Quasilinear evolutionary equations and continuous interpolation spaces

ARTICLE IN PRESS J. Differential Equations 196 (2004) 418–447 Quasilinear evolutionary equations and continuous interpolation spaces Philippe Clément,a Stig-Olof Londen,b,
and Gieri Simonettc a Department of Mathematics and Informatics, TU Delft 2600 GA Delft, The Netherlands b Institute of Mathematics, Helsinki University of Technology, 02150 Espoo, Finland c Department of Mathematics, Vanderbilt University, Nashville, TN 37240, USA Received August 26, 2002; revised June 17, 2003 Abstract In this paper we analyze the abstract parabolic evolutionary equations Dat ðu  xÞ þ AðuÞu ¼ f ðuÞ þ hðtÞ; uð0Þ ¼ x; in continuous interpolation spaces allowing a singularity as tk0: Here Dat denotes the timederivative of order aAð0; 2Þ: We first give a treatment of fractional derivatives in the spaces Lp ðð0; TÞ; X Þ and then consider these derivatives in spaces of continuous functions having (at most) a prescribed singularity as tk0: The corresponding trace spaces are characterized and the dependence on a is demonstrated. Via maximal regularity results on the linear equation Dat ðu  xÞ þ Au ¼ f ; uð0Þ ¼ x; we arrive at results on existence, uniqueness and continuation on the quasilinear equation. Finally, an example is presented. r 2003 Elsevier Inc. All rights reserved. Keywords: Abstract parabolic equations; Continuous interpolation spaces; Quasilinear evolutionary equations; Maximal regularity
Corresponding author. Institute of Mathematics, Helsinki University of Technology, P.O. Box 1100, 02015 HUT, Finland. Fax: +35-89-45-13-016. E-mail address: stig-olof.londen@hut.fi (S.-O. Londen). 0022-0396/$ - see front matter r 2003 Elsevier Inc. All rights reserved. doi:10.1016/j.jde.2003.07.014 ARTICLE IN PRESS P. Cl!ement et al. / J. Differential Equations 196 (2004) 418–447 419 1. Introduction In a recent paper, [7], the quasilinear parabolic evolution equation du þ AðuÞu ¼ f ðuÞ; dt uð0Þ ¼ x; was considered in continuous interpolation spaces. The analysis was based on maximal regularity results concerning the linear equation du þ Au ¼ f ; dt uð0Þ ¼ x: In particular, the approach allowed for solutions having (at most) a prescribed singularity as tk0: Thus the smoothing property of parabolic evolution equations could be incorporated. In this paper we show that the approach and the principal results of [7] extend, in a very natural way, to the entire range of abstract parabolic evolutionary equations Dat ðu  xÞ þ AðuÞu ¼ f ðuÞ; uð0Þ ¼ x: Here Dat denotes the time-derivative of arbitrary order aAð0; 2Þ: As in [7], our basic setting is the following. Let E0 ; E1 be Banach spaces, with E1 CE0 ; and assume that, for each u; AðuÞ is a linear bounded map of E1 into E0 which is positive and satisfies an appropriate spectral angle condition as a map in E0 : Moreover, AðuÞ and f ðuÞ are to satisfy a specific local continuity assumption with respect to u: Problems of fractional order occur in several applications, e.g., in viscoelasticity [10], and in the theory of heat conduction in materials with memory [17]. For an entire volume devoted to applications of fractional differential systems, see [16]. Our paper is structured as follows. We first (Section 2) define, and give a brief treatment of, fractional derivatives in the spaces Lp ðð0; TÞ; X Þ and then (Section 3) consider these derivatives in spaces of continuous functions having a prescribed singularity as tk0: In Section 4 we characterize the corresponding trace spaces at t ¼ 0 and show how these spaces depend on a: In Section 5 we consider the maximal regularity of the linear equation Dat ðu  xÞ þ Au ¼ f ; uð0Þ ¼ x; ð1Þ where again aAð0; 2Þ and where the setting is the space of continuous functions having at most a prescribed singularity as tk0: To obtain maximal regularity we make a further assumption on E0 ; E1 : In Section 6 we analyze the nonautonomous, A ¼ AðtÞ; version of (1). Here we assume that for each fixed t the corresponding operator admits maximal regularity and deduce maximal regularity of the nonautonomous case. In Sections 7 and 8 we combine our results of the previous sections with a contraction mapping technique to obtain existence, uniqueness, and continuation ARTICLE IN PRESS 420 P. Cl!ement et al. / J. Differential Equations 196 (2004) 418–447 results on Dat ðu  xÞ þ AðuÞu ¼ f ðuÞ þ hðtÞ; uð0Þ ¼ x: Finally, in Section 9, we present an application of our results to the nonlinear equation Dat ðu  u0 Þ  ðsðux ÞÞx ¼ hðtÞ; xAð0; 1Þ; tX0; with u ¼ uðt; xÞ; uð0; xÞ ¼ u0 ðxÞ; aAð0; 2Þ; Dirichlet boundary conditions, s monotone increasing and sufficiently smooth. This equation occurs in nonlinear viscoelasticity, and has been studied, e.g., in [10,12]. Parabolic evolution equations, linear and quasilinear, have been considered by several authors using different approaches. Of particular interest to our approach are the references, among others, [1,2,8,15]. The reader may consult [7] for more detailed comments on the relevant literature. It should also be observed that we draw upon results of [4], where (1) is considered in spaces of continuous functions on ½0; T; i.e., without allowance for any singularity at the origin. 2. Fractional derivatives in Lp We recall [20, II, pp. 134–136] the following definition and the ensuing properties. Let X be a Banach space and write gb ðtÞ ¼ 1 b1 t ; GðbÞ t40; b40: Definition 1. Let uAL1 ðð0; TÞ; X Þ for some T40: We say that u has a fractional derivative of order a40 provided u ¼ ga
f for some f AL1 ðð0; TÞ; X Þ: If this is the case, we write Dat u ¼ f : Note that if a ¼ 1; then the above condition is sufficient for u to be absolutely continuous and differentiable a.e. with u0 ¼ f a.e. Tradition has that the word fractional is used to characterize derivatives of noninteger order, although a may of course be any positive real number. The fractional derivative (whenever existing) is essentially unique. Observe the consistency; if u ¼ ga
f ; and aAð0; 1Þ; then f ¼ Dat u ¼ dtd ðg1a
uÞ: Thus, if u has a fractional derivative of order aAð0; 1Þ; then g1a
u is differentiable a.e. and absolutely continuous. Also note a trivial consequence of the definition; i.e., Dat ðga
uÞ ¼ u: Suppose aAð0; 1Þ: By the Hausdorff–Young inequality one easily has that if the fractional derivative f of u satisfies f ALp ðð0; TÞ; X Þ with pA½1; 1aÞ; then ARTICLE IN PRESS 421 P. Cl!ement et al. / J. Differential Equations 196 (2004) 418–447 p : Furthermore, if f ALp ðð0; TÞ; X Þ; with p ¼ a1 ; then uALq ðð0; TÞ; X Þ for 1pqo1ap uALq ðð0; TÞ; X Þ for If qA½1; NÞ: f ALp ðð0; TÞ; X Þ with a1 op; then 1 ap uAh0-0 ð½0; T; X Þ [20, II, p. 138]. In particular note that uð0Þ is now well defined and that one has uð0Þ ¼ 0: (By hy0-0 we denote the little-Hölder continuous functions having modulus of continuity y and vanishing at the origin.) The extension of the last statement to higher order fractional derivatives is obvious. Thus, if u has a fractional derivative f of order aAð1; 2Þ and f ALp with 1 ða  1Þ1 op; then ut Aha1p : 0-0 We also note that if uAL1 ðð0; TÞ; X Þ with Dat uALN ðð0; TÞ; X Þ; aAð0; 1Þ; then a a uAC0-0 ð½0; T; X Þ: The converse is not true, for uAC0-0 ð½0; T; X Þ the fractional derivative of order a of u does not necessarily even exist. To see this, take vAL
[20, vAC a ð½0; T; X Þ: Without loss of I, p. 43], then [20, II, Theorem 8.14(ii), p. 136] D1a t 1a generality, assume Dt v vanishes at t ¼ 0: Assume that there exists f AL1 ðð0; TÞ; X Þ such that Dt1a v ¼ t1þa
f : But this implies (convolve by ta ) v ¼ 1
f ; which does not in general hold for vAL
[20, I, p. 433]. The following proposition shows that the Lp -fractional derivative is the fractional power of the realization of the derivative in Lp : Proposition 2. Let 1ppoN and define def DðLÞ ¼ W01;p ðð0; TÞ; X Þ; and def Lu ¼ u0 ; uADðLÞ: Then L is m-accretive in Lp ðð0; TÞ; X Þ with spectral angle p2: With aAð0; 1Þ we have La u ¼ Dat u; uADðLa Þ; where in fact DðLa Þ coincides with the set of functions u having a fractional derivative in Lp ; i.e., n o DðLa Þ ¼ uALp ðð0; TÞ; X Þ j g1a
uAW01;p ðð0; TÞ; X Þ : Moreover, La has spectral angle ap 2: We only briefly indicate the proof of this known result. (Cf. the proof of Proposition 5 below.) ARTICLE IN PRESS 422 P. Cl!ement et al. / J. Differential Equations 196 (2004) 418–447 The fact that L is m-accretive and has spectral angle p2 is well known. See, e.g., [3, Theorem 3.1]. The representation formula given in the proof of Proposition 5 and the arguments following give the equality of La and Dat : The reasoning used to prove [4, Lemma 11(b)] can be adapted to give that La has spectral angle ap 2: p We remark that if X has the UMD-property then (in L ðð0; TÞ; X Þ with 1opoN) we have DðLa Þ ¼ DðDat Þ ¼ ½Lp ðð0; TÞ; X Þ; W01;p ðð0; TÞ; X Þa : See [9, p. 20] or [19, pp. 103–104], and observe that powers in Lp ðð0; TÞ; X Þ: d dt admits bounded imaginary 3. Fractional derivatives in BUC1l Let X be a Banach space and T40: We consider functions defined on J0 ¼ ð0; T having (at most) a singularity of prescribed order at t ¼ 0: Let J ¼ ½0; T; mAð0; 1Þ; and define BUC1m ðJ; X Þ ¼ fuACðJ0 ; X Þjt1m uðtÞABUCðJ0 ; X Þ; lim t1m jjuðtÞjjX ¼ 0g; tk0 with def jjujjBUC1m ðJ;X Þ ¼ sup t1m jjuðtÞjjX : ð2Þ tAJ0 (In this paper, we restrict ourselves to the case mAð0; 1Þ: The case m ¼ 1 was considered in [4].) It is not difficult to verify that BUC1m ðJ; X Þ; with the norm given in (2), is a Banach space. Note the obvious fact that for T1 4T2 we may view BUC1m ð½0; T1 ; X Þ as a subset of BUC1m ð½0; T2 ; X Þ; and also that if uABUC1m ð½0; T; X Þ for some T40; then (for this same u) one has lim jjujjBUC1m ð½0;t;X Þ ¼ 0: ð3Þ tk0 Moreover, one easily deduces the inequality jjujjLp ðJ;X Þ pcjjujjBUC1m ðJ;X Þ ; mAð0; 1Þ; 1ppoð1  mÞ1 ; and so, for these ðm; pÞ-values, BUC1m ðJ; X ÞCLp ðJ; X Þ; with dense imbedding. To see that this last fact holds, recall that CðJ; X Þ is dense in Lp ðJ; X Þ and that obviously CðJ; X ÞCBUC1m ðJ; X Þ: ARTICLE IN PRESS P. Cl!ement et al. / J. Differential Equations 196 (2004) 418–447 423 We make the following fundamental assumption: a þ m41: ð4Þ To motivate this assumption, suppose we require (as we will do) that both u and Dat u lie in BUC1m and that uð0Þ (¼ 0) is well defined. The requirement Dat uABUC1m 1 : On the other hand, if implies, by the above, Dat uALp ðð0; TÞ; X Þ; for 1ppo1m a 1 p Dat uALp with a1 op then uAh0-0 and uð0Þ ð¼ 0Þ is well defined. Thus our requirements motivate the assumption that the interval ða1 ; ð1  mÞ1 Þ be nonempty. But this is (4). Therefore, under the assumption (4), the following definition makes sense. def a BUC1m ðJ; X Þ ¼ fuABUC1m ðJ; X Þj there exist xAX and f ABUC1m ðJ; X Þ such that u ¼ x þ ga
f g: ð5Þ a We keep in mind that if uABUC1m ðJ; X Þ; then (assuming (4)) uð0Þ ¼ x and u is Hölder-continuous. a We equip BUC1m ðJ; X Þ with the following norm: jjujjBUC a 1m def ðJ;X Þ ¼ jjujjBUC1m ðJ;X Þ þ jjDat ðu  xÞjjBUC1m ðJ;X Þ : ð6Þ Lemma 3. Let a40; mAð0; 1Þ; and let (4) hold. Space (5), equipped with norm (6), is a a Banach space. In particular, BUC1m ðJ; X ÞCBUCðJ; X Þ: a Proof. Take fwn gN n¼1 to be a Cauchy-sequence in BUC1m ðJ; X Þ: Then, by (6), and as BUC1m ðJ; X Þ is a Banach space, there exists wABUC1m ðJ; X Þ such that jjwn  def wjjBUC1m ðJ;X Þ -0: Moreover, fn ¼ Dat ðwn  wn ð0ÞÞ converges in BUC1m ðJ; X Þ to some function z: We claim that wð0Þ is well defined and that z ¼ Dat ðw  wð0ÞÞ: To this end, note that wn ðtÞ  wn ð0Þ ¼ ga
fn ¼ ga
z þ ga
½fn  z: ð7Þ We have limn-N jjt1m ½fn ðtÞ  zðtÞjjX ¼ 0; uniformly on J: Thus, by (4), limn-N jjga
½fn  zjjX ¼ 0; uniformly on J: So, uniformly on J; lim ½wn ðtÞ  wn ð0Þ ¼ ga
z: n-N ARTICLE IN PRESS 424 P. Cl!ement et al. / J. Differential Equations 196 (2004) 418–447 N For each fixed t40; fwn ðtÞgN n¼1 converges to wðtÞ in X : Thus fwn ð0Þgn¼1 must converge in X and by (4) and (7) we must have wn ð0Þ-wð0Þ as n-N: For the proof of the last statement, use the considerations preceding the theorem. & Our next purpose is to consider in more detail differentiation on def X̃ ¼ BUC1m ðJ; X Þ and to connect the fractional powers of this operation with that of taking fractional derivatives. First consider the derivative of integer order. Take a ¼ 1 in (5), (6), (thus a þ m41Þ and define n o def 1 DðL̃Þ ¼ uABUC1m ðJ; X Þ j uð0Þ ¼ 0 ; and L̃u ¼ u0 ðtÞ; uADðL̃Þ: We have Lemma 4. (i) DðL̃Þ is dense in X̃; (ii) L̃ is a positive operator in X̃; with spectral angle p2:  def  Proof. (i) Clearly, Ỹ ¼ uAC 1 ðJ; X Þ j uð0Þ ¼ 0 CDðL̃Þ: It is therefore sufficient to prove that Ỹ is dense in X̃: Observe that ỸCC0-0 ðJ; X ÞCX̃: It is well known that Ỹ is dense in C0-0 ðJ; X Þ with respect to the sup-norm (which is stronger than the norm in X̃). So it suffices to prove that C0-0 ðJ; X Þ is dense in X̃: Let uAX̃: There exists vAC0-0 ðJ; X Þ such that uðtÞ ¼ tm1 vðtÞ; tAð0; T: Set, for n large enough, ( 0; tA½0; 1n; vn ðtÞ ¼ vðt  1nÞ; tAð1n; T; un ðtÞ ¼ tm1 vn ðtÞ; tAð0; T; un ð0Þ ¼ 0: Then un ðtÞAC0-0 ðJ; X Þ; and sup jjt1m ½uðtÞ  un ðtÞjjX ¼ sup jjvðtÞ  vn ðtÞjjX tAð0;T tAð0;T   1 p sup jjvðtÞjjX þ sup jjvðtÞ  v t  jjX -0; n 1 1 0ptpn notpT as n-N: It follows that C0-0 ðJ; X Þ is dense in X̃ and (i) holds. ARTICLE IN PRESS 425 P. Cl!ement et al. / J. Differential Equations 196 (2004) 418–447 (ii). First, note that X̃CL1 ðJ; X Þ and that for every lAC and every f AL1 ðJ; X Þ; the problem lu þ u0 ¼ f ; uð0Þ ¼ 0; has a unique solution uAW01;1 ðð0; TÞ; X ÞCC0-0 ð½0; T; X Þ; given by uðtÞ ¼ t Z exp½lðt  sÞ f ðsÞ ds; tAJ: 0 We use this expression to estimate sup jljt1m jjuðtÞjjX ; sup jarg ljpy tAð0;T in case f AX̃ and yA½0; p2Þ: Thus jjlt 1m 1m uðtÞjjX p t p def t Z 0 jljexp½Rlðt  sÞsm1 ds jjf jjX̃ 1 1m t cos y Z t ðRlÞ exp½Rlðt  sÞsm1 dsjjf jjX̃ : 0 def We write Z ¼ Rl40; t ¼ Zs; to obtain 1 1m ðcos yÞ t Z t ðRlÞ exp½Rlðt  sÞsm1 ds 0 ¼ ðcos yÞ1 ðZtÞ1m Z Zt exp½Zt þ ttm1 dtpcy ; 0 where cy is independent of Z40; t40: To see that the last inequality holds, first observe that the expression to be estimated only depends on the product Zt (and on m; y). Then split the integral into two parts, over ð0; Zt2 Þ; and over ðZt2 ; ZtÞ; respectively (cf. [2, p. 106]). We conclude that the spectral angle of L̃ is not strictly greater that p2: Finally, assume that the spectral angle is less than p2: Then L̃ would generate an analytic semigroup. To obtain a contradiction, observe that L̃ is the restriction to X̃ def 1;1 of L̃1 considered on L1 ðð0; TÞ; X Þ; where DðL̃1 Þ ¼ W0-0 ðð0; TÞ; X Þ; L̃1 u ¼ u0 ; uADðL̃1 Þ: Thus the analytic semigroup TðtÞ generated by L̃ would be the restriction to X̃ of right translation, i.e., ðTðtÞf ÞðsÞ ¼ f ðs  tÞ; 0ptps; 0; sot: ARTICLE IN PRESS 426 P. Cl!ement et al. / J. Differential Equations 196 (2004) 418–447 But X̃ is not invariant under right translation. By this contradiction, (ii) follows and Lemma 4 is proved. & Proceeding next to the fractional powers and fractional derivatives we have: Proposition 5. Let a; mAð0; 1Þ: Then def DðL̃a Þ ¼ DðDat Þ ¼ fuAX̃ j u ¼ ga
f for some f AX̃g; and L̃a u ¼ Dat u; for uADðL̃a Þ: Moreover, Dat is positive; densely defined on X̃; and has spectral angle ap : 2 ð8Þ Proof. We first show that ðL̃1 Þa f ¼ ga
f ; ð9Þ for f AX̃: Observe that 0ArðL̃Þ; and that L̃ is positive. Thus Z N 1 1 a a ðL̃ Þ f ¼ L̃ f ¼ sa ðsI þ L̃Þ1 f ds; GðaÞGð1  aÞ 0 where the integral converges absolutely. But ðsI þ L̃Þ1 f ¼ Z t exp½sðt  sÞ f ðsÞ ds; 0ptpT; 0 and so, after a use of Fubini’s theorem,  Z t Z N 1 sa exp½ssds f ðt  sÞ ds: ðL̃1 Þa f ¼ GðaÞGð1  aÞ 0 0 To obtain (9), note that the inner integral equals ga ðsÞ: Let uADðDat Þ: Then u ¼ ga
f ; with Dat u ¼ f AX̃: So, by (9), u ¼ ðL̃1 Þa f ; which implies uADðL̃a Þ and L̃a u ¼ f : Conversely, let uADðL̃a Þ: Then, for some f AX̃; L̃a u ¼ f ; and so u ¼ ðL̃a Þ1 f : By (9), this gives u ¼ ga
f and so uADðDat Þ: We conclude that DðL̃a Þ ¼ DðDat Þ and that L̃a u ¼ Dat u; uADðL̃a Þ: To get that Dat is densely defined, use (i) of Lemma 4 and apply, e.g., [18, Proposition 2.3.1]. The fact that the spectral angle is ap 2 follows, e.g., by the same arguments as those used to prove [4, Lemma 11(b)]. & Analogously, higher order fractional derivatives may be connected to fractional powers. We have, e.g., the following statement. ARTICLE IN PRESS 427 P. Cl!ement et al. / J. Differential Equations 196 (2004) 418–447 Proposition 6. Let a; mAð0; 1Þ: Define n o def 1 a DðD1þa Þ ¼ uABUC ð½0; T; X Þ j uð0Þ ¼ 0; u ADðD Þ ; t t 1m t u ¼ Dat ut ; for uADðD1þa Þ: Then and D1þa t t L̃1þa u ¼ Dat ut ; uADðD1þa Þ: t Moreover, L̃1þa is positive, densely defined on X̃ with spectral angle with (cf. (9)), ðL̃1þa Þ1 f ¼ g1þa
f ; for f AX̃: ð1þaÞp 2 and For the proof of Proposition 6, first use Proposition 5 and the definition D1þa u¼ t uADðD1þa Þ: To obtain the size of the spectral angle one may argue as in the t proof of [5, Lemma 8(a)]. Dat ut ; 4. Trace spaces Let E1 ; E0 be Banach spaces with E1 CE0 and dense imbedding and let A be an isomorphism mapping E1 into E0 : Take aAð0; 2Þ; mAð0; 1Þ: Further, let A as an operator in E0 be nonnegative with spectral angle fA satisfying fA op 1  a : 2 Assume (4) holds and write J ¼ ½0; T: We consider the spaces def Ẽ0 ðJÞ ¼ BUC1m ðJ; E0 Þ; ð10Þ a Ẽ1 ðJÞ ¼ BUC1m ðJ; E1 Þ-BUC1m ðJ; E0 Þ; ð11Þ def and equip Ẽ1 ðJÞ with the norm h i def jjujjẼ1 ðJÞ ¼ sup t1m jjf ðtÞjjE0 þ jjuðtÞjjE1 ; tAð0;T where f is defined through the fact that uAẼ1 ðJÞ implies u ¼ x þ ga
f ; for some f AẼ0 ðJÞ: Without loss of generality, we take jjyjjE1 ¼ jjAyjjE0 ; for yAE1 ; and note that by Lemma 3, Ẽ1 ðJÞ is a Banach space. We write def def Ey ¼ ðE0 ; E1 Þy ¼ ðE0 ; E1 Þ0y;N ; yAð0; 1Þ; ARTICLE IN PRESS 428 P. Cl!ement et al. / J. Differential Equations 196 (2004) 418–447 for the continuous interpolation spaces between E0 and E1 : Recall that if Z is some number such that 0pZop  fA ; then xAEy iff lim jlj-N;jarg ljpZ jjly AðlI þ AÞ1 xjjE0 ¼ 0; ð12Þ and that we may take def jjxjjy ¼ sup jarg ljpZ;la0 jjly AðlI þ AÞ1 xjjE0 as norm on Ey (see [13, Theorem 3.1, p. 159] and [14, p. 314]). Our purpose is to investigate the trace space of Ẽ1 ðJÞ: We define g : Ẽ1 ðJÞ-E0 by gðuÞ ¼ uð0Þ; def and the trace space gðẼ1 ðJÞÞ ¼ ImðgÞ; with def jjxjjgðẼ1 ðJÞÞ ¼ inffjjvjjẼ1 ðJÞ j vAẼ1 ðJÞ; gðvÞ ¼ xg: It is straightforward to show that this norm makes gðẼ1 ðJÞÞ a Banach space. Define 1m m# ¼ 1  a for mAð0; 1Þ; aAð0; 2Þ with a þ m41: Observe that this very last condition is equivalent to m40 # and that ao1 implies mom; # whereas aAð1; 2Þ gives mom: # Thus 0omomo1; # aAð0; 1Þ; 0omomo1; # aAð1; 2Þ: Obviously, if a ¼ 1; then m# ¼ m: We claim Theorem 7. For mAð0; 1Þ; aAð0; 2Þ; a þ m41; one has gðẼ1 ðJÞÞ ¼ Em# : Proof. The case a ¼ 1 is treated in [7]. Thus let aa1 and first consider the case aAð0; 1Þ: Let xAEm# : We define u as the solution of u  x þ ga
Au ¼ 0; tAJ; ð13Þ tAJ: ð14Þ or, equivalently, as the solution of Dat ðu  xÞ þ Au ¼ 0; ARTICLE IN PRESS P. Cl!ement et al. / J. Differential Equations 196 (2004) 418–447 By Clément et al. [4, Lemma 7], u is well defined and given by Z 1 exp½ltðla I þ AÞ1 la1 x dl; t40; uðtÞ ¼ 2pi G1;c 429 ð15Þ A Here cAðp2; minðp; pf a ÞÞ and def Gr;c ¼ freit j jtjpcg,freic j roroNg,freic j roroNg: Note that limtk0 jjuðtÞ  xjjE0 ¼ 0: We assert that limtk0 jjt1m Dat ðu  xÞjjE0 ¼ 0; i.e., that Z lim t1m exp½ltAðla I þ AÞ1 la1 x dl ¼ 0 ð16Þ t-0 G1;c in E0 : To show this assertion, we take t40 arbitrary and rewrite the expression in def (16) ð ¼ IÞ as follows: I ¼ t1m ¼ Z exp½ltAðla I þ AÞ1 la1 x dl G1 t ;c  s exp½s t G1;c Z am# A n s a o1  x sm ds: I þA t ð17Þ The first equality followed by analyticity; to obtain the second we made the variable def transform s ¼ lt and used the definition of m: # Now recall that xAEm# and use (12) in (17) to get (16). Observe also that by the above one has sup jjt1m Dat ðu  xÞjjE0 pcjjxjjEm# ; ð18Þ tAJ0 where c ¼ cðm; cÞ but where c does not depend on T: By (14), (16), (18), sup jjt1m AuðtÞjjE0 pcjjxjjEm# ; tAJ0 lim jjt1m AuðtÞjjE0 ¼ 0: tk0 ð19Þ Continuity of AuðtÞ and Dat ðu  xÞ in E0 for tAð0; T follows from (15). One concludes that Em# CgðẼ1 ðJÞÞ: ð20Þ Observe that we also have: If xAEm# ; and u solves ð13Þ; then uAẼ1 ðJÞ: ð21Þ ARTICLE IN PRESS 430 P. Cl!ement et al. / J. Differential Equations 196 (2004) 418–447 Conversely, take xAgðẼ1 ðJÞÞ and take vAẼ1 ðJÞ such that vð0Þ ¼ x: Then def H0 ðtÞ ¼ t1m Dat ðv  xÞABUC0-0 ðJ; E0 Þ; def H1 ðtÞ ¼ t1m AvðtÞABUC0-0 ðJ; E0 Þ: def It follows that, with H ¼ H0 þ H1 ; Dat ðv  xÞ þ AvðtÞ ¼ tm1 HðtÞ: ð22Þ We take the Laplace transform ðl40Þ of tm1 HðtÞ (take HðtÞ ¼ 0; t4T), to obtain, in E0 ; Z T exp½lttm1 HðtÞ dt ¼ lm 0 Z lT exp½ssm1 H 0 s ds ¼ oðlm Þ l ð23Þ for l-N: For the last equality, use HAC0-0 ðJ; E0 Þ: Obviously, (23) holds with H replaced by H0 : Hence, by the way H0 was defined and after some straightforward calculations, ṽ  l1 x ¼ la oðlm Þ for l-N: ð24Þ Take transforms in (22), use (23), (24) to obtain Aðla I þ AÞ1 x ¼ l1a oðlm Þ; and so, in E0 ; lam# Aðla I þ AÞ1 x-0; l-N: Hence xAEm# : The case aAð1; 2Þ follows in the same way. Again, define u by (13) (or (14)) but now use [5, Lemma 3] instead of [4, Lemma 7]. Note that one in fact takes ut ð0Þ ¼ 0: Relations (15)–(19) remain valid and (20) follows. The proof of the converse part also carries over from the case where aAð0; 1Þ: & We next show that uAẼ1 ðJÞ implies that the values of u remain in Em# : In particular, we have: Theorem 8. Let mAð0; 1Þ; aAð0; 2Þ and let (4) hold. Then Ẽ1 ðJÞCBUCðJ; Em# Þ: ARTICLE IN PRESS P. Cl!ement et al. / J. Differential Equations 196 (2004) 418–447 431 Proof. Take uAẼ1 ðJÞ: By Theorem 7, uð0ÞAEm# : We split u into two parts, writing u ¼ v þ w where v; w satisfy Dat ðv  uð0ÞÞ þ AvðtÞ ¼ 0; vð0Þ ¼ uð0ÞAEm# ; Dat w þ AwðtÞ ¼ tm1 hðtÞ; wð0Þ ¼ 0: ð25Þ ð26Þ The function hABUC0-0 ðJ; E0 Þ is defined through Eqs. (25), (26). We consider the equations separately, beginning with the former. The claim is then that vAẼ1 ðJÞ-BUCðJ; Em# Þ: Take transforms in (25), use analyticity and invert to get, for t40; Z 1 vðtÞ  uð0Þ ¼  exp½ltl1 Aðla I þ AÞ1 uð0Þ dl; 2pi G1 t ;c and so Zm# AðZI þ AÞ1 ðvðtÞ  uð0ÞÞ Z 1 ¼  exp½ltl1 Aðla I þ AÞ1 Zm# AðZI þ AÞ1 uð0Þ dl: 2pi G1 t ;c Thus, using uð0ÞAEm# ; 1 m# jjZ AðZI þ AÞ ðvðtÞ  uð0ÞÞjjE0 pe ¼ e Z Z jexp½ltl1 j djlj G1 t ;c jexp½tjjtj1 djtjpce; G1;c where e40 arbitrary, and ZXZðeÞ sufficiently large. The conclusion is that ½vðtÞ  uð0ÞAEm# ; for all t40: Moreover, jjvðtÞ  uð0ÞjjEm# pcjjuð0ÞjjEm# ; and so jjvðtÞjjEm# pjjvðtÞ  uð0ÞjjEm# þ jjuð0ÞjjEm# p½c þ 1jjuð0ÞjjEm# : Continuity in Em# follows as in the proof of [4, Lemma 12f]. We infer that vABUCðJ; Em# Þ: The fact that vAẼ1 ðJÞ is stated in (21). We proceed to (26). By assumption, uAẼ1 ðJÞ: Hence, w ¼ u  vAẼ1 ðJÞ: We claim that wABUCðJ; Em# Þ: To show this, first note that wAẼ1 ðJÞ; wð0Þ ¼ 0; implies that Dat w ¼ tm1 hðtÞ; where hABUC0-0 ðJ; E0 Þ; ð27Þ ARTICLE IN PRESS 432 P. Cl!ement et al. / J. Differential Equations 196 (2004) 418–447 and where suptAJ jjhðtÞjjE0 pjjwjjẼ1 ðJÞ : So, after convolving (27) by t1þa and estimating in E0 ; Z t 1 jjwðtÞjjE0 pðGðaÞÞ jjwjjẼ1 ðJÞ ðt  sÞ1þa sm1 dspGð1  aÞtaþm1 jjwjjẼ1 ðJÞ : ð28Þ 0 Moreover, jjwðtÞjjE1 ¼ jjAwðtÞjjE0 ptm1 jjwjjẼ1 ðJÞ : ð29Þ We interpolate between the two estimates (28),(29). To this end, recall that def Kðt; wðtÞ; E0 ; E1 Þ ¼ inf wðtÞ¼aþb jjajjE0 þ tjjbjjE1 ; a t wðtÞ t fix t; and choose a ¼ tþt a wðtÞ; b ¼ tþta : Then, by (28), (29), 2Gð1  aÞttaþm1 Kðt; wðtÞ; E0 ; E1 Þp jjwjjẼ1 ðJÞ : t þ ta So, without loss of generality, jjwðtÞjjEm# ¼ sup tm# Kðt; wðtÞ; E0 ; E1 Þ tAð0;1 p sup tAð0;1 2Gð1  aÞt1m# taþm1 jjwjjẼ1 ðJÞ : t þ ta It is not hard to show that from this follows: jjwðtÞjjEm# p2Gð1  aÞjjwjjẼ1 ðJÞ ; tAJ: ð30Þ Finally observe that the same estimate holds with J ¼ ½0; T replaced by J1 ¼ ½0; T1  for any 0oT1 oT; and recall (3). Thus wðtÞ is continuous in Em# at t ¼ 0: To have continuity for t40 it suffices to observe that since wAẼ1 ðJÞ; then wABUC1m ðJ; DðAÞÞ; and so, (with DðAÞ ¼ E1 ) a fortiori, wACðð0; T; Em# Þ: Thus wABUCð½0; T; Em# Þ: Adding up, we have u ¼ v þ wABUCðJ; Em# Þ: Theorem 8 is proved. & Corollary 9. For uAẼ1 ðJÞ with gðuÞ ¼ 0 one has jjujjBUCðJ;Em# Þ p2Gð1  aÞjjujjẼ1 ðJÞ : ð31Þ ARTICLE IN PRESS 433 P. Cl!ement et al. / J. Differential Equations 196 (2004) 418–447 Proof. It suffices to note that if uAẼ1 ðJÞ; with gðuÞ ¼ 0; then v in (25) vanishes identically and u ¼ w; (w as in (26)) and to recall (30). & Next, we consider Hölder continuity. Theorem 10. Let mAð0; 1Þ; aAð0; 2Þ; a þ m41: Then Ẽ1 ðJÞCBUC a½1s½1m ðJ; Es Þ; 0pspm: # Note that if a þ m42; then the Hölder exponent exceeds 1; provided s40 is sufficiently small. Proof. The case a ¼ 1 was in fact covered in [7]. The case s ¼ m# was already considered above. In case s ¼ 0; the claim is Ẽ1 ðJÞCBUC aþm1 ðJ; E0 Þ: To see that this claim is true, note that if uAẼ1 ðJÞ; then Dat ðu  uð0ÞÞ ¼ tm1 hðtÞ; where hABUC0-0 ðJ; E0 Þ and suptAJ jjhðtÞjjE0 pjjuðtÞjjẼ1 ðJÞ : Then jjuðtÞ  uð0ÞjjE0 pGð1  aÞtaþm1 jjujjẼ1 ðð0;tÞÞ : ð32Þ So we have the desired Hölder continuity at t ¼ 0 for s ¼ 0: The case t40 is straightforward and left to the reader. There remains the case sAð0; mÞ: # By the Reiteration theorem, Es ¼ ðE0 ; Em# Þs ; and m# by the interpolation inequality, s 1 # s # jjuðtÞ  uðsÞjjEs pcjjuðtÞ  uðsÞjjE0 m jjuðtÞ  uðsÞjjmEm# ; Hence, for s ¼ 0; using (32) and the fact that jjuðtÞjjEm# is bounded, jjuðtÞ  uð0ÞjjEs pct s ½aþm1½1m#  We leave the case 0osot to the reader. ¼ cta½1s½1m : & 5. Maximal regularity Let E1 ; E0 ; A be as in Section 4. Let mAð0; 1Þ; aAð0; 2Þ; a þ m41: We have shown that given uAẼ1 ðJÞ we have uð0ÞAEm# : Also, by definition, if uAẼ1 ðJÞ; then def f ¼ Dat ðu  uð0ÞÞ þ AuAẼ0 ðJÞ: ARTICLE IN PRESS 434 P. Cl!ement et al. / J. Differential Equations 196 (2004) 418–447 We now consider the converse question, i.e., the maximal regularity. We ask whether there exists c40 such that h i jjujjẼ1 ðJÞ pc jjf jjẼ0 ðJÞ þ jjxjjEm# ; where u solves Dat ðu  xÞ þ Au ¼ f : By (21) and linearity we may obviously take x ¼ 0: Thus we let u solve Dat u þ Au ¼ f ; uð0Þ ¼ 0; ð33Þ with f AẼ0 ðJÞ; and claim that uAẼ1 ðJÞ: This will follow only under a particular additional assumption on E0 ; E1 : We first need to formulate some definitions. We write, for oX0; n def def Ha ðE1 ; E0 ; oÞ ¼ AALðE1 ; E0 Þ j Ao ¼ oI þ A  is a nonnegative closed operator in E0 with spectral angle opð1  a2Þ and def Ha ðE1 ; E0 Þ ¼ [ Ha ðE1 ; E0 ; oÞ: oX0 Note that as Ha ðE1 ; E0 ; o1 ÞCHa ðE1 ; E0 ; o2 Þ; for o1 oo2 ; we may as well take the union over, e.g., o40: Also note that Ha ðE1 ; E0 Þ is open in LðE1 ; E0 Þ: Furthermore, we let def Mam ðE1 ; E0 Þ ¼ fAAHa ðE1 ; E0 Þj Dat u þ Au ¼ f ; uð0Þ ¼ 0; has maximal regularity in Ẽ0 ðJÞg: Observe that using the assumption a þ m41 one can show that if Dat u þ Au ¼ f has maximal regularity in Ẽ0 ðJÞ; then Dat u þ ðoI þ AÞu ¼ f has maximal regularity in Ẽ0 ðJÞ for any oAR: We equip Mam ðE1 ; E0 Þ with the topology of LðE1 ; E0 Þ and make the following assumptions on E0 ; E1 : Let F1 ; F0 be Banach spaces such that E1 CF1 CE0 CF0 ; ð34Þ and assume that there is an isomorphism à : F1 -F0 such that à (as an operator in F0 ) is nonnegative with spectral angle fà satisfying fà op 1  a ; 2 ð35Þ ARTICLE IN PRESS P. Cl!ement et al. / J. Differential Equations 196 (2004) 418–447 435 and such that for some yAð0; 1Þ; def E0 ¼ Fy ¼ ðF0 ; F1 Þy0;N ð36Þ Ax ¼ Ãx ð37Þ and such that for xAE1 : Our claim is that if f AẼ0 ðJÞ ¼ BUC1m ðJ; Fy Þ; then Aw lies in the same space and we have a norm estimate. Specifically: Theorem 11. Let mAð0; 1Þ; aAð0; 2Þ; a þ m41: Assume (34), let à be as in (35) and suppose (36), (37) hold. Then AAMam ðE1 ; E0 Þ: Proof. We define F̃0 ¼ BUC1m ðJ; F0 Þ; F̃1 ¼ BUC1m ðJ; F1 Þ: Then ðF̃0 ; F̃1 Þy ¼ BUC1m ðJ; ðF0 ; F1 Þy Þ ¼ BUC1m ðJ; E0 Þ ¼ Ẽ0 ðJÞ: To get the first equality above one recalls the characterization of F0 ; F1 ; and that by Clément et al. [4, Lemma 9(c)] the statement holds for m ¼ 1: The cases mAð0; 1Þ follow by an easy adaptation of the proof of [4, Lemma 9(c)]. The second equality above is (36), the third is the definition of Ẽ0 ðJÞ: Write, for aAð0; 2Þ; def * ðAuÞðtÞ ¼ ÃuðtÞ; def * ðBuÞðtÞ ¼ Dat uðtÞ; * def uADðAÞ ¼ F̃1 ; n o a * def uADðBÞ ¼ u j uABUC1m ð½0; T; F0 Þ; uð0Þ ¼ 0 : One then has, using (8), (35), and Proposition 6, * is positive; densely defined in F̃0 ; with spectral angle op 1  a ; A 2 * is positive densely defined in F̃0 with spectral angle ¼ pa: B 2 * B * are resolvent commuting and 0ArðAÞ-rð * * Moreover, the operators A; BÞ: Consider the equation * þ Au * ¼ f; Bu ð38Þ ARTICLE IN PRESS 436 P. Cl!ement et al. / J. Differential Equations 196 (2004) 418–447 where f AẼ0 ðJÞ: By the Da Prato–Grisvard Method of Sums (in particular see [6, * * such that (38) holds, and such Theorem 4]) there exists a unique uADðAÞ-Dð BÞ * * that Au; BuAẼ0 with * jjAujj Ẽ0 pcjjf jjẼ0 ; where c is independent of f : Thus, recall (37), the function u satisfies (33), uAẼ1 ðJÞ; and there exists c such that jjujjẼ1 ðJÞ pcjjf jjẼ0 ðJÞ : Observe that c ¼ cðTÞ but can be taken the same for all intervals ½0; T1 ; with T1 pT: & 6. Linear nonautonomous equations As earlier, we take mAð0; 1Þ; aAð0; 2Þ; a þ m41; and define m# ¼ 1  1m a : Consider the equation u þ ga
BðtÞu ¼ u0 þ ga
h: ð39Þ We prove Theorem 12. Let E0 ; E1 be as in Section 4, let TAð0; NÞ; J ¼ ½0; T and assume that BACðJ; Mam ðE1 ; E0 Þ-Ha ðE1 ; E0 ; 0ÞÞ; u0 AEm# ; hAẼ0 ðJÞ: ð40Þ Then there exists a unique uAẼ1 ðJÞ solving (39) such that BðtÞuðtÞAẼ0 ðJÞ and there exists c40 such that jjujjBUC1m ðJ;E1 Þ þ jjDat ðu  u0 ÞjjẼ0 ðJÞ pc jju0 jjEm# þ jjhjjẼ0 ðJÞ : Proof. From (40) it follows that the norms def jjxjjEm# ¼ sup jjlm# BðsÞðlI þ BðsÞÞ1 xjjE0 l40 are all uniformly equivalent for sA½0; T: ð41Þ ARTICLE IN PRESS P. Cl!ement et al. / J. Differential Equations 196 (2004) 418–447 437 Fix sA½0; T; T 0 Að0; T; and write J 0 ¼ ½0; T 0 : Let uðsÞ ¼ uðsÞ ðtÞ be the solution of Dat ðuðsÞ  u0 Þ þ BðsÞuðsÞ ¼ h; on J 0 : We claim that there exists c1 40; independent of s; T 0 ; such that jjDat ðuðsÞ  u0 ÞjjẼ0 ðJ 0 Þ þ jjBðsÞuðsÞ ðtÞjjẼ0 ðJ 0 Þ pc1 jju0 jjEm# þ jjhjjẼ0 ðJ 0 Þ : ðsÞ ð42Þ ðsÞ To prove (42), write uðsÞ ¼ u1 þ u2 ; where ðsÞ ðsÞ Dat ðu1  u0 Þ þ BðsÞu1 ¼ 0; ðsÞ ðsÞ Dat u2 þ BðsÞu2 ¼ h; ðsÞ u1 ð0Þ ¼ u0 ; ðsÞ u2 ð0Þ ¼ 0: By (18), ðsÞ jjDat ðu1  u0 ÞjjẼ0 ðJ 0 Þ pcjju0 jjEm# ; where c ¼ cðm; cðsÞÞ: By (40), cðsÞ; hence c; can be taken independent of s: By the fact that B takes values in Mam ðE1 ; E0 Þ one has ðsÞ ðsÞ jjDat u2 jjẼ0 ðJ 0 Þ þ jjBðsÞu2 jjẼ0 ðJ 0 Þ pc̃jjhjjẼ0 ðJ 0 Þ ; and from the fact that BACðJ; LðE1 ; E0 ÞÞ one concludes that c̃ can be taken independent of s: Hence claim (42) holds. Choose nX1 such that with q ¼ n1 T one has c1 1 jjBðtÞ  Bððj  1ÞqÞjjLðE1 ;E0 Þ p ; 2 j¼1;y;n;ðj1Þqptpjq max ð43Þ where c1 as in (42). Fix jAf1; 2; y; ng; and assume we have a unique solution u% j1 of (39) on ½0; ðj  1Þq (for j ¼ 1; take u% 0 ¼ u0 ). Then define (recall (11))   Z̃j ¼ uAẼ1 ð½0; jqÞ; uð0Þ ¼ u0 j uðtÞ ¼ u% j1 ðtÞ; 0ptpðj  1Þq : Given an arbitrary vAZ̃j ; we let uj be the unique solution of u þ ga
Bððj  1ÞqÞu ¼ u0 þ ga
h þ ga
½Bððj  1ÞqÞ  BðtÞv on ½0; jq: Clearly, ½Bððj  1ÞqÞ  BðtÞvABUC1m ð½0; jq; E0 Þ: By uniqueness, uj AZ̃j : Denote the map vAZ̃j -uj AZ̃j by Fj : By (42),(43), and observing that v1 ¼ v2 on ½0; ðj  1Þq; 1 jjFj ðv1 Þ  Fj ðv2 ÞjjẼ1 ð½0;jqÞ p jjv1  v2 jjẼ1 ð½0;jqÞ : 2 ARTICLE IN PRESS 438 P. Cl!ement et al. / J. Differential Equations 196 (2004) 418–447 Observe that Z̃j is closed in Ẽ1 ð½0; jqÞ; hence it is a complete metric space with respect to the induced metric. Consequently we may apply the Contraction mapping Theorem and conclude that there exists a unique fixed point of Fj in Z̃j : Denote this fixed point by u% j : Clearly u% j solves (39) on ½0; jq: Proceeding by induction we have the existence of a solution uAẼ1 ðJÞ of (39). The induction procedure also gives c40 such that (41) holds. & 7. Local nonlinear theory We consider the quasilinear equation Dat ðu  u0 Þ þ AðuÞu ¼ f ðuÞ þ hðtÞ; t40; ð44Þ under the following assumptions. Let mAð0; 1Þ aAð0; 2Þ; a þ m41; ð45Þ and define m# as earlier by m# ¼ a1 ða þ m  1Þ: For X ; Y Banach spaces, and g a mapping of X into Y ; write gAC 1 ðX ; Y Þ if every point xAX has a neighbourhood U such that g restricted to U is globally Lipschitz continuous. Let E0 ; E1 be Banach spaces such that E1 CE0 with dense imbedding and suppose ðA; f ÞAC 1 ðEm# ; Mam ðE1 ; E0 Þ u0 AEm# ; hABUC1m ð½0; T; E0 Þ; E0 Þ; for any T40: ð46Þ ð47Þ Observe that by (46), for ũAEm# there exists oðũÞX0 such that def Ao ðũÞ ¼ AðũÞ þ oðũÞIAHa ðE1 ; E0 ; 0Þ-Mam ðE1 ; E0 Þ: We define a solution u of (44) on an interval JCRþ containing 0 as a function u satisfying uACðJ; E0 Þ-Cðð0; T; E1 Þ; uð0Þ ¼ u0 ; and such that the fractional derivative of u  u0 of order a satisfies Dat ðu  u0 ÞACðð0; T; E0 Þ and such that (44) holds on 0otpT: Our result is: Theorem 13. Let (45)–(47) hold, where Em# ¼ ðE0 ; E1 Þm0;N is a continuous interpolation # space. Then there exists a unique maximal solution u defined on the maximal interval of existence ½0; tðu0 ÞÞ; where tðu0 ÞAð0; N; and such that for every Totðu0 Þ one has a (i) uABUC1m ð½0; T; E1 Þ-BUCð½0; T; Em# Þ-BUC1m ð½0; T; E0 Þ; (ii) u þ ga
AðuÞu ¼ u0 þ ga
ðf ðuÞ þ hÞ; 0ptpT; (iii) If tðu0 ÞoN; then ueUCð½0; tðu0 ÞÞ; Em# Þ; ARTICLE IN PRESS P. Cl!ement et al. / J. Differential Equations 196 (2004) 418–447 439 (iv) If tðu0 ÞoN and E1 CCE0 ; then lim sup jjuðtÞjjEd ¼ N; for any dAðm; # 1Þ: tmtðu0 Þ We recall that u defined on an interval J is called a maximal solution if there does not exist a solution v on an interval J 0 strictly containing J such that v restricted to J equals u: If u is a maximal solution, then J is called the maximal interval of existence. In this section, we prove existence and uniqueness of u satisfying (i), (ii) for some T40: The continuation is dealt with in Section 8. Proof of Theorem 13 (i), (ii). Choose o such that Ao ðu0 ÞAHa ðE1 ; E0 ; 0Þ: Then Ao ðu0 ÞAMa ðE1 ; E0 Þ and there exists a constant cu0 ; independent of F ; such that if F AẼ0 ðJÞ and u ¼ uðF Þ solves Dat u þ Ao ðu0 Þu ¼ F ðtÞ; 0otpT; with uð0Þ ¼ 0; then jjujjẼ1 ð½0;TÞ pcu0 ðGð1  aÞÞ1 jjF jjẼ0 ðJÞ : ð48Þ Define BðuÞ ¼ Aðu0 Þ  AðuÞ; uAEm# : Then BAC 1 ðEm# ; LðE1 ; E0 ÞÞ; and so, by (46) there exists r0 40; LX1 such that jjðB; f Þðz1 Þ  ðB; f Þðz2 ÞjjLðE1 ;E0 Þ E0 pLjjz1  z2 jjEm# ; ð49Þ for z1 ; z2 AB% Em# ðu0 ; r0 Þ; and such that 1 ; 12cu0 zAB% Em# ðu0 ; r0 Þ: ð50Þ jjf ðzÞ þ oðu0 ÞzjjE0 pb; zAB% Em# ðu0 ; r0 Þ; ð51Þ jjBðzÞjjLðE1 ;E0 Þ p Define b by and   1 e0 ¼ min r0 ; : 12cu0 L ð52Þ Let ũ solve Dat ðũ  u0 Þ þ Ao ðu0 Þũ ¼ 0; on ½0; T: ð53Þ ARTICLE IN PRESS 440 P. Cl!ement et al. / J. Differential Equations 196 (2004) 418–447 Take t40 small enough so that (ũ as in (53)) e0 jjũ  u0 jjEm# p ; 2 tA½0; t; e0 jjũjjẼ1 ðJt Þ p ; 2 Gð1  aÞt1m pmin   e0 1 ; ; 12cu0 b 12cu0 ðL þ oðu0 ÞÞ jjhjjẼ0 ðJt Þ p e0 ; 12cu0 where Jt ¼ ½0; t: Define n o Wu0 ðJt Þ ¼ vAẼ1 ðJt Þ j vð0Þ ¼ u0 ; jjv  u0 jjCðJt ;Em# Þ pe0 -B% Ẽ1 ðJt Þ ð0; e0 Þ ð54Þ ð55Þ ð56Þ ð57Þ ð58Þ and give this set the topology of Ẽ1 ðJt Þ: Then Wu0 ðJt Þ is a closed subset of Ẽ1 ðJt Þ; and therefore a complete metric space. Moreover, Wu0 ðJt Þ is nonempty, because ũAWu0 ðJt Þ: Consider now the map Gu0 : Wu0 ðJt Þ-Ẽ1 ðJt Þ defined by u ¼ Gu0 ðvÞ; vAWu0 ðJt Þ; where u solves Dat ðu  u0 Þ þ Ao ðu0 Þu ¼ BðvÞv þ f ðvÞ þ oðu0 Þv þ hðtÞ: ð59Þ Our first claim is that this map is well defined. To see this, note that as BAC 1 ðEm# ; LðE1 ; E0 ÞÞ and v is continuous in Em# ; and by the assumption on f ; h it follows that the right-hand side of (59) is in Cðð0; t; E0 Þ: Also, by (50), (51),(53), (56)–(58), sup t1m jjBðvðtÞÞvðtÞ þ f ðvðtÞÞ þ oðu0 ÞvðtÞ þ hðtÞjjE0 0otpt p sup ðt1m jjBðvðtÞÞjjLðE1 ;E0 Þ jjvðtÞjjE1 Þ þ t1m b þ jjhjjẼ0 ðJt Þ 0otpt 1 e0 e0 e0 p jjvjjẼ1 ðJt Þ þ þ p : 12cu0 12cu0 12cu0 4cu0 ð60Þ So the right-hand side of (59) is in Ẽ0 ðJt Þ; and hence, by (21),(48), (53), the map is well defined. ARTICLE IN PRESS P. Cl!ement et al. / J. Differential Equations 196 (2004) 418–447 441 Next, we assert that uAWu0 ðJt Þ: We show first sup jjGu0 ðvÞðtÞ  u0 jjEm# pe0 : ð61Þ Gu0 ðvÞ ¼ ũ þ G̃u0 ðvÞ; ð62Þ tA½0;t Split Gu0 ðvÞ: where G̃u0 ðvÞ solves (zero initial value) Dat ðG̃u0 ðvÞÞ þ Ao ðu0 ÞG̃u0 ðvÞ ¼ BðvÞv þ f ðvÞ þ oðu0 Þv þ hðtÞ: By (31), (48), (60), sup jjG̃u0 ðvÞðtÞjjEm# p 2Gð1  aÞjjG̃u0 ðvÞjjẼ1 ðJt Þ tA½0;t p 2cu0 jjBðvÞv þ f ðvÞ þ oðu0 Þv þ hjjẼ0 ðJt Þ p2cu0 e0 e0 ¼ : 4cu0 2 ð63Þ Combining (54) and (63) we have (61). Next, we assert that jjGu0 ðvÞjjẼ1 ðJt Þ pe0 : To show this, split as in (62) and recall (55),(63). So Gu0 ðvÞAWu0 ðJt Þ: Finally, we claim that Gu0 is a contraction. We have, by linearity and (31), (48), (49), (50), jjGu0 ðv1 Þ  Gu0 ðv2 ÞjjẼ1 ðJt Þ pcu0 jjBðv1 Þv1  Bðv2 Þv2 jjẼ0 ðJt Þ þ cu0 jjf ðv1 Þ  f ðv2 ÞjjẼ0 ðJt Þ þ cu0 oðu0 Þjjv1  v2 jjẼ0 ðJt Þ pcu0 jj½Bðv1 Þ  Bðv2 Þv1 jjẼ0 ðJt Þ þ cu0 jjBðv2 Þ½v1  v2 jjẼ0 ðJt Þ þ cu0 t1m ½L þ oðu0 Þ sup jjv1 ðtÞ  v2 ðtÞjjEm# t 1 jjv1  v2 jjẼ1 ðJt Þ 12 1 þ 2Gð1  aÞcu0 t1m ½L þ oðu0 Þjjv1  v2 jjẼ1 ðJt Þ p jjv1  v2 jjẼ1 ðJt Þ ; 2 pcu0 Ljjv1  v2 jjẼ1 ðJt Þ 2Gð1  aÞjjv1 jjẼ1 ðJt Þ þ where the last step follows by (52) and(56). Thus the map v-Gðu0 Þv is a contraction and has a unique fixed point. We conclude that there exists u satisfying (i), (ii), for some T40: ARTICLE IN PRESS 442 P. Cl!ement et al. / J. Differential Equations 196 (2004) 418–447 We proceed to the proof of uniqueness. Assume there exist two functions u1 ; u2 ; both satisfying (i), (ii) on ½0; T for some T40 and u1 ðtÞ not identically equal to u2 ðtÞ on ½0; T: Define   t1 ¼ sup tA½0; T j ð44Þ has a unique solution in Ẽ1 ð½0; tÞ : def Then 0pt1 oT: Also, for any tAðt1 ; T there exists a solution u of (44) on Jt ¼ ½0; t; such that uðtÞ ¼ u1 ðtÞ on ½0; t1  but u does not equal u1 everywhere on t1 otpt: Let, for tAðt1 ; T; Jt ¼ ½0; t; n Wu1 ðJt Þ ¼ vAẼ1 ðJt Þ j vðtÞ ¼ u1 ðtÞ; 0ptpt1 ; o jjv  u1 jjCðJ ;E Þ pe0 -B% Ẽ ðJ Þ ðu1 ðtÞ; e0 Þ: t m# 1 t Give this set the topology of Ẽ1 ðJt Þ: Then Wu1 ðJt Þ is a complete metric space which is nonempty because u1 AWu1 ðJt Þ: Consider the map Gu1 : Wu1 ðJt Þ-Ẽ1 ðJt Þ defined bu u ¼ Gu1 ðvÞ for vAWu1 ðJt Þ; where u solves Dat ðu  u0 Þ þ Ao ðu1 ðt1 ÞÞuðtÞ ¼ BðvðtÞÞvðtÞ þ f ðvðtÞÞ þ oðu1 ðt1 ÞÞvðtÞ þ hðtÞ; def with BðvðtÞÞ ¼ Aðu1 ðt1 ÞÞ  AðvðtÞÞ and where we have chosen oðuðt1 ÞÞ such that Ao ðu1 ðt1 ÞÞAHa ðE1 ; E0 ; 0Þ: By (46), Ao ðu1 ðt1 ÞÞAMam ðE1 ; E0 Þ: Proceed as in the existence part to show that the map Gu1 is welldefined, and that for t sufficiently close to t1 one has that Gu1 maps Wu1 ðJt Þ into itself. Finally show that the map is a contraction if t  t1 is sufficiently small and so the map has a unique fixed point. On the other hand, any solution of (44) is a fixed point of the map, provided t (depends on the particular solution) is taken sufficiently close to t1 : A contradiction results and uniqueness follows. Thus we have shown that (i), (ii), and uniqueness hold for some T40: 8. Continuation of solutions We proceed to the final part of the proof of Theorem 13. Suppose we have a unique solution u of (44) on Jt ¼ ½0; t; for some t40; such that uACðJt ; Em# Þ-Ẽ1 ðJt Þ: ARTICLE IN PRESS 443 P. Cl!ement et al. / J. Differential Equations 196 (2004) 418–447 Take T4t and let n def Z ¼ wACð½0; T; Em# Þ j wðtÞ ¼ uðtÞ; tA½0; t; ðt  tÞ1m Dat ðw  u0 ÞABUCððt; T; E0 Þ; jj½t  t1m Dat ðw  u0 ÞjjE0 -0; tkt; o ½t  t1m wABUCððt; T; E1 Þ; jj½t  t1m wjjE1 -0; tkt : ð64Þ Choose e0 sufficiently small. Define def Zu ¼ fwAZ j jjw  uðtÞjjCð½t;T;Em# Þ pe0 ; jjwjjẼ1 ð½t;TÞ pe0 g: ð65Þ Choose oðuðtÞÞ so that Ao ðuðtÞÞAHa ðE1 ; E0 ; 0Þ: For vAZu ; consider ð0ptpTÞ; Dat ðu  u0 Þ þ Ao ðuðtÞÞuðtÞ ¼ AðuðtÞÞvðtÞ  AðvðtÞÞvðtÞ þ f ðvðtÞÞ þ oðuðtÞÞvðtÞ þ hðtÞ: Let uv be the corresponding solution. If uv ¼ v; then we have a solution of (44) on ½0; T; identically equal to u on ½0; t: This solution may however have a singularity for tkt: We may repeat the existence proof above to obtain a unique fixed point (of the map v-uv ) ûðtÞ; 0ptpT; in Zu if T is sufficiently close to t: Clearly, û ¼ u on ½0; t: Moreover, ûACð½0; T; Em# Þ and so, by (46), AðûðtÞÞ; tA½0; T; is a compact subset of Ha ðE1 ; E0 Þ: Now use the arguments of [1, Corollary 1.3.2 and proof of Theorem # such that 2.6.1; 9, p. 10] to deduce that there exists a fixed oX0 def # Ao# ðûðtÞÞ ¼ AðûðtÞÞ þ oIAH am ðE1 ; E0 ; 0Þ for every tA½0; T: Also, def Ao# ðtÞ ¼ Ao# ðûðtÞÞACð½0; T; LðE1 ; E0 ÞÞ and so Ao# ðtÞ satisfies (40) (recall that a þ m41 is assumed.) In addition, ˆ def fðtÞ ¼ f ðûðtÞÞABUCð½0; T; E0 ÞCẼ0 ð½0; TÞ; # ûðtÞACð½0; T; Em# ÞCẼ0 ð½0; TÞ: o Then note that û solves ˆ þo # ûðtÞ þ hðtÞ; Dat ðu  u0 Þ þ Âo# ðtÞuðtÞ ¼ fðtÞ tA½0; T; ð66Þ and that the earlier result on nonautonomous linear equations can be applied. But by this result there is a unique function û1 ðtÞ in BUC1m ð½0; T; E1 Þ solving (66) on ½0; T: ARTICLE IN PRESS 444 P. Cl!ement et al. / J. Differential Equations 196 (2004) 418–447 Moreover, there certainly exists T1 4t such that û1 considered on ½0; T1  is contained in Zu (in the definition of Zu ; take T ¼ T1 ). Thus we must have û1 ¼ û on ½t; T1  and so û does not have a singularity as tkt: The solution u may therefore be continued to ½0; T1 ; for some T1 4t; so that (i), (ii) are satisfied on ½0; T1 : (iii) Suppose 0otðu0 ÞoN; and assume uAUCð½0; tðu0 ÞÞ; Em# Þ: Then limtmtðu0 Þ exists in Em# : Define ũðtÞ ¼ uðtÞ; tA½0; tðu0 ÞÞ; ũðtÞ ¼ lim uðtÞ; t ¼ tðu0 Þ: tmtðu0 Þ # sufficiently large, Then ũACð½0; tðu0 Þ; Em# Þ: Define, for o BðtÞ ¼ Ao# ðũðtÞÞ; ˜ ¼ f ðũðtÞÞ þ o # ũðtÞ; fðtÞ 0ptptðu0 Þ: By (46) and the compactness arguments above we have that BðtÞ satisfies the assumptions required in our nonautonomous result. Consider then ˜ þ hðtÞ; Dat ðv  u0 Þ þ BðtÞv ¼ fðtÞ 0ptptðu0 Þ: By the earlier result on linear nonautonomous equations, there exists a unique vAẼ1 ð½0; tðu0 ÞÞ which solves this equation on ½0; tðu0 Þ: By uniqueness, vðtÞ ¼ uðtÞ; 0ptotðu0 Þ: But vAUCð½0; tðu0 Þ; Em# Þ and so vðtðu0 ÞÞ ¼ ũðtðu0 ÞÞ; hence vðtÞ ¼ ũðtÞ; 0ptptðu0 Þ: Thus Dat ðv  u0 Þ þ AðvðtÞÞvðtÞ ¼ f ðvðtÞÞ þ hðtÞ; 0ptptðu0 Þ: By earlier results we may now continue the solution past tðu0 Þ and so a contradiction follows. (iv) Suppose tðu0 ÞoN and assume lim suptmtðu0 Þ jjuðtÞjjEd oN for some d4m: # Consider the set uð½0; tðu0 ÞÞÞ: This set is bounded in Ed ; hence its closure is compact in Em# : Take any t%Að0; tðu0 ÞÞ: Consider Dat ðu  u0 Þ þ Ao ðuðt%ÞÞ ¼ ½Aðuðt%ÞÞ  AðvðtÞÞvðtÞ þ f ðvðtÞÞ þ oðuðt%ÞÞvðtÞ þ hðtÞ; and the solution u (which we have on ½0; tðu0 ÞÞ) on ½0; t%: Now let t% play the role of t in (64), and define the set from which v is picked as in (65). Then, as in the considerations following (64), (65), we obtain a continuation of uðtÞ to ½t%; t% þ d; where d ¼ dðuðt%ÞÞ40: (By uniqueness, on ½t%; tðu0 ÞÞ this is of course the solution we already have.) On the other hand, d depends continuously on uðt%Þ: But the closure of S 0pt%otðu0 Þ uðt%Þ is compact in Em# ; and so dðuðt%ÞÞ is bounded away from zero for 0pt%otðu0 Þ: Hence the solution may be continued past tðu0 Þ (take t% sufficiently close to tðu0 Þ) and a contradiction follows. ARTICLE IN PRESS P. Cl!ement et al. / J. Differential Equations 196 (2004) 418–447 445 9. An example In this last section we indicate briefly how our results may be applied to the quasilinear equation u ¼ u0 þ ga
ðsðux Þx þ hÞ; tX0; xAð0; 1Þ; ð67Þ with u ¼ uðt; xÞ; and uðt; 0Þ ¼ uðt; 1Þ ¼ 0; tX0; uð0; xÞ ¼ u0 ðxÞ: As was indicated in the Introduction, this problem occurs in viscoelasticity theory, see [10]. We require sAC 3 ðRÞ; with sð0Þ ¼ 0; ð68Þ and impose the growth condition 0os0 ps0 ðyÞps1 ; yAR; ð69Þ for some positive constants s0 s1 : Take F0 ¼ fuAC½0; 1 j uð0Þ ¼ uð1Þ ¼ 0g; and F1 ¼ fuAC 2 ½0; 1 j uðiÞ ð0Þ ¼ uðiÞ ð1Þ ¼ 0; i ¼ 0; 2g: We fix m# ¼ 12; then m ¼ 1  a2; and a þ m41 holds. With yAð0; 12Þ; let E0 ¼ ðF0 ; F1 Þy0;N ¼ fu j uAh2y ½0; 1; uð0Þ ¼ uð1Þ ¼ 0g; ð70Þ E1 ¼ fuAF1 j u00 AE0 g: ð71Þ and Then Em# ¼ E12 ¼ fu j uAh1þ2y ½0; 1; uð0Þ ¼ uð1Þ ¼ 0g: We take, for uAE12; vAE1 ; AðuÞv ¼ s0 ðux Þvxx : Then one has AðuÞvAE0 ; and, more generally, that the well defined map v-AðuÞv lies in LðE1 ; E0 Þ for every uAE12 : ARTICLE IN PRESS 446 P. Cl!ement et al. / J. Differential Equations 196 (2004) 418–447 We claim that this map satisfies AðuÞAMam ðE1 ; E0 Þ-Ha ðE1 ; E0 ; 0Þ: To this end one takes (for fixed uAE12 ) def Ãv ¼ s0 ðux Þv00 ; vAF1 ; and observes that this map is an isomorphism F1 -F0 and that Ã; as an operator in F0 ; is closed, positive, with spectral angle 0: Thus Theorem 11 can be applied and our claim follows. The only remaining condition to be verified is that u-AðuÞAC 1 ðE12 ; LðE1 ; E0 ÞÞ: But this follows after some estimates which make use of the smoothness assumption (68) imposed on s: We thus have, applying Theorem 13: Theorem 14. Let aAð0; 2Þ: Take yAð0; 12Þ and E0 ; E1 as in, (70), (71). Let (68), (69) hold. Assume hABUCa2 ð½0; T; h2y ½0; 1Þ; with hð0Þ ¼ hð1Þ ¼ 0: Assume u0 Ah1þ2y ½0; 1 with u0 ð0Þ ¼ u0 ð1Þ ¼ 0: Then (67) has a unique maximal solution u defined on the maximal interval of existence ½0; tðu0 ÞÞ where tðu0 ÞAð0; N and such that for any Totðu0 Þ one has uABUCa ð½0; T; h2þ2y ½0; 1Þ-BUCð½0; T; h1þ2y ½0; 1Þ-BUCaa ð½0; T; h2y ½0; 1Þ: 2 2 If tðu0 ÞoN; then lim suptmtðu0 Þ jjuðtÞjjC 1þ2yþd ¼ N for every d40: In particular, since yAð0; 12Þ is arbitrary, we conclude that if lim sup jjuðtÞjjC 1þd oN; ð72Þ tmtðu0 Þ for some d40; then tðu0 Þ ¼ N: Global existence and uniqueness of smooth solutions of (67) under assumptions (68), (69), is thus seen to follow from (72). However, the verification of (72) is in general a very difficult task. For ao43 this task is essentially solved (see [10]). By different methods, the existence, but not the uniqueness, of a solution u satisfying 1;N uAWloc ðRþ ; L2 ð0; 1ÞÞ-L2loc ðRþ ; W02;2 ð0; 1ÞÞ was proved in [12], for the range aA½43; 32: For 32oao2; only existence of global weak solutions has been proved [11]. We do however conjecture that unique smooth, global solutions do exist for the entire range aAð0; 2Þ: ARTICLE IN PRESS P. Cl!ement et al. / J. Differential Equations 196 (2004) 418–447 447 Acknowledgments The first author acknowledges the support of the Magnus Ehrnrooth foundation (Finland). The second author acknowledges the support of the Nederlandse organisatie voor wetenschappelijk onderzoek (NWO). References [1] H. Amann, Linear and Quasilinear Parabolic Problems, Vol. I, Abstract Linear Theory, Birkhäuser, Basel, 1995. [2] S. Angenent, Nonlinear analytic semiflows, Proc. Roy. Soc. Edinburgh Sect. A 115 (1990) 91–107. [3] Ph. Clément, E. Mitidieri, Qualitative properties of solutions of Volterra equations in Banach spaces, Israel J. Math. 64 (1988) 1–24. [4] Ph. Clément, G. Gripenberg, S-O. Londen, Schauder estimates for equations with fractional derivatives, Trans. Amer. Math. Soc. 352 (2000) 2239–2260. [5] Ph. Clément, G. Gripenberg, S-O. Londen, Regularity properties of solutions of fractional evolution equations, In: Evolution Equations and their Applications in Physical and Life Sciences; Proceedings of the Bad Herrenalb Conference, Marcel Dekker Lecture Notes in Pure and Applied Mathematics, Vol. 215, Marcel Dekker, New York, 2000, pp. 235–246. [6] Ph. Clément, G. Gripenberg, V. Högnäs, Some remarks on the method of sums, In Stochastic Processes, Physics and Geometry: new interplays, II, Amer. Math. Soc. (2000) 125–134. [7] Ph. Clément, G. Simonett, Maximal regularity in continuous interpolation spaces and quasilinear parabolic equations, J. Evolution Equations 1 (2001) 39–67. [8] G. Da Prato, P. Grisvard, Sommes d’operateurs linéaires et équations différentielles opérationelles, J. Math. Pures Appl. 54 (1975) 305–387. [9] R. Denk, M. Hieber, J. Prüss, R-boundedness, Fourier multipliers and problems of elliptic and parabolic type, Preprint 2156, Fachbereich Mathematik, Technische Universität, Darmstad, 2001. [10] H. Engler, Global smooth solutions for a class of parabolic integro-differential equations, Trans. Amer. Math. Soc. 348 (1996) 267–290. [11] G. Gripenberg, Weak solutions of hyperbolic-parabolic Volterra equations, Trans. Amer. Math. Soc. 343 (1994) 675–694. [12] G. Gripenberg, Nonlinear Volterra equations of parabolic type due to singular kernels, J. Differential Equations 112 (1994) 154–169. [13] P. Grisvard, Commutativité de deux foncteurs d’interpolation et applications, J. Math. Pures Appl. 45 (1966) 143–206. [14] P. Grisvard, Équations differentielles abstraites, Ann. Sci. École Norm. Sup 2 (4) (1969) 311–395. [15] A. Lunardi, Analytic Semigroups and Optimal Regularity in Parabolic Problems, Birkhäuser, Basel, 1995. [16] D. Matignon, G. Montseny (Eds.), Fractional Differential Systems: Methods and Applications, SMAI, http://www.emath.fr/proc/Vol.5/, 1998. [17] J.W. Nunziato, On heat conduction in materials with memory, Q. Appl. Math. 29 (1971) 187–204. [18] H. Tanabe, Equations of Evolutions, Pitman, London, 1975. [19] H. Triebel, Interpolation Theory, Function Spaces, Differential Operators, North-Holland, Amsterdam, 1978. [20] A. Zygmund, Trigonometric Series, Cambridge University Press, Cambridge, 1959.