Nonlinear Analysis, Theory, Printed in Great Britain.
Methodr
& Applications,
Vol. 6, No.
11, pp. 1207-1220,
1982.
0362-546&/82/l 11207-14 $03.00/O @ 1982 Pergamon Press Ltd.
DISSIPATIVITY IN THE SENSE OF LEVINSON FOR A CLASS OF SECOND-ORDER NONLINEAR EVOLUTION EQUATIONS ALAIN HARAUX Universitk Pierre & Marie Curie, Laboratoire d’Analyse NumCrique, (L.A. 189) Tour 55/65 5kme E., 4 Place Jussieu, 75230 Paris Cedex 05, France (Received
10 February 1982)
Key words and phrases: Levinson dissipativity, nonlinear evolution equations, second order. 0. INTRODUCTION LET V, H BE
AND NOTATIONS
two real Hilbert spaces such that
with continuous and dense embeddings. in V, H and V’. We set
We denote respectively by ( 1, 1111and 11]I* the norms
&=
Inf
U2
II II 2.
uEV,u#O 0u
The duality pairing in V’ x V will be represented by the symbol ( . ). Let Q, E C’(V, [w) and g : Iws x V-, V’ a ‘multivalued’ variable operator. We investigate boundedness properties of the eventual positive trajectories of u” + O’(u) + g(t, u’(t)) 3 0. In section 1 of this work, we prove that under some assumptions on @ and trajectories (defined in an adequate manner) are such that u E Y’(lR+, V) If in addition, s(t, . ) does not vanish too fast as t + t 00, the dynamical in V x H by (1) is dissipative in the sense of Levinson, which means there that for every solution u of (1) on IF!+,we have lim ]Iu(t>IIS R,lim
t--*+=J
(1) g, all the positive f7 W’*m(Iw+,H).
system generated exists R > 0 such
(u’(t) 1s R.
t++@J
This result is applicable to the class of purely dissipative systems considered in [4]. In section 2, we construct strong global solutions of U’I
-
Au + g(u) + P(u’) W(m)
on IO,T[ X Q on IO, T[ X aS21
u=o
(2)
where S2 is a bounded, open domain of Rn with smooth boundary, p is a maximal monotone graph in R x Iw and f, g satisfy some relevant hypotheses. The results of section 1 are then applied to the solutions of (2). In section 3, we show that if p-’ is uniformly continuous: IF!--) R and f, g satisfy additional 1207
1208
A. HARAUX
assumptions, the dynamical system generated by (2) has stronger dissipativity properties, which can be used for example to construct bounded solutions: R ---) V x H. We also prove an existence theorem of periodic solutions under suitable assumptions on f, g and P*
Notations Throughout the course of this paper, we will use the following conventions. The derivative with respect to t of a function u : IO, T[ + X will be denoted by u’(t) (even by u’ if the dependence on t is clear). with a vector function: IO, T[ + VJ’(S2) (m E Z), we If u:]O, T[ x Q + IT8is identified identify k&t and u’. A map between two metric spaces f:X * Y will be said to be ‘bounded’ if, for every B C X which is bounded, f(B) is bounded in Y. Let X be a (real) Banach space and J = IR’or J = [0, m[. We set t+l
Sp(J, X) e Sup
fE
tEJ
I
If(e)I”xde<
t
+m.
A function u : R + X is said to be S1-almost periodic if u E S’(R, X) and z--, u(t + z) is almost-periodic as a function R + s’(lR, X), endowed with the norm defined above. If p is a (possibly multivalued) maximal monotone graph in R x [w, we set
We use the usual terminology concerning Sobolev’s In section 3, the ‘energy’ space will be denoted by
spaces in S2 or IO, T[ x Q.
%Z=VxH. We assume that the reader is accustomed with the properties of the usual wave equation. For the monotonicity properties and weak solutions of purely dissipative systems, we refer systematically to H. B&is [5]. The development of sections 2 and 3 relies in an essential manner on some results of [2,6] and [ 111, respectively. 1. SOME
ABSTRACT
THEOREMS
In this section, we do not study the Cauchy problem associated to (1) but we prove boundedness results for eventual global solutions. We assume that, for a.e. t E lR+, a (possibly multivalued) operator g(t, . ) : V + V’ is defined and moreover that Q and g have the following properties
3fE
syR+), 3c
2
o(vu E
v, a.e. on R+
whenever
z(t) E g( t, u),
a.e. on R+
(5)
Dissipativity in the sense of Levinson for a class of second-order
3p > 0 such that #k(z)
3k E IT([w+, Iw+),
Ess inf k( 19)2 p Ess sup k( e>, [v+ 11 [t,t+ 11 and 3h ~s’(lR+),
nonlinear evolution equations
h > 0 such that for every
dz>
0,
a.e on II%+
[u, z(t)] like in (5)
(6)
we have
a.e on R+. Remark.
If k is continuous,
(6) is easily seen to imply,
k(t + 0) 3 pexp(-MB)k(t), while the above
inequality
implies
W3
(6) with p replaced
in turn
(7)
for some M related
Iftao,
1209
to p
0,
by p exp( - M).
THEOREM 1.1.Let u EVV/$‘(lR+, v) n N$$i([w+, V’) satisfy -u” - Q’(u) E g(t, u’(t)) a.e on R+. Then, u E L"([w+, V) and u’ E Lco(R+, H). Moreover, there exists K: IR’ x [w++ [w+ bounded
such that vt 2 0, IIu(t) II +
and K only depends The proof
upon
of theorem
Iu’(t) I s wll @) ILlu’60 I)
a, p, Cl, C2, p and the S1-norms 1.1. will be a consequence
(8)
of f, h and kh.
of the following
technical
lemma.
LEMMA 1.2.Let 1 =4&'/2d'2p? Assume
that,
for some t E R+, we have t+l (u” + W(u),
It Then,
0
we have sup {ilu’(t) tE[t,t+ l-j
where
u’) dza
P is bounded:
I2 + @(u(z))}
(R’)‘+
s flllcu,
l/B, C,
Cl,
c2,
lip,
a, b, 4
Iw+ and
t+l t+l a= SUP h(r) dz, b = Sup k( r)h( r) dz tao I t t*o I t t+z c = SUP f (9 dr tao I t Proof of lemma 1.2. We introduce p(r) It is clear that p(z) is differentiable
the numerical =
function
3)u’(z) I2+ @(u(r))*
a.e on It, t + I[ and
p’(z) = (u” + W(u), From
(7) and (9), we deduce,
24’)
a.e
with 0 = p? I(ilk(r)h(r) t
/+‘k(z)lu’(@12drs t + j-t+f lu’(z> t
I2dzs
;.
dz
(10)
1210
A. HARAUX
On the other
hand,
if we define 5( z= )
-(u”
+ W(u),
u’(z)) + k(z)h(z),
we have 5 2 0 and which implies t+l
t-f-1 (p’(z) ( dzs
I
((u” + W(u), u’)l dz=
2b.
t
For convenience
we introduce
Q = SUP b’(Z>I, [f,t +r] and P = sup P(T). [V + r] From
(11) we get:
From
(12), it is immediate
t+l
1
Ps2b+i
P(Z) dr*
It
to deduce 1 t-t1 P s 2b + Laa-ll-’ + Q>(u(z>) d?;. 2 1
Since ilu1l* G K’[@(u)
+ Cl], we have
vz E
+ Cl)“*,
1(u(z), u’(r)) 1s &“*Qa-“*(P
[t, t +
I].
As a consequence: t+l
I
t+l
@‘(u(r), u(z)) dz = -
t
I
(u” + z, u) dz
t
t+l =
ul)l:+‘+ [
-[(u,
lu’(z>
I* dz-
I”i(z(z); t
u(z)) dz
s 2&r l’* a- Ii2Q(P + C1)‘/* + aa-’ + d’*(P
+ Cl)‘/* \‘+f {C[ IL+) t
I2+ (z(z), u’(r))] +f(z)}
Or .. t+l
I t
W(u($), u(z>>dr s
+ aa-’ +
a+*(P
2&l/*a-
+ Cl)“*
“*Q(P
[Caa-’
+ C,) ‘I*
+ c].
dz.
Dissipativity in the sense of Levinson for a class of second-order
nonlinear evolution equations
1211
Since Qz G 2P, from (14) we deduce t+l
I
(@‘(U(Z)),
U(Z)) dzs
23’2Ai1’2#2[P(P
+ C#‘*
+ aa-’
t
+ From
(4)
+ c)(pl’* + cy’).
c@‘*(Ca6l
we find 1 t+l - j@(u(r)) 1 t
On combining
dzs
p-ll-’
the two lines above
l”i t
(@‘(U(Z)), u(r)> dr + C$-’
with (13)
we finally obtain:
P s 23i2&112~‘2~-11-1(P + ;Cl) + M(Z)P’” + N(1) M(I) = (Ca/j-1611-1
+ c~~1Z-1)a~1~2
N(l) = C2P_l + ad1Z-1($
+ /3-l) + 2b + d*C:/2(Caa-’
+ c) p-‘l-?
From (15), and since I >23’2&1’2$12p-1, we get a bound P for P, which has the expected This ends the proof of lemma 1.2. Proof of theorem 1.1. Let t 3 0. From lemma 1.2, it is clear that P(t + I) s Max@(t),
As an immediate
consequence,
G Max{&
(16)
THEOREM
(17) and (3)
there
(16)
1.3. In addition
+ b.
(17)
it is clear that (8) holds for some function to (3)-(7),
exists R > 0 such that,
+oO k(t) dt = +m.
for any function
lim ilu(t) 11s R, t--*+m
I (18)
u as in the statement
there
to+1 p’(z) dza
-
/“”
I*dz < la-’ + ad
k(z) dz
to +
to
1u’(r)
of theorem
lim 1u’(t) 1s R. t* +m
Proof of theorem 1.3. Since p(t) 3 0 and (18) holds,
I
K as stated.
assume
I0 Then,
Sup p(t)}. Ostsl
we have Sup p(t) q(O) Ostsl
From
P}.
we find Sup&t) tao
Since p’(t) s h(t)k(t),
form.
exists to s 0 such that
1.1.: (19)
1212
A. HARAUX
Proceeding
as in the proof
of lemma
1.2 with P=
SUP P(z)? [toJo+Il
we find that (15) holds with N(I) and M(f) slightly increased. right-hand side of (15) remains unchanged, we get clearly
SUP
[to,to+4 Then,
like in the proof
P(z>w*
of theorem
Remark.
of theorem
If k E L’(R+),
2. GLOBAL
SOLUTIONS
for some p* 2 I?
1.1, we conclude s Max{P,p*}
Sup&t) tat0 This ends the proof
= p”
1.3.
the conclusion AND
of theorem
1.3. may fail.
DISSIPATIVITY FOR A CLASS EQUATIONS
OF SEMILINEAR
Let Q be a bounded, monotone
graph
open subset of Rn with a smooth boundary in IF!x R, and a function g EVVt$(R). We set G(u) = Iug(B) 0
Throughout
this section,
THEOREM
2.1. We assume
I’ = &2, p a maximal
notation
H = L2(Q), V’ = H-l@),
% = {u E H, u(x) E o(p>
W = V n H@).
a.e in S2}.
exists (M, D) E [w+ x Iws such that:
that there
VUER,G(U)+M~U~~+DZO. If n > 1, we also assume
that there lg’(u)l
(21)
exists s E R+, (n - 2)s c 2 and K E IFi+such that: s K(l + luls>,
a.eon II%.
f E L1(O, T; H) and [uo, uo] E V x %, there E C(0, T; V) n C1(O, T; H) such that
Then, for any
(22)
exists one and only one u(t, x)
u(0) = u(), u’(0) = U(-J u is a weak solution
WAVE
Vu E R.
de,
we will use the following
V = H;(G),
of P in the
Since the coefficient
on [0, T
(23) ] in the sense of [6, theorem
III. 6,
p. III. 141 of the equation u” - Au + p(u’) 3
ft, n) - g(u(t, x)).
V X % X L’(0, T; H)+ Moreover, the map: (UO, ~0, f) -+ u(t) is continuous: Cl@, T; H), and we have the following additional regularity properties. If
(24) C(0, T; V) n
f E L1(O, T; V) and [uo, uo] E VV x V, then u E L”(0, T; w) n vlym(O, T; If)*
Dissipativity in the sense of Levinson for a class of second-order
nonlinear evolution equations
1213
of p If f E wyo, T; H), u. E V and [ -Au0 + &uo)] f7 H # 0, where b is the extension in the duality between V and V’, then u E W1y"(O, T; V) n W2yoc(0, T;H). If we assume that D(p> = R, then for any [uo, uo] E V x % andf E L1(O, T; H), the solution u of (23)-(24) is such that
p”(u’)
E
x Q), u” - Au E L’(]O, T[ x Q>
L'(]O, T[
and u” - Au + g(u) - f(t, x) E +(u’(t,
x)) a.e on 10, n x Sk
Proof of theorem 2.1. From (22) it is easy to deduce V - H and satisfies Vu E V, N.J E V,
that the map u + g(u) is bounded:
(25)
I&) - m I 6 fYll4~Ibll)llu- 41 where P: [w+x IRS*
IRS is bounded.
(a) Uniqueness and continuous dependence on the data Assume that u1 and u2 are two weak solution s in the sense of [6] of the respective
We set w = ui - u2. Then,
u;’ - Au1 + B(d) 3f1(t,4
- g(4)
u!i -
-
AU2
+
P(4)
3f2(u)
equations
g(uz)*
for all t E [0, 7’1:
{IIw(t)II2+ Iw’(t)12Y2 s {II40) II2+ Iw’(o) lY2 + I flfl(s) 0
-f2Ml
ds
+ I t fw1(s) ILIMS)II)lb449 IIh* 0
Thus,
with A =Su. P( llul(t) (1,/uz(t) II), we find 109
{IIw(t)II2+ Iw’(t)12P2 s exp(A mll w(O) II2+ I w’(O) w2 + I0 T Ifl(s> -
I w*
f2b)
(26)
This formula shows clearly that if fi = f2 and w(0) = w’(0) = 0, then w = 0. Also, if we prove that the map: (UO, ~0, f) -+ u is locally bounded: V x % x L1(O, T; H) + C(0, T; V) n C1(O, T; H), the continuity will follow easily from (26). (b) Solution for a regularized system In this paragraph, we assume that
P is
[uo, uo]E
Lipschi tzian,
x Kf
E
qo,
and TY V).
Then the operator w * p(w) is bounded: V- V and Lipschitz continuous: Ii+ Ii. On the other hand, the operator u ---) g(u) is bounded: VV--, V and Lipschitz continuous: V+ H in bounded subsets of MJ. Therefore, by using for example [lo, theorem 10, p. 121 we obtain at once the existence of a solution of (23), (24) in some subinterval [0, S], 6 G T, which is in the space C(0, 6; IV’) n C’(O, 6; V) n C2(0, 6; H).
1214
A. HARAUX
To prove that 6 = T, according to the standard procedure we need to find a priori bounds for u(t) in VVand u’(t) in V. As a first step, we multiply (24) by u’ and integrate over 52, thus obtaining
Let r(t) = IMt) II2+ I ~‘W I2+ 2{Wt)
2+
G(u(t)) I 52
dx + D IS2I} + 1.
Then, because of (21) (22) we have
1 + II40 112 + l u’(t)l2s 40 s Q( IIW 11, Iu’(t) I> with Q: lR+ x lR+ * IF!+ bounded. We set F =Sg If(t) 12dt. A s a consequence
of (27) we derive successively
r’(t) s 2 If(t) I Iu’(t) I + 4M(u(t),W)>
s [I + If(9 II (1+ Iw> 19+ 4M2I40 I2 s (1 + 4M2&l + If(t) I}r(t). Thus, for t E [O, 61, we have
r(t) s Q( Iluoll,Iuol)exp([l + 4M2AT1]T + F)
= R(llhIll,InIl,F)* Now we multiply (24) by -A& for all h > 0 ~$(IJ~~vu’~~ G
(28)
= -A(1 - AA)%‘,
h > 0. Setting JA = (I - AA)-‘, we obtain
+ IJ/~AU~~)
IlfOIIlb’II+ @A w> + lb IIl%w> I7vf E [OY 61
(because JA: H + H is a contraction). It is classical that (AAPV,pw) s 0, VW E H. Also:
IV(&)) I =
k’(4V~
I2 s C’ I Au I km)
In s c”l
Au 12,
where u = u(t), t E [O, 61 and C’ only depends on IIu& (ug( and F. (by (22), w + g’(w) is bounded: V + L”(S2)). Finally, with z(t) = /u’(t)ll’ + IAu(t) I2+ 1, we find z(t) s r(O) exP{C”(lluoll, (uol,F)T
+ j-*lV(9 IIdBJ,vi!E [O,q* 0
(29)
From (28) and (29), we deduce the solution u(t) is global and (28), (29) hold with 6 = T. (c) Solution
when [uo, uo] t
We consider a sequence /%, A > 0.
W x V n % and f E L’(0,
T; V)
fnE C(0, T; V) such that fn + f in L’(0, T; V) and replace p by
Dissipativity in the sense of Levinson for a class of second-order
As a consequence
nonlinear evolution equations
1215
UA,~of
of (b), we find solutions
p&l - AUA,rz + h(4,n) = fn- g(+z) It is clear by (28) and (29) that u A,nremains uniformly bounded in C(0, T; w> n C1(O, T; V). Let i& be a subsequence of the sequence u~i,,~ such that tik+ U in C(0, T; V). Then, clearly, g( &) * g(a) in C(0, T; H). A simple application of [5, theorem 3.16, p. III. 501, allows to conclude that U is a solution of (23), (24). We’ omit the details in checking the hypotheses of [5, theorem 3.161, which are very similar to [ 11, Lemma 34, p. 227-2291. Finally, because of the estimates (28) and (29) which are independent of n and A, we find that u E L”(0, T; W) n W1ym(O,T; V). This proves
the first ‘additional
regularity
property’.
(d) The General case and additional statements Finally, if [uo, uo] E V x %, we consider sequences [uom, ~0~1 E W X V n %, V) with uorn* uo in V, uom+ uo in H and fm-+f in L1(O, T; H). We use (c) to construct the solution urn(t) of
u;- bn + /q&n>3fmW) 1umc9= u()&&(O) = UOm.
lim urn(t) m_,+m
T;
- g(w?J
As a consequence of (26) and since by (27) u, is bounded in C(0, immediately that urn(t) is a Cauchy sequence in C(0, T; V). Let u(t) =
fmE L1(O,
T; V),
we can see
in C(0, T; V) strong.
It is clear that u satisfies (23) (24) since fm(V>- g(4&,x)>-+ f(t, x)- g(u
of section
Q(u) = +(I*
THEOREM 2.2.Let
f E S*([Wf, II).
1 to the solutions
+ 1 G(u) dx, Q We assume
3eO,g(u)o
W(u)
of (24). Then we have = -Au
that g satisfies
-(Ao-
(22) and
E)Iu~*- CJul
3S E IF!+,G(u) s S(1 + ilolul* + ug(u)). Moreover,
we assume
that p satisfies,
+ g(u).
(30)
for some q > 0
(31)
A.HARAUX and one of the following
conditions
(a >n
= 1 and Int(D(@)
3 0
wn
= 2 and ]@‘wl sC(l
+ 1~1’) for some r E Rs
(c) ns3and]/Pwl
there
holds
i,l”),r
exists R E [w+ such that,
for any solution
lim ]Iu(t) 11s R, t++m Remark 2.3. Conditions
(30) are fulfilled
q(u)
u E C([w+y V) n C1(Iw+, H) of (24)
lim Iu’(t)I s R. t+ +=J if, in addition
to (22)
g satisfies
s aluI’+* - E, lfu E R
the condition (32)
for some a > 0, and E E R+. An alternative sufficient condition is g(0) = 0 and g nondecreasing. Proof of theorem 2.2. Since (~0, ~0, f) + u(t) is a continuous map: V X % X L1(O, T; H) + C(0, T; V) f7 C1(O, T; H), it is enough to check the estimates proved in theorem 1.3 in the special case when [UO,V] E IV X V, p”( ~0) E H and f E C1([w+, V). Then, the compu-
tations made at theorems 1.1-1.3 are justified, provided we are able to check (3) (4) (5) (7) (with k(t) = 1). U n d er any of the conditions (b) or (c), the multivalued operator: g(t, u) = pu - f(t) maps V into V’ in a bounded manner for almost every t E R+. Just like in [4], we check for some CE IW’:ilh E Pu, IlhlI* aC(1 + (h, u)), VU E V. Thus, for all u E V and z(t) E g(t, u)
Ilmll* s Ilf(t)ll*+ cc1+ Wan4 + cm 4) s am 4 + IuI2+ Ilf(9II*+ C’lf(0 I* + c* In case (a), we first replace p by PA and make 3c-+ 0 (the associated to u). In both cases, (7) is an easy consequence of (31) since
solution
UAdoes converge
Finally, it is straightforward to check that (22) and (30) imply (3), (4) for some suitable values of a, p, Cl, C2. This ends the proof of theorem 2.2. Remark 2.4. Theorem 2.2. contains as a special case [4, theorem 1.1 p. 4251 when h = 0, and shows that the system considered in [4] is dissipative in the same of Levinson in HA(Q) x L*(Q). This result remains true if A > 0. Also, when A = 0, we may generally get the same results with f E S1(Rf, H) instead of S*, as shown by the following result. THEOREM
2.5.
In the
hypotheses (a) by
of theorem
2.2, let us replace fE S*(R+, H) by fE
t S’(R+, H) and condition
(a’ 1n
= 1 and either
D(p> = IF&
Then
the conclusion
of theorem
or -D(p) = [p, v] with ,u < 0 < v. 2.2 still holds.
Dissipativity in the sense of Levinson for a class of second-order
1217
then u’ is a priori bounded in 1 and D(p) is bounded, is enough to follow the steps of the proof of 0 E Int(D(P>) S1(R+, H). or (c) hold, we prove that for any E > 0, there exists C(E) such
Idea of the proof. If n = Locl(lR+ x B), and the condition theorems (l.l)-(1.3) when f E If n = 1 and D(p) = II%or (b) that Vv E Indeed,
nonlinear evolution equations
V, Vh E pU, llhll*s E< h, v > + C(E).
(33)
if n = 1 and D(p> = IR, we have
I
h(x) v(x) dx 2 f E
lup-l/E
lwl dx
Ijopll&
lb(x) 1dx s E 1 h(x) v(x) dx sz
3 lh(l s E< k v ’ + IQ1sup IP( (UI
Then (33) follows easily since L1(Q) G H-l@). If (b) or (c) holds, we know from [4] that
;+“-s llhll Since Ilh(l*sdlhlly ur + K(E) , (33) follows there
C(l + (h, v))
easily.
From
(33), we deduce
that for every
E > 0,
exists fE E S1([Wf) such that
Ilmll* s E(z(t)74 + E(f (t) 74
+ f E(t)*
(34)
Then, by following the lines of the proofs of theorems (l.l)-( 1.3), on using (34) with E small enough we can get an estimate on P with the weaker hypothesis f E S1(R+, IS). We omit the details which are tedious but very similar in essence to the proof of theorem 2.2. 3. COMPACTNESS
The following
AND EXISTENCE
result is an easy consequence
OF PERIODIC
of theorem
TRAJECTORIES
2.2 and the methods
4 .11. THEOREM 3.1. In addition
to the hypotheses (n - 2)s < 2,
We assume
moreover
p-l:
of Theorem na3,r<-
andif
R + IR is uniformly
2.2, we assume n+2 n - 2’
continuous
t+l
_I
t>O t SUP Then,
all solutions
If@+
h) -f(e>I
de-0
ash+O.
of (24) on lR+ are such that U(t) = [u(t), u’(t)] is uniform1 y continuous:
with a precompact
range
in X.
R+ + X
of [ 11, theorem
1218
A. HARAUX
Proof. Since u(t) is bounded in V and u’(t) is bounded in H, it is clear that g(u(t)) uniformly continuous IT%+ + H. The rest of the proof is a simple application of [ 111.
is
COROLLARY3.2: In addition to the hypotheses of theorem 3.1, assume that f is defined on II%and f: R + H is S1-almost periodic. Then, there exists at least a solution of (24) on IF!,with the range of [u(t), u’(t)] precompact in X. idea of the proof. Let ii(t) by any solution of (24) on lR+ and choose a sequence {tn}, tn + 00 such that f(tn + t) -+ f(t) in S1(R, H). Since ti: R + X is uniformly continuous and
lies in a coqpact
subset
of X, we may select a subsequence {&} of {tn} such that u’(t) respectively in C([a, b]; V) and C([a, b]; H) for all a, b with - 00 < a < b < + 00. Then u(t) is a solution of (24) on R, which is uniformly continuous with the range of [u(t), u’(t)] precompact in ‘X. ii(t + &,J + u(t) and S(t + &) *
Remark 3.3. The proof of corollary 3.2 shows actually that there exists a solution of (24) bounded on R in the ‘positive hull’ of any positive trajectory of (24). We have been until now unable to prove the existence of an almost periodic trajectory. THEOREM3.4. Assume that the hypotheses of either theorem 2.2 or theorem 2.5 are satisfied and fi R + H is T-periodic. If in addition (n - 2)s < 2, there exists at least a T-periodic curve U(t) = [u(t), u’(t)]: II%* X such that U(t) E C(R, ‘X), and u(t) is a weak solution in the sense of [6], of equation (24). Proof of theorem 3.4. We use a Schauder method and monotonicity For E > 0, A-E [0, l] we look for a eventual T-periodic solutions of the
combination of compactness arguments via the Leraytheory. uniform apriori bound in C(0, T; V) x C(0, T; H) for the system
p&A - V&J + EU&,A= 0 I v;3A - Au, ,A( + B(VE 9A>+ As a first step, for h E P(0, solution uE of
&&,A
3f(t)
- g(Au, ,A)*
T: H), we derive a priori estimates for the (unique)
(4 - VE + EU, = 0 c,d.- Au, + p(u,) + &vE= h.
(39 T-periodic
(36)
For E > 0 fixed, the map: h + [uE, u,] is Lipschitz-continuous L’(0, T; H) --) C(0, T; V) x C(0, T; H)* If h is replaced by h, E W’?‘(O, T; H), then as a consequence of [5, theorem 3.14, p. III 411, the solution uE of (36) is strong. Multiplying the second equation of (36) first by vE, then by
Dissipativity in the sense of Levinson for a class of second-order u,,
nonlinear evolution equations
1219
we obtain with zE(t) =h - u; + Au, - EU~ T
I0
W),
bW>
dt c
Tv@), I0
(37)
uxt)> d
T T( - A uE,uE) I0
dts
I0 +
(38)
Tllr,(911*IIUEWIIdt I0
- g[&(t)],
If in (37), we plug in h(t) =f(t)
(h(t), u&Q)>dt + 0TlUr(f)12df I
we obtain
IT (z,(t), M)) dt s IT If(t>I lvxt>I a + cE(I + I,‘luXtV
d)
0
0
(since (21) and (30) imply -ug(u) (22) imply T (g(&), I0
s C(l + u2)) because u, EC(O, T; V) n C1(O, T; H)
u;) dt = ; j=- G(&(t,
x)) d,]’
and
= 0. 0
Then, by the same method as for the proo on
of lemma 1.2, we can use (38) to get an estimate
au,), UJ dt, hence on
IT {; llud12 + ; j-- G(W b} dt* 0
We end up with an estimate of the form
sup tqo, sup ~lluE,*(t)l12 + lUE,A(t> I21< + O” q
O<&Q&o
OsAsl
Since the operator T(u) =f(t) - g(u(t)) is completely continuous: C(0, T; V) n Cl(O, T; H) +P(O, T; H) , we conclude from Leray-Schauder’s theory that (35) has at least a solution uE,~for 0 < E G ~0 and h E IO, 11. If we keep now A = 1 and make E+ 0, we may assume that for some sequence E, + 0, we have u,,~ =i&--,u in C(0, T;H) aswt-, +m and [ u,, z&] remains bounded in C(0, T; X). Then, by compactness, g(z&) converges in C(0, T; H) to g(u) as vz ---) + 00. As a consequence of [5, theorem 3.16, p. III 501, it is clear that u is a weak solution of (24) on [0, 7’j. This ends the proof of theorem 3.4. REFERENCES 1. AMERIO L. & PROUSEG., Abstract Almost Periodic Functions and Functional Equations, Van Nostrand-Reinhold (1971). I 2. BENILANPH & BRI~ZISH., Solutions faibles d’kquations d’kvolution dans les espaces de Hilbert, An&. Inst. Fourier Univ. Grenoble 22, 311-329 (1972). 3. BIROLI M., Bounded or almost periodic solutions of the nonlinear vibrating membrane equation. Ricer&e 22, 19&202 (1973).
Mat.
1220
A. HARAUX
4. BIRIXI M. & HARAUX A. Asymptotic behavior for an almost periodic, strongly dissipative wave equation, J. dif$ Eqns. 38, 422-440 (1980). 5. BRBZISH., Opkateurs Maximaux Monotones et Semi-groupes de Contractions dam les Espaces de Hilbert, North-Holland, Amsterdam/London (1973). 6. BRI?ZIS H., Probltmes unilatCraux, J. Math. pures appl. 51, 1-168 (1972). 7. BRBZISH. & LIONS J. L., Sur certains problkmes unilatCraux hyperboliques, Cr. Acad. Sci. Paris, Ser. A 264, 928-931 (1967). 8. BR~ZIS H. & NIRENBERG L., Forced vibrations for a nonlinear wave equation, Communs pure appl. Math. 31, l-30 (1978). 9. FICKEN F. A. & FLEISHMAN B. A., Initial value problems and time-periodic solutions for a nonlinear wave equation, Communs. pure appl. Math. 10, 331-356 (1957). 10. HARAUX A., Nonlinear evolution equations: global behavior of solutions, Lecture Notes in Mathematics 841, Springer, Berlin (1981). 11. HARAUX A., Almost periodic forcing for a wave equation with a nonlinear, local damping term, to appear. 12. LERAY J. & SCHAUDER J., Topologie et Cquations fonctionnelles, Annls scient. Ecole norm. sup. Paris 51, 4578 (1934). 13. LIONS J. L. & STRAUSS W. A., Some nonlinear evolution equations, Bull Sot. math. France 93, 43-96 (1965). 14. PRODI G., Soluzioni periodiche della equazione delle onde con termine dissipativo non lineare, Rc. Semin. mat. Uniu. Padova 33 (1966). 15. PROUSE G., Soluzioni quasi periodiche della equazione delle onde con termine dissipativo non lineare, I, II, III, IV, Atti Accad. naz. Lincei Rc. 38-39 (1965). 16. STRAUSS W. A., On continuity of functions with values in various Banach spaces, Pacific .I. Math. 19 543-551 (1966).