Professional Documents
Culture Documents
nz 1NIVERSIDAD COMPLUTENSE
^.^,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
I
1
t
'
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 presentation 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 evolution equations, in order to provide the reader with a sufficient background. In
particular, we recall the basic results of vector integration (Chapter 1) and linear 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 equations or evolution systems which may include non-linear interaction or selfinteraction 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)
f(u),
(0.2)
onr,
(0.3)
au _
an 0 our,
(0.4)
vi Preface
ac
= czAui
i = 1, ... , k,
(0.5)
= Au mu,
Utt
(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 energy 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)
vii
Preface
Paris
1998
T. C.
A. H.
Contents
1.
Notation . . . . . . . . . . . . . . . . . . . . . . . . . . xiii
Preliminary results . . . . . . .. . . . . . . . . . . . . . . . 1
1.1.
1.2.
1.3.
1.4.
1
1
2
4
4
7
10
13
m-dissipative operators . . . . . . . . . . . . . . . . . . .
18
2.1.
2.2.
2.3.
2.4.
2.5.
2.6.
18
19
21
22
25
26
26
27
2.
29
30
31
33
35
38
39
42
x Contents
5.
6.
50
50
54
55
56
59
60
61
62
5.1.
5.2.
5.3.
5.4.
5.5.
62
64
65
72
76
Preliminaries
Local existence
Global existence
Blow-up in finite time
Application to a model case
.
78
6.1. Preliminaries
6.1.1. An abstract result
6.1.2. Functionals on Ho (S2)
6.2. Local existence
6.3. Global existence
6.4. Blow-up in finite time
6.5. Application to a model case
78
78
79
82
84
87
89
7.
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
8.
9.
Bibliography
Index
114
120
121
124
124
125
129
130
134
134
135
137
142
142
143
145
146
149
154
154
156
158
164
165
167
169
185
I
I
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)
0 kI
2 1/2
for u E H m(1l)
xiv Notation
H)
= Wo
' 2 (fl)
= ut
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
IIuIIL
IIuIIw
u
1
Preliminary results
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
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
= fn
f (x)cp(x) dx,
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
(u,
D a uD a v dx
IIm
(u, v) =
in
Vu Vv dx,
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.
\1 m /
IIDauIIL1
IcI=7
<C
IkI='m
IIDauIIL-
IIuIIL9a,
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
VF(u) _
F'(u)Vu, if u V N;
if u E N;
0,
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 distributions 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
on
the existence of E^,, C I with IEj < 2', such that fn ,k -# f
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)> with
with
xnk (x)>x'(x) for all x c X.
Proof. Let
(xn)n>o
'
'
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
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
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 covered 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(;^
O<i<j
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
Vector-valued functions 7
1.4.2.
Integrable functions
II J7 9n
Therefore,
xl
fl gn , x
as
II9nA+LIIfnfAI +I
n -.
X.
oo.
If
_ fII. .
integrable.
8 Preliminary results
I f'
un
I(n+ n fn
n-.cx
LI
1.4.3.
If (IL
if p < oo;
if
p= cc.
P = NsSUPtEIII A0111
^I f (t)jjP dt) ,
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
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
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
almost everywhere.
10 Preliminary results
Since gn (t) <_ f.(t) 1 almost everywhere, it follows from the monotone convergence theorem that g is integrable and that j glILP = lim 11gn,jILP. By weak
lower semicontinuity of the norm, we have
119r.JLP <liin inf ^IIfn.^ILv,(I,X);
El
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
Th.f ( t )
/' t+h
T1/t
f ( s ) d s,
jITaf(t)II P < f
JIf(s)jj''ds.
Vector-valued functions 11
Hence
t}h
11f(s)IIIdsdt
Ilf(t)IIIdtds<
f Ilf(t)II P dt.
/ t+h
I11(t) f(t)II + h J
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
fn(t)II,
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
1 f t}h
t
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
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.
12 Preliminary results
f = lim
fw = lim(f, ca) = 0.
J(t)
(t) = f (t), if t E J,
J(t)=0, iftJ.
1(t)
f(t)
(f',
w) _ -(f, co') = -
f (t)w'(t) dt.
Furthermore,
p( t + h) P(t)
cp'(t)
uniformly on R, as h . 0.
Therefore,
`,
_ lim.
= li o
_ 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
D(I)by
(t)
(V(s) - 6(s) f
) da) ds,
to
).
Hence
/'
0 = (T,') _ (T,) xo
Jr
It follows that
(T,) _ xo
J ^;
I
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,
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
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
Proof. We have
Ilf(t) - f(s)I1
1''
sf
Cb, v, (I, X ).
Vector-valued functions 15
/^ t
It follows that
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
j
,
t+h
1p
f( .+h) -f( .)
f^
in LP((a,c),X), as h 1 0.
and we apply PropoProof. 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
(1.1)
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
(s)I'ds.
lifh(t)lip < T I
rt+h
fh(t) <-J J
h
((s)I r dsdt=
1r r5
J j
a-h
w(t)V'dtds
(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 .
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 sequence (x)>i is dense in X*, w(t) does not depend on the sequence h n ,
Vector-valued functions 17
< L.
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
have
)J ( t ) - f()
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)
L In particular, we have
as k oo, and litn inf ^^ ^p,^ k ^1 Ln lim inf
n--oo
koc
(1.3)
(s) ds >
(1.4)
(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
II <_ c,
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
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 properties 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.
20 m-dissipative operators
Iterating this argument n times, there exists a solution for all A E (2 "`Ao, oo),
A E 12(D(A), X).
o II
uED(A).
Proof. We have IIJ) -
u E D(A). We have
Jau-u=Ja(u-(I-AA)u);
and so lIJ\ u ulI
Ja-I
Aau = J)Au.
Thus,
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 domain. There exists a Banach space X, and an m-dissipative operator A in X,
such that
(i) X '
that
J1Au = Jlu u, du E D(A).
Thus,
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
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 .
..
=
y.
and since A is dissipative, we obtain x
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
du 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
(0,v) E G(A*).
(v,Au) = 0,Vu E D(A)
This last property is equivalent to v E D(A*) and A*v = 0; hence the result.
Proof. One has v E (R(A))
u E D(A).
D(A). We have
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 domain. 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
0
Proposition 2.2.6.
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 operators. 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
Proof. Applying Corollary 2.4.9, it suffices to show that if A and A are mdissipative, 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*.
2.5.
V(u, v) E X x X;
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.
aw
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*).
Bu = Au, du E D(B).
VuVvdx.
vAudx=
n
Js z
i
(2.1)
=J
(uv + Vu Vv)dx
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
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
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
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.
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.
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)};
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-
=J
= (vu Vz+muz+ (Aw mw)v) dx
(2.3)
(2.4)
Slv=u f.
(2.5)
30 m-dissipative operators
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 defined in 2.6.3. Theorem 1.1.4 shows that
E Ho (s^), o^u m = u in D'(sf)},
H '(r) = {u E D'(), 5
-
(2.6)
(u,v)-i =
J (v
.V +
(u, v) E D(B).
J vw dx - J uz dx.
Similarly, we have
((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)
1 - V L2 + IIuI1 2 , = VI1 2 .
32 m-dissipative operators
dx,
Cu=Lu, VuED(C).
Proposition 2.6.14. C is self-adjoint < 0.
(u,v)L2 + (u,v)_1.
(2.8)
for all u,v E D(C). It follows that G(C) C G(C*), and so C is self-adjoint
0
(Corollary 2.4.10).
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 mdissipative 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
1
3
The HilleYosidaPhillips Theorem and
applications
3.1.
1
I
Theorem 3.1.1. For all x E X, the sequence u(t) = T(t)x converges uniformly 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
Vt > 0;
(3.1)
(3.2)
(3.3)
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)
(3.5)
= Au(t),
u(0) = x.
Vt > 0;
(3.6)
Finally,
T(t)Ax = AT(t)x.
forallxED(A) andt>0.
Proof.
Step 1.
(3.7)
t
I
In particular,
Ilua(t)II
(3.8)
IIxil,
It follows that
Ilua(t)
<
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
T(t)x I)
T(t)x,,Il ;
and so TA(t)x --> T (t)x as A j 0 uniformly on [0, T] for all T > 0. Properties (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
JO
va (s) ds,
u(t) = x +
fo
T(s)Ax dx.
(3.9)
(3.10)
(3.11)
u(0) = x.
(3.12)
In addition, we have
II Au(t) II
< - 1 1 xII;
(3.13)
(3.14)
Finally,
Au(t) 112 <
2t
(3.15)
(Ax, x),
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
(3.16)
= 2(Aaua(t),ua(t)),
(3.17)
2(Ix112.
(3.18)
5
Hence, with (3.18),
-f
(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
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)
(3.22)
T(0) = I;
(3.23)
(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)={
T(t),
ift>0;
We easily verify (3.21), (3.22), (3.23), and (3.24) for x E D(A), and then for
x E X by density. Finally,
d u(t) (0= Ax -
dt
d+
-t) (0) =
dt (0)
T(t)* = T(t),
for all t E R.
Therefore
Proof. The result is clear for x E D(A). The general case follows from an usual
density and continuity argument.
u E C([o, oo),
x) n C
([0, oo), X );
(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
and
T(t)x - x
Lx=1h mm h
for all x E D(L).
II x - A
T(h)x - x
f0 00 e
-t T(t)x
dt,
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
0
e -t (T (t + h)x - T(t)x) dt
=1 f ^
1 '
h Jh
-
eh
J e -t T(t)xdt
0
1 /
..11o
e -t T (t)x dt
eh
-
h e-tT(t)xdt
v'^
Letting h . 0, we obtain
urn
m T(h -I Jx=Jx-x;
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
xt
= 11 r t T(s)x ds.
xt =
fh
r
- 1 t
it
+hT(s)xds
-J
T(s)x ds
T(s)x ds.
+h T(s)x ds
-1
Finally, the following result shows the uniqueness of the semigroup generated
by an m-dissipative operator with dense domain.
1
I
u(t + ) u(t)
_ S(h) I
u(t)
as h j 0.
= 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.
T(t)xJ =
(ii) T(0)
(iii)
= I;
T(t + s) = T(t)T(s) for all s, t E R;
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
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)
(3.27)
u(0) = V.
(3.28)
In addition, we have
u E C((0, oo), Ho (1));
(3.29)
JJVIIL2, Vt > 0.
(3.31)
IIIUIIL2 <_
IIV^PIIL2,
(3.32)
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
f
0
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 Remark 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
IIS(t)IIc(L2) 5 e
at,
(3.33)
f u(t)
+ 2 I u(t)u'(t)
= 2A f u(t) 2 + 2 f u(t)Au(t)
= 2f u(t)
-2
J IDu(t)I
<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.
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 Proposition 3.5.2. In addition, the following estimates hold.
Proposition 3.5.7. Let 1 < q < p < oo. Then
'I'II L ,
(3.34)
=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.
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
0 onR N \S2.
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
=-
Juu
t
_ -
f u Au = f Du Vu
z
_-
J IVu
- I2
_<0,
and so u(t) < 0, for all t > 0. We then deduce from (3.35) and (3.36) that
Corollary 3.5.10. Let A > 0 be given by (2.2), and let M = e \ l g l 2,N / (4 ">
Then
II S(t)II L(x) <- Me
-at ,
(3.37)
for allt>0.
Proof. LetcpEXandletT>0. For0<t<T,wehave
{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
(3.37')
for all t > 0. Once more, we cannot take Mp = 1. Actually, one can see that,
for p > 2, one has
Oat
for all t > 0, and that this inequality is optimum in the following sense:
{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
j lu(t)IP +
p dt l^
Ivu2 I>
4
p
1d
f lu(t)IP + 2
4 f Iu(t)IP;
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
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.
Consequently,
J M 2 < fn Iv n
11 2 =A f cp 2 =Alim
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
(3.38)
(3.40)
(3.39)
In addition,
J {Jvu(t)1 2 + mju(t)J 2 + u(t)2} = Jst
{IvcpI 2 + mIcpI 2 + ,0 2 } ,
( 3.41)
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.
(3.42)
u(0) = W.
(3.44)
(3.43)
In addition,
Iu(t) 2 =
n IVu(t)I 2 =
(3.45)
J IV I 2 ,
(3.46)
for all t E R.
(3.47)
IIAu(t)IIx = IIA(PIIx,
for all t E R. But, for all V E Ho (S2), we have
JM
+ IIvIIX
(3.48)
IIT(t)II,C(Lp ,LP) C
for all
(4^rItI)N(2-p)'
(3.49)
0.
(3.50)
P a(t)(6)
- I6I 2 Fu(t)(C) =0 in R N ,
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
T(t) W II L- <
1
(47rt) 2
II(P11t1,
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 analytic semigroups; see Goldstein [1], Haraux [3], Pazy [1]. One can build semigroups 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
(4.3)
t. u(0)=x.
u(t) =T(t)x
J0 T(t s)f(s)ds,
(4.4)
w(s
+h
) w(s)
7t s h) {
rr
Inhomogeneous equations 51
= fo
T(s)f (t - s) ds,
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
1 l+h
v(t + h) - v(t) - f t T(h) - I
h
T(t-s) h
f(s)ds + h T
(t+h-s)f(s)ds,
T(h)-I ,
Af
h f
d+ v
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
write
v(t + h) - v(t) =
T(s) f (t + h
-h
f (t - s)
ds+ T(h)
T (t - s)f
(S)
ds,
.f(t+h.) ,f(t.)
h
f'(t.), as h J 0
d+ v
dt (t)
= fo
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).
-
T (h) -I
h
v(t) =
11t
T (t + h - s) f (s) ds
- v(t+h)-v(t) 1
-- [
h
h J t
t T (t - s) f (s) ds
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,
Inhomogeneous equations 53
w(s + h) w(s)
h
_ T(t s h)
(s)}
It follows that w is absolutely continuous from [0, T] to Y. Moreover, the righthand 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
u(t) = x
+ J t (Aun(s) + fn(s)) ds
0
ti
u(t) = x +
JO t (Au(s) + f(s))
for all t E [0,T]. It follows that u E W ' ((0,T),Y) and that
ds,
1 1
u' E L' ((O, T), X ); and so u E W 1 1 ((0, T ), X ). This completes the proof.
'
"
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
A(s)cp(s) ds,
rt
A(s)ds
Jo
Proof. We set
ft
(t) = C l + C2
J A(s)cp(s) ds.
0
ll
t A(s) ds I 1 < 0,
I ))J
and so
that
IIF(y)-F(x)II <_L(M)IIy
-xII, ex,yEBM,
(4.6)
(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:
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);
4.3.1.
TM = 2L(2M
IIxII
< M.
Proposition 4.3.3. Let M > 0 and let x E X be such that
there exists a unique solution u E C([O,TM], X) of (4.9) with T = TM.
Then
Semilinear problems 57
IIxII. We let
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)
tE[0 ,TM I
II,
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
II(t)It
0)II
< IIxII +JO t IIF(u(s))Ilds <M+t M TM( <K.
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
(4.10)
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 differential 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,
v(s) = T(s)u(t)
10
u(s),
v(s - t),
if s E [t, t + TM].
Semilinear problems 59
estimate describes the blow-up phenomenon sharply, since the solutions actually
blow up as (T(x) - t) I'.
-
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;
(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)).
4.3.3.
Regularity
Hence,
+ L(M)
t llu(s
+ h) u(s) I ds
Frthermore, we have
h
T(h)x x = J T(s)Axds;
O
Isometry groups 61
A is m-dissipative with dense domain, and generates the semigroup (T(t)) t >o
given by
II
5
The heat equation
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 particular, 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
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));
u(0) = cp.
(5.1)
(5.2)
(5.3)
JO
(5.4)
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
"
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(.)),
and so (Proposition 1.4.14 and Corollary 1.4.23) u(t) E Ho (S2). A similar estimate shows that actually u E C((0, T], Ho (l1)),
IIu(t)IIH1 < C(1 +t -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
(0,T]. We show
fo
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)
Remark 5.1.4. The same method also shows that if we assume further that
cp E D(B), then
Global existence 65
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.
IIo(u(t) V)IILZ
I(S(t) I)oVIIL2
+Cft
t1 S
JIF(u(s))IIHlds tie
u( 0 ) = W.
(5.5)
(5.6)
'
Proof. For
Multiply (5.5) by -u (t) and integrate over Q. Integrating by parts and applying (5.6), we obtain (see the proof of Lemma 3.5.9)
-
v -(t)
<C
f Iu(t)1u (t) = Cv
t .
)
t < T,
1 0, it follows that
v(t) e ct
(VP
= 0 , Vt E
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 Theorem 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,
t E [0,T(cp)).
Global existence 67
fo
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
We conclude that f (t) < f (0) = 0, and so u < k; hence the result.
of (5.1)(5.3) satisfies
O<t<00
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
for all t >_ 0. Finally, invoking Holder's inequality, for all e > 0, there exists
C(e) such that
Global existence 69
for all v E X. Choosing e such that eC4 < 1/2, it follows that
C3 + KC(e)C4 + 1 sup IIu(s)II,
2 0<8<t
IIu(t)II
for all t > 0. Therefore
0<s<t
Then, there exists A < oo such that, if JIVII < aA, the corresponding solution u
of (5.1)-(5.3) is global and satisfies
and so
IlIt2e U u,
IIu(t)IIL2
)t u (t)
= e U-u)tT(t)^ + f
o
eU-u)(t-9)T(t - s) (e
F(u(s)))ds
upteUu(s)Ij ,
os
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
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.
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.
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,
We have
E a9 {e a
i
eat llu(t)II < MII^II +Mfo e sllu(s)II} +Eds < MII^vII + eMf(t) 1 +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
,
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.
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)
(5.11)
Q = 1.
zp is easily obtained by solving the minimization problem (2.2), using the compactness of the embedding Ho (St) L 2 (c). We may choose a positive minimizing 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
f^
(-P)
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,
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)))
u(t)A +
J g(u(t))V) =
Af(t)
1+E
+ J g(u(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),
I}e i
(5.12)
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.e.
f( t) -E ) > (bt)
'
'
I
I
f(0)
(0)-E
- bt,
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
for
][^.
Proposition 5.4.4. Assume that there exists K > 0 and a> 0 such that
such that
(2 + s)E(cp) < IQI(p 2ev).
Il
'
'
f(zu
+ g(u))ut = f(_Vu.
^
Du t + g(u)ut) =
(5.3). Then
dt in
f J ut + E(u(t)) = E(u(s));
t
(5.13)
(5.14)
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).
in
IVU(t) 1 2 + 2
juj<K}
u(t)g(u(t)) +2
f f
1 ju>K)
u(t)9(u(t))
G(u(t))
>_ 2 f ^Vu(t)1 2 + 21
u(t)g(u(t)) + 2(2 + e)
Juuj>K)
{Iu,<x}
o
(5.15)
On the other hand, observe that, by Proposition 5.2.3, we may let s = 0 in (5.14);
and so
2(2 + e)
fl
G(u(t)) = -2(2 + e)
juj^!K}
+(2 + e)
f1
G(u(t))
juj<Kj
(5.16)
u(t)g(u(t)) - 2(2 + e)
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)
o f u.
(5.19)
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
<
Uo ^^2
^ I
1/2
1/2
12
< 2h(t) (
2u
^ =)
Cf fu
o J 2)
1/2
(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
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 superlinear. 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 possible 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 Sattinger [1]. The nature of blow-up is currently rather well known. See Baras
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
^
'
d
6
The KleinGordon equation
ii
6.1. Preliminaries
In this section, we give some technical tools that are essential in this chapter.
6.1.1.
An abstract result
'
u (t) = T (t)x +
id
2 d IIu(t)II 2
X, and f
C([0,T],X). Let u E
C([0,T],X) be given by
for all t
= ( f(t),U(t)),
[0, T] .
2 dt Ilu(t)11 2
'
= ( u(t),
Hence
IIu(t)11 2 = IIx11 2
(6.1)
Preliminaries 79
6.1.2.
Functionals on Ho (Sl)
(6.2)
(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)
fo x g(s) ds.
(6.5)
(6.6)
9(u),
(6.7)
80
(9(a) 9(x)) do ;
We now assume that, instead of (6.3), g satisfies the following weaker condition:
I9(x) - g(y) I
(6.8)
9=91+92,
where 92 verifies (6.3) and gl verifies (6.3) with a = 0. For example, consider
9(x),
9i(x) = g(1),
I. g(-l),
if Ixi < 1;
if x > 1;
if x < 1.
HH(Q)'.., Lr(l)
(6.10)
Ln'(S2) , H-1O)
(6.11)
Preliminaries 81
Proof.
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 <
for all u, v E Ho (1l).
by (6.10).
C(!I uII2
+ IvIIia)l rt VIIL-+2,
I
I
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)
d V(u(t))
for all t E [0, T] .
_-J
sz
(6.13)
'
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
I
- (g(u(t)),u'(t))x-1,Ho
= - I g(u(t))ut(t)dx.
It follows that
V(u(t)) = V(u(0)) -
IJ
t
st
(6.14)
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)
(6.15)
u tt - Lu + mu = g(u),
(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.
Local existence 83
U(t) = T(t)(,
0) +
JO
(6.18)
E(u(t),ut(t)) = E(co, l0 ),
for all t E [0,T(p,
L)).
g(u(t))ut(t) dx = 0,
C
Remark 6.2.4. Proposition 6.2.3 justifies the study of the Klein-Gordon equation 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.
G(cp) +2
J G(cp) + 2C J u(t)
2,
(6.20)
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 corresponding 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),
Global existence 85
for all x E R. Sobolev's inequalities show that there exists a constant, which we
will still denote by C, such that
v)lIuIIi,
(6.22)
for all u E Ho(S2). Let (cw) E X, with II(^P,V))Ilx <1, and let u be the corresponding solution to (6.15)(6.17). Using the notation of the proof of Proposition 6.3.1, we deduce from (6.20) and (6.22) that, for all t
vf(t)<f(0) _2
G(cp)+Cf(t) 2 .
(6.23)
Observe that
f G(
LI
I
fl
I
I
El
I
I
El
and so, since f(0) < 1, there exists ME [1, oc) such that
f(0) _2
(6.24)
9( x \_ C x 1+E x
O(xa) + a = 6(y a ) + a = 0.
In addition, we have a < X a <a(1 + e)/e (see Figure 6.1).
86
-E2
Fig. 6.1
By (6.23) and (6.24), we have
l+e
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 Proposition 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.
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
_f
if, ut(t) 2
+2
ut(t)utt(t) =2
If
'
(6.25)
for all t E [0,T]. By density, we then show that (6.25) still holds for u E
hence the result. o
(c));
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 ,
foralltE [0,T].
fn
ut(t) 2 2
fn
IVu(t)1 2 2m n u(t) 2 +2
fn
u(t)g(u(t)),
for all t E [0, T]. It suffices to see that, for all w E Ho (S2), we have
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,
f ut (t)
>2
J u(t)
f (t) =2
J u(t)
- 2m
J u(t)
-2 f IVu(t)1 2 - 2m
-2
J Vu(t)!
f u(t)g(u(t))
+ 2(2 + E)
J G(u(t)),
+2
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)
z
ll sz
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
u(t)2
fn u(t)2 > (
4 +E)Y
ut(t)u(t)) 2 > ( 1 +
4) f ' (t) 2 ,
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
(2+e)E(co, ) < IQI(
(2+e)v),
fn u(t)g(u(t))
= f1juj:5K}
> mjci + (2 + E)
f1 juj>Kj
G(u(t))
J G(u(t)) - (2 + E) J
> (m - (2 + e)v) IcI + (2 + e) G(u(t)),
J
>_ m^c + (2 + e)
<Kl
G(u(t))
7
The Schrodinger equation
7.1. Preliminaries
Throughout this chapter, we use the notation of 2.6.5 and 3.5.3. In particular,
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:
X = H '(S2) = H
-
UE
C([0,T],Ha(cl) flC'([0,T],H-1(f));
iu t + Au + g(u)
= 0, Vt E [0, T];
u(0) = `.
Applying Lemma 4,
(7.1)
(7.2)
(7.3)
u(t) = T(t) cp +
ifo
T(t(7.4)
- s)g(u(s)) ds,
u is a solution of
u E L((O,T),
Ho (Q)) n W 1
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,
g(z) = Izlg(lzI),
(7.7)
IZI
g(s) ds,
(7.8)
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)
(7.10)
(7.11)
= f I Vv
dx + V(v),
(7.12)
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)
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
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
dt II u(t)II i2 = 2(-^u(t), ut(t)),
Proof.
Step 1. cp E L 2 (Il). The global existence in L 2 (cl) is a consequence of Theorem 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 Proposition 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.
(7.15)
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.
Step 4. The conservation law (7.14) for cp E D(B). Taking the scalar product
(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
<
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
(7.17)
u in C([0, T}, L 2 (Sl)) as n -, oo.
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)
cp E
E('P) E(u(s)).
(7.19)
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)
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
H 1 (RN) . Lr(RN) with dense embedding and L''(RN) H -1 (RN). It follows that C([0,T],H l (1RN)) LQ((0,T),L''(RN)) and L9'((O,T),Lr (RN)) y
so
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)).
Remark 7.3.3. If (q, r) is an admissible pair, and taking suitable test functions, we easily verify that for all u E L 9 ((0,T),Lr'(R N )), we have
f
IIiIIL9((o,T),L') =supRe
uapdxdt1;
^P
RN
J
O
1.
IlUIIL4((o,T),Lr) = sup {
Re
J0 JR^
u-pdxdt1;
1 }.
(7.20)
II 4 )f(t)IILr
< fo
The classical estimates on Riesz' potentials (see Stein [1], Thm. 1, p. 119) then
give
IIL9'((o,T),L-')'
97
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
s)f(s)ds,
0
'
fo T(t
v)f(a)de)
tt
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-');
(RN))
to C( [0, T],
Step 5. The operator c E (L 1 ((0, T), L 2 (RN)), L 9 ((O,T), L''(RN))), for all admissible 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
0
Re
c i dx dt =
ft
_ Jo J
= JOJ
0
ft
(f(s),T(s t)^p(t)) ds dt
rRe
fRI
4)fipdxdt
and
1_0+ 1-0
ry
o0
ry'
P+
1 - B + 1-0
r' '
Ibf
= iL4)f + f,
(7.21)
and so
dt"Ds(t) = T(t)f(0)
fo
T(s)f'(t s) ds = 4) f (t).
IIrfIIL2 < IIfIIL9'(R,L*' ) We conclude (as in Step 5 of the proof of Proposition 7.3.4) by noting that
+00
+00
E [0, T(cp));
))
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
for allT <T(cp).
C([0,T],H'(R N
)),
Remark. Property (iv) above is a regularity result. As well as being interesting 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 identity (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 Lemmas 7.4.7, 7.4.9, and 7.4.10, Corollary 7.4.8, Steps 3 and 5 of the proof of Proposition 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.
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 replaced 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, using 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 complete 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
estimates as in > oo.
7.4.1.
Some estimates
Izi < 1;
if IzI > 1.
g(z),
=!
l zg(1),
if
9 m (Z)
if
m9(m),
hr,(z)
(We have in particular g l
Iz) < m;
if Iz) > m;
1 9(z),
= 9rn(z) -
k(z)
f
o
IZI
9,,,.(s) ds.
I
j
E () = 2 f
m
IVul 2 dx
+V,n(u).
IC(u)(t)
= fo
(7.22)
K( 1 + Iz2I" + Izi1 a I)Iz2 z1I;
+1. (7.23)
I9m(z) g(z) < KIzl
(7.24)
Ihm(z2) - hm(zi) <- K(Iz2I" + IziIl)Iz2 - ziI;
I9m(z2) gm(zi)
Ik(z2)
(7.25)
z1I;
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)
(7.27)
(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 ). Finally, we observe that
0,
KIzI
if IzI < m;
+i
if IzI > m.
-T)
flul>m
Iulr
Lemma 7.4.3. Let M Goo. There exist C(M) < oo and > 0 such that
IVm(v) - Vm(u)I < C(M) (IIv IVm(u) - V (u)I < C(M)m -1`;
for all u,v E Hl(RN) such that
N U {oo}.
IIUIIHI M
+ IIv - uliy, 2
uIIL2
) ;
( 7.29)
(7.30)
and
IIuIIL
(2 - T) = a;
(7.31)
I
Cm(z)
l
IzI < m ;
if IzI > m;
j 0, Ti
f
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
(7.34)
))
n W l "((0,T),
)}-
t
9
IIu'(a)IIH
-1
dQ < 2K 2 It sI,
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
(7.35)
We have in particular u
and the result follows readily.
Proof.
Lemma 7.4.9 (*). For all M, there exists C(M) such that
(ck+1)
with a/N = 1/2 1/N 1/(2(c + 1)). In particular, we have a(a + 1) < 1;
hence the result.
Lemma 7.4.10 (*) For all M, there exists C(M) with the following properties. 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
II hm(u) ' [I L'((o,T),L 1 ')
IILo((o,T),L2);
(7.36)
(7.37)
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);
Ilhm(U)'IIL((o,T),L)
(7.37) follows.
^(M)IIu^IIL((o,T),LT),
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
Ilu
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
Proposition 7.4.12. For all M, there exists TM > 0 with the following properties. 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
E [0,TM];
In addition, if u and v are the solutions corresponding to the initial data cc and
we have:
Proof. We proceed in
that
IIWPIIHI << M.
(7.38)
(7.39)
D'm(um(t)) = Em(co);
(7.40)
T,,,, > TM. To see this, assume that T. < oo. In that case, we have
IIum(Tm)IIH1 = 2M.
(7.41)
Vm(P))
(7.42)
On the other hand, applying Lemma 7.1.1, (7.26), (7.27), and Sobolev's embeddings, 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 +
It follows readily that there exists TM > 0 depending only on M, such that
Tm
> TM , and so
(7.43)
f T(t - s)g.,,,,(u
0
rn )'(s) ds.
(7.44)
+i
(7.45)
t
+ IIumIIL((,T),LT)
C( p) + K(M)(T + T 1-2k )(IIumIIL00(c,T),L2) + IIumhIL((0,T),LT))
IIumIIL((O,T),L2)
,
where K(M) depends only on M and C depends on II'PIIH2 Choosing if necessary TM smaller (but depending only on M), we may assume that K(M)(TM +
Z '1-2
< 1/2. We then obtain
M )IIUmhhL0o((,TM);L2) + IfmblL'(c,7M);Lr) < 2C(cv).
(7.46)
(7.47)
+ 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
21
(7.49)
m (t)I dx.
(7.50)
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 Proposition 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
Taking TM smaller if necessary (but depending only on M), we may assume that
T2)
C(M) (TM +TM
We deduce (v).
(7.51)
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))
End of the proof of Theorem 7.4.1. Using Proposition 7.4.12 and Lemma 7.4.11,
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 Proposition 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)
m( )
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.
V(w)I < A
Iwi 2 + C
RN
\ JI]^N
Iwi2 /
e+i - 4
( IRNIDwi
2 )
(7.54)
f(t) <E
()
+ CII^jI f(t),
for all t E [0, T()). Therefore, if IIcc IL2 is sufficiently small, we again have
T( p) = oo.
,
Proposition 7.5.3.
that if IIVIIH1 << 6, then we have T(cp) = oo and sup IIu(t)IIH1 < KII(PIIHI
e>o
Proof.
Observe that
(7.55)
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
(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
+ a = B(ya) + a = 0.
In addition, we have a < x , < a(1 + e)/e (see Figure 7.1).
B(xa)
a
-Ex
Fig. 7.1
By (7.56), we have
9(1(t)) +
for all t
M f (0) > 0,
x,M f( o ),
we deduce that
E Mf( 0 ),
for all t
dt f N
RN :
rziu r dx,
115
(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
0 E (r) =re;
p(r) = r 2 e -Er2
We set
fE (t) = J I B E UI 2 dx,
RN
116
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)),
we have
E
(7.59)
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
-E
Ixu(t)[ 2 dx =
IR
N
RN
(7.57).
tHIII'Iu(t,-)IIL2
is continuous, and then I Iu(t, -) E C([0, T(cp)), L 2 (R N )).
(7.60)
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
(7.61)
(Bu, ix Vu t ) = Im
and that
JRN 6x VT,
dx,
h(t) = h(0) +
fo
t (u t ,
(7.62)
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)
JR
+N f 9(g(IwI)jwj-2G(w))
+7
Re
]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)
= Re
(g(w), 20rw r )
20rg(w)w,. =
fR N 20rG(w) ..
r
28 r w r ) = 2
J (NB + rO')G(w).
(7.66)
IR
(NB + r9') I I 2
Re
(Aw,20rw r ) _ (20Vu,V(rur)) 2
fR
rO'Iwr 1 2 .
(7.68)
J^'
N N
(7.69)
fR
9IVU1 2 + N
RN
9(g(1u1)1u1 2G(u))
Re
(7.70)
(r9
u,
t
iu r ) = (r0 E Cp, icp r ) + 1 fe (s) ds,
0
(7.71)
where f f is a bounded function on [0, T] for all T E [0, T (gyp)) and is such that
fE(t) > _2f
(Vu1 2 + N
(g(^u^)lul 2G(u)) as E 1 0.
(7.72)
_ 4E(u(t)) + N
+ (2N + 4)
RN
RN
(7.73)
G(u)).
the function
tH
I xl 2 ju(t, x)I 2 dx
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. Denote 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
RN
IxI2Iu(t,x)l2dx,
(g(w), 20rw,.) = Re
JR 29rg(w)w,. = I
26rG(w) r .
RN
J (NB + rO')G(w).
(7.66)
(NB + r8')IVW1 2
fR N
Re
R, N
RN
rO'Iw r 1 2 .
(7.68)
IR
IR
rG'jw r.I 2 .
(7.69)
h'(t) _ 2
JR
OIVuI 2 + N
RN
0
2 G(u))
RN (g(IuI)Iuj
Re
fR
(7.70)
(7.71)
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
fit
(7.72)
2 f N IVu1 2 +N f
(g(lul)lul 2G(u))
_ 4E(u(t)) + N
R N
RN(7.73)
G(u)).
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. Denote 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.
(7.74)
L2(RN)
Proof. Let cp be as above. Let u be the corresponding solution of (7.4) and set
for t E [0,T(cp)). It follows from (7.58), (7.74), and the conservation of energy
that
(7.75)
f"(t) < 16E(),
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 conservation 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
u be the corresponding solution of (7.4). Set
f(t)
=f
JR
(7.76)
(7.77)
.N
8t 2 V(u)
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
(7.78)
2 fR N
(7.79)
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 ,
-) <Ct z^^+z>,
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
for all t > 0. Therefore, t 2 f (t) is non-decreasing and, (by (7.78)), we have
8t
IIu(t)IIL-+2 <Ct
Na
a+2
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
!
^...
>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 repulsive 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.
8.1.1.
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
(t -sp
)
f (s)^p(s) ds,
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
O(t) < A+ t- fo
(t
s) a s f(s)0(s) ds,
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
It follows that
f (s) ds
t 1
(,-E)t (t -s) S
1
+ to
r (i-)t
1
1 f(s)9(s)
ds
(t s)$
L 1 (p T)
-
We will also make use of the following comparison lemma, which generalizes
Proposition 5.3.1.
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
Jo
almost everywhere in Q.
1 f (t)u (t),
Jo
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
u(t) =T(t)cp+
e<t_<T
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.
e<t<T
JO (t s)
IIf(S) + IILIIu(S)IILrdS,
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
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 +
for all t >_ 0. Suppose that sup Ilu(t)IILI < oo and that sup 1 f (t) + I[L < oo.
t>O
t>O
t>O
Corollary 8.1.5.
t>O
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
(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 Proposition 5.3.3, we may restrict ourselves to the case in which cp > 0, and so u > 0.
From (8.1), it follows that
(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
- s)
O (tR
(1 + Iju(s)IILP)P
-1
ds,
LP
< CM.
o<s<t
tl f (t)P-1,
(8.3)
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))
I C u(t)
<+ _B,
I
IL
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)
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
d Jr
+e
ff
t
(8.4)
ut dx dt
fn I Vu(t)1 dx
2
+ k 2(2 + E)E(u(s));
(8.5)
Set
r= {t> 1;
(8.6)
f ue(t) dx<i}.
2
f IVu(t)1
f uu
t dx
+ Ilu(t) II L
8.7
2.
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)
O<t<T
Remark 8.1.7. If g does not satisfy (8.1), we do not know whether the conclusions 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
for t > 1. Applying (2.2), we then deduce that sup 1ju(t)ii H l < oo. On the other
t>1
Corollary 8.1.4.
131
for all x E R. Let (ep, z)) E X and let u be the corresponding maximal solution 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.
f (t) =
u (t) 2
d x,
for t > 0.
Step 1.
+ (4 + e)
fn u (t)
t
in u (t)
t
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)
),
(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
(8.12)
We claim that
(8.13)
E() > 0 ;
at (ijf (t) - 2 ( 2 + e)E(^P,
Vt > 0,
Vt > 0.
(8.14)
(8.15)
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)
fo
-4
(8.17)
to.
u t (t) 2 dx, Vt >t0.
(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
)
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
(8.20)
/'e+i
sup]
II (u(s), u t (s)) Il x ds < oo.
t>O t
(8.21)
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)
< CII9(u(t))IIL2w(t)
8.22
1/2 ,
for all t > 0. Observe then that N > 3, and so N/(N - 2) <3. Consequently,
II9(z)IIL2 <
8.23
(8.24)
We deduce from (8.24) and (8.21) that, for all 0 < t < s < t + 1, we have
t +1
(ft
(8.25)
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
w(t) <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
E)G(x),
for Ixi
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]).
(8.26)
(8.27)
u(0) _ ,.
(8.28)
In the following subsection, we gather some preliminary results concerning problem (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
be given by
w(t) = T(t) cp + J t T(t - s) f (s) ds.
(8.29)
II
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!
P ([ 0 ,T] ,
((0 T),H -i
( 1 );
Ho( 9 )) n W
(8.30)
(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
(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.
u(t) = T(t)cp +
T(t - s)g(u(s)) ds +
T (t - s)h(s) ds.
fofo
(8.33)
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)).
'
A priori estimates
(8.34)
z(t) = T(t)cp
+ J0
+ J T(t - s)(-g(-z(s))) ds +
o
fo T( t - s)h
(s) ds.
+ MCeAt ^ e A8 Ilu(s)IILo ds
u(t)II L < II'IIL^ + MCA t
+
II
fo
(8.35)
and so
T(t s)h(s) dsII L < KIIhltL((o,t),Lf e 2 (t 9) (t s) 2 ds
"
J
MC' +MCeat f t
0
eA3jju(S)IILds
(8.36)
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,
2 G f(t)+c2llh(t)Ili2
a-c
< 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
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
u + mu + yu t =
u( 0 ) = <P,
ut( 0 )
= 0.
(8.37)
(8.38)
(8.39)
(8.40)
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,
E(u(t), u(t)) + ry
for
ffr f
ut dxds = E(, ) + J
bu t dxds,
(8.41)
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
(8.42)
and let h E L' (R + , L 2 (1l)). Then, for all (cp, 0) E X, we have T(,) = oo.
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
Proof.
+2
J0
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
f'+bf= l {
2 s ut ^E 6 ) IVU1 2
f
-(E-2 IVuI 2 dx-m(e-)f u 2 dx<-(a+m))
2
2 ^
u 2 dx.
'
On the other hand, if 6 <_ eK, by applying (8.43) and (8.42), we obtain
e
ug(u) dx b
sz
f^ u^ dx.
2)
xy < 2x 2 + ay 2 ,
(7 ty 6)2
gy(1 6) f uut dx < E 2
t
<_ e2y
j u 2 dx + 4 in u dx
J u2 dx + ry4 J ut dx.
(8.47)
(8.48)
(8.49)
hu dx < 4 f h2 dx +e 2
n
KS
\
EI
1 ) ut E \(.\+m) 1\\\ 1
c e(ry+ 1)
u2 +
2
/
+1h
\ 4 Y/
/
2 dx.
(8.50)
Note that, for e sufficiently small, we can take 6 = e 2 . (8.50) then reads
e 2 f
f'+el f <(2e 2 )
(
+m
Jsz utdxe((.^+mKCc)
c++1))
41 +
h2dx.
(8.51)
f(t)>_4
s^
utdx+2
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
0
The invariance principle and some
applications
= fl U
3>0 t>s
{St z}.
(9.1)
St (w(z)) C w(z).
(9.2)
(9.3)
St(w(z)) = w(z) 54 0.
hence (9.3).
Theorem 9.1.8. Suppose that U {St z} is relatively compact in Z. Then:
t>o
Z. Then:
(i) . = lim t - 0 4)(Stz) exists;
t>o
0. Let
y E w(z). Applying Theorem 9.1.8(i) again, and then Theorem 9.2.3(ii), we
obtain
Proof. By continuity of St, is closed. By Theorem 9.1.8(i), w(z)
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.
JO
(9.4)
(9.5)
(9.6)
,EP t>O
(9.7)
u( 0 ) = y.
(9.9)
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,
(9.10)
as n -> oo,
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
Theorem 9.3.2.
(9.10).
(z,,)n >o
C Z and zn -4 z
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
< 00,
(9.11)
we know that there exist sufficient conditions to have (9.11); see, for example,
5.3 and 8.1.
Lemma 9.4.1. Let yp and u be as above. Then we have the following proper-
ties:
IlUnk
un, II L <
tll^nk
Wne II LN +
fo
+ Ci
t II9(unk (S))
Ilu nk (s)
9(un, (s)) II L _ ds
ufl
( s ) IIL ds,
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,
for all k, f E N, the following inequality:
Ikknk U., 11 ' < (1 + t-1/2)Ilonk cne IIL 2
+li
t (1 + (t
s)-1/2)IIunk (s)
carne (s )IIL2
ds,
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
1
1
1
1
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
vt dx da + E(v(t)) = E(v(s)).
(9.12)
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
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]).
u(0) = cp,
(9.13)
(9.14)
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 solution of (9.13)(9.15) if and only if U = (u, u t ) is a solution of
U(t) = S(t)() +
(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
where u is the corresponding solution of (9.13)(9.15) (see the proof of Proposition 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.
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
) = {(u, v) E X; Au E L (1), v E Ho (1l)};
5 D(A
A-,(u, v) = (v, Au - mu - yv), for all (u, v) E D(A).
2
I1T7(t)IIc(x) < Me -
' t ,
(9.18)
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
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;
(
Proof. We proceed in four steps. Set U(t) = (u(t), u(t)) and H(t) = F(U(t)),
for t > 0.
Step 1.
W(t) =
(s)H(t s) ds.
Step 2.
Since S2 is bounded, we see by applying Theorem 1.3.2 and Remark 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 subset 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.
'
Consequently,
U {W(t)} C K' + B(0, e),
(9.20)
t>_T
K'= U
0
^
t>T
T.^(s)H(t s) ds
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
other hand, WE C([O, oo), X); hence U {W(t)} is compact and it can also
o<t<T
t>o
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 perturbation 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.
(10.1)
(10.2)
= f (u(t)), t> 0,
(10.3)
(10.4)
(10.5)
t
a]1e -5
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)
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
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"+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)
+ fo
X such that
(10.8)
if we have
(10.9)
Let a
IIT(t)II < Me
be.
(10.10)
(10.11)
with
1 v
<
(10.12)
b1M.
E
X such that
(10.13)
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
e b ( t s) Ilu(s)11 ds.
cp(t) G C1 + C2
Since 8> vM, we conclude that if MI(xlI <_ R o , then T = +oo and (10.15) holds
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
(10.16)
(10.17)
(10.18)
Proof. We have shown in Corollary 3.5.10 that the contraction semigroup To(t)
generated in C o (12) by the equation
u -Au=0 inR + x52,
JSlu=0
onIII+x81
t
(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
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)
such that u(0) = uo tends to cp as t -> oo. Moreover, we have
Vt > 0,
(10.21)
coo
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
I
I
I
I
I
I
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
'
J {IVwI
f
f
{Ow
VW E D(Q), f{VwI 2 + k()w 2 }dx =
z
=
{^Vw^ 2
+ (A/)w 2 }dx
J {^^w^
J IVw (w/cp)V pI
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),
(10.24)
(10.25)
(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.
dtfin
- I 2 dx =
f( ) _ V(^) lu -
}dx. (10.27)
in w.
(10.28)
dJ
t
Ju
2 dx < 2
with
c(x)
J {V(u
(10.29)
.f (P)
(P/2)
if x E w
if x ^{ W.
J w 2 dx.
f1
(10.30)
Proof. We introduce
6 = inf {
J {IVwI 2 + c(x)w
l ^z
2 }dx,
f^
52 w 2 d x = 1 } .
w E H0'(),
JJ
(10.31)
(10.32)
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)
Vt > T,
In fact, (10.27) and the convexity of f also implies that II u(t, .) W11 2 is nonincreasing; hence
(10.34)
for some K > 0. Then, since u and z remain bounded in C 1 , from (10.34), we
deduce that
Vt > 0,
(10.35)
Remark 10.3.5. The main result of Theorem 10.3.2 can be viewed as a property 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.
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
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,
(10.36)
xaSl
z=0 onR +
I
1
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)
If(u)I <C(1+Iul')
-4
JR with f (0) = 0
a.e.onJR
withr>0arbitraryifN=1or2and0<r<N/(N-2)ifN>3.
We obtain the following result.
(10.40)
u=0 onR+x852.
(10.42)
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
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 stationary 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.
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 + ,
(10.44)
coxx + f ( p) = 0.
,
w + f'(^p)w = 0
in (0,L).
Let
in
z 2 dx = 1}.
This is impossible since V) is not identically 0 and yet vanishes on (0, a). Therefore 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
=J
s^
= 77
s^
ro
Jn f'(cp)weV)dx
Js^ 6 (E)1weIV)dx
J w,V)(x)dx fn b(E)jweI0dx
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
=f
,b(z)w(0, x)dx = e
J 1p2(x)dx.
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.
Let S2 be any bounded open domain of RN, and let w be any
Lemma 10.4.2.
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.
By definition,
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) associated with A, (w; A + p(x)I) and let us consider the extension
Ho (Sl) of
w by 0 outside w. Since
fw
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
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].
Bibliography
ADAMS R. A.
[1] Sobolev spaces, Academic Press, New York, 1975.
1984, 1-34.
170 Bibliography
Bibliography 171
I
I
I
I
172 Bibliography
[4] Introduction aux problemes d'e'volution semi-lineaires, Mathematiques et Ap-
Bibliography 173
DAFERMOS C. M.
[1] Contraction semigroups and trend to equilibrium in continuum mechanics,
174 Bibliography
[2] Asymptotic behavior of positive solutions of a nonlinear heat equation, Houston 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 parabolic problems, Proc. Am. Math. Soc. 102 (1987), 131-136.
FIFE P. C.
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 Lecture 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,
Bibliography 177
178 Bibliography
Bibliography
179
[2] On the Cauchy problem for the (generalized) KdV equation, Advances in
Math. Supplementary studies, Studies in Applied Mathematics 8 (1983), 93128.
[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 dissipation, 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, Publication du Laboratoire d'Analyse Numerique #84011, Univ. P. et M. Curie,
Paris, 1984.
I
I
180 Bibliography
Bibliography 181
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 reactiondiffusion equations, Hokkaido Math. J. 12 (1983), 360-370.
MATANO H.
[1] Asymptotic behavior and stability of solutions of semilinear diffusion equations, 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 equations, 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 equations, Applied Math. Sciences #44, Springer, New York, 1983.
182 Bibliography
RABINOWITZ P. H.
[1] Periodic solutions of nonlinear hyperbolic partial differential equations, Communs. 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.
H. Brezis and J.-L. Lions (eds), Research Notes in Math. #70, Pitman, 1981,
326-336.
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 dimensions, J. Diff. Eq. 52 (1984), 378-406.
[4] Decay estimates for the three dimensional Klein-Gordon equation and applications, Communs. Part. Diff. Eq. 14 (1989), 1421-1455.
Bibliography
183
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 applications to geometric problems, Ann. Math., 118 (1983), 525-571.
SJOLIN P.
[1] Regularity of solutions to Schrodinger equations, Duke Math. J. 55 (1987),
699-715.
SMOLLER J.
[1] Shock waves and reaction diffusion systems, Springer, New York, 1983.
STRAUSS W. A.
[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,
1978.
STRICHARTZ M.
[1] Restrictions of Fourier transforms to quadratic surfaces and decay of solutions
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,
TSUTSUMI M.
[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.
[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,
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.
C^
G
1
I
I
I
J
I
'
I
I
I
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, Communs. Math. Phys. 87 (1983), 567-576.
[2] Lyapunov stability of ground states of nonlinear dispersive evolution equations, 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 applications 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), 163166.
YOSIDA K.
[1] Functional Analysis, Springer, New York, 1965.
Index
a priori estimate 58, 135-7
adjoint 22-3
attractor 152
behaviour, asymptotic
120-2, 142
blow-up 72-6, 87-90, 114-21
blow-up alternative 57, 58
Bochner's Theorem 7
boundary regularity 28, 62, 124, 134,
146
compactness 124-41, 143-4,.151
complete metric space 56, 124, 147
conservation of energy 78, 83, 92, 100
continuous, absolutely 13, 14, 15, 53,
55, 139, 141,
contraction
semigroup 39-41
strict 56
convexity 6, 148
damping 50-55, 134-41
dependence, continuous 59, 100
differentiable, differentiability 38, 51,
60, 62, 78
differential inequality 72, 73, 75, 87,
88, 116, 117, 120, 125,
131-3, 137, 139
distribution 2
vector-valued 10
domain
bounded 27, 43, 46, 62, 68, 74, 124,
134, 141, 146
of an operator 18
dynamical system 142, 143, 145
eigenfunction 72
embedding, compact 3, 151
energy, see conservation of energy
equation
inhomogeneous 50-55
non-autonomous 50-55,134-41
parabolic, see heat equation
with second member, see equation, non-autonomous
equilibrium point 144, 148, 152
estimate, uniform 67, 68, 84, 112-13,
130, 136, 139
existence
global 58, 65-71, 76, 83-6, 112-14
local 56-9, 64-5, 76, 82-3, 100-12
extrapolation 163
forcing 50-55,134-41
function
integrable 7-8
measurable 4-6
functional analysis 1
Gagliardo-Nirenberg's inequality 3,
112
generator 39
graph, closed 18, 20
Gronwall's lemma 55, 125
186 Index
Pettis' Theorem 5
point
fixed 57
hyperbolic 156
regularity 28, 33, 35-7, 39, 41-3, 47,
48, 60-1, 62, 78, 100
Schrodinger equation 47-9, 91-123
set, connected 143
smoothing effect 37
Sobolev embeddings 3
Sobolev space
2, 13-17
solution
bounded, see estimate, uniform
global, see existence, global
maximal 57
stationary 157
stability 154
asymptotic 154
exponential 158-64
Strichartz' estimates 96-100
trajectory 142
bounded, 144-7
relatively compact 143
uniqueness 56, 106
variation of the parameter formula 50
wave equation 47, 78-90, 130-3, 137-41, 149-52, 163-4