This content was uploaded by our users and we assume good faith they have the permission to share this book. If you own the copyright to this book and it is wrongfully on our website, we offer a simple DMCA procedure to remove your content from our site. Start by pressing the button below!
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 = -
J
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
Examples in the theory of partial differential equations 45
1
Proof of Proposition 3.5.7. By density, we may assume that cp E D(A). Let = Icpl. Invoking Lemma 3.5.9, for all t > 0, we have -S(t)( < S(t)cp < S(t)(, almost everywhere on Sl; and so (3.35)
I jS(t)^PI )Lp -< II S(t)CII Ln. We define E C(R N ) by ( on S2; {
0 onR N \S2.
Then we set v(t) = K(t) * , and (3.36)
u(t) = u(t)^ Q - S(t)(.
1
We have u E C((0, oo), C(S2)) n C((0, oo), H l (S2)) n C 1 ((0, oo), L 2 (S2)) and Au E C((0, oo), L 2 (St)). Furthermore, u(t) = v(t) > 0 on 3; u t = Du for t > 0; and u(0) = 0. Thus,
dt f (u
- )2
=-
Juu t
-
_ -
f u Au = f Du • Vu z
-
z
-
_-
J IVu
- I2
_<0,
and so u(t) < 0, for all t > 0. We then deduce from (3.35) and (3.36) that
II S(t) FPII LP C II v(t)II Lp• We conclude by applying Lemma 3.5.8.
I ❑
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
II S(t)'PII L°° < IHIL= < e -At e^ n IIIPIIL. For T < t, it follows from (3.34) that II S(t)^PII L- << (47rt) 4 IIS(t - T)cpII L 2 < (47rt)- 4 e-AteATII^GIIL2
❑
46 The Hille-Yosida-Phillips Theorem and applications
Remark. We cannot take M = 1 in (3.37). More precisely, we have
d
{IIS(t)II'C(x)}t-o = 0 ;
and so, if IIS(t)IIc(x) < M'e - µt 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')
II S(t)II e(LP) <- Mpe-\t,
for all t > 0. Once more, we cannot take Mp = 1. Actually, one can see that, for p > 2, one has Oat
II S(t)Il c(LP) <e,,
for all t > 0, and that this inequality is optimum in the following sense:
a
{IIS(t)II^cLp )}t=o =
t
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
fa IvV)2 I < (A+E)
J IV)IP.
To see this, we consider the first eigenfunction cp l of -A in Ho (S2) (see Brezis [21). For e> 0, let 0, : [0, oo) -+ [0, oo) be such that 6 e - 0 in a neighbourhood of 0, OE(x) < 1 and x - 8 E (x) <_ e for all x >_ 0. Set 2/' _ (9( 1 ))2/P . We verify that Ali, E D(f1). Furthermore, Iv 12 , < IV^pil and, last,
V)
—>
Jn
1pi
as E j 0.
Examples in the theory of partial differential equations 47
Consequently,
J M 2 < fn Iv n 11 2 =A f cp 2 =Alim Ei°Jn 3.5.2. The wave equation (or the Klein-Gordon equation) We use the notation introduced in §2.6.3 and §2.6.4, and we denote by (T(t))tER the isometry group generated by A in X, and by (S(t)) tE R the isometry group generated by B in Y. Proposition 3.5.11. Let (cp, V) E X and let u(t) be the first component of
T(t)(p, z/'). Then u is the unique solution of the following problem: E C(R, Ho (fl)) n C 1 (R, L 2 (1)) n C 2 (R, H -1 ( 1l) ); utt-Du+mu=0, for alltER;
(3.38)
u(0) = cp, u t (0) = z/i.
(3.40)
U
(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)
for alit E R. Finally, if(cp,0) E D(A), wehaveu E C'(R,Ho(St))nC 2 (R,L 2 (1l)) and Du E C(R, L 2 (1)). Proof. Let u E D'(R, H -1 (S2)) and set U = (u, u t ). Then U E C(R, D(B)) fl C'(R, Y) if and only if U E C(R, Ho (52)) fl C(R, L 2 (SZ)) fl C 2 (R, H -1 (1)). Furthermore, in that case, (3.39)-(3.40) is equivalent to the equation
U'(t) = BU(t), for all t E R. The result then follows from Propositions 2.6.9, 2.6.10, and 2.6.11, Theorem 3.2.3, and Corollary 3.3.2. Note that (3.41) is equivalent to (3.22). ❑ 3.5.3. The Schrodinger equation We use the notation introduced in §2.6.5, and we denote by (S(t)) tE 1 and (T(t)) tE R the isometry groups generated by B and A. We have G(B) c G(A), and from this it is easy to deduce the following result. Lemma 3.5.12. For all cp E Y, we have S(t)co = T(t)cp, for all t E R. Then we have the following.
48 The Hille-Yosida-Phillips Theorem and applications Proposition 3.5.13. Let cp E H0'(1) and let u(t) = T(t)cp. Then u is the unique solution of the problem u E C(R, Ho (l)) n C l (R, H-1(l)); ju t +,Lu = 0, for all t E R;
(3.42)
u(0) = W.
(3.44)
(3.43)
In addition,
i
f
Iu(t) 2 =
n IVu(t)I 2 =
f IcpI , for all t E R,
(3.45)
2
J IV I 2 ,
(3.46)
for all t E R.
Finally, if Lcp E L Z (cl), then u E C 1 (R, L 2 (S2)) and Au E C(R, L 2 (fl)). Proof. We use Theorem 3.2.3, Corollary 3.3.2, and Lemma 3.5.12. (3.45) is equivalent to (3.22). On the other hand, invoking (3.7) and (3.22), we obtain
(3.47)
IIAu(t)IIx = IIA(PIIx, for all t E R. But, for all V E Ho (S2), we have
II AvII x = II(Av - v) + vIIi = f IVvI 2 -
JM
2
+ IIvIIX
(3.48)
By (3.22), IIu(t)IIx = Ik IIx for all t E R; and so (3.46) follows by putting 0 together (3.47), (3.48), and (3.45). 3.5.4. The Schrodinger equation in R n' We use the notation of §3.5.3, and we assume that Sl = 1[8N. We can state estimates in the spirit of (3.34). Proposition 3.5.14. For all p E [2, oo] and t 54 0. Then T(t) can be extended to an operator belonging to £(LT (RN), L°(R")). In addition, we have
IIT(t)II,C(Lp ,LP) C for all
(4^rItI)N(2-p)'
(3.49)
0.
Proof. Let cp E S(IRN) and let u(t) E C 00 (IR,S(R N )) be defined by Fu(t)(^) = e-'IE12t.F'G(^),
(3.50)
Examples in the theory of partial differential equations 49
forall^ER' andtER. We have idt
P a(t)(6)
- I6I 2 Fu(t)(C) =0 in R N ,
for all t E R; and so iut + 0u = 0 in R N , for all t E R. Therefore we have (Proposition 3.5.13) u(t) = T(t), for all t E R. Now we know that .-1{e-;1e12t}(x) =
f
2
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
Inhomogeneous equations and abstract semilinear problems Throughout this chapter, we assume that X is a Banach space and that A is an m-dissipative operator with dense domain. We denote by (T(t)) t >o the contraction semigroup generated by A. 4.1. Inhomogeneous equations Let T> 0. Given x E X and f: [0,T] --> X, our aim is to solve the problem (4.1) u E C([0,T],D(A)) nC'([O,T],X); (4.2)
u'(t) = Au(t) + f (t), `dt E [0, T];
(4.3)
t. u(0)=x.
As in the case of ordinary differential equations, we have the following result (the variation of parameters formula, or Duhamel's formula). Lemma 4.1.1. Let x E D(A) and let f E C([0, T}, X). We consider a solution uEC([0,T],D(A))nC'([0,T],X) of problem (4.1)—(4.3). Then, we have
u(t) =T(t)x
+
J0 T(t —s)f(s)ds,
(4.4)
t
for all t E [0, T] . Proof. Let t E (0,T]. Set w(s) = T(t — s)u(s), for s E [0, t]. Lets
w(s
+h
E
[0, t] and h E (0, t — s]. We have
) — w(s)
(
— 7t — s — h) {
u(s + — u(s) — T(h) — I u(s)
1
T(t —s) {u'(s)—Au(s)} = T(t —s)f(s) as h . 0. Since T(t — •) f(•) E C([O,t],X), we deduce that w E C 1 ([0,t),X) and that (4.5) w'(s) = T(t — s) f (s), for all s E [0,t). Integrating (4.5) between 0 and obtain (4.4).
rr
< t, and letting T T t, we 0
Inhomogeneous equations 51
Corollary 4.1.2. For all x E D(A) and f E C([O,T],X), problem (4.1)-(4.3) has at most one solution. Remark 4.1.3. For all x E X and all f E C([O,T],X), formula (4.4) defines a function u E C([O, T], X). Now we are looking for sufficient conditions for u given by (4.4) to be the solution of (4.1)-(4.3). Remark 4.1.4. It is clear that if u is a solution of (4.1)-(4.3), then x E D(A). However, this condition is not sufficient. Indeed, assume that (T(t)) tE R is an isometry group, and let y E X \ D(A). Then (see Remark 3.2.4), T(t)y ¢ D(A), for all t E R. Take f(t) = T(t)y, and x = 0 E D(A). It follows easily that (4.4) gives u(t) = tT(t)y D(A), for t # 0. Lemma 4.1.5. For all x E X and f E L 1 ((0,T),X), formula (4.4) defines a function u E C([0,T],X). In addition, we have IIUIIC([O,T],X)
Proof. The result is clear if f E C([0,T],X), and follows by density in the general case. 0 Proposition 4.1.6. Let x E D(A) and let f E C([0,T],X). Assume that at
least one of the following conditions is satisfied: (i) f E L'((0,T),D(A)); (ii) f E W 1,1 ((0,T),X).
Then u given by (4.4) is the solution of (4.1)-(4.3). Proof. We proceed in four steps. Set v(t)
= J t T(t - s)f (s) ds O
= fo
t
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
f
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,
52 Inhomogeneous equations and abstract semilinear problems
and let h J 0. Note that
T(h)-I —, Af h f
in L 1 ((0,T),X), as h . 0, and apply Lemma 4.1.5. It follows that
d+ v dt (t)
=J
t
T(t - s)A f (s) ds + f(t),
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) =
O
T(s) f (t + h
-h
-
f (t - s)
ds+ T(h)
f
h
T (t - s)f
(S)
ds,
and we let h J, 0. Note that (Corollary 1.4.39)
.f(t+h—.) —,f(t—.) h
f'(t—.), as h J 0
in L 1 ((0, t), X) and apply Lemma 4.1.5. It follows that
d+ v dt (t)
= fo
t T(s) f'(t - s) ds + T(t)f(0),
for all t E [0,T). In both cases, we have d+v/dt E C([0,T),X); and so v E
C 1 ([O,T),X)• Step 2. Similarly, we show that (d v/dt) (T) makes sense and is equal to i imv'(t); and so v E C'([O,T],X). -
Step 3. Let t E [0, T) and let h E [0, T - t]. We have
T (h) -I h
v(t) =
11t
I
t
T (t + h - s) f (s) ds
- v(t+h)-v(t) 1 -- [ h h J t
-
f
t T (t - s) f (s) ds
7(t + h- 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,
u'(t) = AT(t)x + Av(t) + f(t) = Au(t) + f(t), for all t E [0, T]. Hence, we have (4.2), and (4.3) is immediate.
❑
Inhomogeneous equations 53
Corollary 4.1.7. Let x E X, f E C([O,T],X) and let u be given by (4.4). Then using the notation of §2.3 and §3.3, u is the unique solution of the problem u E C([O,T],X) nC'([0,T],X); u'(t) = Au(t) + f (t), Vt E [0, T]; u(0) = x. Proof. We apply Lemma 3.3.1 and Proposition 4.1.6.
❑
Corollary 4.1.8. Let x E X, f E C([0,T],X) and let u be given by (4.4). Assume that at least one of the following conditions is satisfied: (i) u E C([0,T],D(A)); (ii) u E C 1 ([0,T],X). Then u is the solution of (4.1)—(4.3). Proof. Assume that (i) holds. By Corollary 4.1.7, we have u' E C([0, T], X); and so u E C i ([0, T], X ), hence the result. Now assume that (ii) holds. By Corollary 4.1.7, we have Au E C([0,T],X); and so (Corollary 2.3.2) u E C([0,T],D(A)); hence the result. ❑ Throughout §4.1, we have supposed that f E C([0,T],X). But in order to give a sense to (4.4), it suffices that f E L 1 ((0,T),X) (Lemma 4.1.5). In this case, we have the following result. Proposition 4.1.9. Let x E X, f E L 1 ((0,T),X) and let u E L 1 ((0,T),X). Assume further that u E L 1 ((0,T),D(A)) or that u E W 1 1 ((0,T),X). Then u verifies (4.4) if and only if u "
u E L 1 ((0,T),D(A)) nW'" i ((O,T),X); u'(t) = Au(t) + f (t), for almost every t E [0,T]; u(0) = x. Proof. First note that if u E W 1 1 ((0,T),X) then u E C([0,T],X) (Corollary 1.4.36) and so the condition u(0) = x makes sense. Let us first show that the assumptions of the theorem are sufficient to have (4.4). To see this, we argue as in Lemma 4.1.1. We consider t E (0, T] and we set w(s) = T(t — s)u(s), for almost every s E (0, t). For all h E (0, t), and for almost every s E (0, t — h), we have '
w(s + h) — w(s) h
_ T(t — s — h)
u(s + h) — u(s) — T(h) — I
(s)}
54 Inhomogeneous equations and abstract semilinear problems
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
s
d (s) = T(t – s) f (s)• d Similarly, we show that w is left differentiable almost everywhere on (0, t) and that
v
(s) = T(t – s)f(s).
Consequently (Theorem 1.4.35), w E W 1 1 ((0,T),Y) and w'(s) = T(t – s)f(s) almost everywhere. We deduce (4.4). Conversely, assume that u satisfies (4.4). Let (fn ) n >o be a sequence of C([0,T], X) such that fn —if in L 1 ((O,T), X) as n –+ oo, and let u n be the corresponding solutions of (4.4). Invoking Corollary 4.1.7, we have '
I
u(t) = Au(t) + fn(t), for all t E [0,T]; and so
u(t) = x
+ J t (Aun(s) + fn(s)) ds 0
for all t E [0,T]. Letting n
–ti
oo, we obtain (Lemma 4.1.5)
u(t) = x +
JO t (Au(s) + f(s)) for all t E [0,T]. It follows that u E W ' ((0,T),Y) and that ds,
1 1
u'(t) = Au(t) + f(t), for almost every t E [0,T]. If u E W 1 1 ((0,T),X), we have Au E L 1 ((O,T),X); and so u E L 1 ((0,T),D(A)) (Corollary 2.3.2). If u E L 1 ((O,T),D(A)), we have ❑ u' E L' ((O, T), X ); and so u E W 1 1 ((0, T ), X ). This completes the proof. '
"
4.2. Gronwall's lemma In this section, we give a result which is essential in the study of semilinear problems; not only for showing uniqueness of solutions but also for finding bounds on the solutions.
Semilinear problems 55
Lemma 4.2.1. (Gronwall's lemma) Let T > 0, A E L l (0, T), A >_ 0 a.e. and C1, C2 > 0. Let cp E L' (0, T), cp > 0 a.e., be such that A E L'(0, T) and cp(t) < C1 + C2
J
t
0
A(s)cp(s) ds,
for almost everyt E (0,T). Then we have cp(t)
< Cl exp (C2
rt
A(s)ds Jo
)
for almost every t E (0,T).
Proof. We set
ft
(t) = C l + C2
J A(s)cp(s) ds. 0
i) is differentiable almost everywhere (since it is absolutely continuous), and we have z/^'(t) < C2 A(t)cp(t) < C 2 A(t)O(t), for almost every t E (0, T). Consequently,
(
dt{ ip(t) exp (C2 l
ll t A(s) ds I 1 < 0,
f
I ))J
and so
J
t
< Cl exp (C2 A(s) ds) O
hence the result, since cp < 0.
❑
Remark 4.2.2. In particular, if Cl = 0, we have cp = 0 a.e.
4.3. Semilinear problems Definition 4.3.1. A function F : X —> X is Lipschitz continuous on bounded subsets of X provided that for all M > 0, there exists a constant L(M) such
that IIF(y)-F(x)II <_L(M)IIy
-xII, ex,yEBM,
where BM is the ball of center 0 and of radius M. Throughout s4.3, F: X —4 X is a Lipschitz continuous function on bounded subsets of X. We denote by L(M) the Lipschitz constant of F in BM for M > 0. In particular, L(M) is a non-decreasing function. of M.
56 Inhomogeneous equations and abstract semilinear problems
Given x E X, we look for T> 0 and a solution u of the following problem: u E C([0,T],D(A)) nC'([O,T],X);
(4.6)
u'(t) = Au(t) + F(u(t)), Vt E [0, T];
(4.7)
u(0) = x.
(4.8)
We also consider a weak form of the preceding problem. Indeed, by Lemma 4.1.1, any solution u of (4.6)—(4.8) is also solution of the following problem: u(t) = T(t)x +
JO t
T(t — s)F(u(s)) ds, Vt E [0, T].
(4.9)
Finally, note that, for all u E C([O,T],X), (4.9) is equivalent (following the notation of Corollary 4.1.7) to the problem u E C([0,T],X) nC'([O,T],X);
u'(t) = Au(t) + F(u(t)), Vt E [0, T]; u(0) = x.
4.3.1.
A result of local existence
We begin with a uniqueness result. Lemma 4.3.2. Let T > 0, x E X, and let u, v E C([0, T], X) be two solutions to problem (4.9). Then u = v. Proof. We set M = sup max{IIu(t)II, llv(t)ll}. We have tE [O,T]
t II F(u(s)) — F(v(s)) II ds < L(M) II u(t) — v(t)II <— I0
fo t I u(s) — v(s) II ds,
and we conclude using Remark 4.2.2. Set
1
TM = 2L(2M
+ IIF( 0 )II) + 2 > 0,
for M > 0. We can state a first result of local existence.
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
Proof. Lemma 4.3.2 proves uniqueness. Let x E X and let M > K = 2M + IIF(0)II and
IIxII. We let
= { u E C([O,TM],X); [In(t)I[ < K,`dt E [0,TM]},
E
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 u, v E E. Since C([0, TM], X) is a Banach space, (E, d) is a complete metric space. For all u E E, we define (D u E C([O,TM],X) by = T(t)x +
f
T(t — s)F(u(s)) ds,
for all t E [0, TM]. Note that for s E [0, TM], we have F(u(s)) = F(0)+(F(u(s)) — F(0)); and so
IIF(0)II IIF(u(s))II <— IIF(0)II + KL(K) < M +TM It follows that II(t)It
0)II < IIxII +JO t IIF(u(s))Ilds <M+t M TM(
Consequently, we have F : E --4 E. Furthermore, for all u, v E E, we have II(t) — (t)II < L(K)
fo IIv(s) — u(s)II ds < TML(K)d(u, v) < 1 d(u, v).
Therefore, 1 is a contraction in E with Lipschitz constant 1/2, and so 1) has a fixed point (Theorem 1.1.1) u E E, which satisfies the requirements of Proposition 4.3.3. ❑ Theorem 4.3.4. There exists a function T : X — (0, oo] with the following
properties: for all x E X, there exists u E C([0, T(x)), X) such that for all 0< T < T(x), u is the unique solution of (4.9) in C([O,T], X). In addition, 2
L(IIF( 0 )II + 2 11u(t)II) ? T(x) — t —2,
for all t E [0, T(x)). In particular, we have the following alternatives: (i) T(x) = oo;
(4.10)
58 In homogeneous equations and abstract semilinear problems
(ii) T(x) < oo and lim Iu(t) ll = oo. ttT(x)
Remark 4.3.5. If property (i) holds, we say that the solution u is global. On the other hand, if (ii) holds, we say that u blows up in finite time. In other words, the alternatives (i)-(ii) mean that the global existence of the solution u is equivalent to the existence of an a priori estimate of IIu(t)II on [0,T(x)). In applications, we establish such a priori estimates by standard methods in the theory of partial differential equations (multipliers, comparison principles, and maximum principles), as well as by various techniques involving 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,
with M = 1Iu(t)j^. Let v E C([O,TM],X) be the solution, given by Proposition 4.3.3, of
+
v(s) = T(s)u(t)
10
T(s - a)F(v(c)) do, ,
for all SE [0, TM] . We then define w E C([0, t + TM], X) by w(s) =
u(s),
ifs E [0, t];
v(s - t),
if s E [t, t + TM].
We verify easily that w is a solution of (4.9) with T = t + TM, which contradicts the definition of T(x), since t + TM > T(x). Remark 4.3.6. It may very well happen that, for the same equation, T(x) < oo for some initial data, and T(x) = oo for others. For example, choose X = R, A = 0, and F(u) = u 3 - u. This choice corresponds to the ordinary differential equation u' = u 3 - u. If (xi <_ 1 we have T(x) = oo, and if xi > 1 we have T(x) < oo. In the last case, (4.10) gives 12iu(t)1 2 > (T(x) - t) ' - 4. This -
Semilinear problems 59
estimate describes the blow-up phenomenon sharply, since the solutions actually blow up as (T(x) - t) I'. -
4.3.2. Continuous dependence on initial data Proposition 4.3.7. Following the notation of Theorem 4.3.4, we have the following properties: (i) T : X -* (0, oo] is lower semicontinuous; (ii) if x n --* x and if T < T(x), then u n -3 u in C([0, T], X), where u n and u are the solutions of (4.9) corresponding to the initial data x n and x. Proof. Let x E X, and let u be the solution of (4.9) given by Theorem 4.3.4. Let 0
M = 2 sup IIu(t)1I, tE[O,TJ
and Tn = sup{t E [0,T(x n )) ; IIun(s)II < 2M,Vs E [0,t]}. For n large enough, we have (IxII < M; and so Tn > TM > 0. For all t < T, t<Ti,,wehave
Ilu(t) — u n(t)II
(Ix — xn ll + L(2M)
fo t IIu(S) — un(S)IIds;
and it follows from Lemma 4.2.1 that
II u(t) — u(t)ii < IIx — x n lieTL( 2 M),
(4.11)
for t < T, t <_ Tn . In particular, we deduce from (4.11) that for n large enough we have
II un(t)II C Al, for t _< min{T, rrn }; and so Tn > min{T, rrn }, i.e. Tn > T. We then have T(x) > T. Applying (4.11) again, we see that u n --> u in C([0,T],X). This completes the proof. ❑ Remark 4.3.8. Actually, T may be discontinuous. For example, choose X = 1R 2 , A = 0, and F(u, v) = (vu 2 , -2). For x = (1, 2) we have T(x) = 1 and the corresponding solution is ((1 - t) -2 , 2(1 - t)). For x E = ((1 + E) -1 , 2) we have T(x e ) = oo and the corresponding solution is ((e + (1 - t) 2 )',2(1 - t)).
60 Inhomogeneous equations and abstract semilinear problems
4.3.3.
Regularity
In some cases, it is possible to give a more precise result on the regularity of solutions of (4.9). In particular, we have the following. Proposition 4.3.9. Assume that X is reflexive. Let T > 0, x E X, and let u E C([0,T],X) be a solution of problem (4.9). Then, if x E D(A), u is the solution of problem (4.6)—(4.8). Proof. Let h> 0 and let t E [0,T — h]. It is easy to see that u(t + h) — u(t) = T(h)x — x+
J
0
t
T(s){F(u(t + h — s) — F(u(t — s))} ds
+
f
h T(t + s)F(u(s)) ds.
Hence,
IIu(t + h) — u(t)II c IIT(h)x — x1j+h sup IIF(u(s))II sE[O,T]
+ L(M)
J
0
t llu(s
+ h) — u(s) I ds
Frthermore, we have h
T(h)x — x = J T(s)Axds; O
and so IIT(h)x — xII < hilAxil. Applying Lemma 4.2.1, we obtain
u(t + h) — u(t)II < Ch,
for 0 < t
A(u, v) = (u', v'), `d(u, v) E D(A).
Isometry groups 61
A is m-dissipative with dense domain, and generates the semigroup (T(t)) t >o given by
I
T(t)(u, v) = (u(t + •), v(t + )), for t > 0, x E R. Next, consider the Lipschitz continuous function F : X --+ X given by
F(u, v) = (v,0), V(u, v) E X. For all (x, y) E X, the corresponding solution (u, v) of (4.9) is given by
t
(u, v)(t) = (x(t +) + t y +(t + •), y(t + •))• Taking (x, y) E D(A) such that y(0) = 0 and y'(0) so (u, v)(t) V D(A), for t 0.
54 0, y+ is not in C'(R), and
4.4. Isometry groups In the case in which A generates an isometry group (see Theorem 3.2.3), and in particular when X is a Hilbert space and A is skew-adjoint, we can also solve (4.7) for t < 0. Indeed, solving the problem u E C([—T,0],X) nC'([—T,0],X);
u (t) = Au(t) + F(u(t)), Vt E [—T, 0]; u(0) = x; is equivalent to solving v E C([0,T],X) nC 1 ([0,T],X);
v'(t) = —Au(t) — F(u(t)), Vt E [0, TI; 1. v(0)=x; setting u(t) = v(—t), for t E [— T, 0]. The second problem is solved by Theorem 4.3.4, since —A is m-dissipative and —F is Lipschitz continuous on bounded sets of X. Notes. One finds generalizations of the results of §4.3 in Segal [1] and Weissler [1]. Also consult Ball [1, 2] for an interesting discussion about the blow-up phenomenon.
I
II
• I
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
I
F(u)(x) = 9(u(x)), for all u E X and x E Sl. It is easy to check that F is Lipschitz continuous on bounded sets of X. In what follows, we denote g and F by the same expression. 5.1. Preliminaries Given cp E X, we look for T> 0 and u solving the problem u E C([0,T], X) n C((0,T], Ho (S2)) n C'((O, T], L 2 (5l));
Au E C((O,T],L 2 (1l));
u t — Au = F(u), Vt E (0, T];
u(0) = cp.
(5.1) (5.2) (5.3)
The result is the following. Proposition 5.1.1. Let cp E X, T> 0, and let u E C([0,T],X). Then u is solution of (5.1)—(5.3) if and only if u satisfies u(t) = T(t) cp +
JO
t
T(t — s)F(u(s)) ds,
(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
v(0) = u(e) E D(B). Hence, we have (Lemma 4.1.1) v(s) = S(s)u(e)
+
10, S(s - o,)F(v(u)) ds,
for all s E [0, t - E]. Applying Lemma 3.5.6, we deduce that u s+e T(s)u(e) +
"
T s-v F u v +e ds,
for all s E [0, t - e]. Since u E C([0,T], X), we have, for all s E [0, t), T(s)u(E) -; T(S)W,
as j0; F(u(. + e)) -' F(u(.)),
uniformly on [0, s] as e j 0. Letting first E 1 0, and then s T t, we deduce (5.4). Conversely, let u E C([0, T], X) be a solution of (5.4). We consider 0 < t ( T. By Proposition 3.5.2, we have T(t) cp E Ho (Sl), and the function s H T(t - s)F(u(s)) belongs to C([0, t), Ho (S2)), with IIT(t - s)F(u(s))IIHI < C(1 + (t - s) -1 " 2 ) E L 1 (0, t);
and so (Proposition 1.4.14 and Corollary 1.4.23) u(t) E Ho (S2). A similar estimate shows that actually u E C((0, T], Ho (l1)), IIu(t)IIH1 < C(1 +t -1 / 2 )
Since g is Lipschitz continuous on bounded subsets of R and since the range of u is bounded, we conclude (Proposition 1.3.5) that F(u(t)) E H(l), and that IIF(u(t))IIH= < C(1 +t -1 1 2 ).
It follows that F(u) is weakly continuous as a map from (0, T] to Ho (St). Take 0
L2
< C(t - s) -1 / 2 (1 + s -1 / 2 ) E L l (0, t);
and so Au(t) E L 2 (1l). Consequently, u(t) E D(B), for all t similarly that u E C((0,T), D(B)). We then set v(s) = u(e + s),
E
(0,T]. We show
64 The heat equation
for O < s t - E. We have v(s) = S(s)u(E) +
fo
S(s - a)F(v(a)) ds, s
for all s E [0, t - e]. We conclude by applying Corollary 4.1.8.
o
Note that here it is not possible to invoke Proposition 4.3.9, since X is not reflexive. Remark 5.1.2. Considering the above estimates in more detail, we obtain, as a consequence of Proposition 3.5.2,
IIVu(t)IILa
+ <- CIcI 1 / 2 (t -1 + t)
;
where C depends only on g and sup IIu(t)II. O
Iiou(t)IIL2 < CIQ1 1 " 2 (1 + t'/ 2 ); IILu(t)IIL 2 < CIcj 1/2 (t -1/2 + t), where C depends only on g, sup Iu(t)II andII^PIIH1• O
Remark 5.1.4. The same method also shows that if we assume further that cp E D(B), then
IIAu(t)IIL2 G CIslIli 2 (1 + t),
where C depends only on g, sup IIu(t)II, and IIcolIHl. O
5.2. Local existence Applying Proposition 5.1.1 and Theorem 4.3.4, we obtain the following result. Theorem 5.2.1. For all cp E X, there exists a unique function u, defined on a maximal interval [0, T()), which is a solution of (5.1)-(5.3) for all T E (0, T(,)). In addition, if T (cp) < oc, then IIu(t) II -4:: as t j T(). Remark 5.2.2. u depends continuously on cp (this follows from Proposition 4.3.7). The following proposition gives improved regularity of u if cp is more regular.
Global existence 65
Proposition 5.2.3. Assume that cp E X fl Ho (S2). Then the solution u corresponding to (5.1)-(5.3) is in C([0,T(cp)),Ho(c)). Suppose further that Ap E L 2 (c); then u E C([0,T(p)),D(B)) fl C'([0,T(cp)), L 2 (c)). Proof. Assume that cp E X f1 Ho (S2), and let t E (0, T(cp)). Applying (5.2),
Proposition 3.16, and (3.31), we obtain
u(t) — VIIH'
f t 1 IIF(u(s))Ilds < II S(t)V — VIIH1 + C /o t-s <—IIS'(t)V—VIIH=+C^-->0 astj0.
Therefore u E C({0, T()), Ho (S2)). In particular, if T < T(p), then u is bounded in Ho (SI) on [0, T], and then so is F(u). Consequently, in the case in which A E L 2 (c), it follows from (5.2) and (3.32) that
IIo(u(t) — V)IILZ
I(S(t) — I)oVIIL2
+Cft
t1 S
JIF(u(s))IIHlds tie
and so u E C([0,T(cp)), D(B)). In particular, u(t) — Ap + F(cp) in L 2 (c) as t 1 0. Furthermore, for t
J
Consequently, (d + u/dt) (0) = t o u t (t); and so u E C'([0, T(cp)), L 2 (c)).
❑
5.3. Global existence We establish here two kinds of results. First, we show that if g satisfies certain conditions for Ix) large, then all solutions of (5.1)-(5.3) are global. Then, in another spirit, some results prove that if g satisfies certain conditions for IxI small, then the solutions of (5.1)-(5.3) with small initial data are global. We begin with the following result (the maximum principle). Proposition 5.3.1. Let T > 0 and cp E X. Let u E C([0,T], X) n C((0,T), H(i)) n C 1 ((0,T),L 2 (fl)) with Au E C((0,T),L 2 (c)), and f E C([0,T],X) be such that ut1=f, VtE (0,T);
1
u( 0 ) = W.
(5.5)
Assume further that there exists a constant C such that
If(t,x)I <— CIu(t,x)I, in [0, T] x Q. Then, if cc > 0, we have u(t) > 0 for all t E [0, T] .
(5.6)
'
66 The heat equation
Proof. For
t E (0, T), we set v(t) = f((t))2.
Multiply (5.5) by -u (t) and integrate over Q. Integrating by parts and applying (5.6), we obtain (see the proof of Lemma 3.5.9) -
v -(t)
< f If(t)Iu (t)
f Iu(t)1u (t) = Cv
Integrating the last inequality, we obtain, for all 0 < s <
(
t . )
t < T,
v(t) < v(s)et-8). Letting s
1 0, it follows that v(t) e ct
I
hence the result.
f
(VP
)
2
= 0 , Vt E
0
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,
for x >_ K. Since g is bounded on [0, K] and by possibly modifying C, it follows that g(x) < C + Cx,
I
for x > 0; and so IIF(u(t)) for all
t E [0,T(cp)).
+11 < C + CIIu(t)11,
Global existence 67
Applying (5.4), Lemma 3.5.9, and Gronwall's lemma, we deduce that ct 5.8) u(t)II < (C+ IIwII)e (
for all t E [0,T(cp)); and so (Theorem 5.2.1) T(cp) = oo. Step 2. In the general case, we set 0 = IcpI and we denote by v the corresponding solution of (5.1)—(5.3). Let T < min{T(),T(cp)} = T(cp). We easily verify that w = v — u fulfills the assumptions of Proposition 5.3.1; hence
u(t) < v(t), for all t E [0, T] . We then use z = v + u, where v is the maximal solution of the equation v = T(t) +
fo
t
T(t — s){—g(—v)} ds.
Since —g(—x) verifies the same assumptions as g(x), v is a global solution (Step 1). Applying Proposition 5.3.1, we obtain that u(t) > —v_(t), for all t E [0, T]; and so
IIu(t)II <— max{ IIv(t) II, 112(t)II for all t E [0, T]. It follows that T(cp) > T(i) = oo. This completes the proof. ❑
Remark 5.3.4. If C = 0 in (5.7), it follows from (5.8) that all solutions of (5.1)—(5.3) satisfy
IIu(t)II s K + II^II, for t >_ 0. Actually, this result can be sharpened by the following proposition, whose proof is quite simple. Proposition 5.3.5. If C = 0 in (5.7), then solutions of (5.1)—(5.3) satisfy
IIu(t)II 5 max {K, IIVII}, fort>0.
68 The heat equation
Proof. As in Proposition 5.3.3, we may restrict ourselves to the case cp > 0. Set k = max{K, Ilcpll} and multiply (5.2) by (u — k)+ E Ho (S2). Integrating by parts and setting f (t) = f ((u(t) — k)+) 2 , it follows that f'(t) < C fn g(u(t))(u(t) — k) + < 0.
We conclude that f (t) < f (0) = 0, and so u < k; hence the result.
❑
If the constant C of (5.7) is sufficiently small, we still have a result of this kind. To see this, we consider A given by (2.2). Proposition 5.3.6. If C < .\ in (5.7), then, for all cp E X, the solution u
of (5.1)—(5.3) satisfies
sup Iu(t)II < 00.
O
Proof. We have xg(x) < C1 + Cx 2 . As in Proposition 5.3.3, we may assume that cp > 0. Let c > 0 and let u be the corresponding solution. We multiply (5.2) by u. Integrating by parts and setting f (t) = f u(t) 2 , it follows that
f'(t) < —2a f(t) + 2f u(t) g (u(t)) <— —2(A — C)f (t) + C1IQI. By Lemma 8.4.6 (see below), we have f(t) f( 0 )+C1 A I ^ IC ; and thus sup IIu(t)IIL2 = K < oo. Let p E (N/2, oo). By (3.34) and (3.37), 0
)
E L 1 (O ,+oo).
We then note that there exists a constant C2 (see the proof of Proposition 5.3.3) such that
II9(u(t))+IILP <_ C2 + Cu(t)j,,
for all t > 0; and so
Ilu(t)II ^ IIkII + IIT(.)IIG(Ln,LO°)(C2 +C O S p IIu(S)IILP) C3 +C4 Sup IIu(S)IIL", O<_s<_t
for all t >_ 0. Finally, invoking Holder's inequality, for all e > 0, there exists C(e) such that
IIVIILP < EIIvII +C'(E)IIVIIL2,
Global existence 69
1
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
IIu(t)II for all t > 0. Therefore
sup IIu(s)II < 2C3 + 2KC(E)C4 i
0<s
for all t > 0; hence the result.
c
Now we show that if g fulfills certain conditions in a neighbourhood of 0, then solutions with small initial data are global. To see this, consider ) given by (2.2). Proposition 5.3.7. Suppose that there exists a > 0 and p < .\ such that xg(x) < µlx1 2 , for jxj < a.
Then, there exists A < oo such that, if JIVII < aA, the corresponding solution u of (5.1)-(5.3) is global and satisfies
IIu(t)II <_ All(pj!e -(a- µ ) t, for t>0. Proof. As in Proposition 5.3.3, it suffices to deal with the case cp > 0. Set T = sup{t E [0,T()); IIu(s)II < a for s E [0,t}} > 0.
Multiply (5.2) by u. Integrating by parts, and letting f (t) = f u(t) 2 , for t E [0, T], it follows that f'(t) <-2f(t) +2
J
and so
u(t)g(u(t)) < -2(.\ - C) f (t);
IlIt2e U u,
IIu(t)IIL2
-
-
for all t E [0, T]. We write e(
)t u (t)
= e U-u)tT(t)^ + f o
t
eU-u)(t-9)T(t - s) (e
3
F(u(s)))ds
70 The heat equation Note that (see the proof of Proposition 5.3.6)
IIe(a—µ)`T(t)IIG(LP, Loo) E L 1 (O,+oo). Consequently, arguing as in the proof of Proposition 5.3.6, we obtain
e (A— µ ) tlIu(t)II <—e tIHwHI+CIISOIIL2+
1 — 2
upteUu(s)Ij , os
for all t E [0, T]. Hence, there exists A such that
sup e a µ sIIu(s)II : AII^jI, (
—
)
O<s
for all t E [0,T]. It follows that if 1IcpII < aA, then T = oo, which completes the proof. 0 Remark 5.3.8. We see that if p 1 0, we may take 5 j a. If g has a higher order near 0, we have a more precise result. Proposition 5.3.9. Assume that there exist p > 0, e > 0, and a > 0 such that
xg(x) < plxI2+E, for xj < a.
Then there exist Q, -y > 0 such that, if 11 pll < 3, then the corresponding solution u of (5.1)-(5.3) is global and 1)u(t) II : 'YII^PIIe - ^ t , for t > 0.
Proof. As in Proposition 5.3.3, we may assume that y > 0. Set
for x > 0, and -^ = min 0 < 0. For all a E (0, ^), there exist 0 < x a < Ya such
that 0(xa) + a = 0(y a ) + a = 0.
Furthermore, we have a < X a
Global existence 71
-EX
Fig. 5.1 Suppose that IIcpjj 0,
1(t) = sup IIu(s)II• o<s
We have
IIF(u(t))+II s µllu(t)II 1 +E, for t E [0, T]. Applying (5.4), (3.37), and Lemma 3.5.9, it follows that
E a9 {e a i eat llu(t)II < MII^II +µMfo 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 ,
f (t) E [0, xxMliroli ),
for all t E [0, T). We conclude as in Proposition 5.3.7, and we obtain the result with ,3 = min{a, C/M} and ry = (1 + E)M/E. ❑
72 The heat equation
5.4. Blow-up in finite time
We begin with a result whose proof is quite simple. Consider the function 0 E D(B), such that Lv)+A')=0,
(5.9)
'>0 on S2,
(5.10)
I
(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)
Then T(cp) < oo. Proof. Denote by u the solution of (5.1)—(5.3) given by Theorem 5.2.1. By
Proposition 5.3.1, we have u(t) > 0 on S2, for all t E [0,T(p)). Set f (t) = i u(t)V) > 0,
n
for t E [0, T (cp)). Applying (5.9) and Lemma 2.34, we obtain, for all t E [0, T(W)), f-(t) = fut(t) = I ('^'U(t) + g(u(t)))
=
J
u(t)A +
J g(u(t))V) =
> — (A + li)f (t) + a J u(t)
— Af(t)
1+E
+ J g(u(t)) )
b
On the other hand, by (5.11) and Holder's inequality, ./ (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)
Blow-up in finite time 73
for all t E (0,T(^p)). Let T = sup{t E (0,T(cp)), f' > 0 on (0,t)} > 0. If T < T(cp), we have f'(T) = 0 and f (T) > f(0), which contradicts (5.12). Thus, we have T = T(c') and f' > 0 on [0,T(cp)). Now let b > 0 be such that (a - S)f ( 0 ) 6 = A + ,Q.
I
We deduce from (5.12) that, for all t E (0, T(c )), f , (t)
a f(t) i+E + f(t)(—(A + Q) + (a — 6)f (t)) > of (t) -' + f (t) (—(a + p) + (a — 8)f ( 0 ) E ) ? (5f (t) 1
>
,
i.e.
_
(
f( t) -E ) > (bt) '
I
'
I I
From this, we easily deduce that 0 < f (t) -E <
e
f(0) (0)-E
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
E(u) = 2 f IDu1 2 - fc(u) , for u E X fl Ho (1), where
o
I
G(x) = f g(s) ds,
for
x
x
E
I
][^.
Proposition 5.4.4. Assume that there exists K > 0 and a> 0 such that
xg(x) > (2 + e)G(x),
74 The heat equation
I
K xg(x) and v = o m K G(x). Let cp E X n Ho (.l) be for lxi > K. Set µ =
such that (2 + s)E(cp) < IQI(p — 2ev).
Then T(cp) < oo. The proof makes use of the following two lemmas. Lemma 5.4.5. Let T > 0 and u E C((0,T),X)nC((0,T),D(B))f1C'((O,T), L 2 (S2)). Then
Il '
'
(Du + g(u))ut + E(u(t)) = E(cp),
for all 0<s
1
f(zu
+ g(u))ut = f(_Vu. ^
Du t + g(u)ut) =
and hence the result.
❑
Lemma 5.4.6. Let cp E X and let u be the corresponding solution to (5.1)-
(5.3). Then dt in
u(t) 2 dx + 2j i Vu(t)1 2 dx = 2 f u(t)g(u(t)) dx;
f J ut + E(u(t)) = E(u(s)); t
(5.13)
(5.14)
z
for all0<s
C
Proof of Proposition 5.4.4. Set f (t) = fo u(t) 2 . By (5.13), for t E (0,T()), we have f'(t) _ —2
in
IVU(t) 1 2 + 2
J
juj
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)
Blow-up in finite time 75
On the other hand, observe that, by Proposition 5.2.3, we may let s = 0 in (5.14); and so 2(2 + e)
fl
G(u(t)) = -2(2 + e) juj^!K}
+(2 + e)
f1
G(u(t)) juj
Vu(t)1 2 - 2(2 +E) {E() - f'f ut} .
(5.16)
Observe further that 2)
u(t)g(u(t)) - 2(2 + e)
JJJ 1luI
4u
j
G(u(t)) > 2j52(µ - (2 + e)v). (5.17)
Putting together (5.15), (5.16), and (5.17), we obtain the following inequality, for 0
J
f'(t) > 2(2+e) f t ut+E J IVu(t)I 2 +2i52I(µ-(2+e)v)-2(2+e)E(cp). (5.18) o st sz In particular,
f'(t)
t
>
2(2
+
e) o f u.
(5.19)
2
Now set h(t)=f t f (s) ds. Applying the Cauchy-Schwarz inequality, we obtain t h'(t) - h'( 0 ) = f (t) - f ( 0 ) = 2 f i uue
o n
t
< Uo ^^2 ^ I
1/2
1/2 12
t
t
< 2h(t) ( 2u ^ =) Cf fu o J 2)
1/2
From (5.19), it follows that
h(t)h"(t) > (1+ 2) (h'(t) - h'(0)) 2 .
(5.20)
For the sake of contradiction, assume that T(^p) = oo. From (5.20), we have (1 + E/2)(h'(t) - h'(0)) 2 >_ (1 + E/4)h'(t) 2 for t >_ to large enough. It follows from (5.20) that (h(t)- e/4 ) 1,
0 ,
(5.21)
for t >_ t o . But h(t) > 0 and h(t) -, / 4 -> 0 as t --> oo. Thus there exists t l > t o such that (h - £/ 4 )'(t l ) < 0. Hence, by (5.21), 0 < h(t) -e / 4 < h(t1) -e / 4 + (t —
76 The heat equation
for t > tj; and so
h(t l)-E/4
t < tl - (h-E/4)'(tl)' for t > t1, which is absurd.
❑
Remark 5.4.7. The condition on g in Proposition 5.4.4 means that g is 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
Application to a model case 77
I
and Cohen [1], Baras and Goldstein [1], Friedman and Giga [1], Friedman and McLeod [1], Giga and Kohn [1, 2], Mueller and Weissler [1], and Weissler [2, 3]. For the behaviour at infinity of solutions, consult Chapters 8 and 9, as well as, for example, Cazenave and Lions [2], Escobedo and Kavian [1, 2], Haraux [1], Henry [1], Kavian [2], Lions [1, 2], Weissler [4], and Esteban [1, 2].
I 1 0 I 1 1 1 ^
'
y
n
d 6 The Klein—Gordon equation ii
1
6.1. Preliminaries In this section, we give some technical tools that are essential in this chapter. 6.1.1.
An abstract result
'
Let X be a Hilbert space, let A be a skew-adjoint operator in X, and let (T(t))tER be the isometry semigroup generated by A. We have the following result.
I
Proposition 6.1.1. Let T > 0, x
u (t) = T (t)x +
f
for all t E [0, T] . Then the function t
H
id
o
2 d IIu(t)II 2
1
E
X, and f
E
C([0,T],X). Let u E
C([0,T],X) be given by
for all t
E
T (t - s)f (s) ds,
Il u(t)11' belongs to C' ([O, T]) and
= ( f(t),U(t)),
[0, T] .
Proof. Suppose that x E D(A) and f E C 1 ([0,T],X). By Proposition 4.1.6, we haven E C([0,T],D(A)) nC 1 ([0,T],X) and
u'(t) = Au(t) + f (t), for all t E [0, T]. Thus,
2 dt Ilu(t)11 2
'
= ( u(t),
u'(t)) = (u(t), Au(t) + f(t)) = (u(t), f (t)).
Hence
IIu(t)11 2 = IIx11 2
+ f (u(s), f(s)) ds.
(6.1)
0
In the general case, we approximate x and f by sequences (x)>o C D(A) and (fn)n >o C C'QO,T],X), and then we pass to the limit to obtain (6.1). The result follows since u and f are continuous functions. o
Preliminaries 79
6.1.2.
Functionals on Ho (Sl)
In this section, f is any open subset of RN. We consider a function g E C(R, R) such that there exists 0 < a < oo and C < oo so that g(0 ) = 0;
(6.2)
I9(x) — g(y) I ^ C(I xI
a
+ Iyl a )Ix — yI, b'x, y E R.
(6.3)
In particular, we have Ig(x) I < ClxI«+ 1 and so, for all p > a + 1, g defines an operator F : L o,(cl) —+ L^ ,(1l) by F(u)(x) = g(u(x)), for almost every x E Q. When there is no risk of confusion, we still denote by g the operator F. Applying (6.2), (6.3) and Holder's inequality, we easily obtain the following result. Proposition 6.1.2. Let a + 1 < p < oc. Then F is Lipschitz continuous from bounded subsets of LP(S2) to LT (1l). More precisely, we have 19(u) — g(v)II
^ C(u1j
+ IIvIIip)IIu — vIILP,
(6.4)
for all u, v E LP(52). For g as above, we define G E C(]R, R) by G(x) =
fo x g(s) ds.
(6.5)
Then G verifies condition (6.3), with a replaced by a + 1. Then, it follows from Proposition 6.1.2 that G allows us to define a functional V, Lipschitz continuous on bounded subsets of La+ 2 (Sl), by V(u) _ —
J G(u(x)) dx, `du E La+ (I) 2
(6.6)
More precisely, we have the following. Proposition 6.1.3. V is a functional of class Cl on L 2 (f ). Its derivative (which is a continuous mapping La+ 2 (52) -> (L 2 (1))' = L (1)) is given by V'(u) = for all u E L"+2(1)
— 9(u),
(6.7)
80
The Klein-Gordon equation
Proof. We have, for all x, y E R, x+y
(9(a) — 9(x)) do ;
G(x + y) — G(x) — y9(x) = f x
,
and so, applying (6.3), IG(x + y) - G(x) - y9(x)I
<- C , (Ixl a + Iyl a )Iyl 2 •
Applying Holder's inequality, we deduce that, for all u, v E L"+2 (1),
v(u + v) — V(u) + (v, g(u)) L ^ }2 L I = I v(u + v) — v(v) + f vg(u) dxI < C'(IIuliLa+2 + IIvllL.+2)IIvllL^+2
,
J
hence the result.
We now assume that, instead of (6.3), g satisfies the following weaker condition: I9(x) - g(y) I
C(1 + Ixl + IyI )Ix - yI, Q
a
(6.8)
for all x, y E R. In that case, we write (6.9)
9=91+92,
where 92 verifies (6.3) and gl verifies (6.3) with a = 0. For example, consider 9(x), 9i(x) = g(1),
I. g(-l),
if Ixi < 1; if x > 1; if x < 1.
On the other hand, if (N - 2)p < 2N, we have
HH(Q)'.., Lr(l)
(6.10)
Ln'(S2) , H-1O)
(6.11)
with dense embedding, and so
The result is the following.
< 4/(N - 2). Proposition 6.1.4. Let g satisfy (6.2) and (6.8), with 0 < a Then g is Lipschitz continuous from bounded subsets of Ho(1l) to H-'(Il).
Preliminaries 81
1
Proof.
By Proposition 6.1.2, gl is Lipschitz continuous from bounded subsets of L 2 (1l) to L 2 (fl), and so from bounded subsets of H(l) to H -1 (Sl). Invoking Proposition 6.1.2 again, 9 2 is Lipschitz continuous from bounded subsets of L 2 (52) to L+ (S2). Applying (6.10) and (6.11), with p = a+2, we deduce that 92 is Lipschitz continuous from bounded subsets of Ho (S2) to H -1 (52); hence the result. ❑ Proposition 6.1.5. Suppose that g satisfies the hypotheses of Proposition 6.1.4 with (N — 2)a< 2. Then g is Lipschitz continuous from bounded subsets of Ho (Q) to L 2 (1)_
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
However, we have (N — 2)p < 2N; hence the result, ❑
We now consider G and V defined by (6.5) and (6.6). We have the following result. Proposition 6.1.6. Suppose that g satisfies (6.2) and (6.8), with a >_ 0 such that (N — 2)a < 4. Then V is a functional of class Cl on Ho (Sl). Its derivative (which is a continuous mapping from Ho (52) —> (Ho (S2))' = H -1 (1)) is given by
V'(u) = —9(u),
(6.12)
I
for all u E Ho (S2). Proof. We apply Proposition 6.1.3 to gi and 92, and we use embeddings (6.10) and (6.11).
❑
Corollary 6.1.7. Suppose that g satisfies the hypotheses of Proposition 6.1.6, with (N — 2)a < 2. Let T > 0 and u E C([0,T],Ho(1l)) fl C'([0,T],L 2 (S2)). Then the mapping t H V(u(t)) is in C'([0,T]), and we have
d V(u(t)) for all t E [0, T] .
_-J
sz
9(u(t))u t (t) dx,
I
(6.13) '
82 The Klein-Gordon equation
1
Proof. Suppose first that u E C' ([0, T], Ho (S2)). Then, for all t E [0, T], d V (u(t)) = (V'(u(t)), u (t))H-1,Ho
I I
=
- (g(u(t)),u'(t))x-1,Ho
= - I g(u(t))ut(t)dx.
n
It follows that V(u(t)) = V(u(0)) -
IJ t
st
gu(s))u t (s) ds.
(6.14)
By density, we deduce that (6.14) is still true when u E C([O,T],Ho(Sl)) n C 1 ([O,T],L 2 (Sl)); hence the result. ❑ 6.2.
Local existence
Throughout this chapter, we follow the notation of §2.6.3, §2.6.4, and §3.5.2. In particular, 1 is any open subset of R N , m> -A, X = Hl) x L 2 (cl) and Y = L 2 (0) x H -1 (0). We consider a function g E C(R,R) which satisfies (6.2) and (6.8) with (N - 2)a < 2. Finally, we consider G and V defined by (6.5) and (6.6). We define the functional E on X and the mapping F: X -> X by
E(u, v) =
211(u, v) II X + V (u)
= 21 f {Ivul 2 + mlul 2 + lv1 2 — 2G(u)}dam, F(u, v) _ (0, g(u)),
for all (u, v) E Ho (St) x L 2 (Il).
It is clear, from Proposition 6.1.5, that g defines a Lipschitz continuous mapping from Ho(S1) to L 2 (Sl), and so F is Lipschitz continuous on bounded subsets of X. Given (gyp, zli) E X, we are looking for T> 0, and u a solution of uE C([ 0, T] , Ho($)) nC 1 ([ 0 ,T],L 2 (f))flC 2 ([ 0 ,T],H -1 (Q));
(6.15)
u tt - Lu + mu = g(u),
(6.16)
U( 0 ) = ^P,
ut( 0 ) = V •
for all t E [0, T];
(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.
Let T > 0 and (p, 1p) E X. Let u E C([0, T], Ho (Sl))n C ([0,T],L 2 (cl)). Then u is solution of (6.15)-(6.17) if and only if U = (u,u t ) is 1solution of Lemma 6.2.1.
Local existence 83
U(t) = T(t)(,
0) +
JO
t
T(t - s)F(U(s)) ds,
(6.18)
for all t E [0,T]. In addition, if A E L 2 (S2) and E Ho(Sl), then we have u E C 1 ([O,T],H0(1)) nC 2 ([O,T],L 2 (SZ)) and Au E C([O,T],L 2 (fl)). Applying Theorem 4.3.4, we deduce a local existence result. Theorem 6.2.2. For all (cp, 0/i) E X, there exists a unique function u, defined on a maximal interval [0, T(, v>)), which is a solution to (6.15)-(6.17) for all T
(cp, V)) E X and let u be the corresponding solution (6.19)
E(u(t),ut(t)) = E(co, l0 ), for all t E [0,T(p,
L)).
Proof. We apply Proposition 6.1.1 and Corollary 6.1.7. It follows that dtE(u(t), ut(t)) = (F(u(t), ut(t)), (u(t), ut(t)))x — for all t E [0,T(cp,0)); hence the result.
f
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.
84 The Klein-Gordon equation
6.3. Global existence As for the heat equation (§5.3), we will state two kinds of result according to the hypotheses on g: global existence of all solutions (i.e. independent of initial data), or global existence of solutions with small initial data. Proposition 6.3.1. Suppose that there exists C < oo such that G(x) _< CIx1 2 for ai: x E R. Then, for all (v, ) E X, we have T(cp, 0) = oo. Proof. Set f(t) = II(u(t),ut(t))IIX, for t E [0,T(cp,)). By (6.19), we have f (t) <- f (0) _2
f
G(cp) +2
f
G(u(t)) <- f (0) -2
J G(cp) + 2C J u(t)
2,
(6.20)
for all t E [0,T(cp,i )). On the other hand, we have t u(s)ut(s) u(t)IIi2 = IH L2 + f t ddt is, Iu(s)I 2 = II^IIi2 +21 o (6.21) 0
IIIIL2 +2/ t f(s). 0
Applying (6.20), (6.21), and Gronwall's lemma, we obtain the result. Remark 6.3.2. If 2C < A+m (A being given by (2.2)), then for all (co, ii) E X the corresponding solution u of (6.15)-(6.17) satisfies sup II(u(t),ut(t))IIx < no. t>o
Indeed, in this case we easily verify that C f u(t) 2 < (1 - e) f (t), with E > 0, and it follows from (6.20) that e f (t) < C'; hence the result. Proposition 6.3.3. Suppose that there exist p < A + m (A being given by (2.2)) and 3 > 0 such that 2G(x) < PIxI 2 for IxJ <_ 3. Then there exists .5,K >0 such that if I I (cp, Ali) II x < 6, we have T (cp, Vi) = no and the 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),
for all x E I1. By possibly taking larger C, we have 2G(x) < plxi 2 + Clxik,
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
2J G(u)