 0, y+ is not in C'(R), and
so (u, v)(t) V D(A), for t 0.

4.4. Isometry groups

In the case in which A generates an isometry group (see Theorem 3.2.3), and
in particular when X is a Hilbert space and A is skew-adjoint, we can also
solve (4.7) for t < 0. Indeed, solving the problem

u E C([T,0],X) nC'([T,0],X);
u (t) = Au(t) + F(u(t)), Vt E [T, 0];
u(0) = x;

is equivalent to solving

v E C([0,T],X) nC 1 ([0,T],X);
v'(t) = Au(t) F(u(t)), Vt E [0, TI;
1. v(0)=x;
setting u(t) = v(t), for t E [ T, 0]. The second problem is solved by Theo-
rem 4.3.4, since A is m-dissipative and F is Lipschitz continuous on bounded
sets of X.

Notes. One finds generalizations of the results of 4.3 in Segal [1] and Weissler [1].
Also consult Ball [1, 2] for an interesting discussion about the blow-up phe-
nomenon.
I
II
5
The heat equation

I Throughout this chapter, we assume that S2 is a bounded subset of R N with


Lipschitz continuous boundary, and we use the notation of 3.5.1. In particu-
lar, X = Co(1) and Y = L 2 (1). In addition, we consider a locally Lipschitz
continuous function g E C(R, R), such that

g(0) = 0.

We define the function F : X --^ X by

I
F(u)(x) = 9(u(x)),

for all u E X and x E Sl. It is easy to check that F is Lipschitz continuous on


bounded sets of X. In what follows, we denote g and F by the same expression.

5.1. Preliminaries

Given cp E X, we look for T> 0 and u solving the problem

u E C([0,T], X) n C((0,T], Ho (S2)) n C'((O, T], L 2 (5l));



Au E C((O,T],L 2 (1l)); (5.1)

u t Au = F(u), Vt E (0, T]; (5.2)
u(0) = cp. (5.3)

The result is the following.

Proposition 5.1.1. Let cp E X, T> 0, and let u E C([0,T],X). Then u is


solution of (5.1)(5.3) if and only if u satisfies

(5.4)
u(t) = T(t) cp +
JO T(t s)F(u(s)) ds,
t

foralltE [0,7'].

Proof. Let u be a solution of (5.1)(5.3), let t E (0,T], and let e E (0,t]. We


set
v(s) = u(e + s),
Preliminaries 63

for 0 < s < t - e. It is clear that v is a solution of (5.2) on [0, t - e] and that
v(0) = u(e) E D(B). Hence, we have (Lemma 4.1.1)

v(s) = S(s)u(e) +
10, S(s - o,)F(v(u)) ds,
for all s E [0, t - E]. Applying Lemma 3.5.6, we deduce that

u s+e T(s)u(e) + " T s-v F u v +e ds,

for all s E [0, t - e]. Since u E C([0,T], X), we have, for all s E [0, t),

T(s)u(E) -; T(S)W,

as j0;
F(u(. + e)) -' F(u(.)),
uniformly on [0, s] as e j 0. Letting first E 1 0, and then s T t, we deduce (5.4).
Conversely, let u E C([0, T], X) be a solution of (5.4). We consider 0 <
t ( T. By Proposition 3.5.2, we have T(t) cp E Ho (Sl), and the function s H
T(t - s)F(u(s)) belongs to C([0, t), Ho (S2)), with

IIT(t - s)F(u(s))IIHI < C(1 + (t - s) -1 " 2 ) E L 1 (0, t);

and so (Proposition 1.4.14 and Corollary 1.4.23) u(t) E Ho (S2). A similar esti-
mate shows that actually u E C((0, T], Ho (l1)),

IIu(t)IIH1 < C(1 +t -1 / 2 )

Since g is Lipschitz continuous on bounded subsets of R and since the range of


u is bounded, we conclude (Proposition 1.3.5) that F(u(t)) E H(l), and that

IIF(u(t))IIH= < C(1 +t -1 1 2 ).

It follows that F(u) is weakly continuous as a map from (0, T] to Ho (St). Take
0 <t <T again. Applying Proposition 3.5.2 again, we obtain LT(t)v E L Z (52),
and that the function s T(t - s)F(u(s)) is weakly continuous as a map from
(0,t) to L 2 (Sl), with

DT(t - s)F(u(s))II L2 < C(t - s) -1 / 2 (1 + s -1 / 2 ) E L l (0, t);

and so Au(t) E L 2 (1l). Consequently, u(t) E D(B), for all t E (0,T]. We show
similarly that u E C((0,T), D(B)). We then set

v(s) = u(e + s),


64 The heat equation

for O < s t - E. We have

v(s) = S(s)u(E) +
fo s
S(s - a)F(v(a)) ds,

for all s E [0, t - e]. We conclude by applying Corollary 4.1.8. o

Note that here it is not possible to invoke Proposition 4.3.9, since X is not
reflexive.

Remark 5.1.2. Considering the above estimates in more detail, we obtain, as


a consequence of Proposition 3.5.2,

IIVu(t)IILa +
<- CIcI 1 / 2 (t -1 + t) ;

where C depends only on g and sup IIu(t)II.


O<t<T

Remark 5.1.3. Similarly, we easily verify that if cp E Ho (Sl), then we have

Iiou(t)IIL2 < CIQ1 1 " 2 (1 + t'/ 2 );


IILu(t)IIL 2 < CIcj 1/2 (t -1/2 + t),
where C depends only on g, sup Iu(t)II andII^PIIH1
O<t<T

Remark 5.1.4. The same method also shows that if we assume further that
cp E D(B), then
IIAu(t)IIL2 G CIslIli 2 (1 + t),
where C depends only on g, sup IIu(t)II, and IIcolIHl.
O<t<T

5.2. Local existence

Applying Proposition 5.1.1 and Theorem 4.3.4, we obtain the following result.

Theorem 5.2.1. For all cp E X, there exists a unique function u, defined on a


maximal interval [0, T()), which is a solution of (5.1)-(5.3) for all T E (0, T(,)).
In addition, if T (cp) < oc, then IIu(t) II -4:: as t j T().

Remark 5.2.2. u depends continuously on cp (this follows from Proposi-


tion 4.3.7).

The following proposition gives improved regularity of u if cp is more regular.


Global existence 65

Proposition 5.2.3. Assume that cp E X fl Ho (S2). Then the solution u


corresponding to (5.1)-(5.3) is in C([0,T(cp)),Ho(c)). Suppose further that
Ap E L 2 (c); then u E C([0,T(p)),D(B)) fl C'([0,T(cp)), L 2 (c)).

Proof. Assume that cp E X f1 Ho (S2), and let t E (0, T(cp)). Applying (5.2),
Proposition 3.16, and (3.31), we obtain

u(t) VIIH' f t 1 IIF(u(s))Ilds


< II S(t)V VIIH1 + C /o t-s
<IIS'(t)VVIIH=+C^-->0 astj0.
Therefore u E C({0, T()), Ho (S2)). In particular, if T < T(p), then u is bounded
in Ho (SI) on [0, T], and then so is F(u). Consequently, in the case in which
A E L 2 (c), it follows from (5.2) and (3.32) that

IIo(u(t) V)IILZ I(S(t) I)oVIIL2 +Cft t1 S JIF(u(s))IIHlds tie

and so u E C([0,T(cp)), D(B)). In particular, u(t) Ap + F(cp) in L 2 (c) as


t 1 0. Furthermore, for t <T(cp), we have

u(t) - cp _ S(t) - I 1 /' t


t t V+t
0
J
S(t - s)F(u(s)) ds --> Acp + F(cp) as t j 0.

Consequently, (d + u/dt) (0) = t o u t (t); and so u E C'([0, T(cp)), L 2 (c)).

5.3. Global existence

We establish here two kinds of results. First, we show that if g satisfies certain
conditions for Ix) large, then all solutions of (5.1)-(5.3) are global. Then, in
another spirit, some results prove that if g satisfies certain conditions for IxI
small, then the solutions of (5.1)-(5.3) with small initial data are global. We
begin with the following result (the maximum principle).

Proposition 5.3.1. Let T > 0 and cp E X. Let u E C([0,T], X) n C((0,T),


H(i)) n C 1 ((0,T),L 2 (fl)) with Au E C((0,T),L 2 (c)), and f E C([0,T],X)
be such that

1
ut1=f, VtE (0,T);
u( 0 ) = W.
(5.5)

Assume further that there exists a constant C such that

If(t,x)I < CIu(t,x)I, (5.6)

in [0, T] x Q. Then, if cc > 0, we have u(t) > 0 for all t E [0, T] .


' 66 The heat equation

Proof. For t E (0, T), we set


v(t) = f((t))2.

Multiply (5.5) by -u (t) and integrate over Q. Integrating by parts and apply-
-

ing (5.6), we obtain (see the proof of Lemma 3.5.9)

v -(t) < f If(t)Iu (t) <C f Iu(t)1u (t) = Cv ( t .


)

Integrating the last inequality, we obtain, for all 0 < s < t < T,
v(t) < v(s)et-8).
Letting s 1 0, it follows that
v(t) e ct f (VP ) 2 = 0 , Vt E
0
hence the result.

I Remark 5.3.2. Applying Proposition 5.3.1 with v = -u, we see that if cp <0,
then we have u(t) < 0 for all t E [0, T].

We give a first result of global existence.

Proposition 5.3.3. Assume that there exist K, C < co such that

xg(x) < CIx1 2 , ( 5.7)


for Ixl <_ K. Then, for all E X, the solution of (5.1)-(5.3) given by Theo-
rem 5.2.1 is global.

Proof. We proceed in two steps.


Step 1. Assume first that cp > 0. It is easy to verify that, for all T < T(p),
h(t) = F(u(t)) satisfies (5.6). Then we have u(t) > 0 for all t E [0,T(cp)).
Furthermore, by (5.7) we have

g(x) < Cx,

for x >_ K. Since g is bounded on [0, K] and by possibly modifying C, it follows


that
g(x) < C + Cx,

for x > 0; and so

I for all t E [0,T(cp)).


IIF(u(t)) +11 < C + CIIu(t)11,
Global existence 67

Applying (5.4), Lemma 3.5.9, and Gronwall's lemma, we deduce that

ct
u(t)II < (C+ IIwII)e ( 5.8)

for all t E [0,T(cp)); and so (Theorem 5.2.1) T(cp) = oo.

Step 2. In the general case, we set 0 = IcpI and we denote by v the corresponding
solution of (5.1)(5.3). Let T < min{T(),T(cp)} = T(cp). We easily verify that
w = v u fulfills the assumptions of Proposition 5.3.1; hence

u(t) < v(t),

for all t E [0, T] . We then use z = v + u, where v is the maximal solution of the
equation

v = T(t) +
fo t T(t s){g(v)} ds.

Since g(x) verifies the same assumptions as g(x), v is a global solution


(Step 1). Applying Proposition 5.3.1, we obtain that

u(t) > v_(t),

for all t E [0, T]; and so

IIu(t)II < max{ IIv(t) II, 112(t)II

for all t E [0, T]. It follows that T(cp) > T(i) = oo. This completes the proof.

Remark 5.3.4. If C = 0 in (5.7), it follows from (5.8) that all solutions


of (5.1)(5.3) satisfy
IIu(t)II s K + II^II,
for t >_ 0. Actually, this result can be sharpened by the following proposition,
whose proof is quite simple.

Proposition 5.3.5. If C = 0 in (5.7), then solutions of (5.1)(5.3) satisfy

IIu(t)II 5 max {K, IIVII},

fort>0.
68 The heat equation

Proof. As in Proposition 5.3.3, we may restrict ourselves to the case cp > 0.


Set k = max{K, Ilcpll} and multiply (5.2) by (u k)+ E Ho (S2). Integrating by
parts and setting f (t) = f ((u(t) k)+) 2 , it follows that

f'(t) < C fn g(u(t))(u(t) k) + < 0.

We conclude that f (t) < f (0) = 0, and so u < k; hence the result.

If the constant C of (5.7) is sufficiently small, we still have a result of this


kind. To see this, we consider A given by (2.2).

Proposition 5.3.6. If C < .\ in (5.7), then, for all cp E X, the solution u


of (5.1)(5.3) satisfies
sup Iu(t)II < 00.
O<t<00

Proof. We have xg(x) < C1 + Cx 2 . As in Proposition 5.3.3, we may assume


that cp > 0. Let c > 0 and let u be the corresponding solution. We multiply (5.2)
by u. Integrating by parts and setting f (t) = f u(t) 2 , it follows that

f'(t) < 2a f(t) + 2f u(t) g (u(t)) < 2(A C)f (t) + C1IQI.

By Lemma 8.4.6 (see below), we have

f(t) f( 0 )+C1 A I ^ IC ;

and thus sup IIu(t)IIL2 = K < oo. Let p E (N/2, oo). By (3.34) and (3.37),
0<t<oo
we have
TI7(t)lTG(Lp,L00 ) E L 1 (O ,+oo).
We then note that there exists a constant C2 (see the proof of Proposition 5.3.3)
such that
II9(u(t))+IILP <_ C2 + Cu(t)j,,
for all t > 0; and so

Ilu(t)II ^ IIkII + IIT(.)IIG(Ln,LO)(C2 +C O S p IIu(S)IILP)


C3 +C4 Sup IIu(S)IIL",
O<_s<_t

for all t >_ 0. Finally, invoking Holder's inequality, for all e > 0, there exists
C(e) such that
IIVIILP < EIIvII +C'(E)IIVIIL2,
Global existence 69

for all v E X. Choosing e such that eC4 < 1/2, it follows that
1
IIu(t)II C3 + KC(e)C4 + 1 sup IIu(s)II,
2 0<8<t

for all t > 0. Therefore

sup IIu(s)II < 2C3 + 2KC(E)C4 i


0<s<t

for all t > 0; hence the result. c

Now we show that if g fulfills certain conditions in a neighbourhood of 0,


then solutions with small initial data are global. To see this, consider ) given
by (2.2).

Proposition 5.3.7. Suppose that there exists a > 0 and p < .\ such that

xg(x) < lx1 2 , for jxj < a.

Then, there exists A < oo such that, if JIVII < aA, the corresponding solution u
of (5.1)-(5.3) is global and satisfies

IIu(t)II <_ All(pj!e -(a- ) t,


for t>0.

Proof. As in Proposition 5.3.3, it suffices to deal with the case cp > 0. Set

T = sup{t E [0,T()); IIu(s)II < a for s E [0,t}} > 0.

Multiply (5.2) by u. Integrating by parts, and letting f (t) = f u(t) 2 , for t E


[0, T], it follows that

f'(t) <-2f(t) +2
J u(t)g(u(t)) < -2(.\ - C) f (t);

and so
IIu(t)IIL2 IlIt2e U u,
-
-

for all t E [0, T]. We write

t
e( )t u (t) = e U-u)tT(t)^ + f eU-u)(t-9)T(t - s) (e 3
F(u(s)))ds
o
70 The heat equation

Note that (see the proof of Proposition 5.3.6)

IIe(a)`T(t)IIG(LP, Loo) E L 1 (O,+oo).

Consequently, arguing as in the proof of Proposition 5.3.6, we obtain

1
e (A ) tlIu(t)II <e tIHwHI+CIISOIIL2+ upteUu(s)Ij ,
2
os

for all t E [0, T]. Hence, there exists A such that

sup e a sIIu(s)II : AII^jI,


(
)

O<s<t

for all t E [0,T]. It follows that if 1IcpII < aA, then T = oo, which completes the
proof. 0

Remark 5.3.8. We see that if p 1 0, we may take 5 j a. If g has a higher


order near 0, we have a more precise result.

Proposition 5.3.9. Assume that there exist p > 0, e > 0, and a > 0 such
that
xg(x) < plxI2+E, for xj < a.

Then there exist Q, -y > 0 such that, if 11 pll < 3, then the corresponding solution
u of (5.1)-(5.3) is global and

1)u(t) II : 'YII^PIIe - ^ t ,

for t > 0.

Proof. As in Proposition 5.3.3, we may assume that y > 0. Set

for x > 0, and -^ = min 0 < 0. For all a E (0, ^), there exist 0 < x a < Ya such
that
0(xa) + a = 0(y a ) + a = 0.

Furthermore, we have a < X a <a(1 + e)/e (see Figure 5.1).


Global existence 71

-EX

Fig. 5.1

Suppose that IIcpjj <a and set

T = sup{t E [0,T(cp)), Iu(s)II < a for s E [0,t]} > 0,


1(t) = sup IIu(s)II
o<s<t

We have
IIF(u(t))+II s llu(t)II 1 +E,
for t E [0, T]. Applying (5.4), (3.37), and Lemma 3.5.9, it follows that

sllu(s)II} +Eds < MII^vII + eMf(t) 1 +E


E a9 {e a i
eat llu(t)II < MII^II +Mfo e -

Thus,
0(1(t)) + > 0,
for all t E [0, T). If we assume further that MII^PII < , then this implies that
f (t) E [0, xMII II) U (yMII u oc) Since f (0) E [0, xMJIWII) and f (t) is continuous
,

in t, we have
f (t) E [0, xxMliroli ),

for all t E [0, T). We conclude as in Proposition 5.3.7, and we obtain the result
with ,3 = min{a, C/M} and ry = (1 + E)M/E.
72 The heat equation

5.4. Blow-up in finite time


We begin with a result whose proof is quite simple. Consider the function
0 E D(B), such that

Lv)+A')=0, (5.9)
'>0 on S2, (5.10)

IQ = 1.

zp is easily obtained by solving the minimization problem (2.2), using the com-
(5.11)

pactness of the embedding Ho (St) L 2 (c). We may choose a positive mini-


mizing sequence, and obtain (5.10). is the first eigenfunction of 'L in Ho (Sl)
(see, for example, Brezis [2], Theorem VIII.31, p. 192).

Proposition 5.4.1. Suppose that there exist a, /8, e > 0 such that g(x) >
ax 1 +E Ox, for x > 0. Let cp E X, cp > 0 on St, be such that

(-P)
f^
Then T(cp) < oo.

Proof. Denote by u the solution of (5.1)(5.3) given by Theorem 5.2.1. By


Proposition 5.3.1, we have u(t) > 0 on S2, for all t E [0,T(p)). Set

f (t) = i u(t)V) > 0,


n

for t E [0, T (cp)). Applying (5.9) and Lemma 2.34, we obtain, for all t E [0, T(W)),

f-(t) = fut(t) = I ('^'U(t) + g(u(t)))

= J u(t)A + J g(u(t))V) = Af(t) + J g(u(t)) )


1+E b
> (A + li)f (t) + a J u(t)

On the other hand, by (5.11) and Holder's inequality,

I}e i
./ (t) <
(fSZ u(t)1+E ) U2 m ) l+f
< (f u(t)1+E 4^ )

and so
f'(t) ? f(t)((A+0)+of(t)E), (5.12)
Blow-up in finite time 73

for all t E (0,T(^p)). Let T = sup{t E (0,T(cp)), f' > 0 on (0,t)} > 0. If
T < T(cp), we have f'(T) = 0 and f (T) > f(0), which contradicts (5.12). Thus,
we have T = T(c') and f' > 0 on [0,T(cp)). Now let b > 0 be such that

(a - S)f ( 0 ) 6 = A + ,Q. I
We deduce from (5.12) that, for all t E (0, T(c )),
f , (t) > a f(t) i+E + f(t)((A + Q) + (a 6)f (t))

> of (t) -' + f (t) ((a + p) + (a 8)f ( 0 ) E ) ? (5f (t) 1 ,

i.e. I
_ f( t) -E ) > (bt)
'
( '

From this, we easily deduce that I


0 < f (t) -E <
f(0)
(0)-E
e e - bt,
I
for all t E (0,T(cp)); and so EbT(cp) < f(0) E; hence the result.
-

Remark 5.4.2. It is important to note that the above argument only shows
that ebT(cp) < f(0) f and not that EbT(cp) = f(0) - E. For further discussion
-

concerning this question, see Ball [1, 2].

Remark 5.4.3. If we take E X such that ( > 0, then for k > 0 large enough
V = k( satisfying the assumptions of Proposition 5.4.1, and so T(a) < oo.

Now we show a second blow-up result, using a different method. We need


the functional E defined by

E(u) = 2 f IDu1 2 - fc(u) ,

for u E X fl Ho (1), where


I

o
G(x) = f g(s) ds,
x

for x E ][^.I
Proposition 5.4.4. Assume that there exists K > 0 and a> 0 such that

xg(x) > (2 + e)G(x),


74 The heat equation

K xg(x) and v = o m K G(x). Let cp E X n Ho (.l) be


for lxi > K. Set =<K
I such that
(2 + s)E(cp) < IQI(p 2ev).

Then T(cp) < oo.

The proof makes use of the following two lemmas.

Lemma 5.4.5. Let T > 0 and u E C((0,T),X)nC((0,T),D(B))f1C'((O,T),


L 2 (S2)). Then
Il' (Du + g(u))ut + E(u(t)) = E(cp),
'

for all 0<s<t<T.

Proof. By density, we may restrict ourselves to the case in which u E Cl ((0, T),
D(A)) (by possibly replacing u by h ft +h Jau(s) ds). Then we have

1 f(zu + g(u))ut = f(_Vu.

and hence the result.


^
Du t + g(u)ut) =

Lemma 5.4.6. Let cp E X and let u be the corresponding solution to (5.1)-


(5.3). Then

u(t) 2 dx + 2j i Vu(t)1 2 dx = 2 f u(t)g(u(t)) dx; (5.13)


dt in

f J ut + E(u(t)) = E(u(s));
t (5.14)
z
for all0<s<t<T(cp).

Proof. (5.13) is obtained by multiplying (5.2) by u t , integrating over Il, and


then integrating by parts.
We obtain (5.14) by applying Lemma 5.4.5 and (5.2). C

Proof of Proposition 5.4.4. Set f (t) = fo u(t) 2 . By (5.13), for t E (0,T()), we


have

f'(t) _ 2
in IVU(t) 1 2 + 2
J juj<K}
u(t)g(u(t)) +2 f f
1 ju>K)
u(t)9(u(t))

>_ 2 f ^Vu(t)1 2 + 21 u(t)g(u(t)) + 2(2 + e) G(u(t))


o {Iu,<x} Juuj>K)
(5.15)
Blow-up in finite time 75

On the other hand, observe that, by Proposition 5.2.3, we may let s = 0 in (5.14);
and so

2(2 + e) G(u(t)) = -2(2 + e) G(u(t))


fl juj^!K} f1 juj<Kj

+(2 + e) Vu(t)1 2 - 2(2 +E) {E() - f'f ut} . (5.16)

Observe further that

2) u(t)g(u(t)) - 2(2 + e) G(u(t)) > 2j52( - (2 + e)v). (5.17)


JJJ 1luI<K} 4u j<K}

Putting together (5.15), (5.16), and (5.17), we obtain the following inequality,
for 0 <t <T():

f'(t) > 2(2+e) f t


o st J
ut+E J IVu(t)I 2 +2i52I(-(2+e)v)-2(2+e)E(cp). (5.18)
sz
In particular,
t
f'(t) > 2(2 + e)o f u. (5.19)
2
Now set
h(t)=f
t f (s) ds.

Applying the Cauchy-Schwarz inequality, we obtain


t
h'(t) - h'( 0 ) = f (t) - f ( 0 ) = 2 f i uue
o n
t 1/2 t 1/2
12 t 1/2
< 2u < 2h(t) (
Uo ^^2
^ I Cf fu
^ =) o J 2)
From (5.19), it follows that

h(t)h"(t) > (1+ 2) (h'(t) - h'(0)) 2 . (5.20)

For the sake of contradiction, assume that T(^p) = oo. From (5.20), we have
(1 + E/2)(h'(t) - h'(0)) 2 >_ (1 + E/4)h'(t) 2 for t >_ to large enough. It follows
from (5.20) that
(h(t)- e/4 ) 1, 0 , (5.21)
for t >_ t o . But h(t) > 0 and h(t) -, / 4 -> 0 as t --> oo. Thus there exists t l > t o
such that (h - / 4 )'(t l ) < 0. Hence, by (5.21),

0 < h(t) -e / 4 < h(t1) -e / 4 + (t


76 The heat equation

for t > tj; and so


h(t l)-E/4
t < tl - (h-E/4)'(tl)'

for t > t1, which is absurd.

Remark 5.4.7. The condition on g in Proposition 5.4.4 means that g is su-


perlinear. Indeed, in the case in which there exists xo > 0 such that G(xo) > 0,
it means that x - ( 2 +E)G(x) is non-decreasing on [xo, oo) (if G(xo) > 0 for a
certain xo < 0, then x - ( 2 +E)G(x) is non-increasing on (-oo, xo]). In this case,
G(x) > axe - bx 2 , for x >_ so. Thus, if we take ( E X f1 Ho (Sl) such that
C > 0, then E(k) -* -oo as k --- oo. In particular, for k > 0 large enough the
assumptions of Proposition 5.4.4 are fulfilled and so T(k) < oo.

5.5. Application to a model case

We choose g(x) = alxIax, with a > 0 and a 0. We consider cpX, and


we denote by u the corresponding solution of (5.1)-(5.3). Then we have the
following results.
If a < 0, then T() = oo, and u is bounded in X (Proposition 5.3.5).
If a > 0, then T(cp) = oc if lI^PII is small enough (Propositions 5.3.7 or 5.3.9).
On the other hand, for some we have T(cp) < oo (Remarks 5.4.3 and 5.4.7),
and in that case IIu(t)JI > b(T(cp) - t) - * (Theorem 4.3.4).

Notes. We have studied the heat equation in the space Co(1). It is also pos-
sible to study it in the spaces C'a(il) (see Friedman [1], and Ladyzhenskaya,
Solonikov and Ural'ceva [1]) and in the spaces LP(fl) (see Weissler [2]). In LR(1l),
we observe certain singular phenomena, such as non-uniqueness (see Baras [1],
Brezis, Peletier and Terman [1], and Haraux and Weissler [1]).
In some cases, the regularizing effect allows one to solve the Cauchy problem
with singular initial data, such as measures (see Brezis and Friedman [1]). We
can also consider more general non-linearities, depending on the derivatives of
u, and more general elliptic operators than the Laplacian. See, for example,
Friedman [1], Henry [1], and Ladyzhenskaya, Solonikov and Ural'ceva [1]. Some
non-linearities with singularity at 0 have also been studied; see Aguirre and
Escobedo [1]. Concerning systems, consult, for example, Dias and Haraux [1],
Fife [1], and Smoller [1].
Various versions of the maximal principle for the heat equation can be found
in Protter and Weinberger [1]. For the linear and non-linear regularizing effects,
consult Friedman [1], Haraux and Kirane [1], Henry [1], Kirane and Tronel [1],
and Ladyzhenskaya, Solonikov and Ural'ceva [1].
For more blow-up results, consult Fujita [1], Levine [1], and Payne and Sat-
tinger [1]. The nature of blow-up is currently rather well known. See Baras
Application to a model case 77
I

and Cohen [1], Baras and Goldstein [1], Friedman and Giga [1], Friedman and
McLeod [1], Giga and Kohn [1, 2], Mueller and Weissler [1], and Weissler [2, 3].
For the behaviour at infinity of solutions, consult Chapters 8 and 9, as well
as, for example, Cazenave and Lions [2], Escobedo and Kavian [1, 2], Haraux [1],
Henry [1], Kavian [2], Lions [1, 2], Weissler [4], and Esteban [1, 2].

I
1
0
I
1
1
1
^ ' y

n
d
6
The KleinGordon equation

ii 6.1. Preliminaries

1 In this section, we give some technical tools that are essential in this chapter.

6.1.1. An abstract result

' Let X be a Hilbert space, let A be a skew-adjoint operator in X, and let (T(t))tER
be the isometry semigroup generated by A. We have the following result.

I
Proposition 6.1.1. Let T > 0, x E X, and f E C([0,T],X). Let u E
C([0,T],X) be given by

u (t) = T (t)x +
f o
T (t - s)f (s) ds,

for all t E [0, T] . Then the function t H Il u(t)11' belongs to C' ([O, T]) and
id
2 d IIu(t)II 2 = ( f(t),U(t)),

for all t E [0, T] .

1 Proof. Suppose that x E D(A) and f E C 1 ([0,T],X). By Proposition 4.1.6,


we haven E C([0,T],D(A)) nC 1 ([0,T],X) and

u'(t) = Au(t) + f (t),

for all t E [0, T]. Thus,

2 dt Ilu(t)11 2 = ( u(t), u'(t)) = (u(t), Au(t) + f(t)) = (u(t), f (t)).


'
Hence
IIu(t)11 2 = IIx11 2 + f (u(s), f(s)) ds.
0
(6.1)

In the general case, we approximate x and f by sequences (x)>o C D(A) and


(fn)n >o C C'QO,T],X), and then we pass to the limit to obtain (6.1). The
result follows since u and f are continuous functions. o
Preliminaries 79

6.1.2. Functionals on Ho (Sl)

In this section, f is any open subset of RN. We consider a function g E C(R, R)


such that there exists 0 < a < oo and C < oo so that

g(0 ) = 0; (6.2)
I9(x) g(y) I ^ C(I xI a + Iyl a )Ix yI, b'x, y E R. (6.3)

In particular, we have Ig(x) I < ClxI+ 1 and so, for all p > a + 1, g defines an
operator F : L o,(cl) + L^ ,(1l) by

F(u)(x) = g(u(x)),

for almost every x E Q. When there is no risk of confusion, we still denote by g


the operator F. Applying (6.2), (6.3) and Holder's inequality, we easily obtain
the following result.

Proposition 6.1.2. Let a + 1 < p < oc. Then F is Lipschitz continuous from
bounded subsets of LP(S2) to LT (1l). More precisely, we have

19(u) g(v)II ^ C(u1j + IIvIIip)IIu vIILP, (6.4)

for all u, v E LP(52).

For g as above, we define G E C(]R, R) by

G(x) =
fo x g(s) ds. (6.5)

Then G verifies condition (6.3), with a replaced by a + 1. Then, it follows from


Proposition 6.1.2 that G allows us to define a functional V, Lipschitz continuous
on bounded subsets of La+ 2 (Sl), by

V(u) _
J G(u(x)) dx, `du E La+ (I) 2 (6.6)

More precisely, we have the following.

Proposition 6.1.3. V is a functional of class Cl on L 2 (f ). Its derivative


(which is a continuous mapping La+ 2 (52) -> (L 2 (1))' = L (1)) is given by

V'(u) = 9(u), (6.7)

for all u E L"+2(1)


80 The Klein-Gordon equation

Proof. We have, for all x, y E R,


x+y

G(x + y) G(x) y9(x) = f (9(a) 9(x)) do ; ,

and so, applying (6.3),

IG(x + y) - G(x) - y9(x)I <- C , (Ixl a + Iyl a )Iyl 2

Applying Holder's inequality, we deduce that, for all u, v E L"+2 (1),

v(u + v) V(u) + (v, g(u)) L ^ }2 L I = I v(u + v) v(v) + f vg(u) dxI


< C'(IIuliLa+2 + IIvllL.+2)IIvllL^+2 ,

J
hence the result.

We now assume that, instead of (6.3), g satisfies the following weaker condi-
tion:
C(1 + Ixl + IyI )Ix - yI, (6.8)
I9(x) - g(y) I a Q

for all x, y E R. In that case, we write


(6.9)
9=91+92,

where 92 verifies (6.3) and gl verifies (6.3) with a = 0. For example, consider

9(x), if Ixi < 1;

9i(x) = g(1), if x > 1;


I. g(-l), if x < 1.

On the other hand, if (N - 2)p < 2N, we have

HH(Q)'.., Lr(l) (6.10)

with dense embedding, and so

Ln'(S2) , H-1O) (6.11)

The result is the following.

< 4/(N - 2).


Proposition 6.1.4. Let g satisfy (6.2) and (6.8), with 0 < a
Then g is Lipschitz continuous from bounded subsets of Ho(1l) to H-'(Il).
Preliminaries 81 1
Proof. By Proposition 6.1.2, gl is Lipschitz continuous from bounded subsets
of L 2 (1l) to L 2 (fl), and so from bounded subsets of H(l) to H -1 (Sl). Invok-
ing Proposition 6.1.2 again, 9 2 is Lipschitz continuous from bounded subsets of
L 2 (52) to L+ (S2). Applying (6.10) and (6.11), with p = a+2, we deduce that
92 is Lipschitz continuous from bounded subsets of Ho (S2) to H -1 (52); hence the
result.

Proposition 6.1.5. Suppose that g satisfies the hypotheses of Proposition 6.1.4


with (N 2)a< 2. Then g is Lipschitz continuous from bounded subsets of
Ho (Q) to L 2 (1)_ I
Proof. The result is immediate for g l . Set p = 2(a + 2). Applying (6.3) and
Holder's inequality, it follows that

1192(U) 92(V)I1L2 < C(!I uII2 , + IvIIia)l rt VIIL-+2,


I
for all u, v E Ho (1l). However, we have (N 2)p < 2N; hence the result,
by (6.10).

We now consider G and V defined by (6.5) and (6.6). We have the following
result.

Proposition 6.1.6. Suppose that g satisfies (6.2) and (6.8), with a >_ 0 such
that (N 2)a < 4. Then V is a functional of class Cl on Ho (Sl). Its derivative
(which is a continuous mapping from Ho (52) > (Ho (S2))' = H -1 (1)) is given by

V'(u) = 9(u), (6.12) I


for all u E Ho (S2).

Proof. We apply Proposition 6.1.3 to gi and 92, and we use embeddings (6.10)
and (6.11).

Corollary 6.1.7. Suppose that g satisfies the hypotheses of Proposition 6.1.6,

I
with (N 2)a < 2. Let T > 0 and u E C([0,T],Ho(1l)) fl C'([0,T],L 2 (S2)).
Then the mapping t H V(u(t)) is in C'([0,T]), and we have

d V(u(t)) _-J sz
9(u(t))u t (t) dx, (6.13)
'

for all t E [0, T] .


82 The Klein-Gordon equation

1 Proof. Suppose first that u E C' ([0, T], Ho (S2)). Then, for all t E [0, T],

d V (u(t)) = (V'(u(t)), u (t))H-1,Ho

I = - (g(u(t)),u'(t))x-1,Ho = - I g(u(t))ut(t)dx.
n

I It follows that

V(u(t)) = V(u(0)) - IJ t
st
gu(s))u t (s) ds. (6.14)

By density, we deduce that (6.14) is still true when u E C([O,T],Ho(Sl)) n


C 1 ([O,T],L 2 (Sl)); hence the result.

6.2. Local existence

Throughout this chapter, we follow the notation of 2.6.3, 2.6.4, and 3.5.2.
In particular, 1 is any open subset of R N , m> -A, X = Hl) x L 2 (cl) and
Y = L 2 (0) x H -1 (0). We consider a function g E C(R,R) which satisfies (6.2)
and (6.8) with (N - 2)a < 2. Finally, we consider G and V defined by (6.5)
and (6.6). We define the functional E on X and the mapping F: X -> X by

E(u, v) = 211(u, v) II X + V (u)


= 21 f {Ivul 2 + mlul 2 + lv1 2 2G(u)}dam,
F(u, v) _ (0, g(u)), for all (u, v) E Ho (St) x L 2 (Il).

It is clear, from Proposition 6.1.5, that g defines a Lipschitz continuous mapping


from Ho(S1) to L 2 (Sl), and so F is Lipschitz continuous on bounded subsets of
X. Given (gyp, zli) E X, we are looking for T> 0, and u a solution of

uE C([ 0, T] , Ho($)) nC 1 ([ 0 ,T],L 2 (f))flC 2 ([ 0 ,T],H -1 (Q)); (6.15)


u tt - Lu + mu = g(u), for all t E [0, T]; (6.16)
U( 0 ) = ^P, ut( 0 ) = V (6.17)

Applying Corollary 4.1.7 and Proposition 4.3.9, and arguing as in the proof of
Proposition 3.5.11, we obtain the following result.

Lemma 6.2.1. Let T > 0 and (p, 1p) E X. Let u E C([0, T], Ho (Sl))n
C ([0,T],L 2 (cl)). Then u is solution of (6.15)-(6.17) if and only if U = (u,u t )
is 1solution of
Local existence 83

U(t) = T(t)(, 0) + T(t - s)F(U(s)) ds, (6.18)


JO t

for all t E [0,T]. In addition, if A E L 2 (S2) and E Ho(Sl), then we have


u E C 1 ([O,T],H0(1)) nC 2 ([O,T],L 2 (SZ)) and Au E C([O,T],L 2 (fl)).

Applying Theorem 4.3.4, we deduce a local existence result.

Theorem 6.2.2. For all (cp, 0/i) E X, there exists a unique function u, defined
on a maximal interval [0, T(, v>)), which is a solution to (6.15)-(6.17) for all
T <T(p,-%). If, in addition, T(p,i,b) < oo, then u(t)II H 1 +ltu t (t)R L 2 -4 cc as
t T T(0).

Finally, we have conservation of energy.

Proposition 6.2.3. Let (cp, V)) E X and let u be the corresponding solution
of (6.15)-(6.17). Then,

E(u(t),ut(t)) = E(co, l0 ), (6.19)

for all t E [0,T(p, L)).

Proof. We apply Proposition 6.1.1 and Corollary 6.1.7. It follows that

dtE(u(t), ut(t)) = (F(u(t), ut(t)), (u(t), ut(t)))x f g(u(t))ut(t) dx = 0,

for all t E [0,T(cp,0)); hence the result. C

Remark 6.2.4. Proposition 6.2.3 justifies the study of the Klein-Gordon equa-
tion in the space X. Indeed, the energy E is related to the X-norm and, as we
will see in the next sections, the conservation of the energy (6.19) allows us,
under certain hypotheses on g and (p, vi), to obtain estimates for the solution
in X (and so global existence), or results about blow-up in finite time.

Remark 6.2.5. By using 4.4, we can solve problem (6.15)-(6.17) for T < 0
as well as for T> 0. Actually, note that u is a solution of (6.16) on [-T, 0] with
u(0) = cp and u t (0) = Ali if and only if v(t) = u(-t) is solution of (6.16) on [0,T]
with v(0) = cp and vt (0) = - V.
84 The Klein-Gordon equation

6.3. Global existence

As for the heat equation (5.3), we will state two kinds of result according to
the hypotheses on g: global existence of all solutions (i.e. independent of initial
data), or global existence of solutions with small initial data.

Proposition 6.3.1. Suppose that there exists C < oo such that G(x) _< CIx1 2
for ai: x E R. Then, for all (v, ) E X, we have T(cp, 0) = oo.

Proof. Set f(t) = II(u(t),ut(t))IIX, for t E [0,T(cp,)). By (6.19), we have

f (t) <- f (0) _2 f G(cp) +2 f G(u(t)) <- f (0) -2


J G(cp) + 2C J u(t) 2, (6.20)

for all t E [0,T(cp,i )). On the other hand, we have

u(t)IIi2 = IH L2 + f t ddt is, Iu(s)I 2 = II^IIi2 +21t u(s)ut(s)


o (6.21) 0

IIIIL2 +2/ t f(s).


0

Applying (6.20), (6.21), and Gronwall's lemma, we obtain the result.

Remark 6.3.2. If 2C < A+m (A being given by (2.2)), then for all (co, ii) E X
the corresponding solution u of (6.15)-(6.17) satisfies sup II(u(t),ut(t))IIx < no.
t>o
Indeed, in this case we easily verify that C f u(t) 2 < (1 - e) f (t), with E > 0, and
it follows from (6.20) that e f (t) < C'; hence the result.

Proposition 6.3.3. Suppose that there exist p < A + m (A being given


by (2.2)) and 3 > 0 such that 2G(x) < PIxI 2 for IxJ <_ 3. Then there exists
.5,K >0 such that if I I (cp, Ali) II x < 6, we have T (cp, Vi) = no and the correspond-
ing solution u of (6.15)-(6.17) satisfies sup II (u(t), ut(t))II x < I)II x

Proof. The hypotheses on g imply that there exists a constant C < no and
k > 2 with (N - 2)k < 2N, such that

2IG(x)I < C(IxI 2 + Ixlk),

for all x E I1. By possibly taking larger C, we have

2G(x) < plxi 2 + Clxik,


Global existence 85 LI
for all x E R. Sobolev's inequalities show that there exists a constant, which we
will still denote by C, such that
I
2J G(u) <p f u 2 + CIIuIII,
fl
for all u E Ho (1). Furthermore, since p < A + m, we easily verify that there
exists v > 0 such that
pfu2 < (1 v)lIuIIi,
I
for all u E Ho (a), and consequently

(6.22)
I
2 f G(u) < (1 - v)IIuIIHI +CIIuIIHi,

for all u E Ho(S2). Let (cw) E X, with II(^P,V))Ilx <1, and let u be the corre- El
sponding solution to (6.15)(6.17). Using the notation of the proof of Proposi-
tion 6.3.1, we deduce from (6.20) and (6.22) that, for all t

f G(cp)+Cf(t) 2 .
I
vf(t)<f(0) _2 (6.23)

Observe that I
f G( C(II (^v,VG)IIX + II(V, V,)IIX) < C(f(0) 2 + f(0) k );

and so, since f(0) < 1, there exists ME [1, oc) such that
El
f(0) _2 f G(cp) < vM f (0). (6.24) 1

Set k/2 = 1 + E (e > 0) and


I
_
9( x \_ C x 1+E x

v all a E (0, X) there exists 0 < x a, < Ya


for x > 0, and set X = min 6 < 0. For
such that
O(xa) + a = 6(y a ) + a = 0.
I
In addition, we have a < X a <a(1 + e)/e (see Figure 6.1).

1
1 86 The KleinGordon equation

-E2

Fig. 6.1

By (6.23) and (6.24), we have

6(1(t)) + Mf(0) > 0,

for all t E [0, T(cp, v')). Consequently, if we suppose that M f (0) <, we have
f (t) E 0
[ , xMf ( 0 )) U (yM f (0), oo),

and since f is continuous and f (0) < xM f( o ), we deduce that

l+e
f (t) xMf(0) < Mf ( 0 ),

for all t E [0,T(^p, 4 )). The result follows, with 5 = min{1, X/M} and K =
(1 + e)/EM.

Remark 6.3.4. If A = 0 (for example, if 52 = RN), we may apply Proposi-


tion 6.3.3 only if G(x) < 0 for 1st small. Then, we can replace the hypotheses
G(x) < ^x^ 2 by G(x) < / I x l 2 +E, for lxi small (E > 0), the proof being the same.

Remark 6.3.5. When 11 is bounded, we may assume that the conditions on


G involved in the statements of Proposition 6.3.1 and Remark 6.3.2 hold only
for lxi large. The proof is the same; see Proposition 6.4.4.
Blow-up in finite time 87

6.4. Blow-up in finite time

The main result of this section is the following.

Proposition 6.4.1. Suppose that there exists e > 0 such that

xg(x) ? (2 + e)G(x),
for all x E R. Then, if (<p, ) E X and E(cp, 0) < 0, we have T(cp, ,b) < oo.

The proof of Proposition 6.4.1 requires the following lemmas.

Lemma 6.4.2. Let T> 0 and let u E C([0,T],Ho(c)) fl C 1 ([0,T],L 2 (fl)) fl


C 2 ([0, T], H -1 (12)). Then the function t H f u(t) 2 belongs to C 2 ({0, T]) and we
have
d2
J u(t) 2 dx = 2 J u(t) 2 dx + 2(u(t), utt(t))H.,H- 1,

for alit E [0, T] .

Proof. Set
f (t)
n
_f
u(t) 2 , Vt E [0, Tj,

and assume first that u E C 2 ([0, T], Ho (S2)). Then

f"(t) = 2 +2 f J ut(t) 2 + 2 (u(t), utt(t))Ho,H -1,


if, ut(t) 2 ut(t)utt(t) =2

for all t E [0, T]. It follows that

f (t) = f ' ( 0 ) + 2f t
'

o
If
ut(s) 2 +(u(s),Utt(s))Ha,H- 1 } ds, (6.25)

for all t E [0,T]. By density, we then show that (6.25) still holds for u E
C([0,T], Ho (52)) fl C'([O, Tj, L 2 (5l)) fl C2 ([0,T], H - 1 (c));
hence the result. o

Lemma 6.4.3. Let T> 0 and let u E C([O, T], Ho (S2)) n C l ([0, T], L 2 (1l)) fl
C 2 ([0,T],H -1 (52)) be the solution of (6.16). Set

f(t) _f u(t) 2 ,

for all t E [0, T]. Then

f"(t) =2
fn ut(t) 2 2
fn f
IVu(t)1 2 2m n u(t) 2 +2
fn u(t)g(u(t)),

foralltE [0,T].
88 The Klein-Gordon equation

Proof. We apply Lemma 6.4.2 and (6.16). It follows that

f"(t) = 21 u(t)2 + 2(u(t), Au(t) - mu(t) + g(u(t)))Ho,H-^,


t

for all t E [0, T]. It suffices to see that, for all w E Ho (S2), we have

(w, OW)Ho H-1 = J IDwI2.

This is a consequence of Lemma 2.6.2 if tw E L 2 (S2), and follows by density


otherwise.

Proof of Proposition 6.4.1. Let (cp, ') E X be such that E(<p, V)) <0, and set

f(t) = f u(t)2,

for all t E [0,T(cp,i')). By Lemma 6.4.3, we have

f (t) =2
f ut (t) 2 -2 f IVu(t)1 2 - 2m
J u(t) 2 +2
f u(t)g(u(t))
>2
J u(t) 2 -2
J Vu(t)! 2 - 2m
J u(t) 2 + 2(2 + E)
J G(u(t)),
for all t E [0, T(, ii')). Applying Proposition 6.2.3, we deduce that

f " (t) > E { f IVu(t)1 2 + m f u(t) 2 I+(4+E) u t (t) 2 -2(2+e)E(cp,z/^), (6.26)


ll sz z fn
for all t E [0, T(, si)). For the sake of contradiction, assume that T(cp,ip) = oo.
In particular, we deduce from (6.26) that

f"(t) > -2(2 + E)E( W , 7p) > 0, dt > 0.

It follows that f (t) --> oo as t -- oo. On the other hand, applying (6.26) and the
Cauchy-Schwarz inequality, it follows that

f (t)f"(t) > (4+E) fn u(t)2 4 +E)Y ut(t)u(t)) 2 > ( 1 + 4) f ' (t) 2 ,


fn u(t)2 > ( o

and then
f(t) i <0, `dt>0.
We conclude as for Proposition 5.4.4 (inequality (5.21) and below).
Application to a model case 89

In the case in which ci is bounded, we can weaken the hypotheses of Propo-


sition 6.4.1.

Proposition 6.4.4. Suppose that 52 is bounded and that there exist K < oo ,
and e > 0 such that
xg(x) ? (2 + e)G(x),

for IxI > K. Set = min xg(x) and v = max G(x). Then, if (cp, 0) E X
- lxj<x lxl<K
satisfies

we have T(cp, 0) < oo.


(2+e)E(co, ) < IQI( - (2+e)v),
I
Proof. We argue as in Proposition 6.4.1, using

fn u(t)g(u(t)) = f1juj:5K}
u(t)g(u(t)) + 4uj>K} u(t)g(u(t))

> mjci + (2 + E) G(u(t))


f1 juj>Kj
>_ m^c + (2 + e)
J G(u(t)) - (2 + E) J <Kl
G(u(t))

J
> (m - (2 + e)v) IcI + (2 + e) G(u(t)), I

for all t E [0, T(cp, )).

Remark 6.4.5. For some comments on the hypotheses of Proposition 6.4.1


and 6.19, see Remark 5.4.7. In particular, if there exists xo >_ K such that
G(xo) > 0, and if we write (<, i) E X with C > 0 a.e. on S2, we have E(k(, k97) -4
-oo as k -> oo. In particular, for k sufficiently large, (kc, ki1) satisfies the
d
conditions of Propositions 6.16 or 6.19, and so T(kc, krl) < oo.

6.5. Application to a model, case

We choose g(x) =` aIxI' x, with a # 0, a > 0, and (N - 2)a < 2. We consider


(gip, Ali) E X and we denote by u the corresponding solution of (6.15)-(6.17).
Then we have the following results.
0
If a < 0, then T(,) = oo and (u, u t ) is bounded in Ho (S2) x L 2 (fl) (Re-
mark 6.3.2).
90 The Klein-Gordon equation

If a > 0, then T (cp, i1) = oo if 1l ( ') IIx is small enough (Proposition 6.3.3).
In addition, for some (p,) E X, we have T (cp, i/I) < oo (Remark 6.4.5), and in
that case II(u(t),u t (t))11x > 6(T(^,') -t) a (Theorem 4.3.4).

Notes. For more about local and global existence, in the framework of Chap-
ter 6, consult Browder [1] and Heinz and Von Wahl [2], and for more about
blow-up phenomenon, see Levine [2, 3], J. B. Keller [1], and Glassey [1, 3]. In
the general case (ci RN) the behaviour at infinity of solutions is well known
only in the dissipative case. See 9.4 and, for example, Haraux [1, 2]. In the con-
servative case, we only have some partial results, often limited to dimension 1.
See Brezis, Coron, and Nirenberg [1], Cabannes and Haraux [1, 2], Cazenave
and Haraux [2, 31, Cazenave, Haraux, Vazquez, and Weissler [1], Friedlander [1],
C. Keller [1], Payne and Sattinger [1], and Rabinowitz [1, 2]. On conservation
laws, see Serre [1].
For S2 = RN, there exist estimates of the same kind as in 7.3; see Brenner [1,
2], Ginibre and Velo [6, 9], and Marshall, Strauss, and Wainger [1]. These
estimates allow us, for the local existence, to replace in (6.8) the condition
(N - 2)a < 2 by the weaker condition (N - 2)a < 4. See Ginibre and Velo [8,
10] and Jorgens [1]. If condition (6.8) is not satisfied, we only know how to build
solutions in the case xg(x) < 0, but we do not know whether uniqueness holds.
See Strauss [1,2], as well as the very interesting numerical study of Strauss and
Vazquez [1].
Again for S2 = RN, we know how to investigate the dispersive properties
of the linear equation to show the global existence for small initial conditions,
with non-linearities that depend on derivatives of u (Klainerman and Ponce [1]).
If the order of the non-linearity at 0 is not sufficiently high, blow-up in finite
time may occur for arbitrarily small initial data. See Balabane [1, 2], Sideris [2,
3], Hanouzet and Joly [1], and John [1-3]. For some non-linearities, there exist
solutions of the form u(t, x) = eiwtcp(x). These solutions are called stationary
states. See, for example, Berestycki, Gallouet, and Kavian [1], Berestycki and
Lions [1], Berestycki, Lions, and Peletier [1], and Jones and Kiipper [1]. The
behaviour at infinity of solutions is rather well known in the repulsive case, in
which the solutions behave asymptotically as the solutions of the linear equation.
See Brenner [1, 2], Ginibre and Velo [8, 10], Morawetz and Strauss [1], Reed and
Simon [1], and Sideris [1]. In the attractive case, we know mainly how to study
the stability of certain stationary states. See Berestycki and Cazenave [1], Blan-
chard, Stubbe, and Vazquez [1], Cazenave [3], Cazenave and Lions [1], Grillakis,
Shatah, and Strauss [1], C. Keller [1], Payne and Sattinger [1], and Shatah and
Strauss [1].
7
The Schrodinger equation

7.1. Preliminaries

Throughout this chapter, we use the notation of 2.6.5 and 3.5.3. In particular,
X = H '(S2) = H 1 (SI,C) and Y = L 2 (cl) = L 2
(SI,C). The isometry groups
- -

generated by A and B are both denoted by (T(t)) tER


(see Lemma 3.5.12). Given
g : Ho (cl) -* H -1 (c), Lipschitz continuous on bounded subsets and
E Ho (S2),
we are looking for T> 0, and u a solution of the following problem:

UE C([0,T],Ha(cl) flC'([0,T],H-1(f)); (7.1)


iu t + Au + g(u) = 0, Vt E [0, T]; (7.2)
u(0) = `. (7.3)
Applying Lemma 4, and Corollary 4. 1.8, we obtain the following result.

Lemma 7.1.1. Let T > 0, cp E Ho (S1) and let u E


C([0, T], Ho (ii)). Then u
is a solution of (7.1)-(7.3), if and only if u is solution of

u(t) = T(t) cp + t - s)g(u(s)) ds,


T(t(7.4)
ifo
for all t E [0,T].

On the other hand, applying Proposition 4.1.9, we obtain a sharpened version


of Lemma 7.1.1, which will also be used in what follows.

Lemma 7.1.2. Let T> 0, cp E Ho (SI) and let u


E L 011 ((0,T), Ho (I )). Then
u is a solution of
u E L((O,T), Ho (Q)) n W 1 ((0, T), H- 1(Q)); (7.5)
iu t + Du + g(u) = 0, a.e. t E [0, T]; (7.6)
u(0) = y , (7.3)
if and only if u is solution of (7.4).

Remark 7.1.3. If (7.1)-(7.3) can be solved for T> 0, then in general it can
also be solved for T < 0 (see 4.4). If g satisfies certain symmetry properties,
92 The Schrodinger equation

it is es pecial ly clear. Indeed, if we suppose that g(u) = g(u), then v given by


v(t) = u(t) is a solution of (7.2) on [0, T] if and only if u is a solution of (7.2)
on [T, 0].

7.2. A general result

Assume that g : R + 4 ll is (globally) Lipschitz continuous and that g(0) = 0.


We extend g to the complex plane by setting

g(z) = Izlg(lzI), (7.7)

for all z E C, z # 0. We also define the function G by


IZI
G(z) = j g(s) ds, (7.8)
o

for all z E C. Then, g defines a mapping L 2 (S2) > L 2 (1l), which we still denote
by g *, by
g(v)(x) = g(v(x)),
(7.9)

for all v E L 2 (1l) and for almost all x E f. In addition, g is Lipschitz on L 2 (S2)
(see 6.1.2). Furthermore, if we set

V (v) = J G(v(x)) dx, Vv E L 2 (1l), (7.10)

then V E C 1 (L 2 (1Z),l18) and

V'(v) = g(v), t/v E L 2 (1l).


(7.11)

Finally, we define the functional E E C l (Ho (Sl),1[8) by

dx + V(v), (7.12)
E(v) = f I Vv 2

for all v E Ho (1). We have the following result.

Theorem 7.2.1. For all cp E L 2 (S2), there exists a unique function u E


C([0, oc), L 2 (Sl)) which is a solution to (7.4) for all T < oo; furthermore,

IIu(t)IIL2 = II(PIIL2, (7.13)

* This is a common abuse of notation, g denotes both a mapping C + C and


a mapping L2 --> L 2 . However, the context makes it clear which is used.
A general result 93

for all t >_ 0. If we assume further that E Ho (S2), then u E C([0, oo), Ho (S2)) n
C 1 ([O,00),H - '(1l)) and I
E(u(t)) = E(9), (7.14)
for all t >_ 0. If in addition Acp E L 2 (12), then Au E C([0, oo), L 2 (1Z)) and
u E C'([O,00),L Z (Sl))

The proof makes use of the following lemma. I

Lemma 7.2.2. Let T > 0 and let u E C([0,T],D(B)) fl C'([O,T],L 2 (52)).


Then the function t '-* Vu(t)II belongs to C 1 ([O,T]) and we have
;

I
dt II u(t)II i2 = 2(-^u(t), ut(t)),

for all t E [0,T]. I

Proof. The result is clear if u E C'([O, T], D(B)), and is obtained by density in
the general case (see, for example, the proof of Proposition 6.1.1).

Proof of Theorem 7.2.1. We proceed in seven steps.


Step 1. cp E L 2 (Il). The global existence in L 2 (cl) is a consequence of Theo-
rem 4.3.4, since K(M) is bounded.

Step 2. cp E D(B). If cp E D(B), the regularity is a consequence of Proposi-


tion 4.3.9. In that case, u is solution of (7.2) for all T < oo, and (7.2) is satisfied
in L 2 (1l).
Step 3. The conservation law (7.13). If (p E D(B), and taking the scalar
product (in L 2 (cl)) of (7.2) with u, we obtain

2 dt IIu(t) IIi2 = (u t (t), u(t)) = Iiou(t), u(t)) + (i g (u(t)), u(t)) = 0. (7.15)

Indeed, for all w E D(B), we have (see Lemma 2.6.2)

(i^u(t), u(t)) = Ref iL ww = _Ref iIVwl 2 = 0;


I

and (applying (7.8))

(ig(u(t)), u(t)) = Re
Jsz ig(w)iJ = Reig(IwI)Iwf = 0.
fn
Then (7.13) follows from (7.15). In the general case y E L 2 (52), we obtain (7.13)
by density, applying Proposition 4.3.7.
I

94 The Schrodinger equation

Step 4. The conservation law (7.14) for cp E D(B). Taking the scalar product
(in L 2 (Sl)) of (7.2) with u t , we obtain

0 = Ref iIutI 2 = (iut, ut) _ -(AU, u t ) - (g(u), ut),

for all t E [0, oo). Applying Lemma 7.2.2 and Corollary 6.1.7, we deduce that
(d/dt)E(u(t)) = 0; hence (7.14).

Step 5. If E Ho (S2), then u is weakly continuous as a map from [0, oo) to


Ho (S2) and we have
E(u(t)) < E(cp), (7.16)

for all t E [0, oo). Indeed, consider a sequence (W n ) n>o C D(B) such that
cp n --i cp in Ho (1) as n -+ oo, and let u n be the corresponding solutions of (7.2).
Let T > 0. Since g is Lipschitz continuous on L 2 (St), it follows from (7.13)
and (7.14) that u n, is bounded in L ((0, T), Ho (S2)). By Proposition 4.3.7, we
also have
u in C([0, T}, L 2 (Sl)) as n -, oo. (7.17)
u(t) in L 2 (52) as n --> oo, and
In particular, for all t E [0, T], we have u(t) u(t)
^[U n (t)[[H1 is bounded. Therefore,

u(t) -k u(t) in Ho (1) as n -+ oo. (7.18)

Applying Proposition 1.4.24, we deduce that u E L((0,T),Ho(fl)). Since


u E C([0, T], L 2 (f)), it follows that u is weakly continuous from [0, T] to Ho(Sl).
Applying (7.17), (7.18), and the weak lower semicontinuity of the norm in Ho (S2),
we deduce (7.16) from (7.14). We conclude observing that T is arbitrary.
Step 6. The conservation law (7.14) for cp E Ho (S2). For all s E [0, oo), we set

v(t) = u(s - t), `dt E [0,s].

On checking, it is immediate that v is solution of (7.2) on [0, s]. Then, we may


apply (7.16). It follows that

I E('P) E(u(s)). (7.19)

By putting together (7.18) and (7.19), we obtain (7.15).


' Step 7. If cp E Ho(11), we have u E C([0,00),Ho(Q)) n C'([0,co),H -1 (SI))
Indeed, since g is continuous on L 2 (Sl), we deduce from (7.14) that the function

t
t [u(t)11Hi belongs to C([0, oo)). Since u is weakly continuous from [0, 00)
to Ho (S2), we then have u E C([0, oo), Ho (S))); and so by Corollary 4.1.8, u E
C l ([0, oo), H -1 (I )). This completes the proof.
The linear Schrodinger equation in R N 95

Remark 7.2.3. Theorem 7.2.1 applies only if the non-linearity g is mild. In-
deed, if we consider g(u) = IuI "u (a >_ 0), we may use it only if a = 0. However,
Theorem 7.2.1 will be useful to prove a more general result for S2 = R N ( 7.4). It
is then necessary to specify the dispersive properties of the Schrodinger equation
in R N These properties are described in the next section.
.

7.3. The linear Schrodinger equation in R N

We suppose that SZ = R N , and we consider T > 0. We are going to apply


the results of 3.5.4 to give estimates for the solutions of the inhomogeneous
Schrodinger equation
iut+Au +f =0;
{
u(0)=gyp.

To do this, we define the operators 4?, 'I', and O t (for t E [0, T]) by

4? f (t) = 1 t T (t - s) f (s) ds, dt E [0, T];


IY f(s)
= J T(s - t) f (t) dt, Vs E [0, T];
s
T

Ot,f (s) = f T(s - v)f(o) do, ds E [0, T];


o

for all f E L 1 ((0,T),H -1 (RN)). We easily verify (see Lemma 4.1.5) that',
'I', and O t are continuous from L I ((0,T),H -1 (RN)) to C([0,T],H -1 (RN)), and
from L 1 ((0,T),H'(R' )) to C([0,T],H 1 (R"))

Definition 7.3.1. We say that a pair (q, r) of positive numbers is admissible


if the following properties hold:

(i)2<r<2N/(N-2) (2<r<ooifN=2,2<r<ooifN=1);

(ii)2/q = N (1/2 - 1/r).

Observe that if (q, r) is an admissible pair then we have q E (2, oo] (q E [4, oo]
if N = 1). The pair (oo, 2) is always admissible.

Remark 7.3.2. If (q, r) is an admissible pair, then we have in particular


so
H 1 (RN) . Lr(RN) with dense embedding and L''(RN) H -1 (RN). It fol-
lows that C([0,T],H l (1RN)) LQ((0,T),L''(RN)) and L9'((O,T),Lr (RN)) y
L 1 ((0,T),H -1 (R ')). In particular, the operators 4D, 'Y, and O t are continuous
from LQ'((0,T),L" (RN)) to C([0,T],H-1(RN)).
96 The Schrodinger equation

Remark 7.3.3. If (q, r) is an admissible pair, and taking suitable test func-
tions, we easily verify that for all u E L 9 ((0,T),Lr'(R N )), we have

f
IIiIIL9((o,T),L') =supRe uapdxdt1;
JO RN J
L 9 ((0,T), Lr (RN)), IkPIIL9'((o,T),Lr') 1.
,

w E

By truncation and regularization, we also have

( f T f
IlUIIL4((o,T),Lr) = sup { Re u-pdxdt1;
J0 JR^
^P E L 9' (( 0 ,T),L r' (R N )) nC([O,T],Hl(RN)), lfplILa'((O,T),Lt') 1 }.

The main result of this section is the following.

Proposition 7.3.4. Let (y, p) be an admissible pair according to Defini-


tion 7.3.1 and let f E L-Y'((0,T),LP (RN)). Then 4?f E C([0,T], L 2 (R")).
In addition, for any admissible pair (q, r), we have Pf E L 4 ((0,T),Lr'(R N ))
Finally, there exists a constant C, depending only on y and q, such that

II'Pf II L11((o,T),L) < C(7, q) II f II L-' ((o,T),LP' )' (7.20)

for all f E Lry ((O,T), LP (II^ 1`')).

Proof. The proof proceeds in six steps.

Step 1. For all admissible pairs (q,r), the operator 4' E C(Lq'((0,T),LT(RN)),
L 9 ((0,T),L''(R N ))), with norm depending only on q. By density, we need only
consider the case in which f E C([0,T],Lr'(R N )). In this case, it follows from
Proposition 3.5.14 that (Pf E C([O,T], L''(R')), and that for t E [0, T] we have

II 4 )f(t)IILr t It sI -- # f ) IIf(s)I1L, , ds < fT It sI Q^ 11f (s)IILr , ds.


< fo
The classical estimates on Riesz' potentials (see Stein [1], Thm. 1, p. 119) then
give
iii II L9((o,T),Lr) < CII f IIL9'((o,T),L-')'

where C depends only on q; hence the result.


The linear Schrodinger equation in R' 97 1

Step 2. Similarly, T and O t are continuous from L 9 '((0,T),L"(R N )) to


L 9 ((0,T),Lr(R N )) with norms depending only on q.

Step 3. The operator E G(Lq'((0,T),L'(l[8'v )),C([O,T],L Z (R N ))), for all


admissible pairs (q, r), with norm depending only on q. By density, we need
only consider the case in which f E C([0,T],L''(R N )), and using a regularizing
sequence, we may assume further that f E C([0,T],H l (IR N )). In that case, we
have f E C([0,T], H l (R")). Applying Corollary 3.2.6, and denoting by (, )
the scalar product in L 2 (R N ), we obtain, for all t E [0, T],

IIf(t)IIL2=( f t T(t
= I
0
s)f(s)ds,
fo T(t v)f(a)de)

' f t (T(t s) f (s),T(t a) f (Q)) da ds

f
tt

=f f(f(s),T(s a)f(o)) dads = (f (s), Ot,f (s)) ds.

Applying Holder's inequality, first in space and then in time, and using Step 2,
it follows that

IIr(t)IIL2 <-IIfIIL9'((O,T),Lr')Ilot,IIIL9O,T),Lr) <-C(q)IIfIIL4'((o,T),L-');


I
hence the result, since t is arbitrary.
Step 4. Similarly, 'I' and O t are continuous from ((0, T), U' (RN)) to C( [0, T],
L 2 (RN)), with norms depending only on q.

Step 5. The operator c E (L 1 ((0, T), L 2 (RN)), L 9 ((O,T), L''(RN))), for all ad-
missible pairs (q, r), with norm depending only on q. Let cp E C([0, T], Hl (RN))
nC([O,T],L , (RN)) be such that IIcpjIL9(lo,T),L-) < 1. In particular, (f E
C([O,TJ, L 2 (R n')), and (by Step 4) IIW,vIIL((o,T),L2) < C(q). We have

fT r T

0
Re
J
R
N c i dx dt =
JT(f (t), co(t)) dt
ft
_ (T(t s) f (s), cp(t)) ds dt
J
_ Jo J
0 o
,T

I
ft
(f(s),T(s t)^p(t)) ds dt

= JOJ
T T

(f (s), T(s t)^p(t)) dt ds


T s

= f (f (s), `T a(s)) ds.


98 The Schrodinger equation

Applying the Cauchy-Schwarz inequality, and then Step 4, it follows that


fT
rRe 4)fipdxdt < IIfIIL1((O,T),L2)II'c IIL((o,T),L2)
0 fRI
< C(q)II f II L1((O,T),L2) II (PII L9'((O,T),L. )
< C(q) II f II L1((o,T),LZ).
Then it suffices to apply Remark 7.3.3.
Step 6. Conclusion. Let ('y, p) be an admissible pair. By Steps 1 and 3, 'D is
a continuous operator Lry'((O,T),LP (R N )) -> C([0,T],L 2 (R N )) and L 7 '((0,T),
L (R N )) --> Lry((0,T),LP(R N )). Let (q,r) be an admissible pair such that
2 < r < p, and let 0 E [0, 1] be such that
1=0+ 1-6 1_0+ 1-0
and
r p 2 q ry o0
Applying Holder's inequality in space and then in time, we obtain

II^f IIL4((0,T),Lr) <_ II4'f II L-,((O,T),LP) II ^f II L4(O,T),L2) < CII f II L7'((O,T),LP')'

where C depends only on -y and q. Then, 'I' is continuous Lry'((0,T), LP (R N )) -^


L 9 ((0,T), Lr'(R N )). Now let (q, r) be an admissible pair such that p < r, and
let p E (0, 1) be such that

1 P+ 1 - and 1 =u+ 1-,u


ry' 1 q' p' 2 r'
By Steps 1 and 5, 4 is a continuous operator L 9 '((0,T),L' (R n')) --> L 4 ((0,T),
LT(RN)) and L 1 ((0,T),L 2 (R N )) -* L 4 ((0,T),Lr'(R N )). By interpolation (see
Bergh and Lofstrom [1], Thm. 5.1.2 p. 107), 1 is a continuous operator L((0, T),
L 6 (IR N )) -> LQ((0,T),Lr'(RN)), for all (a,6) such that
1=0+ 1-0 and 1 - B + 1-0
o 1 q' 6 2 r' '
with 0 E (0,1). Choosing 0 = p, we deduce that 4) E L(L 1 '((0,T), LP (R N )),
L 9 ((0, T), Lr(R' ))), which completes the proof.

Corollary 7.3.5. Let (y, p) be an admissible pair. Let f E C([0,T], L 2 (R' ))


nW 1,7' (( 0 ,T),L P (R N )). Then F1 E C([0, T], H2(RN)) flCl([0,T],L2(RN))

Proof. We have in particular f E W l " l ([0,T],H -1 (1[x')), and so (applying


Proposition 4.1.6) Ff E C([0, T], H'(R n )) f1 C 1 ([0, T], H -i (R' )) and

d Ibf = iL4)f + f,
(7.21)
The linear Schrodinger equation in RN 99

for all t E [0, T]. In addition, we have

^f(t) = f t T(s)f(t s)ds,


0

and so

dt"Ds(t) = T(t)f(0) + t T(s)f'(t s) ds = 4) f (t).


fo
This relation is clear if f E C 1 ([O,T],L''(R' )) C C l ([O,T],H -1 (R 1")), and is
obtained by density in the general case. Applying Proposition 7.3.4, it follows
that 'f E C 1 ([O,T],L 2 (RN)). By (7.21), we have A-Df E C([O,T],LZ(]f N));
and so 4? f E C([O,T), H 2 (IRN)).

In the spirit of Proposition 7.3.4, we give the following result.

Proposition 7.3.6. Let cp E L 2 (1RN). For all admissible pairs (q,r) we have
T()cp E LQ(1R, L' (RN)). Moreover, there exists a constant C, depending only
on q, such that
II7()coIIL9(R,L*) < C(q)IIpIIL2,
for all cp E L 2 (RN)

Proof. The proof is similar to that of Proposition 7.3.4, and so we only sketch
it briefly. Set

+00 +00
A1 (t) _ f T(t s)f(s)ds, rf = f T(t)f(t)dt.

We show that (see Step 1 of the proof of Proposition 7.3.4)

IIA1IIL9(R,L.) <C(q)IIfIILq' RLr' , ( )

Next we deduce (see Step 3 of the proof of Proposition 7.3.4) that

IIrfIIL2 < IIfIIL9'(R,L*' ) -

We conclude (as in Step 5 of the proof of Proposition 7.3.4) by noting that

I
+00 +00

f (T(t), (t)) = I ( f T(t)(t)) < C(q)IIIIL2IIIIL9'(^Lr'),

and applying Remark 7.3.3. M


100 The Schrodinger equation

7.4. The non - linear Schrodinger equation in RI'': local existence


We extend the result of Theorem 7.2.1 to more general non-linearities. From
now on, g is a function R+ --* R such that g(0) = 0 and such that there exist
K < oo and a > 0 with (N 2)a < 4, such that

Ig(y) g(x)I < K(l + Ixi" + IyI a )Iy x1,


for all x, y E R+ . We extend g to the complex plane by formula (7.8) and
we define the function G : C * R + by formula (7.9). g defines a mapping
Lj + 1 Lj , by formula (7.10). We verify (see 6.1.2) that g is continuous from
Hl(RN) to H l (lfB N ), and that V defined by formula (7.11) is continuous on
H 1 (R N ). Finally, we define the functional E on H 1 (R') by formula (7.12). The
main result of this section is the following theorem.

Theorem 7.4.1. There exists a function T : H 1 (R N ) --, (0, 00] that satisfies
the following properties. For all cp E H' (RN), there exists a function u E
C([0,T(cp)), H 1 (R N )) which is the unique solution of (7.1)(7.3) for all T <
T(cp). In addition,

(i) ifT(cp) < oo, then lju(t)IlH1 ; 00;


(ii) 11u(t)IIL2 = IIoIIL2, for all t E [0, T('));
(iii) E(u(t)) = E(cp), for all t E [0, T(cp));

(iv) if cc E H 2 (TR"), then u E C([0,T(cp)),H 2 (R N )) n C 1 ([0,T(cp)),L 2 (R N )).


Finally, the function T : H'(R") --> (0, oo] is lower semicontinuous, and u
depends continuously on cp in the following sense: if cp,, --+ in H l (RN), and
if u,,, and u denote the corresponding solutions of (7.1)(7.3), then u,,, > u in
C([0,T],H'(R N )),for allT <T(cp).

Remark. Property (iv) above is a regularity result. As well as being inter-


esting in its own right, this result will be indirectly useful in 7.6 (blow-up in
finite time) and 7.7 (behaviour at infinity) to establish the fundamental iden-
tity (7.58). This property has been included in Theorem 7.4.1 since its proof
requires the approximation argument which we use for the local existence in
Hl(R N ). Doing this, the proof of Theorem 7.4.1 turns out to be much more
complicated. Consequently, at the first reading, the reader should omit Lem-
mas 7.4.7, 7.4.9, and 7.4.10, Corollary 7.4.8, Steps 3 and 5 of the proof of Proposi-
tion 7.4.12, and Step 3 of the proof of Corollary 7.4.13 (identified by an asterisk).
These results, whose aim is to prove property (iv), are somewhat technical and
may obscure the proof of the local existence in H'(R N )
In fact, the proof of the local existence itself is not easy. We sketch it briefly
and point out the difficulties.
The non-linear Schmdinger equation in ][fi n': local existence 101

In 7.2, it appears clearly that the methods developed so far do not allow
to solve (7.1)(7.3) if g is superlinear (see Remark 7.2.3). In R n', we can get
round this difficulty by using the inequalities proved in 7.3. Recall, however,
that these inequalities are specific to R N .
The idea of the proof of Theorem 7.4.1 is the following. We approximate g in
a convenient sense by a sequence (g n,),, >o of globally Lipschitz non-linearities,
which satisfy the same assumptions as g uniformly with respect to m. In a
preliminary section (7.4.1), we apply the estimates of 7.3 to establish various
inequalities for g i and for g. In general, they derive from Proposition 7.3.4 and
Holder's inequality. We apply these inequalities in 7.4.2. First, (Lemma 7.4.11),
we obtain immediately a uniqueness result (note that, at this stage, we do not
use approximation but only Proposition 7.3.4 and Holder's inequality). Next
(Proposition 7.4.12) applying Theorem 7.2.1 to problem (7.1)(7.3) with g re-
placed by g.,,,,, we build solutions u,,,, of the approximate problems. On a small
time interval, we estimate these solutions Urn , uniformly with respect to m, us-
ing mainly conservation of energy. These estimates allow us to pass to the limit
as m > oo, and to obtain a local solution u. Solving the backward problem,
we demonstrate that energy is conserved (Corollary 7.4.13). Finally, we com-
plete the proof showing alternative (i) and continuous dependence. To establish
property (iv), we may suppose that cp E H 2 (R') and deduce estimates of U rn
in H 2 (R N ) which are independent of m. We then pass to the limit in these

I
estimates as in > oo.

7.4.1. Some estimates

The following notation will be useful in what follows. We define k E C(C, C) by

=!
g(z), if Izi < 1;
k(z)
l zg(1), if IzI > 1.

for all m E N, we define g,,, E C(C, C) and h,n E C(C, C) by

9 m (Z)
1 9(z),
m9(m),
if Iz) < m;
if Iz) > m;
I
hr,(z) = 9rn(z) - k(z) j
(We have in particular g l = k.) Also, let
900=9;
h00=gk.
Form E N U {oo}, we define G m E C(C, R) by
IZI
Gm(z)
f
o
9,,,.(s) ds.
102 The Schrodinger equation

Observe that for all m E N, 9 m is globally Lipschitz continuous. For all u E


H l (R N ), we set
Vm(u) = J Gm(Iul);

E () = 2 f
m N IVul 2 dx +V,n(u).
We set V^ = V and E.. = E. Finally, for all T> 0, u E L-((0,T),H 1 (R N ))
and mENU{oo},weset

Qm(u)(t) = f T(t s)9m(u(s)) ds;


o
xmlu)(t) = f T(t s)h m (u(s)) ds;
o

IC(u)(t) t T(t s)k(u(s)) ds;


= fo
for alit [0,T], and we let H =71, 9.=G.
By possibly modifying the value of K, we readily verify the following inequal-
ities:

I9m(z2) gm(zi) K( 1 + Iz2I" + Izi1 a I)Iz2 z1I; (7.22)


+1. (7.23)
I9m(z) g(z) < KIzl
Ihm(z2) - hm(zi) <- K(Iz2I" + IziIl)Iz2 - ziI; (7.24)
Ik(z2) k(zi) < K^z2 z1I; (7.25)
for all m E N U {oo} and all z, zi, Z2 E C. In what follows we set r = a + 2 and
a = 4(a + 2)/(Na), so that (a,T) is admissible.

Lemma 7.4.2. Let M < oc. There exists C(M) < oo and v > 0 such that

II k(v) k(u))1 L2 < C(M)I )v uIIL^; (7.26)


II hm(v) hm(u)II L,' < C(M)II v UIILT; (7.27)
II9m(u) 9(u)IIL*' < C(M)m "; (7.28)

for all u,v E H I (R N ) such that (IuIIHl < M and IIvIIHl < M and for all m E N

Proof. (7.26) is an immediate consequence of (7.25). (7.27) is a consequence


of (7.24), Holder's inequality, and of the embedding H l (R N ) y LT (R N ). Fi-
nally, we observe that

0, if IzI < m;
I9m(z) - 9(z)^ < { +i
KIzI if IzI > m.
The non-linear Schrodinger equation in RN: local existence 103

Let r > T be such that (N - 2)r < 2N. Then we have

IR N 19m(U) - g(u)IT < K 412:m IuIT < Km -(r -T)

flul>m
Iulr

< CKm - (r - T) II uII Hl

From this, we deduce (7.28).

Lemma 7.4.3. Let M Goo. There exist C(M) < oo and > 0 such that

IVm(v) - Vm(u)I < C(M) (IIv - uIIL2 + IIv - uliy, 2 ) ; ( 7.29)


IVm(u) - V (u)I < C(M)m -1`; (7.30)

for all u,v E Hl(RN) such that IIUIIHI M and IIvIIHI <_ M and all m E
N U {oo}.

Proof. We observe that, by (7.23),

Icm(z2) Gm(zi)I < K(Iz2 + Izil + Iz 2 I + 1 + (zila +1 )Iz2 z 1 I;


and so, by Holder's inequality,

Ivm(V)-Vm(u)I <- C( IIuIIL2+IIVIIL2) IIV - UIIL2+C(IIuIIL 1 +IIvjj L 1 )IIv- ulILr.


On the other hand, for all u E H 1 (RN), we have (Theorem 1.3.4)

IIuIIL ,
< IItII'HI IiuII7, with a = N (2 - T) = a; (7.31)

and hence (7.29). (7.30) is proved as (7.28), observing that

j 0, Ti
f IzI < m ;
I
l
Cm(z) - G(z)I < KIzI , if IzI > m;

for all m E N.

Lemma 7.4.4. Let M < oo and let (q, r) be an admissible pair. There exist
C(M, q) < oo and v > 0 such that

II K(v) - K(u)II L9((o,T),Lr.) < C(M, q)T II v - UIIL-((o,T),L2); (7.32)


IINm(v) - xm(u)II L9((o,T),L.) < C(M, q)T' -2 /II v - uII Lfl((o,T),L.); (7.33)
IIGm(u) - c(u)II L9((o,T),L-) < C(M, q)T m - ' ; (7.34)
104 The Schrodinger equation

for all T > 0, all m E N U {oo} and all u,v E L 00 ((0,T),H l (R N )) such that
IIUIILo((O,T),H1) < M and IIvllL((O,T),H1) < M.

Proof. We apply (7.20), next Holder's inequality on (0, T), and finally inequal-
ities (7.26), (7.27), or (7.28).

Lemma 7.4.5. Let T > 0 and let u E L((0,T),H 1 (R N )) n W l "((0,T),


H - ' (RN)). Set

K = max {IIuIILoo((O,T),H 1 ), I u II L((O,T),H -1


'
)}-

Then u E C([0,T],L 2 (R N )) and

IIu(t) u(s) IIL2 < 2KIt _812 ,

for all t, s E [0, T].

Proof. We have u E C([0,T],H -1 (R N )) (Corollary 1.4.36), and so U: [0,T] -->


H'(1) is weakly continuous. In addition, we have, for all t, s E [0, T],

u(t) u(s) IIL 2 = ( u(t) u(s) , u(t) u(s))H-i Hi < 2KII u(t) u(s) IIH -1

< 2K f 9
t IIu'(a)IIH -1 dQ < 2K 2 It sI,

C
hence the result.

Lemma 7.4.6. Let T > 0, u E LOO((0,T), HI (lI N)) n C([0,T], L 2 (R N )). Let
r >2 be such that (N 2)r < 2N. Then u E C([0,T], L""(lR' )) and

II'IIL-((0,T),L*) <_ IIUIIL-I(O,T)H1)IIUIIL( (O,T),L2)), (7.35)

where C does not depend on u.

Proof. Applying (7.31) to w = u(t + h) u(t), we obtain the continuity and


then, letting w = u(t), we obtain (7.35).

Similarly, we show the following lemma.

Lemma 7.4.7 (*) Let T> 0, u E L((0,T),H 2 (lRN)) n C([0,T],L 2 (R N ))


Let r >2 be such that (N 4)r < 2N. Then u E C([O,T], L'(RN))

Corollary 7.4.8 (*). If u E L((0,T),H 2 (R N )) n C([0,T], L 2 (R N )), then


g(u) E C([ 0 ,T] , L2(RN))
The non-linear Schrodinger equation in R N : local existence 105

Proof. We have in particular u E C([0, T], L 2 (a+l)(R"')) n C([0,T], L 2 (R N )),


and the result follows readily.

Lemma 7.4.9 (*). For all M, there exists C(M) such that I

Il9m(u)IIL2 <_ 2IIAullL2 + C(M),

for all m E N and all u E HI(RN) such that IIuUIx1 <M.

Proof. It suffices to estimate h.,,,. If N < 2, we have H 1 (RN) L 2 (c, + 1 )(R ),


and then I[h m (u)II L 2 < C(M). If N >_ 3, we may suppose that a >_ 2/(N 2),
so that 2(a + 1) > 2N/(N 2). In that case, we have (Theorem 1.3.4)

Ilhm(u)IIL2 < Cc-4- a+l)


< cHDu1IL(+1)Ilull(1 (ck+1)

< Cll, UlIL2a+1) M( 1 a)(+ 1 )


'

with a/N = 1/2 1/N 1/(2(c + 1)). In particular, we have a(a + 1) < 1;
1
hence the result.

Lemma 7.4.10 (*) For all M, there exists C(M) with the following prop-
erties. For all T > 0, all m E N U {oo} and all u E LO((0,T),H l (R N )) n
YT' 1 ' 0 o((0,T),L 2 (R N )) nW l,- ((0,T),LT(R rs')) such that IIUIILOO((o,T),H 1 ) < M,
we have k(u) E W 1 'o((0,T),L 2 (R")) and h m (u) E W 1 ' 0"((0,T),LT'(R )); in
addition,

Ilk(u)'IIL1((o,T),L2) <C(M)TIIu IILo((o,T),L2); (7.36)

II hm(u) ' [I L'((o,T),L 1 ') C(M)T 1_2/ II IILL((O,T),Lr). (7.37)

Proof. By Theorem 1.4.40, u is Lipschitz [0, T] -4 L 2 (R N ); and then so is k(u),


by (7.26). Applying Theorem 1.4.40, we thus obtain k(u)' E L((0, T), L 2 (Rn'))
and
Ilk(u)'IILoo((O,T),L2) <C(M)Ilu [ILO((o,T),L2);
hence (7.36). Applying (7.27) we show that h ,(u) E Wyo'((O,T),LT'(Rn')) and
7

that
Ilhm(U)'IIL((o,T),L) ^(M)IIu^IIL((o,T),LT),
(7.37) follows. l
1 106 The Schrodinger equation

7.4.2. Proof of Theorem 7.4.1


We begin with a uniqueness result.

Lemma 7.4.11. Let T > 0 and E H 1 (RN). Let u,v E L 00 ((0,T), H 1 (RN))
be two solutions of (7.4). Then u = v.

Proof. Observe first that by Lemmas 7.1.2 and 7.4.5, u,v E C([O,T],L 2 (RN)).
Let 8 = sup{t E [0,T]; u = v on [0,t]}. If 8 = T, we have u =von [0,T].
If 8 < T, we may suppose 8 = 0, using the transformation t -* t - 8. Set
M = max{IIullL- ((o,T),xl), IIVIILOC((o,T),Hl) }, and let (q,r) be an admissible pair.
We have
u-v=i(1Cu-)Cv)+i(7iu-
and applying (7.32)-(7.33), we obtain, for all t E (0, T]

Ilu VIILq((O,t),L.) <C( M , q)( tllu vlILo((o,t),L2) +t 1-21 IIu


VIIL-((o,t),LT)).

Choosing successively (q, r) = ( oo, 2) and (q, r) = (Q, r), making the sum and
then taking t sufficiently small, we obtain

Ilu VII L((o,t),LZ) + In VIIL. ((O,t),L , ) = 0 ;


We therefore have the desired contradiction with the hypothesis 0 = 0.

Proposition 7.4.12. For all M, there exists TM > 0 with the following prop-
erties. For all cp E H 1 (RN) such that II pIIH' <_ M, there exists a solution
u E L 00 ((0,TM);H i (RN)) f1 W" ((0,TM);H - i(TI ')) of (7.4). This solution
satisfies:

(i) IIUIIL-((o,TM );Hi) <- 2M;


(ii) Ilu(t)IIL 2 = IICPIIL2, for alit E [0,TM];
(iii) E(u(t)) < E(cp), for alit E [0,TM];

(iv) if cc E H 2 (1R"'), then u E C([0, TM], H 2 (R n )) fl C' ([0, TM], L 2 (RN)).

In addition, if u and v are the solutions corresponding to the initial data cc and
we have:
(v) In VIILOO((0,Tna);L2) < KII CP Y'IIL -,
where K is a constant which does not depend on M.

Proof. We proceed in six steps. We consider M > 0 and cc E H 1 (lRN) such


that IIWPIIHI << M.
The non-linear Schrodinger equation in RN: local existence 107

Step 1. Construction of approximate solutions. For all m E N, the function g,,,


is globally Lipschitz continuous. Thus we may apply Theorem 7.2.1. There exists
a unique function U rn E C([0, oo), H 1 (RN)) fl C'([0, oo), H -1 (R"')), a solution
of the following equation

um(t) = T (t)v + igm(um)(t), (7.38)

for all t > 0. In addition, we have

I[U rn (t)L2 = IIc IIL 2 ; (7.39)


D'm(um(t)) = Em(co); (7.40)

for all t > 0.


Step 2. A priori estimates on the solutions. Let

Tm. = sup{t > 01 H U IIC([o,t],H')<


_ 2M}.

Our intention is to show that there exists TM, depending only on M, such that
T,,,, > TM. To see this, assume that T. < oo. In that case, we have

IIum(Tm)IIH1 = 2M. (7.41)

But, by (7.39) and (7.40), we have

IIum(t)I^HI = II^PI^ 2r i +2 (V.(um(t)) - Vm(P)) (7.42)

On the other hand, applying Lemma 7.1.1, (7.26), (7.27), and Sobolev's embed-
dings, we obtain, for all t E [0, Tm ],

IIumIIL((0,t),H -1 ) ^ II bUmf[Loo((O,t),Jf_ 1 ) + Itk( Urn) IIL((0,t),H -1 )


+ tIhrn(Um)IILoo((0,t),H_ 1 )
^ 1[Um 1IL (( 0 ,t),H l ) + I k(um) 1 IL o ((0 ,t),L2 )
+ C' II h m (u m) 1 ILoo((o,t),LT')
2M + 2MC(M) + 2MC"C(M).

By Lemma 7.4.5 and (7.29), it follows that there exists D(M), depending only
on M, such that

2 IVm(um(t)) - Vr(cp)I < D(M)(t 1i2 +t 1 / 2-1 i),

for all t E [0, T,,,]. Putting this into (7.42) and applying (7.41), we obtain

4M 2 < M 2 + D(M)( T,m 2 +


108 The Schrodinger equation

It follows readily that there exists TM > 0 depending only on M, such that
Tm > TM , and so
sup{IIu,,,(t)IIH1; t E [0,TM]} < 2M, (7.43)

which is the desired estimate.

Step 3 (*). The case cp E H 2 (RJ"). In that case, we know (Theorem 7.2.1) that
u rn E C([0,00),H 2 (R N )) nC 1 ([0,00),L 2 (R N )). Since g,,,, is Lipschitz, we have
in particular g m,(u,,,,) E W 1, O0 ((0,TM),L 2 (R N )) and, for all t E [0,TM]:

u;,, (t) = T (t) (iAcp + ig rn (cp)) + i f T(t - s)g.,,,,(u


0
rn )'(s) ds. (7.44)

From Propositions 7.3.4 and 7.3.6, it follows that u,,,,, E W 1, ((O,TM);L T (R N ))


Then, we write (7.44) as

u' (t) = T(t)(iAcp + ig, n,(p)) + i J t T(t - s)k(um )'(s) ds


(7.45)
+i
I

t T(t - s)h m (u,)'(s) ds.

By Lemma 7.4.9, App + g(p) is bounded in L 2 (R N ). We estimate the first


term on the right-hand side of (7.45) by Proposition 7.3.6 and the integrals as
a consequence of Proposition 7.3.4 and Lemma 7.4.10. It follows that, for all
T<TM,

IIumIIL((O,T),L2)+ IIumIIL((,T),LT)
C( p) + K(M)(T + T 1-2k )(IIumIIL00(c,T),L2) + IIumhIL((0,T),LT))
,
,

where K(M) depends only on M and C depends on II'PIIH2 Choosing if neces-


sary TM smaller (but depending only on M), we may assume that K(M)(TM +
Z '1-2
M )-
< 1/2. We then obtain

(7.46)
IIUmhhL0o((,TM);L2) + IfmblL'(c,7M);Lr) < 2C(cv).

According to the equation satisfied by U rn , we also obtain

/urhhL-((,T,,);L2) ^ II u'M II L-((,TM);L 2 ) + II gn(um.)I )L-((,TM);L2).

Next, we apply (7.46) and Lemma 7.4.9. It follows that

(7.47)
IIoLrhIL00((o,TM);L2) < D(cp),

where D depends only on II cPII H 2 .


The non-linear Schrodinger equation in R N : local existence 109

Step 4. Convergence of the sequence (u,,,,),,,,> o . Let in and p be two integers


We write that, for all t > 0,

u rn (t) - u(t) =i ( IL ( u m)(t) - K ( UP)( t )) + i (flr( U, )( t ) - llm(uP)(t))


+i( 7-tm(uP)(t) xP(uP)(t))

Apply Lemma 7.4.4 successively with (q, r) _ (oo, 2) and (q, r) = (a, T). We
deduce that there exists C(M), depending only on M, such that

hi u m -up lIL((o,TM);L 2 )+chum uphIL-((O,TNM);Lr) <C(M)(1+TM)(m "+p


+ C(M) (TM + TM 21 ) (hkum UPI[L((O,TM);L 2 ) + (jum UPII L((O,TM);L-)).

Taking possibly TM smaller (but depending only on M as above) we may suppose


that
C(M) (TM +T 2k) <1

Therefore (u.m ) m,, o is a Cauchy sequence in LO((O,TM); L 2 (RN)) n L((0,TM);


LT (RN)) and there exists u E C([0, TM ], L 2 (RN)) n L((O,TM ); LT (R N )) such
that

um --j u in C([0, TM], L 2 (R N )) n L(0, TM); L T (R N )) as m oo. (7.48)

Furthermore, (7.43) and Corollary 1.4.24 imply that u E L 00 ((0,TM);H 1 (R N ))


and estimate (i) follows. Invoking Lemma 7.4.4, we may immediately pass to
the limit in equation (7.38) and conclude that u is a solution of (7.4) on [0, TM].
The conservation law (ii) is a consequence of (7.39) and (7.48). Lemmas 7.4.3
and (7.48) prove that

V,l.(un (t)) --->V(u(t)) as m * oo, (7.49)

for all t E [0,TM]. Next let t E [0,TM]. Since u,(t) k u(t) in H l (R N ) as


m > oo, we also have

J N IVu(t)^ dx < liiminf JR N ^Vu m (t)I dx. (7.50)

Applying (7.49) and (7.50), we obtain (iii).


Step 5 (*). Returning to the case H2(RN). By ( 7.47), u m is bounded
in L 00 ((0,TM );H 2 (RN)), and so (Corollary 1.4.24) u E L((0,T,^,r );HZ(II2n')).
In particular (Corollary 7.4.8), g(u) E C([O,TM ], L 2 (R" )).
Applying estimate (7.46) and Corollary 1.4.42, we obtain in addition that
U E W100((0, TM); L 2 (R N )) n W 10 (( 0 ,Tm); L T (R N )). Applying Lemma 7.4.10,
r 110 The Schrodinger equation

we have k(u) E WC 3 ((O,T,yI);L 2 (1I n')) and h(u) E WN 0 "((0,Tm);L T (R N )).


Finally, Corollary 7.3.5 implies u E C l ([O,TM], L 2 (R N )) fl C([O,TM], H 2 (R N )).

Step 6. The continuous dependence with respect to cp. Let cp and be such
that Ik IIHI <M and II^'IIHI <_ M. Let u and v be the corresponding solutions
of (7.4). We have

uv=T(.)(cpV))+i(Ku)Cv)+i(liu

on [0, TM]. We estimate the first term on the right-hand side with Proposi-
tion 7.3.6 and the following two terms by Lemma 7.4.4, by taking successively
(q, r) = (oo, 2) and (q, r) = (v, ,r). We deduce that there exist K < oo and
C(M), depending only on M, such that

IIu VII L((o,TM);L 2 ) + IIu VIIL((O,TM);L*) < KIIp iljL 2


+c(M) ( TM +TM 2k ) (IIu VII L((o,TNM);L2) + IIu VII L((O,T.);L')).

Taking TM smaller if necessary (but depending only on M), we may assume that

T2)
C(M) (TM +TM

We deduce (v).

Corollary 7.4.13. Let T> 0, cp E H l (R N ), and let u E L 00 ((0,T),H l (lR t '))


be a solution of (7.4). Then u E C([0,T],H l (R N )) f1 C 1 ([0,T],H' 1 (R N )), and
we have:
(i) IIu(t)IIL2 = IIcpll L 2, for all t E [0,T];
(ii) E(u(t)) = E(cp), for alit E [0,T].
In addition, if cc E H 2 (R"), then u E C([0,T], H 2 (R N )) f1 C 1 ([0, T], L 2 (R N )).

Proof. We proceed in three steps.


Step 1. The conservation laws (i) and (ii). Let u be as in the statement of
the corollary. Observe that u : [0, T] + H'(R N ) is weakly continuous. Set
M = (IUIILc((o,T),H1) and

6 = sup{t E [0, T]; (i) and (ii) hold on [0,t]}.

We want to show that 6 = T. To do this, we argue by contradiction, assuming


that 6 < T. Then, we may consider only the case 0 = 0, since otherwise we use
the transform t * t 6. In particular, there exists 6 E (0,TM), such that

lIIa(6)lIL IIcPIILzI + IE(u(6)) E(cp)) > 0,


2 (7.51)
The non-linear Schrodinger equation in RN: local existence 111

where TM is given by Proposition 7.4.12. Denote by v the solution of (7.4) given


by Proposition 7.4.12. Since 6 E (0, TM), we have n = v on [0, b] (Lemma 7.4.11).
In particular, we have II u( 6 )IIL 2 = II'IIL2 and E(u(b)) < E(cp) ((ii) and (iii) of
Proposition 7.4.12); and so (7.51) is equivalent to

E(u(b)) < E(p). (7.52)

Set w = u(t - 6) for all t E [0, 6]. Since u satisfies (7.6), w satisfies

iwt+pw-I-g(w)=0,

for almost all t E (0, 6). Therefore (Lemma 7.1.2) w is solution of (7.4) on [0, d],
with cp replaced by w(0). But (Iw(0)II H 1 = IIu(6)II H 1 <_ M. It follows that w
coincides with the solution given by Proposition 7.4.12 on [0, d]. In particular,
we have E(w(b)) < E(w(0)), and so E(cp) < E(u(b)), which contradicts (7.52).
Consequently, we have (i) and (ii).
Step 2. The continuity in Hl(RN). Since u : [0,T] -a HI(RN) is weakly
continuous, (i) implies that u: [0,T] -> L 2 (RN) is continuous. By (7.29), V(u) :
[0, T] --+ R is also continuous. We then deduce from (ii) that IIuIIHI : [0, T] -* R
is continuous. It is now clear that u E C([0,T],H 1 (RN)). Since g : Hl(RN) --->
H -1 (RN) is continuous, we have u E C([0,T],Hl(RN)) nC l ([0,T],H -1 (RN))
Step 3 (*). The H 2 (RN) regularity. Let m E N and 9 > 0 be such that 9 < TM
and m9 = T. u coincides with the solution of (7.4) given by Proposition 7.4.12
on [0,9], and so we have u E C([0,T],H 2 (RN)) n C l ([0,T],L 2 (RN)). Iterate,
replacing cp successively by u(j9), 1 <_ j < m - 1, in order to obtain u E
C([o,T], H 2 (R ')) n Cl([o,T], L 2 (RN)) o

End of the proof of Theorem 7.4.1. Using Proposition 7.4.12 and Lemma 7.4.11,
and arguing as in the proof of Proposition 4.3.4, we show the existence of a
maximal solution which satisfies (i). Properties (ii), (iii), and (iv) are conse-
quences of Corollary 7.4.13. It remains to show the continuous dependence. To
see this, we consider E H'(R'), and a sequence ('p,,,,) m>o , such that cp,,,, -+ cP
in H l (RN), as in -+ oo. Let u and U rn be the corresponding maximal solutions
of (7.4). It suffices to show that, for all T E [0,T(')), we have T(',,,,) > T for
m large enough, and that U rn -> u in C([0,T], H 1 (R N )), as m -> oo. Set

M = 4IIuiIL((o,T),H1);
0(m) = sup{t E [0,T(cp m )),t < T; IIurIILoo((O,t),H1) <M}.
Let k E N and 6 > 0 be such that 6 < TM and k6 = T. Applying Proposi-
tion 7.4.12, and since 9(m) <T, we easily obtain that

IIu - u m 1 ILoc(o ,8(r) ; L2) < K^"II'P - 'PmIIL2 (7.53)


112 The Schrodinger equation

Let us show that 0(m) = T, for m large enough. Without loss of generality,
we may assume that 0(m) -> 0 E [TM, T], as m -4 oo. Let t E (0, 0). U rn (t)
is defined for m large enough and, according to (7.53), we have u m (t) --. u(t)
in L 2 (R N ), as m -p oo. On the other hand, we have E(u rn (t)) = E(cp m ) -

m( )
E(cp) = E(u(t)). It follows (see Step 2 of the proof of Corollary 7.4.13) that

for m large
H (RN) as lib > IW.
u t - u i1 in l(RN),
enough, and so
particular,
(Proposition
In art,cu,ai
7.4.12(i))

0(m) >_
iiavc u i <_ M/2
we IlUm(t)IIH
have
min{T, t + TM/2}.
t <0 being arbitrary, it follows that 0(m) = T, for m large enough.
Thus, it remains to show that U rn -* u in C([0,T], H l (R N )), as m -* oo. We
argue by contradiction, and we suppose that there exists a sequence (t m ) m >o
in [0,T] and e > 0, such that IIu n (t m ) - u(tm)IIH 1 > e. We may assume
t m -> t e [0,T], which implies that IIu,,,,(t rn ) - u(t)II H l >_ E/2. By (7.53) and
since u E C([0, T], L 2 (R N )), we have u m (t m ) -> u(t) in L 2 (R N ), as m - oo.
Furthermore, we have E(u,,,,(t rn )) = E(cp m ) --> E(cp) = E(u(t)). It follows
(see Step 2 of the proof of Corollary 7.4.13) that u m (t) u(t) in H l (R N ), as
m -* oc; hence the contradiction. This completes the proof of Theorem 7.4.1.

7.5. The non-linear Schri dinger equation in R N : global existence

We suppose that ci = R N and that g satisfies the assumptions of Theorem 7.4.1.


As for the heat and Klein-Gordon equations, we are going to establish two kinds
of results: global existence for all initial data when g satisfies some additional
growth assumptions and global existence for small initial data, without any
additional hypothesis on g.

Proposition 7.5.1. Suppose that there exist 0 < 3 < 4/N and two constants
A and B such that
G(z) < AIz1 2 + BIzja+ 2

for all z E C. Then, for all cp E H l (1R"), we have T (co) = oo, and sup I u(t) IIjp <
t>_o
00.

Proof. Observe first that, for all w E H 1 (R N ), we have

IV(w)I < A f N Iu1 2 +B f N I u Ip+2

Applying Theorem 1.3.4, we deduce that

f r e+i - 4
V(w)I < A
JIwi 2 + C
RN \ JI]^N
Iwi2 / ( IRNIDwi
2 ) (7.54)
The non-linear Schrodinger equation in RN: global existence 113

Now let cp E H l (R N ) and let u be the corresponding solution of (7.4). Let


f(t) = IIVu(t)I^L z , for t E [0,T(cp)). By the conservation of energy, we have

f (t) <E(') + 2IV(u(t))I,

for all t E [0, T(cp)). Applying (7.54), and the conservation of the norm in
L 2 (RN), we deduce that

f(t) < E(p) + C(cp) f (t) 4p , I

for all t E [0,T(cp)); hence the result, since No < 4.

Remark 7.5.2. If 3 = 4/N, the above inequality becomes

f(t) <E ()
+ CII^jI f(t),
for all t E [0, T()). Therefore, if IIcc IL2 is sufficiently small, we again have
I
T( p) = oo.
,

Proposition 7.5.3. Let g be as in Theorem 7.4.1. There exist 6, K > 0 such


that if IIVIIH1 << 6, then we have T(cp) = oo and sup IIu(t)IIH1 < KII(PIIHI
e>o
Proof. Observe that
G(z) < AIz1 2 + BIzI'+z
for all z E C. We may as well assume that a > 0. It follows that I

2 Iv(w)I <- CIIwIIi2 +DIIwIIt 2 , (7.55)


for all w E H l (R N ). Now let cp E H l (R N ) and let u be the corresponding I
solution of (7.4). Let f(t) = II u(t) II H l for t E [0,T()). By the conservation
laws (ii) and (iii), we have

f(t) <_ f( 0 ) + 2 IV(^P)I +2IV(u(t))I,


I
for all t E [0, T()). Applying (7.55) and conservation of the L 2 norm, and
letting e = a/2, it follows that

f(t) < f(0)+Cf(0)+Df(0) 1+f +Cf(0)+Df(t) 1

for all t E [0, T()). Therefore, if we suppose that f(0) _< 1 then, letting
M=1+2C+ D, we have

f(t) <Mf(0)+Df(t) 1 +E, (7.56)

for all t E [0,T(,p)). Set 9(x) = Dxl+E x for x > 0, and let X = min0 < 0.
For all a E (0, X) there exist x a and ya, with 0 <x < ya such that
114 The Schrodinger equation

+ a = B(ya) + a = 0.
B(xa)
In addition, we have a < x , < a(1 + e)/e (see Figure 7.1).
a

G -

-Ex

Fig. 7.1

By (7.56), we have
9(1(t)) + M f (0) > 0,
for all t e [0, T (p) ). Consequently, if we suppose that M f (0) <, then we have

f(t) E [ 0,x Mf( 0 )) U (YMf(0) 00).


In addition, since f is continuous and f (0) < x,M f( o ), we deduce that

f(t) <_ XMf(o) <_ 1 E E Mf( 0 ),

for all t E [0,T(cp)). The result follows, with S = min{1, X/M} and K =
(1 + e)M/e.

7.6. The non - linear Schrodinger equation in RN: blow-up in finite


time
The blow-up results for the non-linear Schrodinger equation are based on the
following proposition.

Proposition 7.6.1. Let cp E Hl(RN) and let u E C([0,T(cp)),H'(R")) be the


corresponding solution of (7.4). If I () E L 2 ), then we have I u(t, ) E
(['

C([0,T(cp)), L 2 (RN)). In addition, the function t '+ f IxI 2 Iu(t,x)I' dx belongs


to C Z ([O,T(cp))), and we have
The non-linear Schrodinger equation in RN : blow-up in finite time 115

dt f N
Ix1 2 1u(t, x)I 2 dx = -4(iu r , ru) = 4Im J N rziu r dx, (7.57)
d2
IxI 2 Iu(t,x)j 2 dx = 16E(u(t))
dt 2 f N
- 4N f g(Iu(t)^)1u(t)j dx + 8(N + 2)f G(u(t)) dx, (7.58)
^N

for alit E [0,T(cp)).

Remark. Identity(7.58) is very important. It is useful not only for estab-


lishing blow-up results, but also for studying the global behaviour of solutions
(Section 7.7).

The proof of (7.57) is rather simple. Formally, it suffices to multiply the


equation by r 2 u. However, this is not correct because of the term r 2 . We
overcome this difficulty multiplying the equation by e - " 2 r 2 u and passing to the
limit as e J 0 (Lemma 7.6.2). The proof of (7.58) is much more complicated.
Formally, we would multiply the equation by ru, though this is not possible for
two reasons: firstly, because of the term in r; secondly, u is only in H 1 (RN) so
the equation makes sense in H -1 (Rn'), but u,. is only in L 2 (RN). Therefore,
we cannot multiply the equation by u,. (for this we need u r. in H l (RN)). To
show (7.58), we then proceed as follows: we approximate cp by a sequence of
functions co of H 2 (RN) , and we denote by un the corresponding solutions. We
multiply the equation satisfied by u, 1 by e - 9r8,.u,, and then we let E j 0. We
obtain (7.58) for the solution u, (Lemma 7.6.3) and then we let n -> oo.

Lemma 7.6.2. Let p E H 1 (R") and let u E C([0,T(cp)),H 1 (R' )) be the


corresponding solution of (7.4). If () E L 2 (RN), we have Ju(t, ) E
C([0,T(,p)), L 2 (RN)). In addition, the function t
f xj 2 lu(t,x)j 2 dx is in
C 1 ([0,T(cp))), and identity (7.57) holds.

Proof. For e > 0, we define the functions 8 and p E by

z
0 E (r) =re;

p(r) = r 2 e -Er2

We set
fE (t) = J I B E UI 2 dx,
RN
116 The Schrodinger equation

for t E [0,T(cp)). It is clear that f E C'([0,T(cp))), and that, for all t E [0,T(co)),
E

we have
f(t) = 2 (Beu, OeUt)Hl,H -1 = 2 (peu, ut) = 2 (peu, iAu + i9(u))
= 2(p f u,iLu) = -2(p 8 Vu,iVu) -2(Vp,u,iVu) (7.59)
= -2(V p,u, iVu) = -2 (p'u, iur)

We easily verify that p'' (r) < CO.,(r), and so we obtain

f(t) <2CIIu(t)IIxlfe(t)1 /2

It follows immediately that f(t) is bounded as E J. 0, uniformly on [0, T], for all
T < T(cp). By monotone convergence, we deduce that f IxI 2 Iu(t,x)I 2 dx < oo
for all t E [0,T(cp)), and that u(t, )11 L 2 is bounded on any interval [0,T],
with T < T(p). In particular, t i I Iu(t, ) is weakly continuous from [0, T())
to L2(RN). Integrating (7.59) between 0 and t E [0, T(cp)), we obtain

fe(t) = ff( 0 ) -4
J0 ((1 - Er )e
t 2 -E

for all t E [0,T(cp)). Letting E J 0, it follows that


t

JRNN
Ixu(t)[ 2 dx =
IR Ixcpl2 dx 4 (ru, iur.),
0

for all t E [0,T(p)). We deduce (7.57). In addition, the function

tHIII'Iu(t,-)IIL2

is continuous, and then I Iu(t, -) E C([0, T(cp)), L 2 (R N )).

Lemma 7.6.3. Let cp E H 2 (R N ) and let u E CC[0,T(cy)),H 2 (R N )) be the


corresponding solution of (7.4). If I Icy() E L 2 (R N ), we have I ( u(t, ) E
C([0,T(ep)),L 2 (R N )). In addition, the function t > f IxI 2 Iu(t,x)I 2 dx is in
C 2 ([0,T(cp))), and (7.58) holds.

Proof. We proceed in five steps.

Step 1. Let 8 E S(l[8 N ) be a radially symmetric real-valued function. Let


T> 0 and u E C([0,T],H'(lR N ))nC l ([0,T],L 2 (R r')). Set h(t) = (r9u,iu,.), for
t E [0,T]. Then h E C 1 ([0,T]) and

h'(t) = (u t , i(20ru,. + (NB + rO')u)), (7.60)


The non-linear Schrodinger equation in R N : blow-up in finite time 117

for all t E [0,T]. Indeed, suppose first that u E C 1 ([0,T],H 1 (lR N )). In that
case, it is clear that h E C 1 ([O,T]) and we have

h'(t) = (rOu t , iu r ) + (rOu,iu tr ) = (u t ,irOu r ) + (Bu,ix Vu t ). (7.61)

Observe that f
(Bu, ix Vu t ) = Im
JRN 6x VT, t dx,

and that

Ox V i = V (xOuut ) NOuut Ox Vu rO'uut .

We then have

(Bu, ix Vu t ) _ (u t , i(Ox Vu + (NO + rO')u)) = (u t , i(Oru r + (NO + rO')u)).

Applying (7.61), we deduce (7.60), and then

h(t) = h(0) + t (u t , i(20ru r + (NO + 8')u)), (7.62)


fo
for all t E [0,T]. By density, we obtain (7.62) as well as (7.60) for u E
C([ 0 ,TJ, H 1 (R N )) n C' ([ 0 ,T], L 2 (R N )).
Step 2. Let u be as in the statement of the lemma and set h(t) = (rOu, iu r.),
for t E [0,T]. Then, by Step 1, we have h E C 1 ([0,T(cp))) and

h'(t) = (Au + g(u), 20ru r + (NO + rO')u), (7.63)

for all t E [0, T}.

Step 3. Let w E H 2 (RN) be such that I Iw() E L 2 (RN). Then we have

(Ow + g(w), 20rw r +(NB + rO')w) = 2


JR N

+N f 9(g(IwI)jwj-2G(w))

+7 N

Re
rO'(g(Jwj)jw) 2G(w) 2 ^wT[ 2 )

r((N + 1)0' + rO")w T w. (7.64)


J
]RN

Note first that, by density, it suffices to consider the case w E D(RN). In that
case, we have

(g(w), (NO + rO')w) = f (NO + r0')g(Jwj)Iw^; (7.65)


118 The Schrodinger equation

= Re
(g(w), 20rw r )
J N 20rg(w)w,. =
fR N 20rG(w) ..
r

But 20rG(w) r = 29xOG(w) = V (x20G(w))-2(N0+r9')G(w). Consequently,

(g(w), 28 r w r ) = 2
J (NB + rO')G(w). (7.66)

On the other hand, we have

(Aw, (NB + r8')w) _ (NB + r9') I I 2


IR N

Re r((N + 1)0' + r0")w r w; (7.67)


J N

(Aw,20rw r ) _ (20Vu,V(rur)) 2 rO'Iwr 1 2 . (7.68)


fR N

Note also that the following identity holds:

Re(20Vw V(rwr )) = Re(20Vw 0(x V5 ))


= ((2 N)9 r8')IVwI 2 + V _ (zOIVwi 2 )

From (7.68), we then deduce that

(1w, 20r w r.) =


J^' ((N 2)0 + r8')IVW1 2 2 / rO'Iw r 1 2 .
N N JJ1R
(7.69)

Applying (7.65), (7.66), (7.67), and (7.69), we obtain (7.64).


Step 4. Let u be as in the statement of the lemma and set h(t) = (r0u, iu r.),
for t E [0,T(cp)). Then, by Steps 2 and 3, we have

h'(t) = 2
fR 9IVU1 2 + N
J 9(g(1u1)1u1 2G(u))

+
J
RN
r9'(9(juj)juj 2G(u) 2Iu r 1 2 )

Re
(7.70)

JRN r((N 1)9' + r8")u u, r

for all t E [0, T(cp)).


Step 5. Take 0 E = e 2 , fore > 0. We have 10,1 1 and 0, * 1 as E J 0. We
easily verify that Ir0EI <_ C, r9E 40 as e J. 0, and r[(N+ 1)0E +r0E]I < C and
that Ir[(N+1)0 +r0E]I +0 as E 1 0. We deduce immediately from (7.70) that
t
(r9 u, iu r ) = (r0 E Cp, icp r ) + 1 fe (s) ds, (7.71)
0

The non-linear Schrodinger equation in R N : blow-up in finite time 119

where f f is a bounded function on [0, T] for all T E [0, T (gyp)) and is such that

fE(t) > _2f N (Vu1 2 + N


J N (g(^u^)lul 2G(u)) as E 1 0. (7.72)

Observe that, due to conservation of energy, we have

_2f jvul 2 +N f (g(IuD)!ul - 2G(u))


N N

_ 4E(u(t)) + N
J g(^uJul (2N + 4) f G(u).
N

Letting e 1 0 in (7.71), we obtain the following:

(ru, iu r ) = (i , r) f ( 4 E(u(t)) N J g(Iul)lul


RN (7.73)
+ (2N + 4)
f RN
G(u)).

Now, (7.58) follows readily from (7.57) and (7.73).

Proof of Proposition 7.6.1. Applying Lemma 7.6.2, it remains to verify that


the function
tH I xl 2 ju(t, x)I 2 dx
J N

is in C 2 ([0,T(cp))), and that we have (7.58). Let T < T(cp) and let (cp m ),,,, >o be a
sequence of functions in H 2 (R N ), such that cp 71 4 cp in H 1 (R N ), as m > oo. De-
note the corresponding solutions of (7.4) by u rn . We know (Theorem 7.4.1) that,
for m sufficiently large, we have T(cp m ) > T and u,,,, , u in C([0,T], H l (R N )),
as in 4 oo. We write identity (7.73) for u,n and t E [0,T] and we let in + oo.
We deduce that u satisfies (7.73); hence the result.

Theorem 7.6.4. Let g be as in Theorem 7.4.1, and suppose further that

sg(s) > (2+ N)) G(s), (7.74)

for all s >_ 0. Then if cp E H l (R N ) is such that yep() E L 2 (R') and E(cp) < 0,
we have T(ep) < oo.

Proof. Let cp be as above. Let u be the corresponding solution of (7.4) and set

1(t) = I IxI2Iu(t,x)l2dx,
RN
118 The Schrodinger equation

26rG(w) r .
(g(w), 20rw,.) = Re
JR 29rg(w)w,. = I
N RN

But 20rG(w), = 20xOG(w) =V.(x20G(w))-2(N6+rO')G(w). Consequently,

(g(w), 20,w,) = 2
J (NB + rO')G(w). (7.66)

On the other hand, we have

(Aw, (NB + r9')w) _ (NB + r8')IVW1 2


fR N
Re
J r((N + 1)0' + r0")w w; (7.67)
R, N
r

(Aw,20rw r ) _ (20Vu,V(ru r ))-2


f
RN
rO'Iw r 1 2 . (7.68)

Note also that the following identity holds:

Re(20Vw V(rw,.)) = Re(20Vw 0(x Vw))


= ((2 N)8 rO')IVwI 2 + V (xOIOwI 2 ).

From (7.68), we then deduce that

rG'jw r.I 2 . (7.69)


(Aw, 20 r W,) =
IR N
((N 2)0 + rO')IVwl 2 2
IR N

Applying (7.65), (7.66), (7.67), and (7.69), we obtain (7.64).

Step 4. Let u be as in the statement of the lemma and set h(t) = (r0u, lU r ),
for t E [0, T(cp)). Then, by Steps 2 and 3, we have

h'(t) _ 2
JR OIVuI 2 + N
I 0
RN (g(IuI)Iuj
2 G(u))

f
N

+ RN
r9 '(g(jul )ju^ 2G(u) 2Iu r.I 2 ) (7.70)

Re
fR N
r((N 1)0' + r8")u r u,

for all t E [0,T(^p)).


Step 5. Take 0 E = e 2 , fore > 0. We have I B E I < 1 and B E --4 1 as e J 0. We

easily verify that I r0 I <_ C, r0 40 as e 1 0, and r [(N + 1)0 + r0E'] I <C and
that jr[(N+1)0'' +r0E]I --*0 as e 1 0. We deduce immediately from (7.70) that
t
(rB E u,iu,) = (r0e ,i^Pr) + f fe(s) ds, (7.71)
o
The non-linear Schrodinger equation in R N : blow-up in finite time 119 1

where fE is a bounded function on [0, T] for all T E [0, T (cp)) and is such that

2
fe(t) 9-2 N IVu1 2 + N JR N (g(lul)lul 2G(u)) as e j 0. (7.72)
fit
Observe that, due to conservation of energy, we have

2 f N IVu1 2 +N f N (g(lul)lul 2G(u)) I

_ 4E(u(t)) + N
J g(lul)Jul (2N +4) fG(u).
]RN

Letting E 1 0 in (7.71), we obtain the following:

(ru, iu r.) = (rev, i^vr) ft (E(u(t)) NJ g(luIuI


RN(7.73)
+ (2N + 4)
fR N
G(u)).

Now, (7.58) follows readily from (7.57) and (7.73).

Proof of Proposition 7.6.1. Applying Lemma 7.6.2, it remains to verify that


the function
t f N Ix1 2 Iu(t,x)I 2 dx
is in C 2 ([0, T())), and that we have (7.58). Let T <T() and let (n)m>0 be a
sequence of functions in H 2 (R N ), such that cc > W in H l (R N ), as m + oo. De-
note the corresponding solutions of (7.4) by u rn . We know (Theorem 7.4.1) that,
for in sufficiently large, we have T(cp T..) > T and u rn > u in C([0,T],H l (R N )),
as m ^ oo. We write identity (7.73) for u,,,, and t E [0, T] and we let m + oo.
We deduce that u satisfies (7.73); hence the result.

Theorem 7.6.4. Let g be as in Theorem 7.4.1, and suppose further that

sg(s) > (2 + N I G(s), (7.74)

for all s >_ 0. Then if cc E H I (R N ) is such that I Icp() E L2(RN) and E(cc) < 0,
we have T (gyp) < oo.

Proof. Let cp be as above. Let u be the corresponding solution of (7.4) and set

f(t) =Ixl 2 1u(t,x)I 2 dx,


fRN
120 The Schrodinger equation

for t E [0,T(cp)). It follows from (7.58), (7.74), and the conservation of energy
that
f"(t) < 16E(), (7.75)

for all t E [0,T(cp)). From (7.75), we deduce that

f(t) <1(0) + t f'(o) + 8t 2 E( V ),

for all t E [0,T(cp)); this implies that T(cp) < oo, since f (t) > 0 and E(cp) < 0.

Remark 7.6.5. If there exists x > 0 such that G(x) > 0, (7.74) implies that
G(s) > (s/x) 2 + 4 I N G(x) for s > x. In particular, if we take E H l (R N ), then
E(kcp) < 0 for k large enough; and so if I Icp() E L Z (R N ), then T(kcp) < oc.

7.7. A remark concerning behaviour at infinity

Identities (7.57) and (7.58) allow us to prove directly the pseudo-conformal con-
servation law, which provides information about the behaviour at infinity in
time of the solutions, in some cases (see 7.8 below). The following proposition
is related to this conservation law.

Proposition 7.7.1. Let cp E H'(R N ) be such that I Icp() E L 2 (R N ) and let


u be the corresponding solution of (7.4). Set

f(t) =f N I(x + 2itV)u(t, x) 12 dx 8t2


JR N
G(u(t, x)) dx, (7.76)

for alit E [0,T(cp)). Then f E C 1 ([0,T(cp))) and

f'(t) = 4t
J (Ng(IuI)IuI 2(N + 2)G(u)),
.N
(7.77)

for alit E [0, T(cp)).

Proof. Developing the right-hand side of (7.76), we obtain

f(t) = IIxuIIi2 + 4t(ru, iu r ) + 4t 2 IIVuIIi2


8t 2 V(u)
= IIxuIIi2 + 4t(ru, iu r ) + 8t 2 E(cp).
It follows immediately from Proposition 7.6.1 that f E C 1 ([0, T())) and that
identity (7.77) holds.

Remark 7.7.2. If g(s) = As 1 + 4 /N, (7.77) means f'(t) = 0, and then we have
f(t) = f(0) = f IxVI2
Application to a model case 121

Remark 7.7.3. Let u be as in the statement of Proposition 7.7.1, and set

v(t, x) = e ' U u(t, x),


-

for x E R N and t E [0,T(cp)). It is clear that

I(x+2itV)u(t)I 2 = 4t 2 IVv(t)I 2 > 4 t 2 IVIu(t)I I 2 ;

and, consequently,
f (t) = 8t 2 E(v(t)) > 8t 2 E(Iu(t)I). (7.78)

7.8. Application to a model case

We choose g(s) = als^as, with a > 0, (N 2)a < 4, and a 54 0. We consider


cp E H 1 (IRN) and we denote by u the corresponding maximal solution of (7.4),
which exists by Theorem 7.4.1. We then have the following results.
If a < 0, then T(cp) = oc and u is bounded in H 1 (R N ) (Proposition 7.5.1).
If a > 0 and if a < 4/N, then T(cp) = oc and u is bounded in H 1 (RN)
(Proposition 7.5.1).
If a > 0 and if a = 4/N, then T(cp) < oo for some initial data cp (Theo-
rem 7.6.4 and Remark 7.6.5); on the other hand, if II PIIL 2 is sufficiently small
then T(o) = oo and u is bounded in H'(Rn') (Remark 7.5.2).
If a> 0 and if a > 4/N, we have T(cp) Goo for some special initial data cp
(Theorem 7.6.4 and Remark 7.6.5); on the other hand, if II PIIH1 is small enough,
then T(cp) = co and u is bounded in H 1 (RN) (Proposition 7.5.3).

Observe that in the case in which a <0 and if I Icp() E L 2 (RN), (7.77) shows
that the solutions converge to 0 as t + oo, in certain spaces LP(RN). Indeed, it
follows from (7.77) that

f'(t) = -4(4 Na)t a + Iu(t) I-+ 2 =4(4 Na)tV(u(t)), (7.79)


2 fR N

for all t >_ 0. In particular, if a > n,, we have f'(t) <0, and so, with (7.78), it
follows that
8t2E(I u(t)I) s f Ix^I 2 ,
for all t > 0. If we apply Theorem 1.3.4, we obtain, in particular,

I^u(t)IIL-+z <CIIVIu(t)IIIL(2 )^^^^IL2N(2 -) <Ct z^^+z>,


122 The Schredinger equation

for all t >_ 0. Observe that we obtain the same negative exponent of t as for the
linear equation (Proposition 3.5.14).
If a < 4/N, it follows from (7.79) and (7.78) that

f'(t) <4(4- Na)tE(I u(t) I) < (2_ Na) t f (t),

Na-4
for all t > 0. Therefore, t 2 f (t) is non-decreasing and, (by (7.78)), we have

8t E(Iu(t)I) <_ f(1),

for all t > 1. Consequently, we again obtain


_ Na

IIu(t)IIL-+2 <Ct a+2

i.e. the same negative exponenent of t as for the linear equation.

Notes. For Sl 54 RN, N > 2, a few results are known. See Brezis and Gal-
louet [1], Kavian [1], Y. Tsutsumi [2, 3], and Yao [1]. Note also that Cazenave
and Haraux's results [1] apply in any open subset St C R N .
For 7.3, see Yajima [1], and Strichartz [1]. A regularizing effect of Hs(R")-
type also exists; see Constantin and Saut [1-3], Sjolin [1], and Vega [1]. The
problem of local existence (7.4) has been studied extensively. Our presentation
is based on Kato [3] and Cazenave and Weissler [1, 3]. See also Baillon, Cazenave,
and Figueira [1], Cazenave [2], Ginibre and Velo [1, 2, 5], Hayashi [1], Hayashi
and Tsutsumi [1], Lin and Strauss [1], and Weinstein [1]. The Cauchy problem
has also been studied in H 9 (R n') (s # 1); see Cazenave and Weissler [2, 4],
Ginibre and Velo [4], and Y. Tsutsumi [4]. There exists a regularizing effect for
the non-linear equation; see Hayashi, Nakamitsu, and Tsutsumi [1, 2], Hayashi
and Ozawa [2], and Kato [3].
Various kinds of non-linearities (possibly non-local) have been considered.
See, for example, Baillon, Cazenave, and Figueira [2], Baillon and Chadam [1],
Cazenave [2], Cazenave and Haraux [1], Cazenave and Weissler [1], Chadam
and Glassey [1], Dias and Figueira [1], Ginibre and Velo [3], Klainerman and
Ponce [1], Lange [1], and Schochet and Weinstein [1]. For more about blow-up
in finite time, consult M. Tsutsumi [1], Glassey [2], and Weinstein [4].
For some non-linearities, there exist solutions of the form u(t, x) = e u)t ^p(x).
These solutions are called stationary states. See, for example, Berestycki, Gal-
louet and Kavian [1], Berestycki and Lions [1], Berestycki, Lions, and Peletier [1],
and Jones and Kiipper [1].
The behaviour at infinity of solutions is rather well known in the repulsive
case, if the solutions behave asymptotically as the solutions of the linear equa-
tion. See Ginibre and Velo [1, 2, 7, 8, 10], Hayashi [2], Hayashi and Ozawa [1],
Application to a model case 123

Lin and Strauss [1], Reed and Simon [1], and Y. Tsutsumi [1]. In the attractive
case, we only know how to study the stability of some stationary states. See
Berestycki and Cazenave [1], Blanchard, Stubbe, and Vazquez [1], Cazenave [3],
Cazenave and Lions [1], Grillakis, Shatah, and Strauss [1], Jones [1], Shatah and
Strauss [1], and Weinstein [2, 3]. For more references concerning these questions,
consult Cazenave [4].

S
! l
^...
Bounds on global solutions

The study of the behaviour at infinity of global solutions is one of the most
important problems in the study of non-linear evolution equations. The problem
can be formulated as follows. If u(t), 0 <t < oo is a solution of an equation of
the form
u'(t) = Au(t) + F(t, u(t)),

for t >_ 0, how does u(t) behave as t > oo? The results concerning the behaviour
at infinity of global solutions are, in general, based on compactness properties
of U {u(t)}, and in particular on bounds for U {u(t)} (see Chapter 9). In some
t >o t>O
cases, the solutions are bounded by construction (see 5.3, 6.3, 7.5). On the
other hand, even for linear equations, global solutions may be not bounded. For
example, u(t, x) = e t sin(xrx) is a solution of

u t = u xX + 2ir 2 u, in cl = ( 0,1);
u=0, in 852;

and u is bounded in no function space as t --f oo.


The available methods allow us to study the behaviour of the solutions in
the following two situations:
Semilinear autonomous equations (see 8.1, 8.2, 5.3, 6.3, and 7.5).
Semilinear non-autonomous equations, with dissipation (8.4) or with repul-
sive interaction (see 8.3).

8.1. The heat equation

We use the notation of Chapter 5. In particular, f is a bounded open subset of


RN with Lipschitz continuous boundary, X = Co(1) and (T(t)) t >o denotes the
semigroup associated with the heat equation. g is a locally Lipschitz continuous
function IR 4 ]R such that g(0) = 0, and we consider G and E to be defined as
for Proposition 5.4.4.
In the following subsection, we gather some preliminary results that will be
useful in establishing the main result.

The heat equation 125

8.1.1. A singular Gronwall's lemma: application to the heat equa-


tion

Lemma 8.1.1. Let T > 0, A >_ 0, 0 < /3 <_ a, and 1 _< y <_ oo. Let
cp E C([0,T]), cp > 0 and f E Lry(0,T), f > 0. Suppose that a + 1/b < 1 and
that
p(t) < At + ti
f (s)^p(s) ds,
(t -sp
)

for all t E (0,T). Then cp(t) < CAt - a, for alit E (0,T), where C depends only
on T, a, p, y, and IIIf II LI (0,T)

Proof. Set zl)(t) = t(t) for all t E [0,T], and let 0(t) = sup O(s) for t E
O<s<t
[0, T]. It is clear that 0 E C( [0, T]) and that we have

O(t) < A+ t- fo 1 s) a s f(s)0(s) ds,


(t -

for all t E (0,T). We have t -0 E L 7 '(0,T), and so there exists E E (0,1) such
that
(1 - E) allt "IIL,'(O,ET)IIfIILI(O ,T) <_ 1/2.
- -

It follows that

^i(t) < A + t^B(t) t 1 f (s) ds


(,-E)t (t -s) S
1 r (i-)t 1
+ to
J 1 f(s)9(s)
(t s)$ ds

< A + 20(t) + pt^ a i(1-E)t s f(s)0(s)ds

We deduce immediately that

T
0(t) < A + 2 B(t) + AT` Q j S f (s)B(s) ds;
-

and we obtain the conclusion by applying Lemma 4.2.1, since s, `Y f (s) E -

L 1 (p T)

We will also make use of the following comparison lemma, which generalizes
Proposition 5.3.1.
126 Bounds on global solutions

Lemma 8.1.2. Let T > 0, p > 2N/(N + 2) (p >_ 1 if N = 1, p >1 if


N = 2), cp E X, u E C([0,T],X) n L 2 ((0,T),Ho(cl)) nW 1 " 2 ((0,T),H -1 ) and
f E Ll((0,T),L'(Sl)) such that

Jtl u(0)
ut = Au + f in H'(1), almost everywhere in (0, T);
= W.

Suppose further that there exists a constant C such that

f (t, x) < CIu(t, x)I almost everywhere in S2,

for almost every t E (0,T). Then if cp > 0 on Sl, we have u(t) > 0 on 52, for all
t E [0, T] .

Proof. Note first that the hypotheses imply that f E L 1 ((0, T),H 1 (S )), and
-

so the equation makes sense in H -1 (52). Now, the proof of Proposition 5.3.1 can
be adapted immediately, since

dt fz
u (t) 2 dx = 2 (ut(t), u (t))H-1,Hl = 2 I IVu (t)1 2 2
Jo 1 f (t)u (t),
Jo
almost everywhere in Q. 1

Lemma 8.1.3. Let T > 0, a, yy >_ 1 be such that N/(2Q) + 1/y < 1. Let
cp E X, u E C([0,T],X) and f E L' ((0,T),L(S2)) be such that

e<t_<T
u(t) =T(t)cp+
fo tT(ts){f (s)u(s)}ds,
for all t E [0,T]. Then for all E E (0,T], we have sup Jju(t)IILo <_ K, where K

depends only on e,o,ry, 1IccII L l and 11 f + IjL1((o,T),L^)

Proof. We proceed in two steps. Note first that the hypotheses imply fu E
L((0,T), L(1)), so the above integral makes sense, since by (3.34) T(t) can
be extended to an operator in (LP, Lp) for all p E [1, oo). On the other hand,
invoking Lemma 8.1.2, we may restrict ourselves to the case in which cp > 0, and
so u > 0 (see the proof of Proposition 5.3.3).

Step 1. Let 1 < p < r < oo be such that

Nil r1<1 and -+-<1.


The heat equation 127

Let us show that, under the hypotheses of the lemma, we have

sup IIu(t)IIL- < C,


e<t<T

where C is a constant that depends only on II f), e, v, y, r, p,T, and


+ 1IL7((o,T),L
IIVIIL' . Indeed, by Holder's inequality, (3.34) and Lemma 3.5.9, we have

I[u(t)IILr <t- v r II^vIILP+


( - )
t

JO (t s)
IIf(S) + IILIIu(S)IILrdS, I
for all t E (0, T]. We conclude by applying Lemma 8.1.1.

Step 2. Let m be an integer such that m < a < m+1. Let E > 0 and
6 = e/(m + 1). Applying Step 1 with p = 1 and r = a/(a 1), we conclude
that u(6) is bounded in L ( i t (cl), with respect to the above parameters.
We iterate this argument, translating the time of 6 at each Step and taking
successively p = Q/(o j) and r = a/(Q j 1), for 1 < j < m 1. We obtain
that u(rn6) is bounded in L/ ( "n ) (Sl) We conclude by applying Step 1 again
with p = o- /(o- m) and r = oo, to find that u is bounded in L(1) on [e,T],
with respect to the above parameters.

Corollary 8.1.4. Let or >_ 1 be such that or > N/2, cc E X, u E C([0, oo), X )
and let f E C([0, oo), L (S2)) be such that
ft
u(t) = T(t) cp +
J
0
T(t s){ f (s)u(s)} ds,

for all t >_ 0. Suppose that sup Ilu(t)IILI < oo and that sup 1 f (t) + I[L < oo.
t>O t>O
Then, for all e > 0, we have sup IIu(t)IIL <00.
t>e

Proof. For all s > 0, we apply Lemma 8.1.3 with T = 1 + e and cp = u(s). We
obtain in particular that u(s + e) is estimated in L (1)-norm, only in terms of
c, s, sup Z[u(t)IIL1 and sup II.f (t) + II L; hence the result, since s >_ 0 is arbitrary.
t>O

Corollary 8.1.5.
t>O

Suppose that g satisfies


I
xg(x) < C(x 2 + (xl"), Vx E R, (8.1)

where p > 2 and (N 2)p < 2N. Then, for all M, there exist t(M) > 0 and
K(M) < oo with the following properties: if cc E X n Ho (1l) is such that
IIccI[x7 < M and if u denotes the corresponding maximal solution of (5.4)
128 Bounds on global solutions

(see Theorem 5.2.1), then T(cp) > t(M) and JIu(t)IILOO < K(M) for all
t E [t(M)/2,t(M)].

Proof. Applying Proposition 5.3.1 and arguing as in the proof of Proposi-


tion 5.3.3, we may restrict ourselves to the case in which cp > 0, and so u > 0.
From (8.1), it follows that

IIg(u(t))+II L <_ A ( 1 + I ju(t)II LP)P ' ,


- (8.2)

for all t E [0,T(cp)). If we set /.3 = N(p - 2)/(2p) E [0,1), we deduce from (3.34),
(8.2), and Lemma 3.5.9 that

u(t)IILP < II ^vIILP +A J t (1 + Iju(s)IILP)P


-1 ds,
O (tR
- s)

for all t E [0,T(cp)). But, using Sobolev's inequalities, we have j II LP < CM.
Setting f (t) = 1 + sup IIu(s)IlLP, we then obtain
o<s<t

f (t) < (1 + CM) + A tl f (t)P-1, (8.3)


for all t E [0,T(cp)). Set

/ (1 _ 3)(1 + CM)2- P 1T
'

T(M) = I 2PA

We easily deduce from (8.3) that f (t) < 2(1+CM) for 0< t < min{T(cp), t(M)}.
Applying (8.1) we conclude immediately that

g(u(t)) <+ _B,


I C u(t) IIL
for 0'< t < min{T(cp), rr(M)}, where B depends on u only through the value of
M. We then write g(u) = hu, with h = g(u)/u and we apply Lemma 8.1.3, to
obtain that, for all e E (0, min{T (cp), T(M)},

sup <K,
e<t<min{T(cp),r(M)}

where K depends on u only through the value of M. It follows that T() > Tr(M)
and we complete the proof taking e = T(M)/2.
The heat equation 129

8.1.2. Uniform estimates


The main result of this section is the following.

Theorem 8.1.6. Suppose that g satisfies (8.1), and that there exist M < oc
and E > 0 such that
xg(x) > ( 2 + e)G(x),
for IxI > M. Let cp E X and let u be the corresponding maximal solution of (5.4)
(see Theorem 5.2.1). Then ifT(<p) = oo, we have sup Ilu(t)IIL < oo.
t>o

Proof. Applying (5.14) and (5.18), we obtain, for all 1 < s < t < oo, the
following inequalities:

ft J ut dxdt = E(u(1)) E(u(t)); (8.4)

ff
0

d Jr u(t) 2 dx > 2(2 + e)



t ut dx dt

+e
fn I Vu(t)1 dx
2 + k 2(2 + E)E(u(s)); (8.5)

where k = 2l521( (2 + e)v), and u, v are defined by Proposition 5.4.4. But if


k 2(2 + e)E(u(s)) > 0, we conclude as in the proof of Proposition 5.4.4 that
T(p) < oo (inequality (5.19) and what follows), which is absurd. We then have
2(2 + E)E(u(t)) > k for all t > 1, and we deduce immediately from (8.4) that

ffudxdt <oo.
(8.6)

Set
r= {t> 1; f ue(t) dx<i}.
2

For t E F, we apply (8.5) with s = 1. It follows that

E
f IVu(t)1 2 dx < 2(2 + e)E(u(1)) - k + 2
f uu t dx
(
8.7 )

2(2 + e)E(u(1)) - k + Ilu(t) II L 2.

From Poincare's inequality, it follows that

Ilu(t)IILZ < 2j IVu(t)1 dx 2 +C(E).

We then deduce from (8.7) that there exists M < oo such that, for all t E F, we
have IIu(t)MH1 < M. By Corollary 8.1.5, we then have

Ilu(t + s)II L- < K(M), (8.8)


130 Bounds on global solutions

for all t E r and s E [r(M)/2,T(M)]. Set E t = is > t,s v F} for t >_ 1. It


follows from (8.6) that there exists 0 < oo such that (E0I < T(M)/4. We then
set T = 0+Tr(M). Since u E C([0,T],X), we have in particular

Sup uu(t)IIC <00.


O<t<T

On the other hand, if t >_ T then there exists s E [t r(M), t 1 T(M)]


such that s E F. Hence we have t E [s + 2T(M), s + T(M)] and so, by (8.8),
IIu(t)IIL" < K(M). Consequently, sup Iiu(t)llL < K(M).
t>T

Remark 8.1.7. If g does not satisfy (8.1), we do not know whether the con-
clusions of Theorem 8.1.6 still hold.

The following proposition sharpens the results obtained in 5.3.

Proposition 8.1.8. Suppose that there exist p < A/2 (A given by (2.2)) and
M < oo such that
G(x) < PX 2 ,

for lxi >_ M. Then, for all cp E X, the corresponding maximal solution u of (5.4)
is global and sup iiu(t)IIL <00.
t>o

Proof. By Lemma 5.3.3, we know that u is global. In addition, (8.4) holds, so


that f

2 J
f IVu(t) 1 2 dx< E(u(1))+C+p ^ u(t) 2 dx,

for t > 1. Applying (2.2), we then deduce that sup 1ju(t)ii H l < oo. On the other
t>1
hand, g(u)/u is bounded in L(SZ), and we complete the proof by applying
Corollary 8.1.4.

8.2. The Klein-Gordon equation

In this section, we use the notation from Chapter 6. In particular, ci is any


open subset of RN, m> A (where A is defined by (2.2)), X = Ho(S2), x L 2 (SZ),
g is a function of C(R, R) such that g(0) = 0, and which satisfies (6.8) with
(N 2)a < 2. G and V are defined by (6.5) and (6.6). E is defined at the
beginning of 6.2. We now state the result.

Theorem 8.2.1. Suppose that N > 3 and that there exists e > 0 such that

xg(x) > (2 + e)G(x),


The KleinGordon equation 131

for all x E R. Let (ep, z)) E X and let u be the corresponding maximal so-
lution of (6.15)(6.17) (see Theorem 6.2.2). Then, if T(cp,) = oo, we have
sups>o II (u(t), u t (t))I1 x < oo.

Proof. We proceed in five steps. Let u be as above and set

f (t) = f u (t) 2 d x,

for t > 0.
Step 1. Some inequalities. By (6.26), we have I

f"(t)>e{ i IVu(t) 1 2 dx +m fu(t)2dx


n J
+ (4 + e)
fn u (t)
t 2 dx 2(2 + -)E(W, 0),

for all t > 0. Therefore

f" (t) > E!I u(t) II H , + (4 + E)


in u (t)
t 2 dx 2(2 + e)E(cp ) (8.9)

for all t > 0. In particular, there exist rl, p > 0 such that

f"(t) ? ref (t) 2(2 + e)E(co, t/5), (8.10)

f"(t) ? II (u(t), ut(t))11 i21i 2(2 + e)E(^P, ), (8.11)

for all t >_ 0. On the other hand, we deduce from the CauchySchwarz inequality
that there exists 6 > 0 such that

II (u(t), ut(t))II X ? 26 In u(t)ut(t) dx = 6 f'(t)I

Then it follows from (8.11) that


i
f"(t) > Sl f '(t) I - 2(2 + e)E(^P, 0), (8.12)

for all t > 0.


Step 2. We claim that

E() > 0 ; (8.13)


I
at (ijf (t) - 2 ( 2 + e)E(^P, , Vt > 0, (8.14)
if(t) < max rl f (0), 2(2 + e)E(cp, 0), Vt > 0. (8.15)
132 Bounds on global solutions

Indeed, if (8.13) does not hold, we know that T(cp, Ali) < oo (Proposition 6.4.1).
On the other hand, if (8.14) does not hold, then there exists t > 0 such that,
setting g(t) = 77f (t) 2(2 + e)E(cp, zL ), we have

g(t) > 0 and g'(t) > 0. (8.16)

Applying (8.10), we also obtain

9'(t) ? gg(t), Vt > 0. (8.17)

We deduce from (8.16) and (8.17) that g(t) -4 oc, as t -+ oc. Therefore, by (8.9),
there exists to > 0 such that

f"(t) > (4 + e) to.


u t (t) 2 dx, Vt >t0.
fo
As for Proposition 6.4.1, this implies that b) Goo, which is absurd. We
then have (8.14), and (8.15) follows immediately.
Step 3. We claim that

61 f'(t)I < max{bI f'(0)I, 2(2 + E)E( cc, 0)}, Vt > 0. (8.18)

Indeed, set h(t) = b f'(t) 2(2 + e)E(cp, ), for all t >_ 0. We deduce from (8.12)
that h'(t) >_ 6h(t). We then have h(t) >_ e s(t- s h(s), for t >_ s >_ 0. If there
)

exists t >_ 0 such that h(t) > 0, then h(t) --* oo, as t -4 oo; and so f (t) ' oo, as
t --> oo. This contradicts (8.15), and so

b f'(t) < 2(2 + e)E(ip, 0), Vt > 0. (8.19)

Now set k(t) = bf'(t) 2(2 + e)E(cp,V)), for t > 0. From (8.12), we have
k'(t) >_ bk(t). Therefore, k(t) <_ e -bt k(0) for t > 0; and so k(t) < max{k(0), 0}.
Consequently, we have

s f'(t) < max{-6 f'(0), 2(2 + e)E(,p, 0) }, Vt > 0. (8.20)

Putting together (8.19) and (8.20), we obtain (8.18).


Step 4. We have
/'e+i
sup] II (u(s), u t (s)) Il x ds < oo. (8.21)
t>O t

To verify (8.21), it suffices to integrate (8.11) between t and t + 1, and next to


apply (8.18).
The Klein-Gordon equation 133

Step 5. Conclusion. Set w(t) = IKu(t),u t (t))IIX, for t > 0. We have (see
Proposition 6.2.3 and Corollary 6.1.7)

w'(t) =2! g(u(t))u t (t)dx < 2II9(u(t))IIL2IIut(t)IIL2


Il (
8.22 )

< CII9(u(t))IIL2w(t) 1/2 ,

for all t > 0. Observe then that N > 3, and so N/(N - 2) <3. Consequently,

II9(z)IIL2 <CII IzI + IZI' IIL2 _< C (ii + IIHIIL )


8.23
< CIIzII H (1 + IIzlIHI) ,
( )

for all z E Ho (S2). It follows from (8.22) and (8.23) that

w'(t) < Cw(t)(1 + w(t)), Vt > 0. (8.24)

We deduce from (8.24) and (8.21) that, for all 0 < t < s < t + 1, we have

t +1
w(s) < Cw(t) exp w(a) do) < Kw(t), (8.25)
(ft

where K depends neither on t nor on s. In particular, for all t >_ 1 and all
T E [0, 1], we have

w(t) < Kw(t-r).

Integrating this last inequality in T on [0, 1] and applying (8.21) again, it follows
that
t
w(t) <K w(s) ds < K'.
Jt -1
where K' does not depend on t. Since sup w(t) < oo, the proof is complete.
o <t<i

Remark 8.2.2. If f is bounded, we may suppose that g only satisfies the


following weaker condition (see Proposition 6.4.4): xg(x) > (2 + E)G(x), for Ixi
large.

Remark 8.2.3. We have supposed that .N >_ 3. Modifying only the end of the
proof (Step 5) we can show that the conclusions of Theorem 8.2.1 remain valid
if N = 2 and a < 4 (a appears in (6.8)), and if N = 1 for any value of a >_ 0
(see Cazenave [1] and Sili [1]).
134 Bounds on global solutions

8.3. The non-autonomous heat equation

In this section, we use the notation of Chapter 5. In particular, S2 is a bounded


open subset of RN with Lipschitz continuous boundary, X = Co(1l), and
(T(t)) t >o denotes the semigroup associated with the heat equation. g is a locally
Lipschitz function R -4 1[8 such that g(0) = 0. On the other hand, we will use
the space H(l) defined in 2.6.4 and 2.6.5. We also consider a > 1 such
that v > N/2. We have in particular Ls -* H -1 . Given T > 0, cp E X, and
h: [0, T] -> L (S2), we are going to study the solutions of the following problem:

U E C([0,T],X) n L 1 ([O,T],Ho(1)) n W' ,1 ((0,T),H -1 (Q); (8.26)

u t (t) _ Lu(t) + g(u(t)) + h(t), almost everywhere in (0, T); (8.27)


(8.28)
u(0) _ ,.
In the following subsection, we gather some preliminary results concerning prob-
lem (8.26)-(8.28).

8.3.1. The Cauchy problem for the non-autonomous heat equation


Lemma 8.3.1. Let T> 0, 1 < p < oo, and f E LP((O,T),L 0 (S2)), and let w
be given by
w(t) = T(t) cp + J t T(t - s) f (s) ds. (8.29)
0

Then, w E C([0,T], L 3 ) n LP((0,T), Ho (9)) n L1 ((0,T),X). Ifp = oo, we have


in addition that w E C([0, TI, X) nC([O, TI, H'(9)).

Proof. Observe first that, by (3.34), the integral appearing in (8.29) does make
sense. On the other hand, in view of (3.34) and (3.31), the result is clear if
f c C([0,T], L (52)). The general case follows by density since, by (3.34), (3.31),
and Young's inequality, we have:

II
!lW LP(( 0 ,T),X + 1IwIILP((0,T),H1 < CII f 11 LP((o,T),L
Corollary 8.3.2. Let T > 0, 1 < p < oo, and f E Lp((0,T),L (S2)), and
w E C([0, T], L (0)). Then w is a solution of (8.29) if and only if w is a solution
of the following problem:

1 w E C([0 ,T],L ( l)) n L P ([ 0 ,T] , Ho( 9 )) n W


((0 T),H -i
( 1 ); (8.30)

1!
wt (t) = Lw(t) + f (t), almost everywhere in (0,T); (8.31)

w(0) = 0. (8.32)

Proof. Denote by (S(t)) t >o the semigroup generated in H -1 (52) by the operator
C considered in Proposition 2.6.14. It is clear that (S(t))t>o coincides with
u

The non-autonomous heat equation 135 1


(T(t)) t >o on L(52). In particular, note that f E LP((O,T),H -i (St)) and then
apply Lemma 8.3.1 and Proposition 4.1.9.

Corollary 8.3.3. Let T > 0, cp E X, and h E L((0,T),L(S2)), and let


u E C([0,T], X). Then, u is a solution of (8.26)-(8.28) if and only if u is a
solution of

u(t) = T(t)cp +
fofo
T(t - s)g(u(s)) ds + T (t - s)h(s) ds. (8.33)

In addition, u E C((0,T],Ho(Q)) fl L Z ((0,T),Ho(Q)) nW 1 2 ((0,T),H''(S2)).


" I
Proof. We apply Lemma 8.3.1 and Corollary 8.3.2 to f (t) = g(u(t)) + h(t)
and w(t) = u(t) - T(t)<p. Thus, we obtain the equivalence between (8.33)
and (8.26)-(8.28). On the other hand, it follows easily from (3.18) that T()co E
L 2 ((0,T),Ho(1l)); which implies that u E W 1 2 ((0,T),H '(SI)). By (3.29), we
' -

have T()cp E C((0,T],Ho(SZ)); and so u E C((0,T],Ho(c)).

Now we can state a result of local existence for problem (8.26)-(8.28).

Proposition 8.3.4. Let E X, h E LOO ([0, oo), L(1l)). Then, there exists a
unique maximal solution u E C([0,T(cp)),X) of (8.33). We haveT(p) > 0, and
ifT(<p) < oo, then llu(t)IIL- -> oo as t T T(cp).

Proof. Applying Lemma 8.3.1, we easily adapt the proof of Theorem 4.3.4.

Remark 8.3.5. We see that the condition g(0) = 0 is not necessary to solve
problem (5.1)-(5.3) or problem (8.26)-(8.28). Indeed, if g(0) # 0, we can replace
g by g - g(0) and h by h + g(0)1c.

8.3.2. A priori estimates

Proposition 8.3.6. Under the hypotheses of Proposition 8.3.4, and if there


exist M, C > 0 such that

xg(x) < Cx 2 , for lxi > M, (8.34)

then T(cp) = oo.

Proof. Let w and z be the maximal solutions of the following problems: I


w(t) = T(t) cp+ + J T(t - s)g(w(s)) ds T(t - s)h(s) ds,
0 + J0
ft t
+ J T(t - s)(-g(-z(s))) ds +
z(t) = T(t)cp -

o fo T( t - s)h - (s) ds.


136 Bounds on global solutions

Applying Lemma 8.1.2, we easily verify that

-z(t) < u(t) < w(t),

for all t >_ 0 such that u, w, and z are defined. In addition, applying Lemma 8.1.2
again, we readily obtain w(t) > 0 and z(t) >_ 0. We may restrict ourselves to
the case in which cp > 0 and h >_ 0, and so u >_ 0. We deduce from (8.34) that
there exists C' > 0 such that (see Step 1 of the proof of Proposition 5.3.3)

Iig(u(t)) + IIL <C'+CIIu(t)IIL , ,


for t E [0,T(cp)). Applying (3.34), (3.37), and Lemma 3.5.9, we obtain

+ MCeAt ^ e A8 Ilu(s)IILo ds
u(t)II L < II'IIL^ + MCA t
(8.35)
+ II
fo L T(t - s)h(s) ds(IL-,

for t E [0, T(co)). Next, observe that, by (3.34) and (3.37),

IIT(ts)h(s)(IL <Ke 2(t 9)(ts) 2 IIh(s)IIL-;


- -

and so

T(t s)h(s) dsII L < KIIhltL((o,t),Lf e 2 (t 9) (t s) 2 ds


"
-

J 0
K'II hII L((O,t),L

Therefore, it follows from (8.35) that

MC' +MCeat f t
II u(t)II L <_ IHIL + 0
eA3jju(S)IILds

+ K' II hII L-((o,t),L-). (8.36)

We conclude by applying Lemma 4.2.1. D

The main result of this section is the following.

Theorem 8.3.7. Suppose that g satisfies (8.34) with C < A, and let h E
L(ll+, L(2)) n L([8 + , L 2 (Sl)). Then, for all cp E X, the maximal solution u
of (8.33) is global and satisfies sup Iu(t)IILo <00.
t>o

Proof. We know (Proposition 8.3.6) that T(cp) = oo. To establish the bound,
as in the proof of Proposition 8.3.6 we may assume that cp > 0 and h > 0,
The dissipative non-autonomous Klein-Gordon equation 137

and so it >_ 0. Multiplying the equation by u, integrating by parts, and setting


f (t) = fo u(t) 2 dx, it follows that

f'(t) < -A f(t) +


I
n
u(t)g(u(t)) dx + i u(t)h(t) dx.
n
But xg(x) < Cl + Cx 2 and

I u(t)h(t)dx < Ilu(t)IIL211h(t)IIL2 <_


a-c
2 G f(t)+c2llh(t)Ili2

< 2 f(t)+C3.

Consequently,

f'(t) c -A 2 C f(t)+C3.
It follows (see Lemma 8.4.6 below) that sup Hu(t)IIL2 < 00. We conclude as in
t >o
the proof of Proposition 5.3.6, using (see the proof of Proposition 8.3.6 above)
ft
T(t s)h(s) ds < K II h11 L((O,T),L),
'

for t E [0, T].

Remark 8.3.8. Applying the estimates of Lemma 8.3.1, we easily show that,
for all E > 0, we have sup Ilu(t)IIH1 < 00.
t >E

8.4. The dissipative non-autonomous Klein-Gordon equation

In this section, we follow the notation of Chapter 6. In particular, 1 is any


open subset of RA', m> -A (where A is defined by (2.2)), X = Ho (1l) x L 2 (S2),
(S(t) ) tER is the isometry group associated with the Klein-Gordon equation in
Y = L 2 (l) x H -1 (f ), and g is a function of C(lR,l1) such that g(0) = 0, and
which satisfies (6.8) with (N - 2)ci < 2. G and V are defined by (6.5) and (6.6).
E is defined at the beginning of 6.2. We denote by F the function defined
by F((u,v)) = ( 0,g(u)), for (u,v) E X. F is Lipschitz continuous on bounded
subsets of X. We also consider 'y > 0, and we define the operator r E G(X)
given by I, ((u, v)) = (0, ryv), for (u, v) E X. For T > 0, (cp, 0) E X, and
h: [0, T] ---> L 2 (fl), we are going to study the solutions of the following problem:

it E C([ 0 ,T],Ho(l)) C'([O,T],L 2 ( 1 ))nW 2 ' 1 (( 0 ,T),H -1 ( 1 )); (8.37)


utt - u + mu + yu t = g(u) + h in H -1 (S2), a.e. in (0, T); (8.38)
u( 0 ) = <P, ut( 0 ) = 0. (8.39)
138 Bounds on global solutions

We have a result which is similar to Lemma 6.2.1.

Lemma 8.4.1. Let T>0, (cp, Vi) E X, and h E L 1 ((0, T), L 2 (Sl)), and let u E
C([0, TI, Ho(Sl))nC'([O,T],L 2 (Sl)). Define H E L'((0,T),X), H(t) = (0, h(t)),
for almost every t E [0,T]. Then u is a solution of (8.37)-(8.39) if and only if
U = (u, u t ) is a solution of

U(t) = S(t)(, b) + f t S(t - s){F(U(s)) - F(U(s) + H(s)} ds, (8.40)


o

for all t E [0, T].

Proof. We apply Proposition 4.1.9 in the space Y.

Proposition 8.4.2. Let h E Li ,(R+ , L 2 (Sl)). For all (cp, V,) E X, there exists
a unique maximal solution u E C([0,T(cp,0)],H01 (1l)) n C 1 ([0,T(p,)],L 2 (SZ))
of (8.37)-(8.39). We have T (cp, Ji) > 0, and if T (cp, zJ) < oo, then

jj(u(t),ut(t))IIx -; oc as t I T(^P,' )

In addition,

ffr f
t
E(u(t), u(t)) + ry ut dxds = E(, ) + J bu t dxds, (8.41)
z o ^
for all t E [0, T(,p, )).

Proof. We apply the method of Theorem 4.3.4 to solve (8.40), and we apply
Lemma 8.4.1 to show (8.41). We note (see the proof of Proposition 4.3.7) that
( u depends continuously on h and, by density, we need only consider the case in
which h E C(l[8 + , L 2 (cl)). In that case, we apply Proposition 6.1.1 and (6.13),
and we obtain

dt
E(u(t),
J dx + f hut dx;
ut(t)) _ -ryu

and hence (8.41).

Corollary 8.4.3. Suppose that there exists C such that

G(x) < Cx 2 , Vx E l[8, (8.42)

and let h E L' (R + , L 2 (1l)). Then, for all (cp, 0) E X, we have T(,) = oo.
The dissipative non-autonomous Klein-Gordon equation 139

Proof. Set f (t) _ II (u(t),u t (t))II' , for t E [0,T(co, u')). It follows from (8.41)
and (8.42) that

f (t) < 2E(cp, ') +2


J0 t ^Ih(s) IIL2 f (s) 1 / 2 ds + 2C
J
o
t f (s) ds,

for t E [0, T(cp, l)); hence the result, applying Lemma 4.2.1.

El
Remark 8.4.4. If 1 is bounded, we may assume that (8.42) holds only for (xI
large (see Proposition 6.4.4).

The main result of this section is the following.

Theorem 8.4.5. Suppose that y > 0, that g satisfies (8.42) with 2C < . + m,
and that there exist K > 0 and c < .A + m - KC (A given by (2.2)) such that

xg(x) - KG(x) <2, Vx E R. (8.43)

Let h E L (R+ , L 2 (S2)). Then, for all (cc,) E X, we have T (cp, 7b) = oo, and the
corresponding maximal solution u of (8.37)-(8.39) satisfies sup 1(u(t), u t (t)) x <
t>o
oo. -

Proof. We know that T(co, zl^) = oo (Corollary 8.4.3). Take E > 0 and set

f (t) = E(u(t), u t (t)) + e


I n
uut dx, b't > 0.

It is easy to verify that f is absolutely continuous and that we have

f'(t) =
J {-(ry - e)ut - eIVu^
R
2 - emu 2 - Equu t + eug(u) + ehu + hu t } dx,

almost everywhere. Let 6 E (0, 2E). We have

f'+bf= l { ^ 2 s ut ^E 6 ) IVU1 2
( l
-m E - 2 u 2 - e(7 - 6)uu t + eug(u) - 6G(u) + ehu +hu t dx. (8.44)

Observe that, by (2.2),

2 ^
f
-(E-2 IVuI 2 dx-m(e-)f u 2 dx<-(a+m))
2 ^
u 2 dx.
2 ^

' 140 Bounds on global solutions

On the other hand, if 6 <_ eK, by applying (8.43) and (8.42), we obtain

e
f ug(u) dx b
sz J f
G(u) dx < ec u 2 dx + (eK 6) G(u) dx
f
I < ((eK 6)C + ec)
f^ u^ dx.
Thus, it follows from (8.44) that

f'+6f < f {(ye-6)U E((A+m)(1 2)


1 (K S) C c) u 2 e(y 6)uu t + ehu +h u t } dx. (8.45)
e JJJ
Suppose further that 6 <y. Applying

xy < 2x 2 + ay 2 , (8.46)

with a = 2e(ry E)/'y, it follows that


(7 ty 6)2
gy(1 6) f uut dx < E 2 j u 2 dx + 4 in u dx
t

1 <_ e2y
J u2 dx + ry4 J ut dx. (8.47)

Applying (8.46) with a = 1/(2e), we find that

1 E
^
hu dx < 4 f h2 dx +e 2
n
u2 dx; (8.48)

and next, applying (8.46) with a = 2/y,

j but dx < 1ly fJ"h2 dx + 4 ff2 ut dx.


t
(8.49)

Combining (8.45), (8.47), (8.48), and (8.49), we obtain

f'+6f < fit { (2 e 1 ) ut E \(.\+m) 1\\\ 1 2

1 (
KS l c e(ry+ 1)
l u2 +
/
+1h
2 dx. (8.50)
\ EI /\ 4 Y/
Note that, for e sufficiently small, we can take 6 = e 2 . (8.50) then reads

e 2 f
f'+el f <(2e 2 )
Jsz utdxe((.^+mKCc)
I ( +m c++1)) f 2d+(1 +1) fh2d
t
The dissipative non-autonomous Klein-Gordon equation 141

Recall that we assume that A + m - KC - c> 0, and so, if e is small enough,


we have
f' + e 2 f < (
41 + 1
) f h2dx.
(8.51)

In addition, applying (8.42) and (8.46) with a = 2e, we obtain

f(t)>_4
J
s^
utdx+2 IVu1 dx+(2 -C -e 2) j u 2 dx.
Jsi
2
o

Since 2C < A + m, applying (2.2), we see that if e is small enough, then there
exists b > 0 such that
1(t) ? EIi((u(t),ut(t))I1 2 (8.52)

The result is now a direct consequence of (8.51), (8.52), and of the following
lemma.

Lemma 8.4.6. Let T > 0, ,a > 0, and H _> 0. Let f E C([0, T]) be an
absolutely continuous function such that

f' + lif < H,

almost everywhere on (0,T). Then, we have

f (t) < - + e - ^` t f ( 0 ),

for alltE[0,T].

Proof. Set w(t) = eu t (f (t) -H/) for t E [0, T]. We have w'(t) <_ 0 almost
everywhere; and so w(t) < w(0) for all t E [0, T]; hence the result.

Remark 8.4.7. If ci is bounded, we may suppose that (8.42) and (8.43) hold
only for lxi large (see Proposition 6.4.4).

Notes. About 8.1, see Cazenave and Lions [1], Giga [3], and Ni, Sacks, and
Tavantzis [1]; and for 8.2, see Cazenave [1] and Sill [1]. Concerning non-
autonomous problems (8.3 and 8.4), see, for example, Haraux [1, 2].
0
The invariance principle and some
applications

9.1. Abstract dynamical systems

Throughout this section, (Z, d) is a complete metric space.

Definition 9.1.1. A dynamical system on Z is a family {S t } t >o of mappings


on Z such that:

(i) St E C(Z, Z),Vt > 0;

(ii) So = I;
(iii) St +s = St o S9 , Vs, t > 0;

(iv) the function t H St z is in C([0, oo), Z) for all z E Z.

Remark 9.1.2. In what follows, we write St S 8 instead of St o S.

Remark 9.1.3. It is clear that if F is a closed subset of Z such that SF C F


for all t > 0, then {(St)I F } t >o is a dynamical system on (F,d).

Definition 9.1.4. For all z E Z, the continuous curve t '--> St z is called the
trajectory from z.

Definition 9.1.5. Let z E Z. The set

w(z)={yEZ;St n >oo, St z*yarn--+c},

is called the w-limit set of z.

Proposition 9.1.6. We have w(z) = fl U {St z}.


3>0 t>s

Proof. The proof is straightforward, by Definition 9.1.5. o


Liapunov functions and the invariance principle 143

Proposition 9.1.7. For all z E Z and all t > 0, we have

w(St z) = w(z), (9.1)


St (w(z)) C w(z). (9.2)

In addition, if U {St z} is relatively compact in Z, then


t>o

St(w(z)) = w(z) 54 0. (9.3)

Proof. (9.1) is an immediate consequence of Proposition 9.1.6. Let y E w(z).


There exists t o -- oo such that St z ^ y. For all t >_ 0, and setting -r = t i-, + t,
we have S1-,z --+ S t y, and so St y E w(z); hence (9.2). Now suppose that U {S t z}
t>o
is relatively compact in Z. Then there exists a sequence t,,, ---> oo and y E Z
such that St z --> y. Therefore y E w(z) and w(z) 0. It remains to show
that w(z) C St w(z). To see this, consider y E w(z) and t,,, such that
St z > y. Set rr, = t o t. There exists a subsequence r,,, k ---> oc such that
ST,.k z ^ w E w (z). Thus,

St w = St lim ST, z = lim St , k z = y;

hence (9.3).
Theorem 9.1.8. Suppose that U {St z} is relatively compact in Z. Then:
t>o

(i) St (w(z)) = w(z) 0, for all t > 0;


(ii) w(z) is a compact connected subset of Z;
(iii) d(S t z, w(z)) --> 0 as t --> oo.

Proof. (i) is a consequence of (9.3). On the other hand, for all s > 0, U {St z}
t>s
is a relatively compact connected set. By Proposition 9.1.6, w(z) is then the
decreasing intersection of connected and compacts subsets. Hence we have (ii).
To show (iii), assume by contradiction that there exists a sequence t o * 00
and e > 0 such that d(St z,w(z)) >_ e. There exists y E Z and a subsequence
t,lk --> oo such that St k z * y E w(z). Therefore d(St k , w(z)) > 0 as k > oo,
which is absurd.

9.2. Liapunov functions and the invariance principle

Definition 9.2.1. A function 4P E C(Z, R) is called a Liapunov function for


{St } t >o if we have
4)(Stz) < (D(z),

for all z E Z and all t > 0.


144 The invariance principle and some applications

Remark 9.2.2. If 4) is a Liapunov function for {St}t>o then, for all z E Z,


the function t H 4)(St z) is non-increasing.

Theorem 9.2.3. (LaSalle Invariance Principle) Let 4) be a Liapunov func-


tion for {St }t>o, and let z E Z be such that U {St z} is relatively compact in
t>o
Z. Then:
(i) . = lim t - 0 4)(Stz) exists;

(ii) 4)(y)=, for ally Ew(z).

Proof. 4)(Stz) is non-increasing (Remark 9.2.2) and bounded since U {Stz}


t>_o
is relatively compact. Hence we have (i). If y E w(z), there exists a sequence
t n, --> oc such that S t z ^ y. Therefore, 4)(St z) > 4)(y); hence (ii).

Definition 9.2.4. An element z E Z is called an equilibrium point of {St}t>o


if Stz =z for alit >0.

Remark 9.2.5. In practical applications, Theorem 9.2.3 is used mainly to


establish that some trajectories of {S t } t >o converge to equilibrium points.

Definition 9.2.6. A Liapunov function 4) for {St } t >o is said a strict Liapunov
function if the following condition is fulfilled. If z E Z is such that 4)(St z) = 4)(z)
for all t > 0, then z is an equilibrium point of {St}t>o.

Theorem 9.2.7. Let 4) be a strict Liapunov function for {St } t >o, and let
z E Z be such that U {Stz} is relatively compact in Z. Let be the set of
t>o
equilibrium points of {St } t >o. Then:
(i) is a non-empty closed subset of Z;
(ii) d(St z, ) --+ 0 as t (i.e. w(z) C E).

0. Let
Proof. By continuity of St, is closed. By Theorem 9.1.8(i), w(z)
y E w(z). Applying Theorem 9.1.8(i) again, and then Theorem 9.2.3(ii), we
obtain
4)(Sty) = 4) (y), Vt > 0;
therefore y is an equilibrium point. From this, we deduce (i) and then (ii) by
applying Theorem 9.1.8(iii).

Remark 9.2.8. Theorem 9.2.7 means that the set of equilibrium points at-
tracts all the trajectories of {St}t>o.
A dynamical system associated with a semilinear evolution equation 145

Corollary 9.2.9. Suppose that the hypotheses of Theorem 9.2.7 are fulfilled.
Let P = lim 4D(St z) and Ee = {x E E, 4?(x) = Q}. Then Ee is a non-empty closed
subset of Z and d(St z, EQ) > 0 as t + cc (and so w(z) C Ee). If, furthermore,
Ee is discrete, then there exists y E Ej such that St z --+ y as t > oo.

Proof. Since E is closed and 1 is continuous, E is closed. The remaining part


of the corollary is a consequence of Theorems 9.2.3, 9.2.7, and 9.1.8 (ii).

9.3. A dynamical system associated with a semilinear evolution equa-


tion

We consider in this section a Banach space X, an m-dissipative operator A


with dense domain, and a function F : X -- X that is Lipschitz continuous on
bounded subsets. We use the notation of Chapter 4, and in particular we denote
by (T(t)) t > o the contraction semigroup generated by A. We recall that, for all
x E X, there exists a unique maximal solution u E C([0,T*(x)),X) of

u(t) = T(t)x +
JO t T(t s)F(u(s)) ds, `dt E [0,T*(x)). (9.4)

For x E X and t E [0,T*(x)), we set

St x = u(t).

We consider a subset P C X such that there exists M < oc with

T(y) = oo, Vy E P; (9.5)


^IStyll <_ M, Vy E P,dt > 0; (9.6)

We set Z = U U {St y}, and we denote by d the distance induced on Z by the


,EP t>O
norm of X.

Lemma 9.3.1. We have the following properties:


(i) T*(z) = co, Vz E Z;

(ii) IISt zU <M, Vz E Z, Vt > 0;


(iii) StZ E Z, Vz E Z, Vt > 0.

Proof. Let y E P. Then u(t) = St y is the solution of (see 4.3)

u E C([0, cc), X) n C' ([O, oo), Y); (9.7)


u'(t) = Bu(t) + F(u(t)), Vt > 0 (9.8) ;

u( 0 ) = y. (9.9)
146 The invariance principle and some applications

Therefore, for all s > 0, v(t) = u(t + s) is the solution of (9.7), (9.8), and
v(0) = u(s). Thus, St(S s y)) = St (u(s)) = u(t+s) for all s,t >_ 0. Consequently,
we have T*(S s y) = oc for ally E P and all s> 0, and jjStS s yjj < M for ally E P
and all s, t > 0. Now take z E Z. There exists a sequence (t n ) n,> o c [0, oo) and
a sequence (yn,),,, >o C P such that Sy - n - z as n -; oo. Let T < T*(z). By
Proposition 4.3.7, we have

StSt y n ' Stz, as n -> oo, (9.10)

uniformly on [0, T]. In particular, we have l(StzIl _< M, for t E [0, T]. Since
T < T*(z) is arbitrary, we deduce (i), and next (ii). (iii) is then a consequence
of (9.10).

Theorem 9.3.2. {St } t >o is a dynamical system on (Z,d).

Proof. We have So = I. In addition, for all z E Z, if (z,,)n >o C Z and zn -4 z


as n -i oc then, by Proposition 4.3.7, we have

St zn ---> St z, as n -- oo,

for all t > 0. Hence St E C(Z, Z) for all t > 0. Furthermore, since for all y E Z,
u(t) = Sty is the solution of (9.7)-(9.9), we deduce easily that 5t59 = St +s for
all s, t > 0. Finally, we have Stz E C([0, oo), Z) for all z E Z; hence the result.

9.4. Applications to the non-linear heat equation

We follow the notation of Chapter 5. In particular, S2 is a bounded open subset of


RN, with Lipschitz continuous boundary, X = Co (l), and (T(t)) t >o denotes the
semigroup associated with the heat equation. g is a locally Lipschitz function
R -* R such that g(0) = 0, and we consider G and E to be defined as in
Proposition 5.4.4.
Let cp E X be such that T(cp) = oo and let u be the corresponding maximal
solution of (5.1)-(5.3) (see Theorem 5.2.1). If we have

sup J[u(t)lIL- < 00, (9.11)


t >o

then we may apply the results of 9.3, choosing Y = {cp} to associate to cp


a co mplete m etric (Z, d), where d is the distance induced by the norm in X,
Z = IJ {u(t)}, and a dynamical system {St } t > o on (Z, d). On the other hand,
t >1
we know that there exist sufficient conditions to have (9.11); see, for example,
5.3 and 8.1.
Applications to the non-linear heat equation 147

Lemma 9.4.1. Let yp and u be as above. Then we have the following proper-
ties: 1
(i) U {u(t)} is relatively compact in X;
t> o

(ii) for all e > 0, we have sup II u(t) II H 1 < co;


t>E

(iii) for all e > 0, U {u(t)} is relatively compact in Ho(Sl);


t>E

(iv) E is a strict Liapunov function for {St}t>o


t
Proof. The proof proceeds in three steps.
Step 1. Let e > 0 and s > 0. Applying Remark 5.1.2, replacing cp by u(s),
we obtain in particular that IIu(s + E) H1 <_ C(e 1 / 2 + e -1 / 2 ), where C does not
depend on s. Hence we have (ii), since s is arbitrary.

Step 2. To establish (i) and (iii), we need only show that if t o 4 oc, then
there exists a subsequence t flk and w E X n Ho (1l) such that u(t fk ) --> w in
X n Ho (S2) as k 4 oo. Set Tn = t o 1, cp n = u(r) and u n (.) = u(rrn + ). It is
clear that u(t) = u n (1). By Step 1, cp n is bounded in X n Ho (1 ), and so there
exists E LO n Ho (Sl) and a subsequence (nk) such that cp nk 0 in L 2 (1), as
k > oo. Since Il^pnk IILOO is bounded and I ^Pnk 0IIL 2 > 0, it follows from
Holder's inequality that II^Pnk I Lp --> 0, for all 1 < p < oo. In particular,
conk ^ 0 in L N (1), as k --f oo. From (3.37) and (9.11), we deduce that, for all
k, f E N, we have

IlUnk un, II L < tll^nk

t
Wne II LN +

i ^ 2 llWnk one IILN + Ci


fo
t II9(unk (S))

J t Ilu nk (s)
9(un, (s)) II L _ ds

ufl ( s ) IIL ds,


1
0

for all t E (0,1]. Consequently (Lemma 8.1.1), u(t flk ) = u nk (1) is a Cauchy
sequence in X. Let w be its limit. Now applying (3.32) and (9.11), we obtain,
1
1
for all k, f E N, the following inequality:

Ikknk U., 11 ' < (1 + t-1/2)Ilonk cne IIL 2

+li J
0
t (1 + (t s)-1/2)IIunk (s) carne (s )IIL2 ds,
1
for all t E (0,1]; from this, it follows (Lemma 8.1.1) that u(t nk ) = u nk (1) is a
Cauchy sequence in Ho (1); and so that u(t nk ) * w in X n Hp (S2) as k f 00.
We have shown (i) and (iii). 1
148 The invariance principle and some applications

Step 3. E is a strict Liapunov function on Z. Indeed, E is continuous on


X n Ho (S2) and so, by (i) and (iii), E is continuous on (Z, d). Let z E Z and let
v(t) = S t z. It follows from (5.14) that, for all 0 < s < t, we have

Js f z
vt dx da + E(v(t)) = E(v(s)). (9.12)

We then have E(v(t)) <_ E(v(s)), and consequently E is a Liapunov function.


On the other hand, E(v(t)) = E(z) for all t > 0, and we deduce from (9.12) that

Jo
J vt dx du = 0.
s^

Consequently, v t = 0 for almost all t > 0, and it follows from this that v is
constant in L 2 (51), and then is also constant in X. Thus, z is an equilibrium
point and E is a strict Liapunov function. This completes the proof.

Theorem 9.4.2. Let g be as above. Set = {u E D(A); Du+g(u) = 0}, and


o, = {u E ; E(u) = a}, for a E R. Let cp E X and let u be the corresponding
maximal solution of (5.1)-(5.3) (see Theorem 5.2.1). Suppose that T(cp) = oo
and that u satisfies (9.11). Then, we have the following properties:
(i) E(u(t)) converges to a finite limit a, as t --^ oo;

(1 i) a 0;
(iii) dist(u(t), a ) > 0, as t oo, where dist denotes the distance in Xf1Ho (St

Proof. We apply Lemma 9.4.1 and Corollary 9.2.9. It suffices to note that the
set of equilibrium points of the dynamical system associated with u is included
in S.
Remark 9.4.3. If N = 1, we can give a sharper result (see Matano [1]). There
exists w E a such that u(t) w, as t # oc. If N >_ 2, this remains valid if
we suppose that g is analytic (see Simon [1]). In the general case, it remains
true for most of the solutions (see Lions [1, 2]) but, except in some special cases
(see Louzar [1] and Remarks 9.4.4 and 9.4.5 below), we do not know whether it
remains true for any solution, apart from the recent results of Hale and Raugel [1]
and Haraux and Polacik [1].

Remark 9.4.4. If we suppose that xg(x) < Cx 2 , with C < A (A given


by (2.2)), then we verify immediately by applying (2.2) that S = {0}. In that
case, all bounded solutions of (5.1)(5.3) converge to 0 in X n Ho (52) as t > oo.

Remark 9.4.5. If g is strictly concave on (0, oo), fl {u >_ 0} = {0, co}, where
co is the unique positive solution of L = g(cp), cp E Ho (Sl). In that case,
w(u0) is either 0 or ca, for all 'ao > 0 (cf. Haraux [5]).
Application to a dissipative KleinGordon equation 149

9.5. Application to a dissipative Klein Gordon equation

In this section, we use the notation of Chapter 6. In particular, S1 is any open


subset of R N , m > A (where A is defined by (2.2)), X = Ho(S2) x L 2 (f ),
(S(t))t>o is the isometry group associated with the KleinGordon equation in
Y = L 2 (52) x H '(1), g is a function of C(R,R) such that g(0) = 0, and which
satisfies (6.8) with (N 2)a < 2. G and V are defined by (6.5) and (6.6). E
is defined at the beginning of 6.2. We denote by F the function defined by
F((u,v)) = (0,g(u)), for (u,v) E X. F is Lipschitz continuous on bounded
subsets of X. We also consider 'y > 0, and we define the operator r E G(X)
given by F((u, v)) = (0, yyv), for (u, v) E X. For T > 0 and (cp, Vi) E X, we are
going to study the solutions of the following problem:

U E C([0, T], Ho (SZ)) n C 1 ([0, T], L 2 (S2)) n C 2 ([0, T], H ' (S2)); (9.13)
utt 'Lu + mu + 7u t = g(u), b't > 0; (9.14)
u(0) = cp, u t (0) = z/^ (9.15)

We know (Lemma 8.4.1) that u E C([0, T], Ho (S2)) f1 C 1 ([0, T], L 2 (S2)) is a solu-
tion of (9.13)(9.15) if and only if U = (u, u t ) is a solution of

U(t) = S(t)() +
J0
t S(t s){F(U(s) F(U(s))}ds, (9.16)

for all t E [0, T]. We also know (Proposition 8.4.2) that it is possible to solve
locally (9.16) and that the solutions satisfy

E(u(t), u t (t)) +'Yf t f ui = E( , (9.17)


^

for all t E [0, T]. In particular, we have E(u(t), u t (t)) _< E(,); and so, if there
exists C such that 2C < A + m and

G(x) < Cx 2 , Vx E IR,

then we have T(co,0) = oo for all (cp,0) E X and

sup II (u(t), ut(t))II x < 00,


t>o

where u is the corresponding solution of (9.13)(9.15) (see the proof of Proposi-


tion 6.3.1, and Remark 6.3.2).
On the other hand, if S2 is bounded, it suffices that g is such that

G(x) < Cx 2 , for x large.


150 The invariance principle and some applications

For the end of this section, it is useful to formulate (9.16) in a different way. To
do this, define the operator A. on X by

5 D(A ) = {(u, v) E X; Au E L (1), v E Ho (1l)};


7 2

A-,(u, v) = (v, Au - mu - yv), for all (u, v) E D(A).

Lemma 9.5.1. The operator A. y is m-dissipative on X, with dense domain.


In addition, if we denote by (T7 (t)) t >o the contraction semigroup generated by
A y on X, and if we suppose that ry > 0, then there exist M and a > 0 such that

' t , (9.18)
I1T7(t)IIc(x) < Me -

for all t > 0.

Proof. We show that the operator A. y is m-dissipative on X, with dense domain,


as in Proposition 2.6.9. It then remains to establish (9.18). To do this, we argue
as in Theorem 8.4.5. We consider e > 0, (cp, J) E D(A,). We set T^,(t)(co,b) _
(u(t), v(t)) and

.f (t) = 2 J v 2 + 1 Vu1 2 + M f u 2 + e f uv.

We verify that, for E small enough, we have f (t) > 611(u(t), v(t)) and f'+e 2 f <
0. We deduce (9.18), with a = e 2 . El

We verify that U E C([0,T],X) is a solution of (9.16) if and only if U is a


solution of
U(t) = E y (t)(^p,V') +
J
0
t Ty (t - s)F(U(s)) ds, (9.19)

for all t E [0, T]. Let (cp, 0) E X, and let u be the corresponding maximal solution
of (9.13)(9.15). Suppose that T (p, ) = oo and that sup I (u(t), u t (t)) lI x < oo,
t>o
and set Z = U {(u(t),ut(t)}. The results of 9.3 allow us to associate with u
t>_o
a dynamical system {St}t>o on (Z, d), where d is the distance induced by the
norm in X. We have the following result.

Lemma 9.5.2. Suppose that S2 is bounded and that -y > 0. Let (p, /i) E X
be as above. Then, we have following properties:

(i) Z is compact;
( n E is a strict Liapunov function for S }
Application to a dissipative KleinGordon equation 151 1

Proof. We proceed in four steps. Set U(t) = (u(t), u(t)) and H(t) = F(U(t)),
for t > 0.

Step 1. By (9.19), we have U(t) = T,(t)(cp,7P) + W(t), where

W(t) =
f (s)H(t s) ds.

By (Lemma 9.5.1) T,,(t)(^, ) 0 in X, as t oo, and so there exists a compact


subset K l of X such that U {2 (t)(cp, ^l )} C Kl. Then we need only verify that
t >o
there exists a compact subset K 2 of X such that U {W(t)} C K2 .
t>o
Step 2. Since S2 is bounded, we see by applying Theorem 1.3.2 and Re-
mark 1.3.3, as well as the estimates of 6.1.2 (see in particular the proof of
Proposition 6.1.5), that the range by the mapping u H g(u) of a bounded sub-
set of Ho (Sl) is a relatively compact subset of L 2 (Sl). Since u is bounded in
Ho(1l), there exists a compact K of X such that U {H(t)} C K.
t>o
Step 3. Let E > 0, and let T be such that (see Lemma 9.5.1)

IIHIIL(o,00;x) FT IITryIIG(x) <6.

For t >_ T, we then have

T '
W(t) f Ty (s)H(t s) ds1Ix <6.
0

Consequently,
U {W(t)} C K' + B(0, e), (9.20)
t>_T

where we have set

T
K'= U T.^(s)H(t s) ds
t>T
^ 0 1 J

Observe that the mapping (s, x) '-- Ty (s)x is continuous from [0, oo) x X to X.
Consequently, U = U {Ty (t)K} is compact in X. Therefore, F = T conv(U)
O<t<T
is relatively compact in X. Since K' c F, K' is relatively compact in X.
By (9.20), we can cover U {W(t)} by a finite union of balls of radius 6. On the
t>T
other hand, WE C([O, oo), X); hence U {W(t)} is compact and it can also
o<t<T
152 The invariance principle and some applications

be covered by a finite union of balls of radius a. Finally, we can cover U {W(t)}


t>o
by a finite union of balls of radius E. Since e is arbitrary, U {W(t)} is relatively
t>o
compact and, by applying Step 1, we obtain (i).

Step 4. E is continuous on X, and thus also on (Z, d). (9.17) shows that E is
a Liapunov function. Finally, if E is constant on a trajectory (v, v t ) of {St } t >o,
we deduce from (9.17) that v t = 0 for all t >_ 0; therefore v does not depend
on t; z = (v, 0) is then an equilibrium point, and E is strict Liapunov function.
This completes the proof.

Theorem 9.5.3. Suppose that Sl is bounded and that ry > 0. Set = {u E


Ho (S2); ,u mu + g(u) = 0}, and E. = {u E ; E(u, 0) = a}, for a E R. Let
(cp, 0) E X, and let u be the corresponding maximal solution of (9.13)(9.15).
Suppose that T(cp,z//) = oc and that sup Il(u(t),u t (t))llX < oo. Then, we have
t>o
the following properties:
(i) E(u(t), u t (t)) converges to a finite limit ,Q, as t f oo;

(ii) A ^ 0 ;
(iii) IIut(t)IIL2 4 0, as t --4 oo;
(iv) dist(u(t), p) 0, as t > oo, where dist denotes the distance in Ho (S2).

Proof. We apply Lemma 9.4.1 and Corollary 9.2.9. It suffices to observe that
the set of equilibrium points of the dynamical system associated with u is in-
cluded in E.

Remark 9.5.4. If we suppose that xg(x) < Cx 2 , with C < A + m (A given


by (2.2)), then we verify immediately by applying (2.2) that S = {0}. In that
case, all bounded solutions of (9.13)(9.15) converge to 0 in X as t --4 oo. If
g does not satisfy this condition, sufficient conditions to ensure that w(cp, v) is
reduced to one point can be found in the literature (cf. Haraux [4], Hale and
Raugel [1]).

Notes. See Ball [3], Dafermos [1-3], Dafermos and Slemrod [1], Hale [1], Ha-
raux [1, 2], Henry [1], LaSalle [1], and Sell [1]. The w-limit sets also appear in
the theory of maximal attractors. Consult Babin and Vishik [1-3], Ghidaglia
and Temam [1, 2], Hale [2], Haraux [2], and Ladyzhenskaya [2]. The invariance
principle is also very useful in the study of the behaviour at infinity of positive
solutions of reactiondiffusion systems. See, for example, Masuda [1], Haraux
and Kirane [1], and Haraux and Youkana [1]. There has recently been substan-
tial progress on asymptotic behaviour of gradient-like systems as a consequence
Application to a dissipative KleinGordon equation 153

of the work of Hale and Raugel [1] (cf., e.g., Haraux and Polacik [1] where
the condition of Hale and Raugel is used in an essential way). On the other
hand, negative results are beginning to appear in the literature (see Polacik and
Rybakowsky [1]) when the non-linearity depends on x.
10
Stability of stationary solutions

In this chapter, we describe an extension of the Liapunov linearization method


to establish the (local or global) asymptotic stability of equilibria. The per-
turbation argument developed here is applicable to various semilinear evolution
problems on infinite-dimensional Banach spaces. We also discuss the connection
between stability and positivity in the case of the heat equation.

10.1. Definitions and simple examples

Let (X, d) be a complete metric space and {S(t) } t >o a dynamical system on X.

Definition 10.1.1. A trajectory v(t) = S(t)a of the dynamical system


{S(t) t >o is called (positively) stable in the sense of Liapunov if
}

de > 0, 35 > 0 such that


x E X and d(x, a) < S = Vt > 0, d(S(t)x, v(t)) (10.1)

Definition 10.1.2. A trajectory v(t) = S(t)a of the dynamical system


{S(t) t >o is called (positively) asymptotically stable in the sense of Liapunov if
}

it is stable in the sense of Liapunov and

1S > 0 such that x E X, d(x, a) < Si = lim d(S(t)x, v(t)) = 0. (10.2)

In particular, an equilibrium a of {S(t) t >o is called stable (resp. asymptotically


}

stable) in the sense of Liapunov if the constant trajectory v(t) - a satisfies (10.1)
(resp. (10.1) and (10.2)).

In the easiest cases for X = RN, stability of the equilibrium a of an equation

= f (u(t)), t> 0, (10.3)

with f E C' (X, X) can be seen from the linearized equation

z' = ( Df)(a)(z(t)), t > 0. (10.4)

More precisely, the exponential asymptotic stability of a for (10.3) is related


to exponential asymptotic stability of 0 for the linearized equation (10.4). We
recall here the Liapunov stability theorem (for a proof, cf., e.g., Haraux [5]).
Definitions and simple examples 155

Theorem 10.1.3. (Liapunov) Let X be a finite-dimensional normed space,


and f E C 1 (X,X) a vector field on X. Let a E X be such that f(a) = 0 and
assume that

all eigenvalues si, 1 < j < k of D f (a) have negative real parts. (10.5)

Then a is an asymptotically Liapunov stable equilibrium solution of equation


(10.3) in the following sense: for each b < v = min1<^<k{Re(s 3 )}, there exists
p = p(b) > 0 and M(5) > 1 such that, if l]x all <p(6), the solution u of (10.3)
such that u(0) = x is global with

t
Vt > 0, II u(t) all <_ M(b)]Ix a]1e -5 I

In the opposite direction, we have the following result (cf. Haraux [5]).

Proposition 10.1.4. Let X be a finite-dimensional normed space, and f E


C' (X, X) a vector field on X. Let a E X be such that f (a) = 0 and assume
that

all eigenvalues Si, 1 <j < k of Df (a) have positive real parts. (10.6)

Then a is a completely unstable equilibrium solution of (10.3) in the following


sense: there is a neighbourhood w of a in X such that, for each b E X, b a,
the unique solution of (10.3) with initial condition b leaves w for ever if t >_ T
large enough.

Remark. Stability of an equilibrium cannot always be seen on the lineariza -

tion.As a simple example, we may consider u' = f (u) = u 3 , in which cases


D f (0) = 0. When we choose the () sign, 0 is asymptotically (but not ex-
ponentially) stable, while when we choose the (+) sign it becomes completely
unstable. In this last case, 0 is the only global solution.

To illustrate the general ideas of this section, we give two simple examples.

Example 10.1.5. Let f E C 1 (IR) and consider the first order scalar ODE

u'(t) = f (u(t))

It is known (cf., e.g., Haraux [5]) that each bounded global solution u(t) of this
equation on IR+ tends to a limit c with f (c) = 0. The stability of such an
equilibrium c is delicate only when f'(c) = 0. Indeed,
If f'(c) < 0, c is exponentially stable.
Ii 156 Stability of stationary solutions

If f'(c) > 0, c is completely unstable in the sense of Proposition 10.1.4.

As an illustration, the simple first order ODE

u'+u 3 u=0

has exactly three equilibria {-1, 0, 1}. The equilibria 1 and (-1) are exponentially
stable and they attract, respectively, the positive solutions and the negative
solutions of the equation. On the other hand, the equilibrium 0 is completely
unstable in a very strong sense: it attracts no solution except itself.

Example 10.1.6 Let f E C 1 (R), c> 0 and consider the second order ODE

u"(t) + au'(t) = f (u(t)).

It is known (cf., e.g., Haraux [5], Hale and Raugel [1]) that each bounded global
solution u(t) of this equation on 1R+ tends to a limit c such that f (c) = 0 (and
u'(t) tends to 0). The stability here is defined in the sense of the phase space
H x H for the corresponding first order system in (u, u'). The situation is more
complicated than in the previous example:
If f'(c) < 0, then (c, 0) is exponentially stable in the phase space R x R.

If f'(c) > 0, then (c, 0) is unstable in the phase space R x R but attracts
some other trajectories than the equilibrium itself. We have here a typical
example of a hyperbolic point.

As an illustration, the simple second order ODE

u"+u'+u 3 u=0

has exactly three equilibria {(-1, 0); (0, 0); (1, 0)}. The equilibria (1,0) and
(-1,0) are exponentially stable in the phase space H x R. On the other hand,
the equilibrium (0,0) is a hyperbolic point.

10.2. A simple general result

Let X be a real Banach space, let T(t) = e^ t S(t) with c E H, and let (S(t)) t > 0 be
a contraction semigroup on X (it is easy to check that the family of operators
(T(t)) t > o has the semigroup property, cf. Definition 3.4.1), and F : X --> X
locally Lipschitz continuous on bounded subsets. For any x E X, we consider
the unique maximal solution u E C([0,T(x)), X) of the equation

u(t) = T(t)x + f T(t s)F(u(s))ds, Vt E [0, (x)). (10.7)


a
A simple general result 157

By a stationary solution of (10.7) we mean a constant vector a E X such that

a = T(t)a T(t - s)F(a)ds, Vt > 0.


+ fo (10.8)
t

The following result is an easy consequence of the general theory of strongly


continuous linear semigroups, and can easily be verified. Let L = cl + A, where
A is the generator of S(t) (L can be considered the generator of T(t)
in the sense
of Definition 3.4.2). Then we have the following.

Lemma 10.2.1. A vector a E X is a stationary solution of (10.7) if and only


if we have
a E D(L) and La + F(a) = 0. (10.9)

We are now in a position to state the main result of this section.

Theorem 10.2.2. Assume that, for some constants 8> 0, M > 1, we have

Vt > 0, be.
IIT(t)II < Me (10.10)
Let a E X be a stationary solution of (10.7) such that

2Ro > 0, 3v >0:


IIF(u) F(a)II < vllu all for IIu all < Ro , (10.11)
with
1 v < b1M. (10.12)
Then, for all x E X such that

Ijx - all < R i = Ro /M, (10.13)


the solution u of (10.7) is global and satisfies

Vt > 0, II u(t) - all < MII x - all e - r t , ( 10.14)


with y=b-vM>0.

Proof. On replacing u by u - a and F by F - F(a),


we may assume that a = 0
and F(a) = 0 with IIF(u)ll < vjjujj whenever IIujj C Ro . In particular, setting

T = sup{t > 0, IIu(t)II < R o } < oc,


we find that

Vt E [0, T), jju(t)II < MIIxjje -6 t +vM f e b ( t s) Ilu(s)11 ds.


0
158 Stability of stationary solutions

Letting cp(t) = e bt ^Iu(t)11, we obtain


t
cp(t) G C1 + C2
J p(s) ds, for all t E (0, T)
O

with C l = MIIxIl and C2 = vM.


By applying Gronwall's lemma, we deduce that

Vt E [0, T), e at lju(t)lI < MIIxIIe" Mt (10.15)

Since 8> vM, we conclude that if MI(xlI <_ R o , then T = +oo and (10.15) holds
true on [0, oo). This completes the proof of (10.14).

Remark 10.2.3. It is not sufficient for our purposes to state Theorem 10.2.2
with c < 0 (in which case, T(t) itself is a contraction semigroup). Indeed, in the
examples given below in 10.3, the generator of the linearized equation will not
be dissipative in general, especially when working in C0(S2).

10.3. Exponentially stable systems governed by POE

In this paragraph, we show how the stability theorem 10.2.2 can be applied to
partial differential equations.

(a) We first consider the semilinear heat equation

u t -Au+f(u)=0 inR xS2,


(10.16)
u=0 onR + x9Q

where Il is a bounded domain in R N and f is a function of class C l : ]R -4 J


such that
f(0) = 0 and f'(0) > -Ai, (10.17)

where A l = A 1 (fl) is the smallest eigenvalue of (-0) in Ho (Sl). We have the


following simple result.

Proposition 10.3.1. Under the above hypotheses, the stationary solution u


0 of (10.16) is exponentially stable in X = Co(1l) in the following sense: for each
y E (0, A l + f'(0)), there exists R = R(-y) such that for all x E X with lixil < R,
the solution u of (10.16) such that u(0) = x exists and is global, and satisfies

Vt > 0, u(t)I < MiIxIIe - ry t , (10.18)

with M independent of -y and x.


Exponentially stable systems governed by PDE 159

Proof. We have shown in Corollary 3.5.10 that the contraction semigroup To(t)
generated in C o (12) by the equation

JSlu=0
u -Au=0 inR + x52,
t

onIII+x81

satisfies (10.10) with S = Al and some M > 1. It is therefore sufficient to


apply Theorem 10.2.2 with T(t) = e - f' () To(t), since for f E C 1 (R), F(u) _
f (u) - f'(0)u satisfies (10.11) with a = 0 and v arbitrarily small.

(b) Another situation: this time we assume some conditions which are in a sense
opposite to (10.17):

f is strictly convex on [0, oo) and f (0) = 0, fd(0) < -A1(S2) (10.19)

where A 1 (l) is the smallest eigenvalue of (-0) in Ho (52). Here the solution 0
i
I
is unstable and we have the following.

Theorem 10.3.2. (i) There exists one and only one positive solution cp of the
problem:
cp E X n Ho (52), -O p + f(p) = 0. (10.20)

(ii) For each uo E X, no > 0 and not identically 0, the solution u of (10.16)
I
such that u(0) = uo tends to cp as t -> oo. Moreover, we have

Vt > 0, jIu(t, .) - cp(t, .)IILo < C(uo)exp(-yt), (10.21) I


I
where -y> 0 is independent of uo.

Proof. The proof is divided into several steps.


Step 1. We already know by Theorem 9.4.2 and the positivity preserving
property that the solution u(t,.) asymptotes towards the set of non-negative
solutions of (10.20) as t -> oo. We now show that if no # 0, u(t,.) cannot tend

u
to 0 as t - oc. Indeed, assuming that lim j^u(t, .) II L= = 0, then, for each a > 0,
coo
there is T(e) such that

Vt > T(e), f (u(t, x)) < { fd(0) + e}u(t, x) on 52.

Choosing e > 0 so small that -f(0) - e - A 1 (1l) > 0, multiplying the equation
by a positive eigenfunction cp l corresponding to the first eigenvalue A1(Q) of
(-Li) in Ho (1), and then integrating over 52, we find,

d
ju(t,x)^o,(x)dx>O for all t > T(e).
dt
160 Stability of stationary solutions

1 Since the function: t ---> fs , u(t, x)cpl (x) dx is non-decreasing on [T(E), oo) and
tends to 0 as t --> oo, it must vanish identically on [T(E), oo). Because eO 1 is
positive on S2, this implies that u(t,.) = 0 for all t >_ T(E). Then a classical
connectedness argument shows that u0 = 0. Therefore if n o # 0, the w-limit set
of u0 under S(t) contains at least a non-negative solution cp # 0 of (10.20).

Step 2. By the strong maximum principle, we must have f() <0 and then

I cp > 0 in Q. We now prove the following.

Lemma 10.3.3. Let f be as above and cp > 0 in 1 be a solution of the equation

I E C(1l) n Ho (S2), A^p + f (gyp) = 0.

On the other hand, let 0 > 0 satisfy

' E C(S2) n Ho (S2), AV) + f (0) ? 0.

Then either = 0 or b >_ cp.

Proof. We first establish

Vw E Ho (1), + k(ep)w 2 }dx >_ 0, with k(cp) := f(o)/co. (10.22)


J {IVwI 2

In fact, denoting by D(fl) the set of real-valued C functions with compact


support in 52, we have the following sequence of identities

VW E D(Q), f{VwI 2 + k()w 2 }dx =


z
{Ow f 2 + (A/)w 2 }dx

=
f {^Vw^ 2 DAP V(w2/ )}dx.

Since V (w 2 /cp) = 2(wVw/cp) (w 2 / p 2 )O p, we obtain the formula

Vw E D(S2),
J {^^w^ 2 + [f (p)/ p^w 2 }dx =
J IVw (w/cp)V pI 2 dx. (10.23)

This establishes (10.22) when w E V(1). Then, by passing to the limit in (10.22)
in the sense of Ho (52) along a sequence of functions w,,, E V(l) tending to w,
we find that

' bw E Ho (1), f {IVwI 2 + [f (cp)/cp]w 2 }dx >_ 0.

We may, in particular, use (10.22) with w = (p )+ E C(SZ) n Ho (Sl). On the


other hand, we have, by the properties of cp and V):

-o(v^ - 0) + f (^) - f (VG) <_ 0.


Exponentially stable systems governed by PDE 161
]

Multiplying by w = (cp 0)+ and integrating over S2, we find that

I
{ V w l 2 + [(f() f())/( - 0)]w ' }dx < 0. (10.24)

By combining (10.22) and (10.24), we finally obtain

f^{[(f (^P) f())/( V)] [f ( ,P)/ ]}w 2 dx < 0. (10.25)

As a consequence of strict convexity of f, (10.25) now implies that

wO = 5(cp V,)+ = 0, everywhere in f. (10.26)

Since cp > 0 in S2 and E C(Il), the conclusion follows at once from (10.26).

Lemma 10.3.3 implies in particular the uniqueness part of (i). By combining


this with Step 1 we conclude that w(u o ) _ { p}, which means that u(t,.) tends
tocpast >oo.

Step 3. We now establish (10.21). We begin with the identity

dtfin - I 2 dx = -2f {IV(u -x)12 + f( ) _ V(^) lu - 2 }dx. (10.27)

From the convergence of u(t,.) to cp in X, it follows in particular that, fixing


some non-empty open set w contained in a compact subset of 1, we have for
t > T (depending on the solution u),

Vt > T, u(t, x) ? (1/2)cp(x) in w. (10.28)


Now, from (10.27) and (10.28), we easily deduce the inequality

dJt Ju 2 dx < 2
J {V(u 2 + c(x)iu cp^ 2 }dx, (10.29)

with
c(x) 2
.f (P) (P/2) if x E w
0
if x ^{ W.

The result will now become a consequence of the following lemma.

Lemma 10.3.4. There exists 6 > 0 so that, c(x) being given as above, we
have the inequality

Vw E Ho (S2), f {iVwi 2 + c(x)w 2 }dx >_ 6


J w 2 dx.
f1
(10.30)
162 Stability of stationary solutions

Proof. We introduce

6 = inf {
J {IVwI 2 + c(x)w 2 }dx, w E H0'(), f^
52 w 2 d x = 1 } .
JJ
(10.31)
l ^z

Since c E L(1l), a standard argument shows that the infimum in (10.31) is


achieved for w = ( >_ 0, where (E Ho (S2) satisfies f ri ( 2 dx = 1 and is a solution
of the elliptic problem

( E C(1l) n H' (l), A( + c(x)( = 6C. (10.32)

Multiplying by cp and integrating over S2, we immediately obtain

6
J Ccp dx = J (A(+ c(x)C)cp dx = jo (Acc(+ c(x)Ccp)dx
= f[c(x) k(^G)(x)]C(x)^P(x) dx,

where k(cp) = f(cp)/cp. By the strict convexity of f, it now follows that c(x)
k(cp)(x) >_ 0 in Il and c(x) k(cp)(x) > 0 in w. In addition, we have C > 0
everywhere in Il by (10.32) and the strong maximum principle: in particular,
we find 6 > 0. The result (10.30) follows at once by homogeneity.
Proof of Theorem 10.3.2 continued. We deduce from (10.29) and (10.30) the
simple inequality

d
d (Mu(t) GII) 2611 u(t, ) (10.33)

From (10.33), we first deduce that

Vt > T, IIu(t, ) W112 < II u(T, ) (112

In fact, (10.27) and the convexity of f also implies that II u(t, .) W11 2 is non-
increasing; hence

Vt > T, JIu(t, ) W112 < exp(6T) Iluo w112 exp(St)


Kuo cp^^. exp(bt), (10.34)

for some K > 0. Then, since u and z remain bounded in C 1 , from (10.34), we
deduce that
Vt > 0, IIu(t, .) cp^^. < C(uo) exp( yt), (10.35)

by replacing 6 by a slightly smaller positive constant, denoted by y. Hence


Theorem 10.3.2 is completely proven.
Exponentially stable systems governed by PDE 163

Remark 10.3.5. The main result of Theorem 10.3.2 can be viewed as a prop-
erty of global exponential stability of the positive stationary solution cp(x) in the
metric space Z \ {0} = {u E Co (); u >_ 0, u # 0). Here, three remarks are in
order.
1. The constant C(uo) in (10.35) does not remain bounded with IIuoIIL In
fact, let A > 0 arbitrary and select t = T such that exp( -yT)II pII L > A. By
letting uo --> 0 in Co(1), we deduce from (10.5) with t = T the estimate

lim inf{C(uo), uo + 0 in C o (S1)} > A.


I
Since A is arbitrary, we conclude: lim{C(u o ), uo > 0 in C0 (S2)} = oo.

2. Assuming that f : JR 1[8 is odd, locally Lipschitz continuous with
1
f (s) >_ 0 for s > oo, and satisfies (10.19), it also follows from the proof of
Theorem 10.3.2 that the positive stationary solution cp(x) is exponentially stable
in the larger space C 0 (1). Indeed, the linearized equation around u = cp(x) is

z t Oz+f'(cp)z=0 inlR xS2,


z=0 onR + xaSl
(10.36) I
which, using the convexity off on R+, turns out to be exponentially damped in
Co (cl) by the method of Lemma 10.3.4.
3. Theorem 10.3.2 and the two remarks above are applicable, as a typical
case, to the non-linearity
f (u) = clul au Au (10.37)
for some positive constants c, a, and A.

(c) Similarly, we can consider the semilinear wave equation

u tt Au + f (u) +.\ut = 0 in R x Q; u = 0 on I[8 + x aci (10.38)

where S2 is a bounded domain in R N f is a function of class C 1 : JR --> JR such


,

that
f (0) = 0 and f'(0) > A r , (10.39)
in which f is a locally Lipschitz continuous function: 1[8 -4 JR with f (0) = 0
satisfying the growth condition

If(u)I <C(1+Iul') a.e.onJR (10.40)

withr>0arbitraryifN=1or2and0<r<N/(N-2)ifN>3.

We obtain the following result.


I

164 Stability of stationary solutions

Proposition 10.3.6. Under the above hypotheses, the stationary solution


(u, v) - (0, 0) of (10.38) is exponentially stable in X = Ho (S2) x L 2 () in the
following sense: there exist 6 > 0 and R = R(6) such that, for all x E X with
IIxMI < R, the solution u of (10.38) such that u(0) = x is global and satisfies

Vt > 0, JJu(t)II < M(6)^JxJIexp(-6t). (10.41)

Proof. It is well known, and this can be deduced from the proof of Theo-
rem 8.4.5 and Lemma 8.4.6 with H = 0, that the contraction semigroup T0 (t)
generated in X = Hp (S2) x L 2 (cl) by the equation

uttL\u+.\ut =0 inRx
+ 1l,
{u=0 onR+xOR

satisfies (10.10). The method of proof clearly applies to the slightly more general
equation
utt Au + f'(0)u + Au t = 0 in R+ x S2,
(10.42)
I u=0 onR+x852.
In order to apply Theorem 10.2.2 with T(t) the semigroup generated by (10.42),
I all we need to check is that the function F(u, v) = (0,1(u) f'(0)u) satisfies
(10.11) with a = 0 and v arbitrarily small. But this is immediate, since the
function cp(s) = f (s) f'(0)s is o(Is) near the origin and, by (10.40), we have
I(s) J <_ C(Jst"') for s large. Therefore, for each 6 arbitrarily small, we have
< 6^s +C(6)Is, globally on R. The result then follows immediately from
the Sobolev embedding theorems.

10.4. Stability and positivity

We consider the semilinear parabolic equation (10.16), where f is a locally Lip-


' schitz continuous function on the reals. If u is solution of (10.16) uniformly
bounded on ][8+ x SI, the solution u(t,.) asymptotes towards the set of station-
ary solutions as t - oo. In particular, the existence of a bounded trajectory
implies the existence of at least one solution to the elliptic problem (10.20).
Roughly speaking, the stability of a solution cp of (10.20) rests on the sign of
the first eigenvalue 77 = A 1 (L + f'(cp)I) in the sense of Ho (1); more precisely,
if 77 > 0, <p is Liapunov-stable in the uniform norm as well as in any reasonable
usual norm, while if r < 0, cp is not Liapunov-stable. In the more specific case
in which f is strictly convex on [0, oo) with f(0) = 0 and fa (0) < . 1 (cl), where
) 1 (Sl) is the smallest eigenvalue of (-0) in Ho (S2), we have proved in 10.3 that
the unique positive solution cp is in fact exponentially stable in the space Co(S2).
I This is because the strict convexity of f implies that 17 = A 1 (-0 + f'(cp)I) > 0.
Stability and positivity 165

A typical example where the exponential stability of the positive solution occurs
is when
f (u) = cl u^ a u Au

for some constants c, a > 0 and A > ) (1l). When N = 1, 52 = (0, L), and
for .A > (x/2) 2 = .11(51), non-trivial stationary solutions appear as pairs of
opposite functions that have a finite number (n + 2) of zeroes equally spaced
on [0, L] with n < (L/ir)A 1 / 2 1, built from positive solutions of the same
problem on (0, L/(n + 1)). A new pair of solutions appears when the increasing
positive parameter A crosses an eigenvalue (kir/L) 2 = .\k(0, L) of the operator
(u xy ) in Ho (0, L). It has been known for some time (cf., e.g., Chafee and
Infante [1]) that the only stationary solutions that are stable in the sense of
Liapunov are the positive and the negative solution. In higher dimensions the
situation seems much more intricate, but it is still of interest to investigate the
relationship between stability and the absence of zeroes. For instance, in the case
of Neumann boundary conditions, a result of Casten and Holland [1] asserts that
if S2 is star-shaped, any non-constant solution is unstable. Counter-examples of
stable solutions changing sign in f1 are known for both Dirichlet and Neumann
boundary conditions in non-convex domains. On the other hand, even for fI
convex, there is no general instability result for solutions changing sign in the
case of Dirichlet boundary conditions.

10.4.1. The one-dimensional case


Consider, as a motivation, the one-dimensional semilinear heat equation

u t u yx + f (u) = 0 in lR + x (0, L),


(10.43)
u(t, 0) = u(t, L) = 0 on R + ,

where f is a C 1 function: H # R. Any solution u of this problem which is global


and uniformly bounded on ][8+ x (0, L) converges as t + oo to a solution cp of
the elliptic problem

(p E Ho (0, L), coxx + f ( p) = 0.


, (10.44)

Proposition 10.4.1. If cp is a stable solution of (10.44), then cp has a constant


sign on (0, L).

Proof. Indeed, if cp is not identically 0 and vanishes somewhere in (0, L), the
function w := cp' has two zeroes in (0, L) and satisfies

WE C 2 ([0, L]) f1 Ha (0, L), w + f'(^p)w = 0 in (0,L).


166 Stability of stationary solutions

In particular, if 0 < a < ,3 < L are such that w(a) = w(0) = 0, w 54 0 on


(a, 3) and if we set w = (a, Q), we clearly have ) 1 (w; + f'(^p)I) = 0, where
A l (w; ^ + f'(cp)I) denotes the first eigenvalue of ,L + f'(cp)I in the sense of
Ho (w). We introduce the quadratic form J defined by

Vz E Ho (0, L) J(z) := f {zx + f'(^p)z 2 }dx.

Let
:= inf{J(z), z E Ho (0, L),
in z 2 dx = 1}.

Let us also denote, by (, the extension of w by 0 outside w. Because J() _


f^{fix+f'(p)^ 2 }dx = f^ {+f'(p)C 2 }dx = f^ {wx+f'(^p)w 2 }dx = 0, we clearly
have
71 = A (1l; . + f'(^P)I) < 0.
Assuming that 71 = 0 means that a multiple AS = 1i of ( realizes the minimum
of J and therefore is a solution of

E C 2 ([ 0 , L]) n Ho ( 0 , L), 0xx + f'(^P)V) = 0.

This is impossible since V) is not identically 0 and yet vanishes on (0, a). There-
fore 77 < 0. The deduction of instability will now follow in a few lines: assume
that cp is stable in the sense of Liapunov in Co(S2) x Ho(1l) and let u E be the
solution of (10.43) with u,(0) = cp + E0, where V) is a positive solution of
^b E C 2 n Ho ({ 0 , L]), V)x. +f'() = rl0 in (0, L).

Since ij> > 0, the order preserving property implies that u F > V. Now let
W e =u E cc>0.
Because w E --> 0 uniformly as E + 0, we have

f (uE) = f(cc) + f'(cc)we + 6(E)I wEI ,

with 6(E) ^ 0 as e --> 0. On the other hand, we have

d f w,V)(x) dx = j 0(uExx f (u E ))dx

=J O(Pxx + f V)xxw dxJ


.f (cc)') dx


s^ ro s^

Jn f'(cp)weV)dx

Js^ 6 (E)1weIV)dx
= 77
J w,V)(x)dx fn b(E)jweI0dx
(1771/ 2 ) f w,O(x) dx,
Stability and positivity 167

for all t >_ 0 and e > 0 small enough. This inequality, combined with positivity
of z() and w E together with boundedness of w E , implies that

0
=f ^
,b(z)w(0, x)dx = e

Thus Proposition 10.4.1 is proven by contradiction.


J 1p2(x)dx.
^

10.4.2. The multidimensional case

It is interesting to remark that the above instability result does not require
hypotheses on the shape of f. Even the differentiability condition can in fact be
relaxed, since the important point is just boundedness of the potential f'(cp(x)).
In higher dimensions, at present we have no such general results; however, the
previous technique can be extended to some particular cases of special interest.
First, we establish a basic lemma.

Lemma 10.4.2. Let S2 be any bounded open domain of RN, and let w be any
open sub-domain which is not dense in Il. Then, for any potential p E L(1l),
we have
Ai (I; A + p(x)I) < A i (w; A + p(x)I).

Proof. We introduce the quadratic form J, defined by

dz E Ho (1), J(z) :_ f {^Vzl 2 + p(x)z2}dx.


I
By definition,

rl := inf{J(z), z E Ho (0, L), z 2 dx = 1} = A + p(x)I).

Let w E Ho (w) denote a normalized eigenvector of A + p(x)I in Ho (w) asso-


ciated with A, (w; A + p(x)I) and let us consider the extension Ho (Sl) of
w by 0 outside w. Since

J(S) = f {IV I 2 +p(x)(2}dx = / {IV 1 2 +p(x)1 2 }dx I


fw
=
f
w
{IVwI 2 +p(x)w 2 }dx=. 1 (w; A+p(x)I),
we clearly have ?) < A, (w; A + p(x)I).
I
Assuming that 77 = a l (w; A+p(x)I) means that a real multiple of ( realizes
the minimum of J and therefore AC + p(x) = 0. This is impossible, since C
is not identically 0 and yet vanishes on a non-empty open subset of Q.
168 Stability of stationary solutions

An especially interesting special case is when ci is a rectangle in R 2 , and if f


is a C' odd function on the reals. For instance, if f (u) = cu 3 Au with a large,
in addition to the unique positive solution, (10.20) will have more complicated
solutions arising by odd extensions from the positive solutions in cellular sub-
domains which divide the domain into m x p equal rectangular regions. Such so-
lutions, which are the analogue of odd-periodic extensions in the one-dimensional
case, are also Liapunov-unstable. More precisely, we have the following.

Proposition 10.4.3. Let cp be a solution of (10.20) in a rectangle S2 = (0, a) x


(0, b) and assume that there is a sub-rectangle R = (0, a/p) x (0, b/q) with p, q
integers > 1 and max(p, q) > 1 such that cp has a constant sign in R and cp = 0
on 8R. Then cp is unstable.

Proof. Assume, for instance, that p > 1. It is clear that the trace of (p on
R' = (a/p, 2a/p) x (0, b/q) coincides with the odd reflection of cc with respect
to the line x = a/p, and in addition cp^ R is even with respect to the line x =
a/2p. Therefore, w := &,o/ax vanishes on the boundary of the sub-rectangle
w = (a/2p, 3a/2p) x (0, b/q) and satisfies

w E C 2 (w) n Ho (w), Aw + f' (p)w = 0 in w.

In particular, since w does not vanish inside w, we have A 1 (w; A+f'(p)I) _


0. Then Lemma 10.4.2 gives

1\1(52; A + f'(1P)I) < ,A 1 (w; A + f'(cc)I) = 0.

Therefore cp is unstable. u

There is another quite interesting result for the case in which S2 is a sphere
in R N : in Comte, Haraux, and Mironescu [1] the following result is obtained.

Proposition 10.4.4. If S2 is a ball in 1R', N >_ 2, and cp is a Liapunov-stable


solution of (10.20), we have either cp = 0, or (p is spherically symmetric with
constant sign in Ii.

Notes. Theorem 10.2.2 is also valid when (T(t)) t > o is a general Co-semigroup;
see Haraux [5]. The exponential stability property (10.21) can also be established
in a non-autonomous framework; see Haraux [6].

Further remarks. We have not addressed the NavierStokes equation. The


reader may consult Constantin and Foias [1], Foias and Temam [1], Fujita and
Kato [1], Giga [1, 2], Heywood [1], Ito [1], Kato [1], Ladyzhenskaya [1], Leray [1-
3], Temam [1], and Lions [3, 4]. For the KortewegDe Vries equation, see, for
example, Gardner, Greene, Kruskal, and Miura [1], and Kato [2].
Bibliography

ADAMS R. A.
[1] Sobolev spaces, Academic Press, New York, 1975.

AGUIRRE J. and ESCOBEDO M.


[1] A Cauchy problem for u t Au = uP with 0 <p < 1. Asymptotic behavior
of solutions, Ann Fac. Sci. Toulouse 8 (1986-87), 175-203.
BABIN A. V. and VISHIK M. I.
[1] Regular attractors of semi-groups and evolution equations, J. Math. Pures
Appl. 62 (1983), 441-491.
[2] Attractors of some evolution equations and estimates of their dimensions,
Uspekhi Math. Nauk. 38 (1983), 133-187 (in Russian).
[3] Attracteurs maximaux dans les equations aux derivees partielles, in Nonlinear
partial differential equations and their applications, College de France Seminar,
vol. 7, H. Brezis and J.L. Lions (eds), Research Notes in Math. #122, Pitman,
1984, 1-34.
BAILLON J.-B. and CHADAM J. M.
[1] The Cauchy problem for the coupled SchrodingerKleinGordon equations,
in Contemporary developments in continuum mechanics and partial differential
equations, North Holland Math. Studies, North Holland, 1978.

BAILLON J.-B., CAZENAVE T., and FIGUEIRA M.


[1] Equation de Schrodinger non lineaire, C. R. Acad. Sci. Paris 284 (1977),
869-872.
[2] Equation de Schrodinger avec non-linearite integrale, C. R. Acad. Sci. Paris
284 (1977), 939-942.
BALABANE M.
[1] Non existence de solutions globales pour des equations des ondes non lineaires
a donnees de Cauchy petites, C. R. Acad. Sci. Paris 301 (1985), 569-572.
[2] Ondes progressives et resultats d'explosion pour les systemes du premier
ordre, C. R. Acad. Sci. Paris 302 (1986), 211-214.
BALL J. M.
[1] On the asymptotic behavior of generalized processes, with applications to
nonlinear evolution equations, J. Diff. Eq. 27 (1978), 224-265.
170 Bibliography

[2] Finite-time blow-up in nonlinear problems, in Nonlinear evolution equations,


M. G. Crandall (ed.), Academic Press, New York, 1978, 189-205.
[3] Remarks on blow-up and nonexistence theorems for nonlinear evolution equa-
tions, Q. J. Math. Oxford 28 (1977), 473-486.
BARAS P.
[1] Non unicite des solutions d'une equation d'evolution non lineaire, Ann. Fac.
Sci. Toulouse 5 (1983), 287-302.
BARAS P. and COHEN L.
[1] Complete blow-up after T,,,, o x for the solution of a semilinear heat equation,
,

J. Funct. Anal. 71 (1987), 142-174.


BARAS P. and GOLDSTEIN J.
[1] The heat equation with a singular potential, Trans. Am. Math. Soc. 284
(1984), 121-139.
BERESTYCKI H. and CAZENAVE T.
[1] Instabilite des etats stationnaires dans les equations de Schrodinger et de
KleinGordon non lineaires, C. R. Acad. Sci. Paris 293 (1981), 489-492.
BERESTYCKI H. and LIONS P.-L.
[1] Nonlinear scalar field equations, Arch. Rat. Mech. Anal. 82 (1983), 313-375.
BERESTYCKI H., GALLOUET T., and KAVIAN O.
[1] Equations de champs scalaires Euclidiens non-lineaires dans le plan, C. R.
Acad. Sci. Paris 297 (1983), 307-310.
BERESTYCKI H., LIONS P.-L., and PELETIER L. A.
[1] An O.D.E. approach to the existence of positive solutions for semilinear
problems in IR' , Indiana Univ. Math. J. 30 (1981), 141-157.
2

BERGH J. and LOFSTROM J.


[1] Interpolation spaces, Springer, New York, 1976.
BLANCHARD P., STUBBE J. and VAZQUEZ L.
[1] On the stability of solitary waves for classical scalar fields, Ann. Inst. Henri
Poincare, Physique Theorique 47 (1987), 309-336.
BRENNER P.
[1] On LP Lp estimates for the wave equation, Math. Z. 145 (1975), 251-254.
[2] On spacetime means and everywhere defined scattering operators for non-
linear KleinGordon equations, Math. Z. 186 (1984), 383-391.
BREZIS H.
[1] Operateurs maximaux monotones et semi-groupes de contractions dans les es-
paces de Hilbert, Lecture Notes #5, NorthHolland, Amsterdam, London, 1972.
[2] Analyse fonctionnelle, thorie et applications, Masson, Paris, 1983.
Bibliography 171 1
BREZIS H. and FRIEDMAN A.
[1] Nonlinear parabolic equations involving measures as initial conditions, J.
Math. Pures Appl. 62 (1983), 73-97.
BREZIS H. and GALLOUET T.
[1] Nonlinear Schrodinger evolution equations, Nonlinear Anal. T.M.A. 4 (1980),
677-681.
BREZIS H., CORON J.-M., and NIRENBERG L.
[1] Free vibrations for a nonlinear wave equation and a theorem of P. Rabinowitz,

I
Communs. Pure Appl. Math. 33 (1980), 667-689.
BREZIS H., PELETIER L. A., and TERMAN D.
[1] A very singular solution of the heat equation with absorption, Arch. Rat.

I
Mech. Anal. 95 (1986), 185-209:
BROWDER F.
[1] On nonlinear wave equations, Math. Z. 80 (1962), 249-264.
CABANNES H. and HARAUX A.
[1] Mouvements presque-periodiques d'une corde vibrante en presence d'un ob-
stacle fixe, rectiligne ou ponctuel, Int. J. Nonlinear Mech. 16 (1981), 449-458.
[2] Almost periodic motion of a string vibrating against a straight, fixed obstacle,
Nonlinear Anal. T.M.A. 7 (1983), 129-141.
CASTEN R. J. and HOLLAND C. J.
[1] Instability results for reaction diffusion equations with Neumann boundary
conditions, J. Diff. Eq. 27 (1978), 266-273.
CAZENAVE T.
[1] Uniform estimates for solutions of nonlinear KleinGordon equations, J.
Funct. Anal. 60 (1985), 36-55.
[2] Equations de Schrodinger non lineaires en dimension deux, Proc. R. Soc.
Edin. 84 (1979), 327-346.
[3] Stable solutions of the logarithmic Schrodinger equation, Nonlinear Anal. T.
M. A. 7 (1983), 1127-1140.
[4] An introduction to nonlinear Schrodinger equations, Textos de Metodos Mate-
maticos #22, I.M.U.F.R.J., Rio de Janeiro, 1989.

CAZENAVE T. and HARAUX A.


I
[1] Equation d'evolution avec non-linearite logarithmique, Ann. Fac. Sci. Tou-
louse 2 (1980), 21-51.
[2] Oscillatory phenomena associated to semilinear wave equations in one spatial

I
dimension, Trans. Am. Math. Soc. 300 (1987), 207-233.
[3] Some oscillation properties of the wave equation in several space dimensions,
J. Funct. Anal. 76 (1988), 87-109.
172 Bibliography

[4] Introduction aux problemes d'e'volution semi-lineaires, Mathematiques et Ap-


plications #1, Ellipses, Paris, 1990.
CAZENAVE T. and LIONS P.-L.
[1] Orbital stability of standing waves for some nonlinear Schrodinger equations,
Communs. Math. Phys. 85 (1982), 549-561.
[2] Solutions globales d'equations de la chaleur semi lineaires, Communs. Part.
Diff. Eq. 9 (1984), 955-978.
CAZENAVE T. and WEISSLER F. B.
[1] Some remarks on the nonlinear Schrodinger equation in the subcritical case, in
New methods and results in nonlinear field equations, Ph. Blanchard, J.-P. Dias,
and J. Stubbe (eds), Lect. Notes in Phys. #347, Springer, 1989, 59-69.
[2] Some remarks on the nonlinear Schrodinger equation in the critical case, in
Nonlinear semigroups, partial differential equations, and attractors, T. L. Gill
and W. W. Zachary (eds), Lect. Notes in Math. # 1394, Springer, 1989, 18-29.
[3] The Cauchy problem for the nonlinear Schrodinger equation in H 1 , Manus-
cripta Math. 61 (1988), 477-494.
[4] The Cauchy problem for the critical nonlinear Schrodinger equation in Hs,
Nonlinear Anal, T.M.A. 14 (1990), 807-836.
[5] The structure of solutions to the pseudo-conformally invariant nonlinear
Schrodinger equation, Proc. R. Soc. Edin. 117A (1991), 251-273.
[6] Rapidly decaying solutions of the nonlinear Schrodinger equation, Communs.
Math. Phys. 147 (1992), 75-100.
CAZENAVE T., HARAUX A., VAZQUEZ L., and WEISSLER F. B.
[1] Nonlinear effects in the wave equation with a cubic restoring force, Comput.
Mech. 5 (1989), 49-72.
CHADAM J. M. and GLASSEY R. T.
[1] Global existence of solutions to the Cauchy problem for time-dependent
Hartree equations, J. Math. Phys. 16 (1975), 1122-1130.
CHAFEE N. and INFANTE L. I.
[1] A bifurcation problem for a nonlinear partial differential equation of parabolic
type, Applic. Anal. 4 (1974), 17-37.
COMTE M., HARAUX A., and MIRONESCU P.
[1] Multiplicity and stability topics in semilinear parabolic problems, Publ. Lab.
Anal. Numerique R96038 (1991).
CONSTANTIN P. and FOIAS C.
[1] NavierStokes equations, Chicago University Press, 1989.
CONSTANTIN P. and SAUT J.-C.
[1] Effets regularisant locaux pour des equations dispersives generales, C. R.
Acad. Sci. Paris 304 (1987), 407-410.
Bibliography 173

[2] Local smoothing properties of dispersive equations, J. Am. Math. Soc. 1


(1988), 413-439.
[3] Local smoothing properties of Schrodinger equations, Indiana Univ. Math.
J. 38 (1989), 791-810.
COURANT R. and HILBERT D.
[1] Methods of mathematical physics, vol. 2, Interscience, John Wiley, 1962.

CRANDALL M. G. and LIGGETT T.


[1] Generation of semigroups of nonlinear transformations on general Banach
spaces, Am. J. Math. 93 (1971), 265-298.
CRANDALL M. G. and PAZY A.
[1] Semigroups of nonlinear contractions and dissipative sets, J. Funct. Anal. 3
(1969), 376-418.
DAFERMOS C. M.
[1] Contraction semigroups and trend to equilibrium in continuum mechanics,
Lect. Notes in Math. #503, Springer, 1976, 295-306.
[2] Almost periodic processes and almost periodic solutions of evolution equa-
tions, Proceedings of a University of Florida International Symposium, Academic
Press, New York, 1977, 43-57.
[3] Asymptotic behavior of solutions of evolution equations, in Nonlinear evolu-
tion equations, M. G. Crandall (ed.), Academic Press, New York, 1978, 103-123.

DAFERMOS C. M. and SLEMROD M.


[1] Asymptotic behavior of nonlinear contraction semi-groups, J. Funct. Anal.
12 (1973), 97-103.
DIAS J.-P. and FIGUEIRA M.
[1] Conservation laws and time decay for the solutions to some nonlinear Schro-
dingerHartree equations, J. Math. Anal. Appl. 84 (1981), 486-508.
DIAS J.-P. and HARAUX A.
[1] Smoothing effect and asymptotic behavior for the solutions of a nonlinear
time dependent system, Proc. R. Soc. Edin. 87 (1981), 289-303.
DINCULEANU N.
[1] Vector measures, Pergamon Press, New York, 1967.

DUNFORD C. and SCHWARTZ J. T.


[1] Linear operators, Interscience, New York, 1958.

ESCOBEDO M. and KAVIAN O.


[1] Variational problems related to self-similar solutions of the heat equation,
Nonlinear Anal. T.M.A. 11 (1987), 1103-1133.
174 Bibliography

[2] Asymptotic behavior of positive solutions of a nonlinear heat equation, Hous-


ton J. Math. 13 (1987), 39-50.

ESTEBAN M. J.
[1] On periodic solutions of superlinear parabolic problems, Trans. Am. Math.
Soc 293 (1986), 171-189.
[1] A remark on the existence of positive periodic solutions of superlinear para-
bolic problems, Proc. Am. Math. Soc. 102 (1987), 131-136.

FIFE P. C.
[1] Mathematical aspects of reacting and diffusing systems, Lecture Notes in
Biomathematics #28, Springer, New York, 1979.

FOIAS C. and TEMAM R.


[1] Some analytic and geometric properties of the solutions of the NavierStokes
equations, J. Math. Pure Appl. 58 (1979), 339-368.

FRIEDLANDER L.
[1] An invariant measure for the equation u tt u XX + u 3 = 0, Communs. Math.
Phys. 98 (1985), 1-16.

FRIEDMAN A.
[1] Partial differential equations of parabolic type, Prentice-Hall, 1964.
FRIEDMAN A. and GIGA Y.
[1] A single point blow-up for solutions of semilinear parabolic equations, J. Fac.
Sci. Univ. Tokyo, Sect. IA, 34 (1987), 65-79.

FRIEDMAN A. and McLEOD B.


[1] Blow-up of positive solutions of semilinear heat equations, Indiana Univ.
Math. J. 34 (1985), 425-447.

FUJITA H.
l
[1] On the blowing-up of solutions of the Cauchy problem for u t = L u + u+
J. Fac. Sci. Univ. Tokyo 13 (1966), 109-124.

FUJITA H. and KATO T.


[1] On the NavierStokes initial value problem, Arch. Rat. Mech. Anal. 16
(1964), 269-315.

GARDNER C. S., GREENE J., KRUSKAL M. D., and MIURA R. M.


[1] KortewegDe Vries equation and generalizations, methods for exact solutions,
Communs. Pure Appl. Math. 28 (1974), 97-133.

GHIDAGLIA J.-M. and TEMAM R.


[1] Proprietes des attracteurs associes a des equations des ondes amorties, C. R.
Acad. Sci. Paris 300 (1985), 185-188.
Bibliography 175

[2] Attractors for damped nonlinear hyperbolic equations, J. Math. Pures Appl.
66 (1987), 233-319.

GIGA Y.
[1] The Navier-Stokes initial value problem in LP and related problems, in Lec-
ture Notes in Num. Appl. Anal., 5 (1982), 37-54.
[2] Weak and strong solutions of the Navier-Stokes initial value problem, Publ.
Res. Inst. Math. Sci. (Kyoto Univ.) 19 (1983), 887-910.
[3] A bound for global solutions of semilinear heat equations, Communs. Math.
Phys. 103 (1986), 415-421.
GIGA Y. and KOHN R. V.
[1] Asymptotically self-similar blow-up of semilinear heat equations, Communs.
Pure Appl. Math. 38 (1985), 297-319.
[2] Characterizing blow up using similarity variables, Indiana Univ. Math. J.
36 (1987), 1-40.
[3] Nondegeneracy of blowup for semilinear heat equations, Communs. Pure
Appl. Math. 62 (1989), 845-885.
[4] Removability of blowup points for semilinear heat equations, in Differential
equations, Xanthi, 1987, G. Papanicolaou (ed.), Lecture Notes in Pure and
Appl. Math. 118, Marcel Dekker, New York, 1989, 257-264.
GILBARG D. and TRUDINGER N. S.
[1] Elliptic partial differential equations of the second order, Springer-Verlag,
Berlin, 1983.
GINIBRE J. and VELO G.
[1] On a class of nonlinear Schrodinger equations, J. Funct. Anal. 32 (1979),
1-71.
[2] On a class of nonlinear Schrodinger equations, special theories in dimensions
1, 2 and 3, Ann. Inst. Henri Poincare 28 (1978), 287-316.
[3] On a class of nonlinear Schrodinger equations with non local interaction,
Math. Z. 170 (1980), 109-136.
[4] On the global Cauchy problem for some nonlinear Schrodinger equations,
Ann. Inst. Henri Poincare, Analyse Non Lineaire 1 (1984), 309-323.
[5] The global Cauchy problem for the nonlinear Schrodinger equation revisited,
Ann. Inst. Henri Poincare, Analyse Non Lineaire 2 (1985), 309-327.
[6] The global Cauchy problem for the nonlinear Klein-Gordon equation, Math.
Z. 189 (1985), 487-505.
[7] Scattering theory in the energy space for a class of nonlinear Schrodinger
equations, J. Math. Pure Appl. 64 (1985), 363-401.
[8] Time decay of finite energy solutions of the nonlinear Klein-Gordon and
Schrodinger equations, Ann. Inst. Henri Poincare, Physique Theorique 43
(1985), 399-442.
176 Bibliography

[9] The global Cauchy problem for the nonlinear Klein-Gordon equation II, Ann.
Inst. Henri Poincare, Analyse Non Lineaire 6 (1989), 15-35.
[10] Conformal invariance and time decay for nonlinear wave equations, Ann.
Inst. Henri Poincare, Physique Theorique 47 (1987), 221-276.
[11] Sur une equation de Schrodinger non-lineaire avec interaction non-locale, in
Nonlinear partial differential equations and their applications, College de France
Seminar, vol. 2. H. Brezis and J.L. Lions (eds), Research Notes in Math. #60,
Pitman, 1982, 155-199.
[12] Scattering theory in the energy space for a class of nonlinear wave equations,
Communs. Math. Phys. 123 (1989), 535-573.
GLASSEY R. T.
[1] Blow up theorems for nonlinear wave equations, Math. Z. 132 (1973), 183-
203.
[2] On the blowing up of solutions to the Cauchy problem for nonlinear Schro-
dinger equations, J. Math. Phys. 18 (1977), 1794-1797.
[3] Finite-time blow up for solutions of nonlinear wave equations, Math. Z. 177
(1981), 323-340.
GOLDSTEIN J.
[1] Semigroups of linear operators and applications to partial differential equa-
tions, Oxford University Press, New York, 1986.
GRILLAKIS M., SHATAH J. and STRAUSS W. A.
[1] Stability theory of solitary waves in the presence of symmetry I, J. Funct.
Anal. 74 (1987), 160-197.
HALE J. K.
[1] Ordinary differential equations, Wiley-Interscience, New York, 1969.
[2] Asymptotic behavior and dynamics in infinite dimensions, in Nonlinear dif-
ferential equations, Hale J. K. and Martines-Amores P. (eds), Research Notes in
Math. #132, Pitman, 1985, 1-42.
HALE J. K. and RAUGEL G.
Convergence in gradient-like systems with applications to PDE, Z. Math. Phys.
43 (1992), 63-124.
HANOUZET B. and JOLY J.-L.
[1] Explosion pour des problemes hyperboliques semi lineaires avec second mem-
bre compatible, C. R. Acad. Sci. Paris 301 (1985), 581-584.
HARAUX A.
[1] Nonlinear evolution equations: global behavior of solutions, Lecture Notes in
Math. #841, Springer, 1981.
[2] Semi-linear wave equations in bounded domains, Mathematical reports, J.
Dieudonne (ed.), Vol. 3, Part 1, Harwood, London, 1987.
Bibliography 177

[3] Linear semi-groups in Banach spaces, in Semigroups, theory and applications,


H. Brezis, M. G. Crandall, and F. Kappel (eds), Pitman Research Notes in
Mathematics series #152, Longman, 1986, 93-135.
[4] Nonlinear vibrations and the wave equation, Textos de Metodos Matematicos
#20, I.M.U.F.R.J., Rio de Janeiro, 1987.
[5] Systemes dynamiques dissipatifs et applications, R.M.A. #17, P. G. Ciarlet
and J.-L. Lions (eds), Masson, Paris, 1991.
[6] Exponentially stable positive solutions to a forced semilinear parabolic equa-
tion, Asymptotic Anal. 7 (1993), 3-13.
HARAUX A. and KIRANE M.
[1] Estimations C l pour des problemes paraboliques semi-lineaires, Ann. Fac.
Sci. Toulouse 5 (1983), 265-280.
HARAUX A. and POLACIK P.
[1] Convergence to a positive equilibrium for some semilinear evolution equations
in a ball, Acta. Math. Univ Comenianae LXI 2 (1993), 129-141.
HARAUX A. and WEISSLER F. B.
[1] Non uniqueness for a semilinear initial value problem, Indiana Univ. Math.
J. 31 (1982), 167-189.
HARAUX A. and YOUKANA A.
[1] On a result of K. Masuda concerning reactiondiffusion equations. Tohoku
Math. J. 40 (1988), 159-163.
HAYASHI N.
[1] Classical solutions of nonlinear Schrodinger equations, Manuscripta Math.
55 (1986), 171-190.
[2] Asymptotic behavior of solutions to time-dependent Hartree equations, Arch.
Rat. Mech. Anal. 100 (1988), 191-206.
HAYASHI N. and OZAWA T.
[1] Time decay of solutions to the Cauchy problem for time-dependent Schro-
dingerHartree equations, Communs. Math. Phys. 110 (1987), 467-478.
[2] Smoothing effect for some Schrodinger equations, J. Funct. Anal. 85 (1989),
307-348.
HAYASHI N. and TSUTSUMI M.
[1] Classical solutions of nonlinear Schrodinger equations in higher dimensions,
Math. Z. 177 (1981), 217-234.
HAYASHI N. and TSUTSUMI Y.
[1] Scattering theory for Hartree type equations, Ann. Inst. Henri Poincare,
Physique Theorique 46 (1987), 187-213.
178 Bibliography

HAYASHI N., NAKAMITSU K., and TSUTSUMI M.


[1] On solutions of the initial value problem for the nonlinear Schrodinger equa-
tions in one space dimension, Math. Z. 192 (1986), 637-650.
[2] On solutions of the initial value problem for the nonlinear Schrodinger equa-
tions, J. Funct. Anal. 71 (1987), 218-245.
[3] Nonlinear Schrodinger equations in weighted Sobolev spaces, Funk. Ekva 31
(1988), 363-381.

HEINZ E. and VON WAHL W.


[1] Zu einem Satz von F. E. Browder uber nichtlineare Wellengleichingen, Math.
Z. 141 (1974).

HENRY D.
[1] Geometric theory of semilinear parabolic equations, Lecture Notes in Math.
#840, Springer, New York, 1981.

HERRERO M. A. and VELAZQUEZ J. J. L.


[1] Blow-up behaviour of one-dimensional semilinear parabolic equations, Ann.
Inst. Henri Poincare, Analyse Non Lineaire 10 (1993), 131-189.
[2] Flat blow-up in one-dimensional semilinear heat equations, Diff. Int. Eq. 5
(1992), 973-997.
[3] Blow-up profiles in one-dimensional, semilinear parabolic problems, Com-
muns. Part. Diff. Eq. 17 (1992), 205-219.
[4] Plane structures in thermal runaway, Israel J. Math. 81 (1993), 321-341.
[5] Generic behavior of one-dimensional blow-up patterns, Ann. Scuola Norm.
Sup. Pisa 19 (1992), 381-450.
[6] Approaching an extinction point in one-dimensional semilinear heat equations
with strong absorption, J. Math. Anal. Appl. 170 (1992), 353-381.
[7] Some results on blow up for some semilinear parabolic problems, in Degen-
erate diffusion (Minneapolis, 1991), IMA Vol. Math. Appl. 47, Springer, New
York, 1993, 105-125.

HEYWOOD J. G.
[1] The NavierStokes equations: On the existence, regularity and decay of so-
lutions, Indiana Univ. Math. J. 29 (1980), 639-681.

ITO S.
[1] The existence and the uniqueness of regular solution of non-stationary
NavierStokes equation, J. Fac. Sci. Univ. Tokyo 9 (1961), 103-140.

JOHN F.
[1]Blow-up for quasilinear wave equations in three space dimensions, Communs.
Pure Appl. Math. 34 (1981), 29-52.
[2] Formation of singularities in one-dimensional nonlinear wave propagation,
Communs. Pure Appl. Math. 27 (1974), 377-405.
u

Bibliography 179

[3] Blow-up of solutions of nonlinear wave equations in three space dimension,


Manuscripta Math. 28 (1979), 235-268.
I
JONES C. and KUPPER T.
[1] On the infinitely many solutions of a semilinear elliptic equation, SIAM J.

I
Math. Anal. 17 (1986), 803-836.
JONES C. K. R. T.
[1]An instability mechanism for radially symmetric standing waves of a nonlinear
Schrodinger equation, J. Diff. Eq. 71 (1988), 34-62.
[2] Instability of standing waves for nonlinear Schrodinger equations, Ergod. Th.
Dynam. Sys. 8 (1988), 119-138.
JORGENS K.
[1] Das Anfangswertproblem in grossen fur eine klasse nichtlinearer Wellengle-
ichungen, Math. Z. 77 (1961), 295-308.
KATO T.
[1] Quasi-linear evolution equations, with applications to partial differential equa-
tions, Lecture Notes in Math. #448, Springer, New York, 1975.
[2] On the Cauchy problem for the (generalized) KdV equation, Advances in
Math. Supplementary studies, Studies in Applied Mathematics 8 (1983), 93-
128.
[3] On nonlinear Schrodinger equations, Ann. Inst. Henri Poincare, Physique
Theorique 46 (1987), 113-129.
KAVIAN O.
[1] A remark on the blowing-up of solutions to the Cauchy problem for nonlinear
Schrodinger equations, Trans. Am. Math. Soc. 299 (1987), 193-205.
[2] Remarks on the time behaviour of a nonlinear diffusion equation, Ann. Inst.
Henri Poincare, Analyse Non Lineaire 4 (1987), 423-452.
KELLER C.
[1] Stable and unstable manifolds for the nonlinear wave equation with dissipa-
tion, J. Diff. Eq. 50 (1983), 330-347.
KELLER J. B.
[1] On solutions of nonlinear wave equations, Communs. Pure Appl. Math. 10
(1957), 523-530.
KIRANE M. and TRONEL G.
[1] Regularite C de solutions de problemes paraboliques semi-lineaires, Pub-
lication du Laboratoire d'Analyse Numerique #84011, Univ. P. et M. Curie,
Paris, 1984.
180 Bibliography

KLAINERMAN S. and PONCE G.


[1] Global, small amplitude solutions to non-linear evolution equations, Com-
muns. Pure Appl. Math. 36 (1983), 133-141.
LADYZHENSKAYA O. A.
[1] The mathematical theory of viscous incompressible flow, Gordon and Breach,
1969.
[2] On the determination of global B-attractors for semi-groups generated by
boundary-value problems for nonlinear dissipative partial differential equations,
preprint, Steklov Inst., Leningrad, 1987 (in Russian).
LADYZHENSKAYA O. A., SOLONIKOV V. A., and URAL'CEVA N. N.
[1] Linear and quasilinear equations of parabolic type, Am. Math. Soc., 1968.

LANGE H.
[1] On some nonlinear Schrodinger equations of Hartree type, Math. Verf. Math.
Physik Ba. 21 (1980), 99-111.
LASALLE J. P.
[1] Asymptotic stability criteria, Proc. Symp. Appl. Math. 13 (1962), 299-307.
LERAY J.
[1] Etude de diverses equations integrales non lineaires et quelques problemes
que pose 1'hydrodynamique, J. Math. Pures Appl. 12 (1933), 1-82.
[2] Essai sur les mouvements plans d'un fluide visqueux que limitent des parois,
J. Math. Pures Appl. 13 (1934), 331-418.
[3] Sur le mouvement d'un fluide visqueux emplissant 1'espace, Acta Math. 63
(1934), 193-248.
LEVINE H. A.
[1] Some nonexistence and instability theorems for formally parabolic equations
of the form Pu t = - Au + f (u), Arch. Rat. Mech. Anal. 51 (1973), 371-386.
[2] Instability and nonexistence of global solutions to nonlinear wave equations
of the form Putt = - Au + f (u), Trans. Am. Math. Soc. 192 (1974), 1-121.
[3] Nonexistence of global weak solutions to some properly and improperly posed
problems of mathematical physics: the method of unbounded Fourier coeffi-
cients, Math. Anal. 214 (1975), 205-220.
LIN J. E. and STRAUSS W. A.
[1] Decay and scattering of solutions of a nonlinear Schrodinger equation, J.
Funct. Anal. 30 (1978), 245-263.
LIONS P.-L.
[1] Asymptotic behavior of some nonlinear heat equations, Nonlinear Phenom-
ena, Physica D5 (1982), 293-306.
Bibliography 181

[2] Structure of the set of steady-state solutions and asymptotic behaviour of


semilinear heat equations, J. Diff. Eq. 53 (1984), 362-386.
[3] Mathematical topics in fluid mechanics, Vol. 1: Incompressible models, Ox-
ford Science Publications, 1996.
[4] Mathematical topics in fluid mechanics, Vol. 2: Compressible models, Oxford
Science Publications, 1998.
} LOUZAR T.
[1] Comportement a l'infini des solutions radiales d'une equation de la chaleur
non lineaire, Thesis, Univ. P. et M. Curie, Paris, 1984.
MARCUS M. and MIZEL V.J.
[1] Every superposition operator mapping one Sobolev space into another is
} continuous, J. Funct. Anal. 33 (1979), 217-229.
MARSHALL B., STRAUSS W. A., and WAINGER S.
[1] LP-L 9 estimates for the Klein-Gordon equation, J. Math. Pures Appl. 59
(1980), 417-440.
MASUDA K.
[1] On the global existence and asymptotic behaviour of solutions of reaction-
diffusion equations, Hokkaido Math. J. 12 (1983), 360-370.
MATANO H.
[1] Asymptotic behavior and stability of solutions of semilinear diffusion equa-
tions, Pub!. Res. Inst. Math. Sci. (Kyoto Univ.), 15 (1979), 401-454.
MORAWETZ C. and STRAUSS W. A.
[1] Decay and scattering of solutions of a nonlinear relativistic wave equation,
Communs. Pure Appl. Math., 25 (1972), 1-31.
MUELLER C. E. and WEISSLER F. B.
[1] Single point blow-up for a general semilinear heat equation, Indiana Univ.
Math. J. 35 (1986), 881-913.
NI W. M., SACKS P. E. and TAVANTZIS J.
[1] On the asymptotic behavior of solutions of certain quasilinear parabolic equa-
tions, J. Diff. Eq., 54 (1984), 97-120.
PAYNE L. and SATTINGER D. H.
[1] Saddle points and instability of nonlinear hyperbolic equations, Israel J.
Math. 22 (1975), 273-303.
PAZY A.
[1] Semi-groups of linear operators and applications to partial differential equa-
tions, Applied Math. Sciences #44, Springer, New York, 1983.
182 Bibliography

POLACIK P. and RYBAKOWSKI K.


[1] Nonconvergent bounded trajectories in semilinear heat equations, J. Diff. Eq.
124 (1983), 525-537.
PROTTER M. H. and WEINBERGER H. F.
[1] Maximum principles in differential equations, Prentice-Hall, 1967.

RABINOWITZ P. H.
[1] Periodic solutions of nonlinear hyperbolic partial differential equations, Com-
muns. Pure Appl. Math. 20 (1967), 145-205.
[2] Free vibrations for a semi-linear wave equation, Communs. Pure. Appl.
Math. 31 (1978), 31-68.
REED M. and SIMON B.
[1] Methods of modern mathematical physics, Vol. II, Academic Press, 1975.
SCHOCHET H. and WEINSTEIN M. I.
[1] The nonlinear Schrodinger limit of the Zakharov equations governing Lang-
muir turbulence, Communs. Math. Phys. 106 (1986), 569-580.
SEGAL I. E.
[1] Nonlinear semi-groups, Ann. Math. 2 (1963), 339-364.
SELL G.
[1] Non autonomous differential equations and topological dynamics, Trans. Am.
Math. Soc. 127 (1967), 241-283.
SERRE D.
[1] A propos des invariants de diverses equations d'evolution, in Nonlinear partial
differential equations and their applications, College de France Seminar, vol. 3,
H. Brezis and J.-L. Lions (eds), Research Notes in Math. #70, Pitman, 1981,
326-336.
SHATAH J. and STRAUSS W. A.
[1] Instability of nonlinear bound states, Communs. Math. Phys. 100 (1985),
173-190.
SIDERIS T.
[1] Global behaviour of solutions to nonlinear wave equations in three space
dimensions, Communs. Part. Diff. Eq. 8 (1983), 1291-1323.
[2] Formation of singularities in solutions to nonlinear hyperbolic equations,
Arch. Rat. Mech. Anal. 86 (1984), 369-381.
[3] Nonexistence of global solutions to semilinear wave equations in high dimen-
sions, J. Diff. Eq. 52 (1984), 378-406.
[4] Decay estimates for the three dimensional Klein-Gordon equation and appli-
cations, Communs. Part. Diff. Eq. 14 (1989), 1421-1455.
M

Bibliography 183

C^
SILI A.
[1] These de 3 Cycle, Universite Pierre et Marie Curie, Paris, 1987.
SIMON L.
[1] Asymptotics for a class of nonlinear evolution equation equations, with ap-
plications to geometric problems, Ann. Math., 118 (1983), 525-571. G '

SJOLIN P.
[1] Regularity of solutions to Schrodinger equations, Duke Math. J. 55 (1987),
699-715. 1
I
SMOLLER J.
[1] Shock waves and reaction diffusion systems, Springer, New York, 1983.
STRAUSS W. A.

I
[1] On weak solutions of semilinear hyperbolic equations, An. Acad. Brasil.
Cien. 42 (1970), 645-651.
[2] Nonlinear invariant wave equations, Lecture Notes in Physics #73, Springer,

I
1978.
[3] Nonlinear wave equations, Regional Conference Series in Mathematics #73,
Am. Math. Soc., Providence, 1989.

J
STRAUSS W. A. and VAZQUEZ L.
[1] Numerical solution of a nonlinear Klein-Gordon equation, J. Comput. Phys.
28 (1978), 271-278.

STRICHARTZ M.
[1] Restrictions of Fourier transforms to quadratic surfaces and decay of solutions I
u
of wave equations, Duke Math. J. 44 (1977), 705-714.
TEMAM R.
[1] NavierStokes equations, North Holland, 1977.
[2] Infinite dynamical systems in mechanics and physics, Appl. Math. Sci. #68,
Springer, New York, 1988.
TSUTSUMI M. I
[1] Nonexistence of global solutions to the Cauchy problem for the damped
nonlinear Schrodinger equation, S.I.A.M. J. Math. Anal. 15 (1984), 357-366.
TSUTSUMI Y.
I
I
[1] Scattering problem for nonlinear Schrodinger equations, Ann. Inst. Henri
Poincare, Physique Theorique 43 (1985), 321-347.
[2] Global solutions of the nonlinear Schrodinger equation in exterior domains,

n
Communs. Part. Diff. Eq. 8 (1984), 1337-1374.
[3] Local energy decay of solutions to the free Schrodinger equation in exterior
domains, J. Fac. Sci. Univ. Tokyo, Sect. 1A, Math. 31 (1984), 97-108.
184 Bibliography

[4] L 2 -solutions for nonlinear Schrodinger equations and nonlinear groups, Funk.
Ekva. 30 (1987), 115-125.
VEGA L.
[1] Schrodinger equations: pointwise convergence to the initial data, Proc. Am.
Math. Soc. 102 (1988), 874-878.
WEINSTEIN M. I.
[1] Nonlinear Schrodinger equations and sharp interpolation estimates, Com-
muns. Math. Phys. 87 (1983), 567-576.
[2] Lyapunov stability of ground states of nonlinear dispersive evolution equa-
tions, Communs. Pure Appl. Math. 29 (1986), 51-68.
[3] Modulational stability of ground states of nonlinear Schrodinger equations,
SIAM J. Math. Anal. 16 (1985), 567-576.
[4] On the structure and formation of singularities of solutions to nonlinear
dispersive evolution equations, Communs. Part. Diff. Eq. 11 (1986), 545-565.
[5] Solitary waves of nonlinear dispersive evolution equations with critical power
nonlinearity, J. Diff. Eq. 69 (1987), 192-203.
WEISSLER F. B.
[1] Semilinear evolution equations in Banach spaces, J. Funct. Anal. 32 (1979),
277-296.
[2] Local existence and nonexistence for semilinear parabolic equations in Lr',
Indiana Univ. Math. J. 29 (1980), 79-102.
[3] Single point blow-up for a semilinear initial value problem, J. Diff. Eq. 55
(1984), 204-224.
[4] Rapidly decaying solutions of an ordinary differential equation with appli-
cations to semilinear elliptic and parabolic partial differential equations, Arch.
Rat. Mech. Anal. 91 (1986), 247-266.
YAJIMA K.
[1] Existence of solutions for Schrodinger evolution equations, Communs. Math.
Phys. 110 (1987), 415-426.
YAO J.-Q.
[1] Comportement a l'infini des solutions d'une equation de Schrodinger non
lineaire dans un domain exterieur, C. R. Acad. Sci. Paris. 294 (1982), 163-
166.
YOSIDA K.
[1] Functional Analysis, Springer, New York, 1965.
Index t

a priori estimate 58, 135-7 of an operator 18


adjoint 22-3 dynamical system 142, 143, 145
attractor 152
eigenfunction 72
behaviour, asymptotic 120-2, 142 embedding, compact 3, 151
blow-up 72-6, 87-90, 114-21 energy, see conservation of energy
blow-up alternative 57, 58 equation
Bochner's Theorem 7 inhomogeneous 50-55
boundary regularity 28, 62, 124, 134, non-autonomous 50-55,134-41
146 parabolic, see heat equation
with second member, see equa-
compactness 124-41, 143-4,.151
tion, non-autonomous
complete metric space 56, 124, 147
equilibrium point 144, 148, 152
conservation of energy 78, 83, 92, 100
estimate, uniform 67, 68, 84, 112-13,
continuous, absolutely 13, 14, 15, 53,
130, 136, 139
55, 139, 141,
existence
contraction
global 58, 65-71, 76, 83-6, 112-
semigroup 39-41
-14
strict 56
local 56-9, 64-5, 76, 82-3, 100-12
convexity 6, 148
extrapolation 163
damping 50-55, 134-41
dependence, continuous 59, 100 forcing 50-55,134-41
differentiable, differentiability 38, 51, function
60, 62, 78 integrable 7-8
differential inequality 72, 73, 75, 87, measurable 4-6
88, 116, 117, 120, 125, functional analysis 1
131-3, 137, 139
distribution 2 Gagliardo-Nirenberg's inequality 3,
vector-valued 10 112
domain generator 39
bounded 27, 43, 46, 62, 68, 74, 124, graph, closed 18, 20
134, 141, 146 Gronwall's lemma 55, 125
186 Index

growth condition 65, 70, 84, 86, 112, Pettis' Theorem 5


119, 127, 130, 133, 138, point
149 fixed 57
hyperbolic 156
heat equation 42-7, 62-77, 124-30,
134-7, 146-9, 158-63, regularity 28, 33, 35-7, 39, 41-3, 47,
164-8 48, 60-1, 62, 78, 100
Hilbert complex space 25-6
Schrodinger equation 47-9, 91-123
Hille-Yosida-Phillips Theorem 40 set, connected 143
smoothing effect 37
Sobolev embeddings 3
invariance principle 143-4
Sobolev space
isometry group 37, 41, 47-9, 61, 78,
2, 13-17
91, 137, 149
solution
bounded, see estimate, uniform
Klein-Gordon equation, see wave
global, see existence, global
equation
maximal 57
stationary 157
Lax-Milgram Theorem 1, 26-7
stability 154
Liapunov function 143-4, 147, 150
asymptotic 154
strict 144, 147, 150
exponential 158-64
Lipschitz condition, local 55, 57, 62,
Strichartz' estimates 96-100
63, 79-81, 100, 145-6, 149
trajectory 142
maximum principle, 65-7 bounded, 144-7
relatively compact 143
w-limit set 142, 143, 145
operator uniqueness 56, 106
closed 18, 20
dissipative 19 variation of the parameter formula 50
m-dissipative 18-32, 33-5, 38-41
self-adjoint 24, 26, 32, 35 wave equation 47, 78-90, 130-3, 137-
skew-adjoint 24, 26, 29-32, 37, -41, 149-52, 163-4
61, 78