INSTITUTO NACIONAL DE MATEMATICA´ PURA E APLICADA DOUTORADO EM MATEMATICA´
ON THE PROPAGATION OF REGULARITY IN SOME NON-LOCAL DISPERSIVE MODELS
Thesis presented to the Postgraduate Program in Mathematics at Instituto Nacional de Matematica´ Pura e Aplicada-IMPA as partial fulfillment of the requirements for the degree of Doctor in Mathematics. Author: Argenis J Mendez G Advisor: Felipe Linares
Rio de Janeiro, Brazil March 12th, 2019 ii
Dedicated to
To all of you ”Soldados de franela” my deepest respect and admiration. iii
Acknowledgments
Firstly, I would like to thank to God and to my family: my parents Argenis and Mirna, for en- couraging me constantly to pursuit this goal as well as to my brother Gustavo, who constantly supported me in this part of my life. I would like to express my sincere gratitude to my advisor Prof. Felipe Linares for the con- tinuous support of my Ph.D study and related research, for his patience, motivation, and im- mense knowledge. His guidance helped me in all the time of research and writing of this thesis. Much´ısimas gracias profesor. Besides my advisor, I would like to thank the rest of my thesis committee: Prof. Ademir Pastor, Prof. Adan´ Corcho, Prof. Hermano Frid, Prof. Luis Gustavo Farah, and Prof. Alexei Mailybaev, for their insightful comments and suggestion that improved in great measure the presentation of this work. To Prof. Juan Guevara, that without doubts was the person who guided me in Venezuela to pursuit this goal, thank you very much professor. To Prof. Mercedes Arriojas and Yamilet Quintana who undoubtedly gave me their vote of confidence to get here, thanks for all. To Marco Mendez,´ for your incredible help at my arrive to Rio de Janeiro, as well as the price- less advice at IMPA. To IMPA, that without doubts became my home for the last 4 years, for providing me with an adequate environment that has allowed me to grow professionally. Also, I would like to thanks my fellows from the Analysis & PDE group, Oscar´ and Ricardo for the stimulating discussions we always have about mathematics. To Anabel for all your support, thank you. I thank my friends at IMPA. In particular, I am grateful to Daniel, Clarena, Bely, Mateus, Miguel, Lazaro, Gil, Diego, Makson etc. To CNPq, for the financial support.
Brasil, Muito obrigado... Contents
Chapter 1. Preliminary 11 1. Notation 11 2. Commutators and Inequalities 12 3. Commutator Expansions 16
Chapter 2. Dispersive generalized Benjamin-Ono equation: local well-posedness & propagation of regularity 18 1. The Linear Problem. 18 2. The Nonlinear Problem 20 3. Refined Strichartz estimate 23 4. Proof of TheoremA 25 5. Existence of Solutions 30 6. Continuity of the Flow 33 7. Propagation of regularity 34
Chapter 3. Fractional Korteweg-de Vries equation: propagation of regularity 68 1. Proof of TheoremC 69
iv CONTENTS 1
Abstract
In this thesis we study some properties of propagation of regularity of solutions associated to some nonlocal dispersive models. The first analysis, corresponds to the dispersive generalized Benjamin-Ono (DGBO) equation, that can be seen as a dispersive interpolation between Benjamin- Ono (BO) equation and the Korteweg-de Vries (KdV) equation. The second one, is the fractional Korteweg-de Vries (fKdV) equation, that unlike the DGBO, this can be interpreted as a dispersive perturbation of the Burger’s equation. Recently, it has been shown that solutions of the KdV equation and BO equation satisfy the following property: if the initial data has some prescribed regularity on the right hand side of the real line, then this regularity is propagated with infinite speed by the flow solution. In the case that attain us, the nonlocal term present in the dispersive part in both equations, the DGBO and the fKdV equations, are more challenging than for KdV and BO equation. In the case of the DGBO, we firstly prove a local well-posedness result associated to the initial value problem, by using a compactness argument. As a second part of the thesis we study the propagation of regularity for solutions of the DGBO equation. To deal with this a new argument is employed. In fact, the approach used to deal with the later equations cannot be applied directly and some new ideas have to be introduced. The new ingredient combines commutator expansions into the weighted energy estimate that allow us to obtain the property of propagation. Finally, the technique used in the propagation of regularity in solutions of the DGBO allow us to verify that the solutions of the fKdV also satisfy the propagation of regularity phenomena. Here the argument is more involved since the dispersion is very weak. CONTENTS 2
Resumo
Nessa tese, nos´ estudamos algumas propriedades de propagac¸ao˜ de regularidade de soluc¸oes˜ associadas a modelos dispersivos nao˜ locais. O primeiro a ser considerado corresponde a equac¸ao˜ dispersiva generalizada Benjamin-Ono (DGBO), que pode ser observada como a interpolac¸ao˜ dis- persiva entre a equac¸ao˜ de Benjamin-Ono (BO) e a equac¸ao˜ de Korteweg-de Vries (KdV). Para a segunda analise,´ trataremos da equac¸ao˜ fracionaria´ de KdV, a qual, diferente da DGBO, pode ser interpretado como uma perturbac¸ao˜ dispersiva da equac¸ao˜ de Burger. Recentemente, foi mostrado que as soluoes˜ das equac¸oes˜ KdV e BO satisfazem a propriedade: se o dado inicial tem alguma regularidade predeterminada do lado direito da reta, entao˜ essa regularidade e´ propagada com velocidade infinita pelo fluxo da soluao.˜ Os casos os quais es- tamos tratando, os termos nao˜ locais apresentados na parte dispersiva em ambas as equac¸oes,˜ a DGBO e a fKdV, sao˜ mais desafiadoras que as apresentadas para as equac¸oes˜ BO e KdV. No caso da DGBO, nos,´ primeiramente, provamos um resultado de boa colocac¸ao˜ local associado ao problema de valor inicial, utilizando um argumento de compacidade. Na segunda parte da tese, nos´ estudamos a propagac¸ao˜ de regularidade nas soluc¸oes˜ da DGBO. A fim de lidar com esse problema, introduzimos uma nova abordagem. De fato, a abordagem utilizada para lidar com estas equac¸oes˜ nao˜ pode ser aplicada diretamente e novas ideias´ precisam ser introduzidas. Esse novo ingrediente combinado com expansoes˜ do comutador em estimativas de energia com peso permitem-nos obter as propriedades de propagac¸ao˜ desejadas. Finalmente, a tecnica´ usada na propagac¸ao˜ de regularidade nas soluc¸oes˜ da DGBO permite-nos verificar que as soluc¸oes˜ da fKdV tambem´ satisfazem tal fenomenoˆ de propagac¸ao˜ da regularidade. Nessa situac¸ao,˜ o argumento e´ mais elaborado haja vista que a dispersao˜ e´ mais fraca. Introduction
The aim of this work is to study some special regularity properties of solutions to the initial value problem (IVP) associated to some nonlocal dispersives models. The first to be considered is the dispersion generalized Benjamin-Ono equation # B u ¡ Dα+1B u + uB u = 0, x, t P R, 0 α 1, (0.1) t x x x u(x, 0) = u0(x), s P where Dx, denotes the homogeneous derivative of order s R, q s ¡B2 s/2 s | |s p Dx = ( x) thus Dx f = cs ξ f (ξ) ,
s B s which in its polar form is decomposed as Dx = (H x) , where H denotes the Hilbert transform, » 1 f (x ¡ y) p q H f (x) = lim dy = (¡i sgn(ξ) f (ξ)) (x), Ñ + π e 0 |y|¥e y wherep¤ denotes the Fourier transform and q denotes its inverse. These equations model vorticity waves in the coastal zone, see Molinet, Saut and Tzvetkov [36] and the references therein. Our starting point is a property established by Isaza, Linares and Ponce [17] concerning the solutions of the IVP associated to the k¡generalized KdV equation # B u + B3u + ukB u = 0, x, t P R, k P N, (0.2) t x x u(x, 0) = u0(x). It was shown by Isaza, Linares and Ponce [17] that the unidirectional dispersion of the k¡generalized KdV equation entails the following propagation of regularity phenomena.
3/4+ THEOREM 0.1 ([17]). If u0 P H (R) and for some l P Z, l ¥ 1 and x0 P R » 8 § § 2 § §2 Bl §Bl § 8 (0.3) xu0 = xu0(x) dx , L2((x ,8)) 0 x0 then the solution of the IVP associated to (0.2) satisfies that for any v ¥ 0 and e ¡ 0 » 8 j 2 sup Bxu (x, t) dx c, ¡ 0¤t¤T x0+e vt
3 CONTENTS 4
l = = } } + }B } 2 8 P ( ] for j 0, 1, 2, . . . , l with c c l; u0 H3/4 (R); xu0 L ((x0, )); v; e; T . In particular, for all t 0, T , l the restriction of u(¤, t) to any interval (x0, 8) belongs to H ((x0, 8)). Moreover, for any v ¥ 0, e ¡ 0 and R ¡ e » » ¡ T x0+R vt 2 Bl+1 x u (x, t) dx dt c, ¡ 0 x0+e vt
l = } } + }B } 2 8 with c c l; u0 H3/4 (R); xu0 L ((x0, )); v; e; R; T . The proof of Theorem 0.1 is based on weighted energy estimates. In detail, the iterative process in the induction argument is based in a property discovered originally by T. Kato [19] in the context of the KdV equation. More precisely, he showed that solution of the KdV equation satisfies » » T R 2 (B u) (x, t) dx dt ¤ c R; T; }u } 2 , x 0 Lx 0 ¡R being this the fundamental fact in his proof of existence of the global weak solutions of (0.2), for k = 1 and initial data in L2(R). This result was also obtained for the Benjamin-Ono equation by Isaza, Linares and Ponce [18] but it does not follow as the KdV case because of the presence of the Hilbert transform. Later on, Kenig et al. [22] extended the results in Theorem 0.1 to the case when the local regularity of the initial data u0 in (0.3) is measured with a fractional indices. The scope to this case is quite more involved, and its proof is mainly based in weighted energy estimates combined with techniques involving pseudo-differential operators and singular integrals. The property described in Theorem 0.1 is intrinsic to suitable solutions of some nonlinear dispersive models (see also Linares, Ponce and Smith [35]). In the context of 2D models, analogous results for the Kadomtsev- Petviashvili II equation (see Isaza, Linares and Ponce [16]) and Zakharov-Kuznetsov (see Linares and Ponce [32]) equations were proved. Before state our main result we will give an overview of the local well-posedness of the IVP (0.1). Following Kato [19] we have that the initial value problem IVP (0.1) is locally well-posed (LWP) in the Banach space X if for every initial condition u0 P X, there exists T ¡ 0 and a unique solution u(t) satisfying
(0.4) u P C ([0, T] : X) X AT where AT is an auxiliary function space. Moreover, the solution map u0 ÞÝÑ u, is continuous from X into the class (0.4). If T can be taken arbitrarily large, one says that the IVP (0.1) is globally well-posed (GWP) in the space X. It is natural to study the IVP (0.1) in the Sobolev space s ¡B2 ¡s/2 2 P H (R) = (1 x) L (R), s R. There exist remarkable differences between the KdV (0.2) and the IVP (0.1). In case of KdV e.g. it posses infinite conserved quantities, define a Hamiltonian system, have multi-soliton solutions and is a completely integrable system by the inverse scattering method see Coifman and Wicker- hauser [7] and Fokas and Ablowitz [9]. Instead, in the case of the IVP (0.1) there is no integrability, CONTENTS 5 but three conserved quantities (see Sidi, Sulem and Sulem[42]), specifically » » I[u](t) = u dx,M[u](t) = u2 dx, R » § § » R § 1+α §2 1 § 2 § 1 3 H[u](t) = §Dx u§ dx ¡ u dx, 2 R 6 R are satisfied at least for smooth solutions. Another property in which these two models differ, relies in the fact that one can obtain a local existence theory for the KdV equation in Hs(R), based on the contraction principle. On the contrary, this cannot be done in the case of the (IVP) (0.1). This is a consequence of the fact that dispersion is not enough to deal with the nonlinear term. In this direction, Molinet, Saut and s Tzvetkov [36] showed that for 0 ¤ α 1 the IVP (0.1) with the assumption u0 P H (R) is not enough to prove local well-posedness by using fixed point arguments or Picard iteration method. Nevertheless, Herr, Ionescu, Kenig and Koch [15] show that the IVP (0.1) is globally well-posed in the space of the real-valued L2(R)¡functions, by using a renormalization method to control the strong low-high frequency interactions. It is not clear that this theory can be used to establish our main result. We need to have a local theory obtained by using energy estimates in addition to dispersive properties of the smooth solutions. In the first step, we obtain the following a priori estimate for solutions of IVP (0.1)
c}B u} 8 x L1 L }u} 8 s }u } s e T x , LT Hx . 0 Hx part of this estimate is based on the Kato-Ponce commutator estimate (see Kato and Ponce [21]). The inequality above reads as follows: in order to the solution u abide in the Sobolev space s H (R), continuously in time, we require to control the term }Bxu} 1 8 . LT Lx First, we use Kenig, Ponce and Vega in [27] results concerning oscillatory integrals, in order to α+1B obtain the classical Strichartz estimates associated to the group S(t) = etDx x , corresponding to the linear part of the equation in (0.1). In second place, the technique introduced by Koch and Tzvetkov [29] related to refined Strichartz estimate are fundamentals in our analysis. More precisely, their method is mainly based in a de- composition of the time interval in small pieces whose length depends on the spatial frequencies of the solution. This approach allowed to Koch and Tzvetkov to prove local well-posedness, for the + Benjamin-Ono equation in H5/4 (R). Succeeding, Kenig and Koenig [23] enhanced this estimate, + which led to prove local well-posedness for the Benjamin-Ono equation in H9/8 (R). Several issues arise when handling the nonlinear part of the equation in (0.1), nevertheless, following the work of Kenig, Ponce and Vega [24], we manage the loss of derivatives by means of a combination of the local smoothing effect and a maximal function estimate of the group α+1B S(t) = etDx x . These observations lead us to present our first result.
THEOREM A. Let 0 α 1. Set s(α) = 9 ¡ 3α and assume that s ¡ s(α). Then, for any 8 8 P s } } ¡ u0 H (R), there exists a positive time T = T u0 Hs(R) 0 and a unique solution u satisfying (0.1) such that s 1 8 (0.5) u P C ([0, T] : H (R)) and Bxu P L ([0, T] : L (R)) . CONTENTS 6
! ) ¡ ÞÑ P s } } Moreover, for any r 0, the map u0 u(t) is continuous from the ball u0 H (R) : u0 Hs(R) to C ([0, T] : Hs(R)) .
TheoremA is the base result to describe the propagation of regularity phenomena. As we mentioned above the propagation of regularity phenomena is satisfied by the BO and KdV equa- tions respectively. These two models correspond to particular cases of the IVP (0.1), specifically by taking α = 0 and α = 1. A question that arises naturally is to determine whether the propagation of regularity phenom- ena is satisfied for a model with an intermediate dispersion between these two models mentioned above. Our second main result give answer to this problem and it is summarized in the following:
P s 3¡α THEOREM B. Let u0 H (R) with s = 2 , and u = u(x, t), be the corresponding solution of the IVP (0.1) provided by TheoremA. + If for some x0 P R and for some m P Z , m ¥ 2, Bm P 2 t ¥ u (0.6) x u0 L ( x x0 ) , then for any v ¥ 0, T ¡ 0, e ¡ 0 and τ ¡ e » » » 8 + ¡ 2 2 T x0 τ vt α+1 j 2 j sup Bxu (x, t)dx + Dx Bxu (x, t)dx dt ¡ ¡ 0¤t¤T x0+e vt 0 x0+e vt » » ¡ T x0+τ vt α+1 2 2 j + Dx HBxu (x, t)dx dt ¤ c ¡ 0 x0+e vt
m for j = 1, 2, . . . , m with c = c T; e; v; α; }u } s ; }B u } 2 8 ¡ 0. 0 H x 0 L ((x0, )) + If in addition to (0.6) there exists x0 P R
1¡α 2 Bm P 2 t ¥ u (0.7) Dx x u0 L ( x x0 ) then for any v ¥ 0, e ¡ 0 and τ ¡ e » » » 8 2 + ¡ 1¡α T x0 τ vt 2 2 Bm Bm+1 sup Dx x u (x, t) dx + x u (x, t) dx dt ¡ ¡ 0¤t¤T x0+e vt 0 x0+e vt (0.8) » » ¡ T x0+τ vt 2 Bm+1 ¤ + x Hu (x, t) dx dt c ¡ 0 x0+e vt ! 1¡α } } 2 Bm ¡ with c = c T; e; v; α; u0 Hs ; Dx x u0 0. 2 8 L ((x0, ))
REMARK 0.2. (I) It will be clear from our proof that the requirement on the initial data, 3¡α 9¡3α + that is u0 P H 2 (R) in TheoremB, can be lowered to H 8 (R). (II) Also it is worth highlighting that the proof of TheoremB can be extended to solutions of the IVP # B ¡ α+1B kB P P + tu Dx xu + u xu = 0, x, t R, 0 α 1, k Z , u(x, 0) = u0(x). CONTENTS 7
(III) The results in TheoremB still holds for solutions of the defocussing generalized disper- sive Benjamin-Ono equation # B ¡ α+1B ¡ B P tu Dx xu u xu = 0, x, t R, 0 α 1, u(x, 0) = u0(x).
This can be seen applying TheoremB to the function v(x, t) = u(¡x, ¡t), where u(x, t) is a solution of (0.1). In short, TheoremB remains valid, backward in time for initial data
u0, satisfying (0.6) and (0.7).
Although the argument of the proof of TheoremB follows in spirit that of KdV i.e. an induc- tion process combined with weighted energy estimates. The presence of the non-local operator α+1B Dx x, in the term providing the dispersion, makes the proof much harder. More precisely, two difficulties appear, in the first place and the most important is to obtain explicitly the Kato smooth- ing effect as in Kato [19], that as in the proof of Theorem 0.1 it is fundamental. In contrast to KdV equation, the gain of the local smoothing in solutions of the dispersive gen- α+1 eralized Benjamin-Ono equation is just 2 derivatives, so as occurs in the case of the Benjamin- Ono equation Isaza, linares and Ponce [18], the iterative argument in the induction process is 1¡α carried out in two steps, one for positive integers m and another one for m + 2 derivative. In the case of the BO equation Isaza, linares and Ponce [18], the authors obtain the smoothing effect basing their analysis on several commutator estimates, such as the extension of the first Calderon’s commutator for the Hilbert transform (see B. Bajsanskiˇ and Coifman [1]). However, their method of proof do not allow them to obtain explicitly the local smoothing as in Kato [19]. The advantage of our method is that it allows obtain explicitly the smoothing effect for any α P (0, 1) in the IVP (0.1). Roughly, we rewrite the term modeling the dispersive part of the h α+2 2 i equation in (0.1), in terms of an expression involving HDx ; χe,b . At this point, we incorporate Ginibre and Velo [13] results about commutator decomposition. This, allows us to obtain explicitly the smoothing effect as in Kato [19], at every step of the induction process in the energy estimate. Concerning the nonlinear part of IVP (0.1) into the weighted energy estimate, several issues arises. Nevertheless, following Kenig et al. [22] approach, combined with the works of Kato-Ponce [21], and the recent work D. Li [30] on the generalization of several commutators estimate, allow us to overcome these difficulties. Next, we present some immediate consequences of TheoremB.
3¡α COROLLARY 0.3. Let u P C [¡T, T] : H 2 (R) be a solution of the equation in (0.1) described by P + ¤ P TheoremB. If there exist n , m Z with m n such that for some τ1, τ2 R with τ1 τ2
» 8 |Bn |2 8 Bm R 2 8 x u0(x) dx but x u0 L ((τ1, )), τ2 then for any t P (0, T) and any v ¡ 0 and e ¡ 0
» 8 |Bn |2 8 x u(x, t) dx , ¡ τ2+e vt CONTENTS 8
and for any t P (¡T, 0) and any τ3 P R » 8 |Bm |2 8 x u(x, t) dx = . τ3 The second part of the results we will present here are based mainly on verifying if propaga- tion of regularity phenomena is also satisfied by real solutions when considering a model similar to that one in (0.1), but taken into consideration less dispersion than that in (0.1). More precisely, we studied the IVP # B u ¡ DαB u + uB u = 0, x, t P R, 0 α 1, (0.9) t x x x u(x, 0) = u0(x), where u = u(x, t) represents a real value function. This model is known in the literature as the fractional Korteweg-de Vries (fKdV) equation. Unlike the DGBO this new model can be seen as a ”dispersive” interpolation between the Burger’s equation (α = 0) and the Benjamin-Ono equation (α = 1). The IVP (0.9) has been the focus of attention of recent studies where the local well posedness theory has been approached as well as the existence of solitary wave solutions, the stability prop- erties of ground states and numerical simulations, these issues have been largely carried out by Linares et al. [34], [33], Frank and Lenzmann [11], Klein and Saut [28] and Angulo [39]. Since, the applicability of the method used in the proof of TheoremB depends strongly on local well-posedness results in the Sobolev scale with the minimal regularity that ensures Bxu P 8 L1([0, T] : L (R)), our first task is to fix the space where will take place the solutions. To our knowledge the best result that fits with our requirements was obtained by Linares, Pilod and Saut [33] when studying weakly dispersive perturbation of Burger’s. More precisely, they obtain the following:
3 ¡ 3α ¡ THEOREM 0.4. Let 0 α 1. Define s(α) := 2 8 and assume that s s(α). Then, for every s u0 P H (R), there exists a positive time T = T(}u0}s,2) (which can be chosen as a non-increasing functions of its argument) and a unique solution u to (0.9) such that
s 1 8 (0.10) u P C ([0, T] : H (R)) and Bxu P L ([0, T] : L (R)) .
It is also important to point out that several improvements have been made in the well- posedness theory of the IVP (0.9). In this sense, we shall mention the recent work announced by Molinet, Pilod and Vento [38], where the authors prove that the IVP (0.9) is locally well-posedness s ¡ 3 ¡ 5α in the Sobolev space H (R), with s 2 4 . Additionally, Molinet, Pilod and Vento [38] proved α/2 ¡ 6 Global well-posedness (GWP) in the energy space H (R), as long as α 7 . Most of these results are based in a method introduced by Molinet and Vento [37] to obtain energy estimates at low regularity for strongly nonresonant dispersive equations. However, the s 3 5α LWP of the IVP (0.9) in the space H (R), s ¡ ¡ , does not ensure the boundeness of }Bxu} 1 8 2 4 LT Lx which is crucial in our analysis as we mentioned above. Since the space where we will consider solutions has already been described, we will proceed to describe our third main result. CONTENTS 9
P sα+ ¡ α THEOREM C. Let u0 H (R) where sα := 2 2 , and u = u(x, t) be the corresponding solution of the IVP (0.9) provided by Theorem 0.4. Suppose that for some x0 P R and some m P Z+, m ¥ 2, Bm P 2 t ¥ u (0.11) x u0 L ( x x0 ) . Then for any v ¥ 0, T ¡ 0, e ¡ 0 and τ ¡ e are satisfied the following relations: » » » 8 T x +τ¡vt 2 αn 2 0 j+α n+1 j 2 ( 2 ) sup BxDx u (x, t) dx + Dx u (x, t) dx dt ¡ ¡ 0¤t¤T x0+e vt 0 x0+e vt » » ¡ T x0+τ vt n+1 2 j+α( 2 ) + HDx u (x, t) dx dt ¤ c, ¡ 0 x0+e vt » » » 8 + ¡ ¡ α 2 T x0 τ vt 2 j 1 2 j+1 sup BxDx u (x, t) dx + Bx u (x, t) dx dt ¡ ¡ 0¤t¤T x0+e vt 0 x0+e vt » » ¡ T x0+τ vt j+1 2 + HBx u (x, t) dx dt ¤ c, ¡ 0 x0+e vt and » » » 8 T x +τ¡vt 0 m+ α 2 Bm 2 2 sup ( x u) (x, t) dx + Dx u (x, t) dx dt 0¤t¤T x +e¡vt 0 x +e¡vt » » 0 0 + ¡ T x0 τ vt α 2 m+ 2 + HDx u (x, t) dx dt ¤ c, ¡ 0 x0+e vt ¤ ¤ ¤ ¡ ¤ ¤ ¤ r 2 s ¡ r¤s where j = 2, 3, , m 1 and n = 0, 1, , α 1, where denotes the greatest integer. If in addition to (0.11)
1¡ α αj 2 Bm P 2 t ¥ u 2 Bm P 2 t ¥ u (0.12) Dx x u0 L ( x x0 ) and Dx x u0 L ( x x0 ) , ¤ ¤ ¤ r 2 s ¡ ¥ ¡ ¡ ¡ for j = 1, 2, , α 1, then for any v 0, T 0, e 0 and τ e
» » » + !2 αj 2 T m+α j 1 Bm 2 2 sup x Dx u (x, t) dx + Dx u (x, t) dx dt 0¤t¤T R 0 R » » 2 T j+1 ! m+α 2 + HDx u (x, t) dx dt ¤ c, 0 R and » » » 8 T x +τ¡vt 1¡ α 2 0 2 Bm 2 Bm+1 sup x Dx u (x, t) dx + x u (x, t) dx dt ¡ ¡ 0¤t¤T x0+e vt 0 x0+e vt » » ¡ T x0+τ vt 2 Bm+1 ¤ + H x u (x, t) dx dt c. ¡ 0 x0+e vt
REMARK 0.5. We shall remark the difference between the index s(α) and sα used in Theorem 0.4 and TheoremB, respectively.
The method of proof used in TheoremC follows in spirit the same lines than that used in the proof of TheoremB. Nevertheless, the main differences relay on the inductive argument. In the case of the DGBO equation, the induction is carried out in two steps one for a positive integer m CONTENTS 10
1¡α and another for m + 2 , 0 α 1. In contrast, in the fKdV is required a bi-induction argument, the first for a positive integer m, followed by a a sequence of steps of the form m + αj/2, j = ¤ ¤ ¤ r 2 s ¡ ¡ 1, , α 1, and a final one step for m + 1 α/2. A direct consequence of the Theorems B and C is that for an appropriated class of initial data, the regularity of the solutions is propagated with infinite speed to the left as time evolves. Also, the time reversibility property implies that the solution cannot have had some regularity in the past. As well as in the case of the DGBO equation, the propagation of regularity, also is satisfied in some related models to the (0.9). More precisely, the following properties are verified:
REMARK 0.6. (I) It will be clear from our proof that the requirement on the initial data, ¡ α + P (2 2 ) sα+ 3 ¡ 3α that is u0 H (R) in TheoremC can be lowered to H (R), where sα = 2 8 . (II) Also, it is worth highlighting that the proof of TheoremC can be extended to solutions of the IVP # B ¡ αB kB P P + tu Dx xu + u xu = 0, x, t R, 0 α 1, k Z , u(x, 0) = u0(x). (III) The results in TheoremC still holds for solutions of the defocussing fractional Korteweg-de Vries equation # B ¡ α+1B ¡ B P tu Dx xu u xu = 0, x, t R, 0 α 1, u(x, 0) = u0(x). This can be seen applying TheoremC to the function v(x, t) = u(¡x, ¡t), where u(x, t) is a solution of (0.9). In short, TheoremC remains valid, backward in time for initial data
u0, satisfying (0.11) and (0.12).
The document is organized as follows: the first chapter is focused in fix the notation used throughout all the document as well as to present several known results concerning commutators estimates joint with a particular decomposition of these. In chapter 2 we make a review of several issues concerning oscillatory integrals that will allow us to obtain refined Strichartz estimates s ¡ 9¡3α required to proof the local well-posedness of the DGBO equation in the space H (R), s 8 . In the second part of chapter 2 we present the proof of TheoremB. Finally, in the chapter 3 we present the proof of TheoremC. As a direct consequence of the TheoremsB andC, one has that for an appropriate class of initial data, the singularity of the solution travels with infinity speed to the left as time evolves. Also, the time reversibility property implies that the solution cannot have had some regularity in the past. Chapter 1 Preliminary
This chapter is devoted to fix the notation used throughout all the document as well as to make a review of some commutators estimates of Kato-Ponce type, interpolation inequalities and finally a pointwise commutator decomposition that will be fundamental of Theorems B and C.
1. Notation
s The following notation will be used extensively throughout this work. The operators Dx = ¡B2 1/2 s ¡B2 s/2 ¡ ( x) and J = (1 x) denote the Riesz potential and the Bessel potentials of order s, respec- tively. + m For m P Z the Sobolev space H ((x0, 8)) coincides with the space of the functions f P 2 j 2 L ((x0, 8)) whose derivatives B f (in the distributional sense) belongs to L ((x0, 8)) for every j P Z+ with j ¤ m. p For 1 ¤ p ¤ 8, L (R) is the usual Lebesgue space with the norm indicated by } ¤ }Lp . In addi- tion, for s P R, we consider the Sobolev space Hs(R) defined as the set of tempered distributions 2 s/2 p P 2 } } } s } ( f , such that (1 + ξ ) f (ξ) L (R) with its norm defined as f Hs = J f 2 , and the space H˙ R) p L consist of tempered distributions such that |ξ|s f (ξ) P L2(R). In this context, we define £ 8 H (R) = Hs(R). s¥0 Let f = f (x, t) be a function defined for every x P R and t in the interval [0, T], with T ¡ 0 or in p the hole line R. Then, if A denotes any of the spaces defined above, we define the spaces LT Ax p and Lt Ax by the norms
» 1/p » T ! 1/p p p } f } p = } f (¤, t)} dt and } f } p = } f (¤, t)} dt , L Ax A L Ax A T 0 t R with the natural modification in the case p = 8. Moreover, we use similar definitions for the mixed q p q p ¤ ¤ 8 spaces Lx Lt and Lx LT with 1 p, q . For A, B operators we indicate the commutator between A and B as [A; B] = AB ¡ BA.
11 2. COMMUTATORS AND INEQUALITIES 12
The set of the even and odd numbers will be denoted by Q1 and Q2 respectively. Additionally, P + for m Z the sets Q1(m) and Q2(m) will denote the set of the even and odd numbers between 1 and m, respectively. For two quantities A and B, we denote by A . B whenever A ¤ cB, for some positive constant c. Similarly, A & B if A ¥ cB, for some c ¡ 0. We say that A and B are comparable when A . B and B . A and it will be indicated as A B. The dependence on the constant c on other parameters are usually clear from the context and we will often suppress this dependence whenever it be possible. For a real number a we will denote by a+ instead of a + e, whenever e is a positive number whose value is small enough.
2. Commutators and Inequalities
In this section, we state several inequalities to be used in the next sections. First, we have an extension of the Calderon commutator theorem (see Calderon [6]) established by B. Bajsanskiˇ et al. [1].
+ THEOREM 1.1. For any p P (1, 8) and any l, m P Z Y t0u there exists c = c(p; l; m) ¡ 0 such that Bl Bm ¤ Bm+l } } (1.1) x [H; ψ] x f c x ψ f Lp . Lp L8 For a different proof see [8] Lemma 3.1. In our analysis the Leibniz rule for fractional derivatives, established by Grafakos and Se- ungly [14], Kato and Ponce [21] and Kenig, Ponce and Vega [25]; will be crucial. Even though most of these estimates are valid in several dimensions, we will restrict our attention to the one- dimensional case.
LEMMA 1.2. For s ¡ 0, p P [1, 8) } s } } } } s } } } } s } (1.2) D ( f g) Lp . f Lp1 D g Lp2 + g Lp3 D f Lp4 with 1 1 1 1 1 P 8 = + = + , pj (1, ], j = 1, 2, 3, 4. p p1 p2 p3 p4 Also, we will state the fractional Leibniz rule proved by Kenig, Ponce and Vega [24]. P P P 8 LEMMA 1.3. Let s = s1 + s2 (0, 1) with s1, s2 (0, s), and p, p1, p2 (1, ) satisfy 1 1 1 = + . p p1 p2 Then,
} s ¡ s ¡ s } } s1 } } s2 } (1.3) D ( f g) f D g gD f Lp . D f Lp1 D g Lp2 .
Moreover, the case s2 = 0 and p2 = 8 is allowed. A natural question about Lemma 1.3 is to investigate the possible generalization of the es- timate (1.3) when s ¥ 1. The answer to this question was given recently by Li [30], where he establishes new fractional Leibniz rules for the nonlocal operator Ds, s ¡ 0, and related ones, including various end-point situations. 2. COMMUTATORS AND INEQUALITIES 13
THEOREM 1.4. Case 1: 1 p 8. ¡ 8 ¥ P Let s 0 and 1 p . Then for any s1, s2 0 with s = s1 + s2, and any f , g S(R), the following hold: 8 1 1 1 (1) If 1 p1, p2 with p = p + p , then 1 2 ¸ ¸ 1 1 β s ¡ Bα s,α ¡ B s,β } s1 } } s2 } D ( f g) f D g gD f D f p1 D g p2 . α! x β! x . L L α¤s β¤s 1 2 Lp (2) If p = p, p = 8, then 1 2 ¸ ¸ 1 1 β s ¡ Bα s,α ¡ B s,β } s1 } } s2 } D ( f g) x f D g x gD f . D f Lp D g BMO, α! β! α s1 β¤s 2 Lp } ¤ } 1 where BMO denotes the norm in the BMO space . (3) If p = 8, p = p, then 1 2 ¸ ¸ s 1 α s,α 1 β s,β s s D ( f g) ¡ B f D g ¡ B gD f }D 1 f } }D 2 g} p . α! β! . BMO L α¤s β s 1 2 Lp The operator Ds,α is defined via Fourier transform2 z y Ds,αg(ξ) = Ds,α(ξ)gp(ξ),
ys,α ¡αBα | |s D (ξ) = i ξ ( ξ ) . 1 ¤ Case 2: 2 p 1. If 1 p ¤ 1, s ¡ 1 ¡ 1 or s P 2N, then for any 1 p , p 8 with 1 = 1 + 1 , any s , s ¥ 0 2 p 1 2 p p1 p2 1 2 with s + s = s, 1 2 ¸ ¸ 1 1 β s ¡ Bα s,α ¡ B s,β } s1 } } s2 } D ( f g) f D g gD f D f p1 D g p2 . α! x β! x . L L α¤s β¤s 1 2 Lp ° REMARK 1.5. As usual empty summation (such as 0¤α 0) is defined as zero.
PROOF. For a detailed proof of this Theorem and related results, see Li [30]. Next we have the following commutator estimates involving non-homogeneous fractional derivatives, established by Kato and Ponce [21]. ¡ P 8 P 8 LEMMA 1.6. Let s 0 and p, p2, p3 (1, ) and p1, p4 (1, ] be such that 1 1 1 1 1 = + = + . p p1 p2 p3 p4
1For any f P L1 (Rn), the BMO semi-norm is given by loc » } } 1 | ¡ | f BMO = sup | | f (y) ( f )Q dy, Q Q Q
n where ( f )Q is the average of f on Q, and the supreme is taken over all cubes Q in R . 2The precise form of the Fourier transform does not matter. 2. COMMUTATORS AND INEQUALITIES 14
Then, } s } }B } } s¡1 } } s } } } (1.4) [J ; f ]g Lp . x f Lp1 J g Lp2 + J f Lp3 g Lp4 and } s } } s } } } } s } } } J ( f g) Lp . J f Lp1 g Lp2 + J g Lp3 f Lp4 . There are many other reformulations and generalizations of the Kato-Ponce commutator in- equalities (cf. [2] and the references therein). Recently Li [30], has obtained a family of refined Kato-Ponce type inequalities for the operator Ds. In particular he showed that 8 ¤ 8 LEMMA 1.7. Let 1 p . Let 1 p1, p2, p3, p4 satisfy 1 1 1 1 1 = + = + . p p1 p2 p3 p4 Therefore, (a) If 0 s ¤ 1, then } s ¡ s } } s¡1B } } } D ( f g) f D g Lp . D x f Lp1 g Lp2 . (b) If s ¡ 1, then } s ¡ s } } s¡1B } } } }B } } s¡1 } (1.5) D ( f g) f D g Lp . D x f Lp1 g Lp2 + x f Lp3 D g Lp4 . For a more detailed exposition on these estimates see section 5 in Li [30]. In addition, a non sharp commutator estimate is required in our analysis.
8 1 8 LEMMA 1.8. Let φ P C (R) with φ P C (R). If f P Hs(R), s ¡ 0, then for any l ¡ s + 1 0 2 s 1 }[ ] } } } ¡ Dx; φ f 2 . φ l,2 f s 1,2.
PROOF. We will give an outline of the proof. s ¡ First, we use the decomposition of the operator Dx for s 0, given by Bourgain and Li [5] in the study of commutator estimates. In fact, ¸ s s ¡ s¡2j Dx = Jx cs,j Jx + Ks ¤ ¤ s 1 j 2 p p where Ks is a bounded integral operator satisfying Ks : L (R) ÝÑ L (R), 1 ¤ p ¤ 8. By using this decomposition and a proof similar to that found in [10, Chapter 6, Lemma 6.16 ], the lemma follows. In addition, we have the following inequality of Gagliardo-Nirenberg type:
LEMMA 1.9. Let 1 q, p 8, 1 r ¤ 8 and 0 α β. Then, } α } } }1¡θ} β }θ D f Lp . c f Lr D f Lq with 1 1 1 ¡ α = (1 ¡ θ) + θ ¡ β , θ P [α/β, 1]. p r q PROOF. See chapter 4 in Berg and Loftr¨ om¨ [3]. 2. COMMUTATORS AND INEQUALITIES 15
Now, we present a result that will help us to establish the propagation of regularity of solutions of (0.1). A previous result was proved by Kenig et al.(c.f [22], Corollary 2.1) using the fact that Jr (r P R) can be seen as a pseudo-differential operator. Thus, this approach allows to obtain an expression for Jr in terms of a convolution with a certain kernel k(x, y) which enjoys some properties on localized regions in R. In fact, this is known as the singular integral realization of a pseudo-differential operator, whose proof can be found in Stein [43] Chapter 4. The estimate we consider here involves the non-local operator Ds instead of Js.
+ LEMMA 1.10. Let m P Z and s ¥ 0. If f P L2(R) and g P Lp(R), 2 ¤ p ¤ 8, with
(1.6) dist (supp( f ), supp(g)) ¥ δ ¡ 0.
Then } Bm s } } } } } g x D f Lp . g Lp f L2 .
PROOF. Let f , g be functions in the Schwartz class satisfying (1.6). Notice that » g(x) y g(x) (Ds Bm f ) (x) = ixξ | |sBm f ( ) x x 1/2 e ξ x ξ dξ (2π) R (1.7) » g(x) s { = | | ( ¡ Bm f )( ) 1/2 ξ τ x x ξ dξ. (2π) R
3 where τh is the translation operator. Moreover, the last expression in (1.7) defines a tempered distribution for s in a suitable class, that will be specified later. Indeed, for z P C with ¡1 Re(z) 0 » » Bm 1 z { (τ¡x x ϕ) (y) | | ( ¡ Bm )( ) = c(z) y @ P S(R) (1.8) 1/2 ξ τ x x ϕ ξ dξ 1+z d , ϕ (2π) R R |y|
¡ 2 with c(z) is independent of ϕ. In fact, evaluating ϕ(x) = e x /2 in (1.8) yields
2z Γ z+1 c(z) = 2 . 1/2 ¡ z π Γ 2 Thus, for every ϕ P S(R) the right hand side in (1.8) defines a meromorphic function for every test function, which can be extended analytically to a wider range of complex numbers z’s, specifically z with Im(z) = 0 and Re(z) = s ¡ 0 that is the case that attains us. By an abuse of notation, we will denote the meromorphic extension and the original as the same. Thus, combining (1.6), (1.7) and (1.8) it follows that » g(x) (τ¡ Bm f ) (y) ( ) ( s Bm ) ( ) = ( ) x x g x Dx x f x c s 1+s dy R |y| 1t|y|¥δu = c(s)g(x) f ¦ (x) |y|s+m+1
3 P ¡ For h R the translation operator τh is defined as (τh f ) (x) = f (x h). 3. COMMUTATOR EXPANSIONS 16
Notice that the kernel in the integral expression is not anymore singular due to the condition (1.6). In fact, in the particular case that m is even, we obtain after apply integration by parts 1t|y|¥δu g(x) (Ds Bm f ) (x) = c(s, m)g(x) f ¦ (x) x x |y|s+m+1 and in the case m being odd y1t|y|¥δu g(x) (Ds Bm f ) (x) = c(s, m)g(x) f ¦ (x). x x |y|s+m+2
Finally, in both cases combining Young’s inequality and Holder’s¨ inequality one gets 1t| |¥ u m s y δ }g B D f } 2 }g} p } f } 2 x x L . L L | ¤ |s+m+1 Lr } } } } . g Lp f L2 where the index p satisfies 1 1 1 = + , 2 p r which clearly implies p P [2, 8], as was required.
Further, we will use extensively some results about commutator additionally to those pre- sented in previous section. Next, we will study the smoothing effect for solutions of the dispersive generalized Benjamin-Ono equation (0.1) following Kato’s ideas [19].
3. Commutator Expansions
In this section we present several tools obtained by Ginibre and Velo [12], [13] which will be the cornerstone in the proof of TheoremB. They include commutator expansions together with their estimates. The basic problem is to handle the non-local operator Ds for non-integer s and in particular to obtain representations of its commutator with multiplication operators by functions that exhibit as much locality as possible. Let a = 2µ + 1 ¡ 1, let n be a non-negative integer and h be a smooth function with suitable 1 P 8 decay at infinity, for instance with h C0 (R). We define the operator 1 (1.9) R (a) = [HDa; h] ¡ (P (a) ¡ HP (a)H) , n 2 n n
¸ ¡ j ¡j µ¡j (2j+1) µ¡j (1.10) Pn(a) = a c2j+1( 1) 4 D h D 0¤j¤n where ¹ 1 2 2 c = 1, c + = a ¡ (2k + 1) and H = ¡H. 1 2j 1 (2j + 1)! 0¤k j 3. COMMUTATOR EXPANSIONS 17
4 It was shown in [12] that the operator Rn(a) can be represented in terms of anti-commutators as follows 1 (1.11) R (a) = [H; Q (a)] + [Da; [H; h]] , n 2 n + + where the operator Qn(a) is represented in the Fourier space variables by the integral kernel 1 p 1 1 a (1.12) Qn(a) ÝÑ (2π) 2 h(ξ ¡ ξ )|ξξ | 2 2aqn(a, t), 1 with |ξ| = |ξ | e2t, and » t 1 2 ¡ 2 2n+1 ¡ qn(a, t) = (a (2n + 1) )c2n+1 sinh τ sinh((a(t τ))) dτ. a 0 Based on (1.11) and (1.12), Ginibre and Velo [13] obtain the following properties of boundedness and compactness of the operator Rn(a).
PROPOSITION 1.11. Let n be a non-negative integer, a ¥ 1, and σ ¥ 0, be such that (1.13) 2n + 1 ¤ a + 2σ ¤ 2n + 3. Then (a) The operator DσR (a)Dσ is bounded in L2 with norm n σ σ ¡1/2 {a+2σ (1.14) }D R (a)D f } 2 ¤ C(2π) (D h) } f } 2 . n L 1 L Lξ If a ¥ 2n + 1, one can take C = 1. (b) Assume in addition that 2n + 1 ¤ a + 2σ 2n + 3. σ σ 2 Then the operator D Rn(a)D is compact in L (R).
PROOF. See Proposition 2.2 in Ginibre and Velo [13]. In fact the Proposition 1.11 is a generalization of a previous result, where the derivatives of operator Rn(a) are not considered (cf. Proposition 1 in Ginibre and Velo [12]).
4 For any two operators P and Q we denote the anti-commutator by [P; Q]+ = PQ + QP. Chapter 2 Dispersive generalized Benjamin-Ono equation: local well-posedness & propagation of regularity
In this chapter we will establish the local well-posedness result that will be used later to prove our main result concerning the DGBO equation. Our argument of proof uses energy estimates. The key point behind this argument is to es- tablish an estimate of the type }Bxu} 1 8 ¤ cT. To reach this goal we prove a refined Strichartz LT Lx estimate for solutions associated to the linear problem. Once we establish the local well-posedness in a suitable Sobolev space that guarantee us that
}Bxu} 1 8 ¤ cT, we proceed to show the arguments behind the proof of Theorem B. LT Lx
1. The Linear Problem.
The aim of this section is to obtain Strichartz estimates associated to solutions of the IVP (0.1). First, consider the linear problem # B u ¡ Dα+1B u = 0, x, t P R, 0 α 1, (2.1) t x x u(x, 0) = u0(x), whose solution is given by
α+1 it |ξ| ξ p (2.2) u(x, t) = S(t)u0 = e u0 q.
We begin studying estimates of the unitary group obtained in (2.2).
2 1 1 ¤ ¤ 8 PROPOSITION 2.1. Assume that 0 α 1. Let q, p satisfy q + p = 2 with 2 p . Then α q } } (2.3) Dx S(t)u0 u0 L2 , q p . x Lt Lx 2 for all u0 P L (R).
18 1. THE LINEAR PROBLEM. 19
PROOF. The proof follows as an application on Theorem 2.1 in Kenig, Ponce and Vega [27].
REMARK 2.2. Notice that the condition in p implies q P [4, 8], which in one of the extremal (p q) = (8 ) cases , , 4 yields α 4 D S(t)u }u } 2 x 0 4 8 . 0 Lx Lt Lx α which shows the gain of 4 derivatives globally in time for solutions of (2.1). 8 LEMMA 2.3. Assume that 0 α 1. Let ϕk be a C (R) function supported in the interval k¡1 k+1 P + α [2 , 2 ] where k Z . Then, the function Hk defined as $ &' 2k i f |x| ¤ 1, α k | |¡ 1 ¤ | | ¤ k(α+1) Hk (x) = ' 2 2 x 2 i f 1 x c2 , % ¡1 1 + x2 i f |x| ¡ c2k(α+1) satisfies §» § § 8 § § i(tξ|ξ|α+1+xξ) § α (2.4) § e ψk(ξ)dξ§ . Hk (x) ¡8 for |t| ¤ 2, where the constant c does not depend on t nor k. Moreover, we have that ¸8 α+1 α | | k( 2 ) (2.5) Hk ( l ) . 2 . l=¡8
PROOF. The proof of estimate (2.4) is given in Proposition 2.6 Kenig, Ponce and Vega [26], and it uses arguments of localization and the classical Van der Corput’s Lemma. Meanwhile, (2.5) follows exactly that of Lemma 2.6 in [33].
¡ 1 THEOREM 2.4. Assume 0 α 1. Let s 2 . Then,
¸8 1/2 2 } ( ) } 8 ¤ | ( ) ( )| } } s S t u0 L2 L ([¡1,1]) sup sup S t u0 x . u0 Hx x t | |¤ ¤ j=¡8 t 1 j x j+1
s for any u0 P H (R).
PROOF. See Theorem 2.7 in Kenig, Ponce and Vega [26].
Next, we recall a maximal function estimate proved by Kenig, Ponce and Vega [26].
¡ 1 ¡ 3 COROLLARY 2.5. Assume that 0 α 1. Then, for any s 2 and any η 4
¸8 1/2 2 η | ( ) | ( + ) } } s sup sup S t v0 . 1 T v0 Hx . | |¤ ¤ j=¡8 t T j x j+1
PROOF. See Corollary 2.8 in Kenig, Ponce and Vega [26]. 2. THE NONLINEAR PROBLEM 20
2. The Nonlinear Problem
This section is devoted to study general properties of solutions of the non-linear problem # B u ¡ Dα+1B u + uB u = 0, x, t P R, 0 α 1, (2.6) t x x x u(x, 0) = u0(x). We begin this section stating the following local existence theorem proved by Kato [20] and Saut, Teman [40].
P s ¡ 3 THEOREM 2.6. (1) For any u0 H (R) with s 2 there exists a unique solution u to (2.6) in s the class C([¡T, T] : H (R)) with T = T(}u0}Hs ) ¡ 0. 1 s r r (2) For any T T there exists a neighborhood V of u0 in H (R) such that the map u0 ÞÝÑ u(t) from V into C ([¡T, T] : Hs(R)) is continuous. s1 1 (3) If u0 P H (R) with s ¡ s, then the time of existence T can be taken to depend only on }u0}Hs .
Our first goal will be obtain some energy estimates satisfied by smooth solutions of the IVP (2.6). We firstly present a result that arises as a consequence of commutator estimates.
8 LEMMA 2.7. Suppose that 0 α 1. Let u P C([0, T] : H (R)) be a smooth solution of (2.6). If s ¡ 0 is given, then
c}B u} 8 x L1 L (2.7) }u} 8 s }u } s e T x . LT Hx . 0 Hx
PROOF. Let s ¡ 0. By a standard energy estimate argument we have that » » » 1 d s 2 s B s s s B (Jxu) dx + [Jx; u] xu Jxu dx + uJxuJx xu dx = 0. 2 dt R R R Hence integration by parts, Gronwall’s inequality and commutator estimate (1.4) lead to (2.7).
REMARK 2.8. In view of the energy estimate (2.7), the key point to obtain a priori estimates in s s H (R) is to control }Bxu} 1 8 at the H (R)¡level. x LT Lx x Additionally to this estimate, we will present the smoothing effect provided by solutions of dispersive generalized Benjamin-Ono equation. In fact, the smoothing effect was first observed by Kato in the context of the Korteweg-de Vries equation (see [19]). Following Kato’s approach joint with the commutator expansions, we present a result proved by Kenig-Ponce-Vega [26] (see Lemma 5.1). The estimative (1.14) yields the following identity of localization of derivatives. P 8 1 P 8 LEMMA 2.9. Assume 0 α 1. Let be ϕ C (R) with ϕ C0 (R). Then, » » § § § § 2 2 α+1 α + 2 § α+1 § § α+1 § 1 ϕ f D B f dx = §D 2 f § + §D 2 H f § ϕ dx x 4 (2.8) R » R 1 + f R0(α + 2) f dx, 2 R where R0(α + 2) is given as in (1.11) . 2. THE NONLINEAR PROBLEM 21
PROOF. First notice that h i α+1B α+1B ¡ α+1B (2.9) Dx x; ϕ f = Dx x(ϕ f ) ϕDx x f .
Using identity (2.9) and integration by parts we obtain » » » h i α+1 B ¡ α+1B α+1B ϕ f Dx ( x f ) dx = f Dx x; ϕ f dx + f Dx x (ϕ f ) dx (2.10) R »R »R h i ¡ α+1B ¡ B α+1 = f Dx x; ϕ f dx x f Dx (ϕ f ) dx, R R applying Plancherel’s identity to the second term in the right-hand side of (2.10) we find » » B α+1 α+1B (2.11) x f Dx (ϕ f ) dx = ϕ f Dx x f dx. R R Thus, gathering (2.10) and (2.11) we obtain » » 1 h i ϕ f Dα+1(B f ) dx = ¡ f Dα+1B ; ϕ f dx x 2 x (2.12) R » R 1 = f HDα+2; ϕ f dx. 2 R α + 1 Now, we employ (1.9), with µ = and n = 0, which yields 2 1 1 [HDs+2; ϕ] = P (α + 2) ¡ HP (α + 2)H + R (α + 2) 2 0 2 0 0 (2.13) α + 2 α+1 1 α+1 α + 2 α+1 1 α+1 = D 2 ϕ D 2 ¡ HD 2 ϕ D 2 H + R (α + 2). 2 x x 2 x x 0 Hence, inserting (2.13) into (2.12) and using Plancherel’s identity we obtain » » § § » § § § + §2 § + §2 + α + 2 § α 1 § 1 α + 2 § α 1 § 1 f [HDα 2; ϕ] f dx = §D 2 f § ϕ (x) dx + §D 2 H f § ϕ (x) dx x 2 x 2 x (2.14) R » R R
+ f R0(α + 2) f dx. R Therefore, gathering (2.12) and (2.14) follows the identity (2.8).
This identity will be useful in establishing the smoothing effect in solutions of the DGBO equa- tion and the fKdV equation.
1 PROPOSITION 2.10. Let ϕ denote a non-decreasing smooth function such that supp ϕ (¡1, 2) 1| P ¤ ¤ ¡ P 8 and ϕ [0,1) = 1. For j Z, we define ϕj( ) = ϕ( j). Let u C([0, T] : H (R)) be a real smooth solution of (0.1) with 0 α 1. Assume also, that s ¥ 0 and r ¡ 1/2. Then, » » § § § § ! !1/2 T § α+1 §2 § α+1 §2 § s+ 2 § § s+ 2 § 1 §Dx u(x, t)§ + §Dx Hu(x, t)§ ϕj(x) dx dt (2.15) 0 R 1/2 }B } } } 8 } } 8 1 + T + xu 1 8 + T u L Hr u L Hs . . LT Lx T x T x 2. THE NONLINEAR PROBLEM 22
s s PROOF. Apply the operator Dx to the equation in (0.1), multiply by Dxuϕj, and integrate in space to obtain » » » 1 d | s |2 ¡ s 1+α+sB s B s (2.16) Dxu ϕj dx ϕjDxu Dx xu dx + Dx(u xu)Dxu ϕj dx = 0. 2 dt R R R Now the term corresponding to the nonlinear part of equation in (0.1), can be handled through Mihlin’s Theorem (see [31] Theorem 2.8 ) and the commutator estimate in (1.4), » » » s B s ¡1 B 1 | s |2 s B s Dx(u xu)Dxu ϕj dx = xuϕj + uϕj Dxu dx + [Dx; u] xu Dxuϕj dx R 2 R R 2 }B u} 8 + }u} 8 }u} s . . x Lx Lx Hx However, Lemma 2.9 allow us to write the second term in the left-hand side of (2.16) as » » § § § § § §2 § §2! α + 2 s+ α+1 s+ α+1 1 s α+1 B s § 2 § § 2 § ϕjDxu D ( xDxu) dx = §Dx u§ + §Dx Hu§ ϕj dx R 4 R (2.17) » 1 s s + DxuR0(α + 2)Dxu dx. 2 R Replacing (2.17) into (2.16) we obtain » » d s 2 2 s s |D u| ϕ dx + 1 + }B u} 8 + }u} 8 }u} s ¡ D uR (α + 2)D u dx x j x Lx Lx Hx x 0 x dt R R » § § § § (2.18) 2 2! + § α+1 § § α+1 § α 2 § s+ 2 § § s+ 2 § 1 & ( ) §Dx u§ + §Dx Hu§ ϕj dx. 4 R
2 Since the operator R0(α + 2) satisfies (1.14) i.e., R0(α + 2) is bounded in the L -norm; we de- duce after integrating in time in (2.18) that » » § § § § ! T § α+1 §2 § α+1 §2 § s+ 2 § § s+ 2 § §Dx u(x, t)§ + §Dx Hu(x, t)§ ϕ(x ¡ j) dx dt (2.19) 0 R }B } } } 8 8 } } 8 1 + T + xu 1 8 + T u L L u L Hs . . LT Lx T x T x Finally, (2.19) and Sobolev’s embedding yield estimate (2.15).
REMARK 2.11. Using a similar argument as the employed in the proof of Proposition 2.10 is possible obtain the smoothing effect provided by DGBO, but instead localized over any interval ¡ symmetric on the real line. This can be achieved replacing the function ϕj by φR, where R 0 and 1 ¡ 1 φR denotes a nondecreasing smooth function such that φR is supported in ( 2R, 2R) with φR = 1 on [¡R, R].
In addition to the smoothing effect presented above, we will need the following localized version of the Hs(R)-norm. For this propose we will consider a cutoff function ϕ, with the same properties that in Proposition 2.10.
PROPOSITION 2.12. Let s ¥ 0. Then, for any f P Hs(R)
8 1/2 ¸ 2 } } 1 f Hs(R) f ϕj . Hs(R) j=¡8 3. REFINED STRICHARTZ ESTIMATE 23
3. Refined Strichartz estimate
Our first goal in establishing the local well-posedness of (2.6), will start off in obtaining Strichartz estimates associated to solutions of
B ¡ 1+αB (2.20) tu Dx xu = F.
PROPOSITION 2.13. Assume that 0 α 1, T ¡ 0 and σ P [0, 1]. Let u be a smooth solution to ¤ (2.20) defined on the time interval [0, T]. Then there exist 0 µ1, µ2 1/2 such that ¡ α σ ¡ α ¡ 3σ µ1 1 + +ε µ2 1 +ε (2.21) }Bxu} 2 8 T J 4 4 u + T J 4 4 F LT Lx . 8 2 2 2 LT Lx LT Lx for any ε ¡ 0.
REMARK 2.14. The optimal choice in the parameters present in the estimate (2.21) corresponds 1¡α to σ = 2 . Indeed, as pointed out Kenig and Koenig in the case of the Benjamin-Ono equation (case α = 0) (see Remarks in Proposition 2.8 [23]) given a linear estimate of the form }B } µ1 } a } µ2 b xu 2 8 T J u 8 2 + T J F LT Lx . LT Lx 2 2 LT Lx the idea is to apply the smoothing effect (2.15) and absorb as many as derivatives as possible of the 1¡α function F. Concerning to our case, the approach requires the choice a = b + 2 ; this particular 1¡α ¡ ¡ choice, σ = 2 , in the estimate (2.21) provides the regularity s 9/8 3α/8 in TheoremA. ° PROOFOF PROPOSITION 2.13. Let f = k fk denote the Littlewood-Paley decomposition of a 8 function f . More precisely we choose functions η, χ P C (R) with
supp(η) tξ : 1/2 |ξ| 2u and supp(χ) tξ : |ξ| 2u, such that ¸8 η(ξ/2k) + χ(ξ) = 1 k=1 and fk = Pk( f ), where
{ p { k p ¥ (P0 f )(ξ) = χ(ξ) f (ξ) and (Pk f )(ξ) = η(ξ/2 ) f (ξ) for all k 1.
Recall that for 1 p 8 the Littlewood-Paley theorem
¸8 !1/2 } } | |2 f p Pk f . Lx k=0 p Lx 3. REFINED STRICHARTZ ESTIMATE 24
¡ ¡ 1 Fix ε 0. Let p ε . By Sobolev embedding and Littlewood-Paley Theorem it follows that
} } 8 } ε } f L . J f p x Lx ¸8 !1/2 |JεP f |1/2 k k=0 p L x 8 1/2 ¸ | ε |2 = J Pk f = p/2 k 0 Lx ¸8 !1/2 ε 2 }J P f } p . . k Lx k=0 Therefore, to obtain (2.21) it enough to prove that for p ¡ 2 ¡ α + σ + α¡σ ¡ α ¡ 3σ + α¡σ 1 4 4 2p 1 4 4 2p }B u } 2 p D u + D F , k ¥ 1. x k L Lx . x k x k T 8 2 2 2 LT Lx LT Lx The estimate for the case k = 0 follows using Holder’s¨ inequality and (2.3). For such reason we fix k ¥ 1, and at these level of frequencies we have that
B ¡ α+1B tuk Dx xuk = Fk. Consider a partition of the interval [0, T] = jPJ Ij where Ij = [aj, bj], and T = bj for some j. kσ 1¡µ | | ¡kσ µ Indeed, we choose a quantity 2 T of intervals, with length Ij 2 T , where µ is a positive number to be fixed. Let q be such that 2 1 1 + = . q p 2 Using that u solves the integral equation » t 1 1 1 (2.22) u(t) = S(t)u0 + S(t ¡ t )F(t ) dt , 0 we deduce that }B } xuk 2 p LT Lx ¸ 1/2 }B }2 = xuk 2 p LI Lx jPJ j
1/2 1 ¡ 1 ¸ ¤ µ ¡kσ 2 q }B }2 T 2 xuk q p LI Lx jPJ j 1/2 » 2 1 ¡ 1 ¸ t µ ¡kσ 2 q ¡ B 2 ¡ B . T 2 S(t aj) xuk(aj) q p + S(t s) x Fk(s)ds . LI Lx P j aj q p j J L Lx Ij 4. PROOF OF THEOREMA 25
In this sense, it follows from (2.3) that $ , & » !2.1/2 1 ¡ 1 ¸ ¡ α 2 ¸ ¡ α }B } µ ¡kσ 2 q q B q B xuk 2 p T 2 Dx xuk(aj) + Dx x Fk(t) dt LT Lx . % - P L2 P Ij L2 $ j J x j J x &' 1/2 1 ¡ 1 ¸ ¡ α 2 µ ¡kσ 2 q 1 q . T 2 ' Dx uk % 8 2 jPJ LT Lx , » 1/2./ ¸ ¡ α 2 µ ¡kσ 1 q + T 2 Dx Fk(t) dt / 2 - jPJ Ij Lx 1 ¡ 1 1/2 ¡ α µ ¡kσ 2 q 1¡µ kσ 1 q . T 2 T 2 Dx uk 8 2 LT Lx » !1/2 1 ¡ 1 1/2 T ¡ α 2 µ ¡kσ 2 q µ ¡kσ 1 q + T 2 T 2 Dx Fk(t) dt 0 2 Lx ¡ α + σ ¡ α + σ ¡ 1/2¡µ/q 1 q q µ(1¡1/q) 1 q q σ . T Dx uk + T Dx Fk . 8 2 2 2 LT Lx LT Lx ¡ α σ ¡ α σ α¡σ ¡ α σ ¡ ¡ α ¡ 3σ α¡σ ¡ 1 Since, 1 q + q = 1 4 + 4 + 2p and 1 q + q σ = 1 4 4 + 2p . We recall that ε p , P P α¡σ ¡ α¡σ+2 ¡ 1 ¡ µ ¡ 1 σ [0, 1] and α (0, 1), then ε + 2p 2p 0. Next, we choose µ1 = 2 q , µ2 = µ(1 q ) with the particular choice µ = 1/2. Gathering the inequalities above follows the proposition. Now we turn our attention to the proof of TheoremA. Our starting point will be the energy es- timate (2.7), that as was remarked above, the key point is to establish a priori control of }Bxu} 1 8 . LT Lx
4. Proof of TheoremA
4.1. A priori estimates. First notice that by scaling, it is enough to deal with small initial data in the Hs¡norm. Indeed, if u(x, t) is a solution of (0.1) defined on a time interval [0, T], for some ¡ 1+α 2+α positive time T, then for all λ 0, uλ(x, t) = λ u(λx, λ t) is also solution with initial data 1+α 2+α u0,λ(x) = λ u0(λx), and time interval 0, T/λ . ¡ s For any δ 0, we define Bδ(0) as the ball with center at the origin in H (R) and radius δ. Since 1+2α 1 } } 2 } } } s } 2 +α+s } s } u 2 = λ u0 2 and D u 2 = λ D u0 2 , 0,λ Lx Lx x 0,λ Lx x Lx then 1 +α s } } s 2 ( + )} } s u0,λ Hx . λ 1 λ u0 Hx , ¤ so we can force uλ( , 0) to belong to the ball Bδ(0) by choosing the parameter λ with the condition " * 1+2α ¡ 1+2α 2 } } 2 λ min δ u0 s , 1 . Hx
Thus, the existence and uniqueness of a solution to (0.1) on the time interval [0, 1] for small initial
} } s data u0 Hx will ensure the existence and uniqueness of a solution to (0.1) for arbitrary large initial 4. PROOF OF THEOREMA 26 data on a time interval [0, T] with
" * ¡ 2(2+α) } } 1+2α T min 1, u0 s . Hx
Thus, without loss of generality we will assume that T ¤ 1, and that
} } } s } ¤ Λ := u0 L2 + Dxu0 L2 δ, where δ is a small positive number to be fixed later. We fix s such that
9 3α 3 α s(α) = ¡ s ¡ 8 8 2 2 and set ε = s ¡ s(α) ¡ 0. 1¡α ¡ ¡ B Next, taking σ = 2 0, F = u xu in (2.21) together with (2.7) yields
¡ α ¡ 3σ + }B } µ1 } s } 8 µ2 1 4 4 ε B xu L2 L8 . T J u L L2 + T J (u xu) T x T x L2 L2 T x ¡ ! s+ α 1 µ1 } } } s } 8 µ2 } B } 2 B T u0 2 + D u 2 + T u xu 2 2 + Dx (u xu) . Lx x LT Lx LT Lx L2 L2 T x (2.23) ¡ c}B u} 8 α 1 x L2 L s+ 2 T Λ + Λ e T x +}Bxu} 2 8 }u0} 2 + Dx (uBxu) . LT Lx Lx L2 L2 T x }B } α¡1 c xu 2 8 s+ L Lx 2 B .T Λ + Λ e T + Dx (u xu) . 2 2 LT Lx
Now, to analyze the product coming from the nonlinear term we use the Leibniz rule for fractional derivatives (1.3) joint with the energy estimate (2.7) as follows
¡ ¡ ¡ s+ α 1 s+ α 1 s+ α 1 D 2 (uB u) uD 2 B u + }B u(t)} 8 D 2 u(t) x x . x x x Lx x 2 2 2 2 2 LT Lx LT Lx Lx L2 T α¡1 α¡1 s+ 2 s+ 2 ¤ uD B u + }B u} 2 8 D u x x x L Lx x 2 2 T 8 2 (2.24) LT Lx LT Lx ¡ s+ α 1 } 2 B } }B } } } 8 . uDx xu L2 L2 + xu L2 L8 u L Hs T x T x T x α¡1 }B } s+ c xu 2 8 2 B L Lx . uDx xu + Λ e T . 2 2 LT Lx 4. PROOF OF THEOREMA 27
To handle the first term in the right hand side above, we incorporate Kato’s smoothing effect estimate obtained in (2.15) in the following way
» » !1/2 α¡1 T α¡1 s+ 2 s+ 2 2 uDx Bxu = |uDx Bxu| dx dt 2 2 R LT Lx 0 » 1/2 ¸8 T 2 s+ α+1 ¤ } }2 2 u(t) L8 Dx Hu(t) dt 0 [j,j+1) 2 j=¡8 L[j,j+1) 1/2 ¸8 2 s+ α+1 ¤ } }2 2 u L8 L8 Dx Hu T [j,j+1) 2 2 j=¡8 LT L[j,j+1) (2.25) 1/2 ¸8 s+ α+1 ¤ } }2 2 u 8 8 sup Dx Hu LT L[ + ) j,j 1 jPZ 2 2 j=¡8 LT L[j,j+1)
8 1/2 ¸ }B } c xu 2 8 } }2 }B } L Lx u 8 8 1 + Λ + xu 2 8 Λ e T . LT L[j,j+1) LT Lx j=¡8
¸8 1/2 c}B u} 8 2 x L2 L }u} 8 8 (1 + Λ) Λ e T x . . LT L[j,j+1) j=¡8
In summary, after gathering the estimates (2.23)-(2.25) yields
8 1/2 }B } ¸ }B } c xu 2 8 c xu 2 8 }B } L Lx } }2 L Lx (2.26) xu 2 8 Λ(1 + Λ) e T u 8 8 + Λ + Λ e T . LT Lx . LT L[j,j+1) j=¡8
Since u is a solution to (2.6), then by Duhamel’s formula it follows that
» t u(t) = S(t)u0 ¡ S(t ¡ s) (uBxu) (s) ds 0
α+1B where S(t) = etDx x . ¡ 1+α Now, we fix η 0 such that η 8 ; this choice implies that
1 α ¡ 1 η + s + . 2 2 4. PROOF OF THEOREMA 28
Hence, Sobolev’s embedding, Holder’s¨ inequality and Corollary 2.5 produce (2.27) 1/2 » 1/2 ¸8 ¸8 t 2 } }2 ¡ B u L8 L8 = S(t)u0 S(t¡s)(u xu) ds T [j,j+1) 0 8 8 j=¡8 j=¡8 LT L[j,j+1) 1/2 » 1/2 ¸8 ¸8 t 2 } }2 ¡ B . S(t)u0 L8 L8 + S(t s)(u xu)(s)ds T [j,j+1) 0 8 8 j=¡8 j=¡8 LT L[j,j+1)
(1 + T)Λ + (1 + T)}uB u} η+1/2 . x L1 H T x } B } η+1/2 B .T Λ + u xu L1 L2 + Dx (u xu) T x L1 L2 T x η+1/2 T Λ + Λ}Bxu} 2 8 + Dx (uBxu) . . LT Lx 2 2 LT Lx Employing an argument similar to the one applied in (2.24) and (2.26) it is possible to bound the last term in the right hand side as follows
¸8 1/2 }B } 8 }B } 8 η+1/2 c xu L2 L c xu L2 L B } }2 8 8 T x T x (2.28) Dx (u xu) . u L L Λ(Λ + 1) e +Λ e . L2 L2 T [j,j+1) T x j=¡8 Next, we define » 1/2 T ¸8 2 2 φ(T) = }Bxu(s)} 8 ds + }u} 8 8 , Lx LT L[j,j+1) 0 j=¡8 which is a continuous, non-decreasing function of T. From obtained in (2.26), (2.27) and (2.28) follows that
c φ(T) c φ(T) c φ(T) φ(T) . Λ(Λ + 1)φ(T) e +Λ e φ(T) + Λ e +Λ + Λφ(T).
Now, if we suppose that Λ ¤ δ ¤ 1 we obtain
φ(T) ¤ cΛ + cΛ ecφ(T) for some constant c ¡ 0. To complete the proof we will show that there exists δ ¡ 0 such that if Λ ¤ δ, then
φ(1) ¤ A, for some constant A ¡ 0.
To do this, we define the function
(2.29) Ψ(x, y) = x ¡ cy ¡ cy ecx .
First notice that Ψ(0, 0) = 0 and BxΨ(0, 0) = 1. Then the Implicit Function Theorem asserts that there exists δ ¡ 0, and a smooth function ξ(y) such that ξ(0) = 0, and Ψ(ξ(y), y) = 0 for |y| ¤ δ. Notice that the condition Ψ(ξ(y), y) = 0 implies that ξ(y) ¡ 0 for y ¡ 0. Moreover, since
BxΨ(0, 0) = 1, then the function Ψ(¤, y) is increasing close to ξ(y), whenever δ is chosen sufficiently small. 4. PROOF OF THEOREMA 29
Let us suppose that Λ ¤ δ, and set λ = ξ(Λ). Then, combining interpolation and Proposition 2.12 we obtain
¸8 !21/2 φ(0) = sup |u(x, 0)| j=¡8 xP[j,j+1) } } . u0 Hs(R) s ¤ c }u } 2 + c }D u } 2 1 0 L 1 x 0 Lx ¡ where we take c c1. Therefore, ¤ cξ(Λ) φ(0) c1Λ cΛ + cΛ e = λ. Suppose that φ(T) ¡ λ for some T P (0, 1) and define
T0 = inf tT P (0, 1) | φ(T) ¡ λu . ¡ t u Hence, T0 0 and φ(T0) = λ, besides, there exists a decreasing sequence Tn n¥1 converging to T0 such that φ(Tn) ¡ λ. In addition, notice that (2.29) implies Ψ(φ(T), Λ) ¤ 0 for all T P [0, 1]. Since the function Ψ(¤, Λ) is increasing near λ it implies that
Ψ(φ(Tn), Λ) ¡ Ψ(φ(T0), Λ) = Ψ(λ, Λ) = Ψ(ξ(Λ), Λ) = 0 for n sufficiently large. This is a contradiction with the fact that φ(T) ¡ λ. So we conclude φ(T) ¤ A for all T P (0, 1), as was claimed. Thus, φ(1) ¤ A. In conclusion we have proved that
» 1/2 T ¸8 }B }2 } }2 } } @ P (2.30) φ(T) = xu(s) 8 ds + u 8 8 u0 Hs , T [0, 1]. Lx LT L[j,j+1) . x 0 j=¡8
4.2. Uniqueness. This subsection is devoted to prove the uniqueness of solutions to the IVP (0.1).
Let u(t) and v(t) be two solutions to the equation in (0.1) with initial conditions u(0) = u0 and v(0) = v0, respectively. Set ! )
K = max }Bxu} 1 8 , }Bxv} 1 8 . LT Lx LT Lx We define w = u ¡ v. Then w satisfies the IVP # B w ¡ Dα+1B w + wB v + uB w = 0, x, t P R, 0 α 1, (2.31) t x x x x w|t=0 = u0 ¡ v0,
After multiply the equation in (2.31) by w and integrate in the x¡ variable we obtain d } }2 } }2 }B } 8 }B } 8 (2.32) w(t) 2 . w(t) 2 xu(t) L + xv(t) L dt Lx Lx x x 5. EXISTENCE OF SOLUTIONS 30 which holds for all t P [0, T]. Applying Gronwall’s inequality in (2.32) yields }B u} 8 +}B v} 8 x L1 L x L1 L } }2 } ¡ }2 T x T x w(t) 2 u0 v0 2 e (2.33) Lx . Lx ¤ c K } ¡ }2 e u0 v0 2 Lx which clearly proves the uniqueness.
2¡ REMARK 2.15. In fact, the estimate (2.33) besides granting the uniqueness in the Lx space, it also provides the L2¡Lipschitz bound of the flow.
5. Existence of Solutions
To establish the existence of local solutions of (0.1) we will follow the approach of Bona and Smith [4] (Section 4 ). So, we will recall some results about approximation of solutions. P Let ρ» S(R) such that (i) ρ(x) dx = 1, »R (ii) xkρ(x) dx = 0 for k P Z+ with 0 ¤ k ¤ tsu + 1. R ¡1 ¡1 Let ρe(x) = e ρ(e x) for any e ¡ 0.
s LEMMA 2.16. Let q ¥ 0, φ P H (R) and for e ¡ 0, φe = ρe ¦ φ. Then, } } ¡r} } ¥ (2.34) φe Hq+r . e φ Hq for all r 0 and } ¡ } q Ñ P (2.35) φ φe Hq¡p = o(e ) as e 0, for all p [0, q].
PROOF. See section 4 in [4].
s P (R) } } s We will consider u0 H . As was done previously, we can always assume that u0 Hx δ for some δ ¡ 0. Then, as was proved in section 4.1, it follows that for e ¡ 0, the solution ue is defined in the interval [0, 1] and satisfies the a priori estimate (2.30). s So, in this direction we fix u0 P H (R) with s ¡ 9/8 ¡ 3α/8, 0 α 1 and the regularized initial data, this is u0,e = ρe ¦ u0. 8 Since u0,e P H (R), we can warranty by means of Theorem 2.6, that for any e ¡ 0, there exists 8 a unique solution ue P C([0, 1]; H (R)) satisfying the condition ue(x, 0) = u0,e(x), x P R. The following lemmas summarize some properties associated to the smooth solutions.
LEMMA 2.17. Let e ¡ 0. Then,
} } s } } s (2.36) u0,e Hx . u0 Hx .
} s B } } } ¡s} } (2.37) D xue 2 8 u0,e 2s e u0 Hs . x LT Lx . Hx . x ¡ α 1 α¡1 s+ 2 ¡(s+ ) B } } α¡1 2 } } s (2.38) Dx xue . u0,e 2s+ . e u0 Hx . 8 2 2 Hx LT Lx 5. EXISTENCE OF SOLUTIONS 31
PROOF. By the sake of brevity we will omit the proof of these estimates. In fact, these one’s are obtained by similar arguments as the ones used in section 4.1 and the estimate (2.34). 1 These estimates are the basis to prove the convergence of the smooth solutions when e, e go to zero. Since u satisfies e # B ¡ α+1B B P tue Dx xue + ue xue = 0, x, t R, 0 α 1, ue|t=0 = u0,e. 1 Let ε ¡ ε ¡ 0. We set v = u ¡ u 1 . Then v satisfies the IVP # ε ε + B v ¡ D1 αB v + vB u + u 1 B v = 0 = 0, x, t P R, 0 α 1, (2.39) t x x x ε e x ¡ 1 v(x, 0) = u0,e(x) u0,e (x). s We will prove that the sequence tueue¡0 is a Cauchy sequence in the space C([0, 1] : H (R)). In this direction, we notice that arguing as in section 4.2 we obtain by Gronwall’s inequality, and (2.35) with q = s, p = 0
} } 8 ¤ cK } ¡ 1 } s (2.40) v L L2 e u0,e u0,e L2 = o(e ). T x xeÑ0 Moreover, combining (2.40) and Lemma 1.9 it follows that
r s¡r s s s¡r } } 8 r } } 8 } } 8 = ( ) ¤ v L Hx . v L Hs v 2 o e for all 0 r s. T T x LT Lx eÑ0 s To prove that tueue¡0 is a Cauchy sequence in C([0, 1] : H (R)), it remains to show the following: 9 ¡ 3α 3 ¡ α PROPOSITION 2.18. Assume that 0 α 1 and 8 8 s 2 2 . Let v be the solution of (2.39). } } ¤ Then, there exists a time T1 = T1 u0 Hs , with 0 T1 T 1 such that x ! )
}v} 8 s ¤ max }v} 8 s , }B v} 2 8 ÝÑ 0. LT Hx LT Hx x L L 1 1 T1 x eÑ0
PROOF. A standard energy estimate shows that » » 1 d 2 s s s s (2.41) }v} s + J (vB u ) J v dx + J (u 1 B v) J v dx = 0. Hx x e e x 2 dt R R The second term in the left side can be handled by using Lemma 1.2 » s s s J (vB u ) J v dx }v} 2 }B u } 8 + }D v} 2 }B u } 8 x e . Lx x e Lx x Lx x e Lx R s +}D B u } 8 }v} 2 }v} s . x x e Lx Lx Hx On the other hand, the estimate (1.4) applied in the third term in the left hand side of (2.41) yields » s s 2 J (u 1 B v)J v dx }B u 1 } 8 }v} s + }u 1 } s }B v} 8 }v} s . e x . x e Lx Hx e Hx x Lx Hx R Thus, after collecting the estimate above d }v} s }B u 1 } 8 + }B u } 8 }v} s dt Hx . x e Lx x e Lx Hx s + }v} 2 }D B u } 8 + }u 1 } s }B v} 8 Lx x x e Lx e Hx x Lx 5. EXISTENCE OF SOLUTIONS 32
Hence, by Gronwall’s inequality and (2.30) one gets that 1/2 }B } 8 +}B 1 } 8 T xue L2 L xue L2 L 1 } } 8 T x T x 1/2} } }B } v L Hs e T u0 Hs xv 2 8 + f (e, e ) T x . x LT Lx 1/2} } 1 T u0 Hs 1/2 e x T }u0}Hs }Bxv} 2 8 + f (e, e ) . x LT Lx where 1 1/2} } 8 } s B } } ¡ 1 } ÝÑ f (e, e ) = T v L L2 Dx xue L2 L8 + u0,e u0,e Hs 0. T x T x x eÑ0 To finish with the estimates it is necessary to establish control on the term involving the partial derivatives of v. So, as was done previously, the estimate (2.21) will be our starting point, in this 1¡α ¡ case with the particular choice of parameters σ = 2 , e = s s(α), and the function
¡ B ¡ 1 B F = v xue ue xv.
In this way α¡1 α¡1 }B } µ1 } } 8 µ2 s+ 2 B µ2 s+ 2 1 B xv 2 8 T v L Hs + T J (v xue) + T J (ue xv) (2.42) LT Lx . T x 2 2 2 2 LT Lx LT Lx P where µ1, µ2 (0, 1/2). 1¡α Since 0 s + 2 1, we apply inequality (1.3) to obtain ¡ α¡1 s+ α 1 s+ 2 B } B } 2 J (v xue) v xue 2 2 + Dx v 2 2 . LT Lx LT Lx L2 L2 (2.43) T x α¡1 s+ 2 }v} 8 s }B u } 2 8 + D B u }v} 8 2 , . LT Hx x e L Lx x x e L Lx T 2 8 T LT Lx and ¡ α¡1 s+ α 1 s+ 2 1 B } 1 B } 2 1 B J (ue xv) ue xv 2 2 + Dx (ue xv) 2 2 . LT Lx LT Lx L2 L2 T x α¡1 s+ 2 }B v} 2 8 }u 1 } 8 s + u 1 D B v , . x L Lx e LT Hx e x x T 2 2 LT Lx where the last term on the right hand side is handled as follows
8 1/2 ¡ ¸ 2 s+ α 1 s+ α+1 1 2 B 1 2 ue Dx xv = ue Dx Hv 2 2 2 2 LT Lx j=¡8 LT L[j,j+1) 1/2 ¸8 (2.44) s+ α+1 ¤ } 1 }2 2 ue 8 8 sup Dx Hv LT L[j,j+1) =¡8 jPZ L2 L2 j T [j,j+1) α+1 η s+ 2 ( + ) } } s H ¡ . 1 T u0 Hx sup Dx v for η 1/2, jPZ 2 2 LT L[j,j+1) being the last inequality a consequence of Duhamel’s formula and Corollary 2.5. Since v is a solution of the equation in (2.39), then we will compare the Kato smoothing effect satisfied by this solution with the last term on the right hand side of (2.44). 6. CONTINUITY OF THE FLOW 33
In this way, by using an approach similar to the one in Proposition 2.10 it follows that s+ α+1 1/2 2 }B } }B 1 } } } 8 sup Dx Hv 1 + T + xue 1 8 + xue 1 8 v L Hs . LT Lx LT Lx T x jPZ 2 2 (2.45) LT L[j,j+1)
} s B } } } 8 } 1 } 8 }B } + D xue 2 8 v 2 + ue L Hs xv 1 8 . x LT Lx LT Lx T x LT Lx Gathering the bounds obtained in (2.42) and (2.43)- (2.45) it follows that 1/2 }B } µ1 µ2 } } 1/2} } µ2 η} } } } 8 xv 2 8 T + T u0 Hs + 1 + T + T u0 Hs T (1 + T) u0 Hs v L Hs LT Lx . x x x T x 1 µ2 + T }u0}Hs }Bxv} 2 8 + h(e, e ), x LT Lx where ¡ 1 + α 1 µ2 s 2 µ2 s h(e, e ) = T D B u }v} 8 2 + T }D B u } 2 8 }v} 8 2 ÝÑ 0. x x e L Lx x x e L L L Lx 2 8 T T x T eÑ0 LT Lx in view of (2.37), (2.38), and (2.40). By gathering the estimates above the proposition follows. As was mentioned before the Proposition 2.18 is the basis to conclude that the sequence 8 t ¡ 1 u s ue ue e¡0 defines a Cauchy sequence in the space L ([0, T] : H (R)). In fact, by using a weak compactness argument it is possible to show that there exists a function u such that
} ¡ } 8 ÝÑ (2.46) u ue L Hs 0. T1 x In fact, from this we deduce that u is a solution of the IVP (0.1) in the distributional sense satisfying the condition (0.5).
6. Continuity of the Flow
To finish with the proof of TheoremA it remains to prove the continuous dependence of the solution upon the initial data. P s 9¡3α Let u0 H (R), with 8 s 2. By using a scaling argument, we can always assume } } s ¤ that u0 Hx δ, being δ a small positive number. In fact, in section5, it was proved that the corresponding solution u to (0.1) is defined in the interval [0, 1] and belongs to u P C([0, 1] : Hs(R)). Thus, the continuous dependence upon the initial data comes down to prove that given ν ¡ 0, s ¡ P (R) } ¡ } s P ([ ] there exists θ 0, such that for any v0 H with v0 u0 Hx ν, then the solution v C 0, 1 : s H (R)) of (0.1) satisfies }v ¡ u} 8 s θ. L1 Hx To achieve this goal we regularize the initial data u0 and v0, by defining v0,e = ρe ¦ v0 and u0,e = ρe ¦ u0, where ρe is the mollifier defined in section5. Next, we consider the solutions 8 ue, ve P C([0, 1] : H (R)) associated to the initial data u0,e and v0,e respectively. Then, it follows from the triangle inequality that
(2.47) }u ¡ v} 8 s ¤ }u ¡ u } 8 s + }u ¡ v } 8 s + }v ¡ v } 8 s . L1 Hx e L1 Hx e e L1 Hx e L1 Hx
Because, in view of (2.46) we can choose e0 ¡ 0 sufficiently small such that
(2.48) }u ¡ u } 8 s + }v ¡ v } 8 s 2θ/3. e0 L1 Hx e0 L1 Hx 7. PROPAGATION OF REGULARITY 34
Next, we proceed to estimate the second term on the right hand side of (2.47). For such purpose ¡ we consider the regularization of the elements ue0 and ve0 , specifically by defining for µ 0, the elements uµ = u ¦ ρ and vµ = v ¦ ρ . The function uµ , respectively vµ , is a solution of 0,e0 0,e0 µ 0,e0 0,e0 µ e0 e0 the equation in (0.1) with initial conditions u0,e0 , respectively v0,e0 . 2 Since u0,e0 and v0,e0 are elements in H (R) by virtue of (2.34) it follows that
s¡2 s¡2 }u } 2 e }u } s e δ 0,e0 Hx . 0 Hx . 0 and s¡2 s¡2 }v } 2 }u ¡ v } 2 + }u } 2 e ν + e δ. 0,e0 Hx . e0 e0 Hx e0 Hx . 0 0
Thus, by choosing δ(e0), ν(e0) ¡ 0 small enough, we obtain by the existence theory of solutions µ µ 8 t u ¡ t u ¡ s that ue0 µ 0 and ve0 µ 0 converge to ue0 and ve0 , respectively in L ([0, 1] : H (R)) as µ goes to zero. 8 s The convergence in L ([0, 1] : H (R)) implies that there exists µ0 ¡ 0 small enough, such that µ0 µ0 µ0 µ0 }u ¡ v } 8 s ¤ u ¡ u 8 + v ¡ v 8 + u ¡ v 8 e0 e0 L1 Hx e0 e0 L Hs e0 e0 L Hs e0 e0 L Hs (2.49) 1 x 1 x 1 x µ0 µ0 ¤ ¡ 8 θ/6 + ue0 ve0 s L1 Hx however, by (2.34) it follows that
}u0,e ¡ v0,e } 8 2 ν. 0 0 L1 Hx Thus, by applying Theorem 2.6, the continuity of the flow solution for initial data in H2(R) implies that for ν ¡ 0 small enough µ0 µ0 ¡ 8 (2.50) ue0 ve0 s θ/6. L1 Hx Finally, combining (2.48), (2.49)) and (2.50) it follows the continuous dependence upon the initial data. This finish the proof of TheoremA. l
7. Propagation of regularity
The aim of this section is to prove TheoremB. To achieve this goal it is necessary to take into account two important aspects of our analysis. First, the ambient space, that in our case is the Sobolev space where the theorem is valid together with the properties satisfied by the real solutions of the dispersive generalized Benjamin-Ono equation. In second place, the auxiliaries weights functions involved in the energy estimates that we will describe in detail. The following is a summary of the local well-posedness and Kato’s smoothing effect presented in the previous sections.
P s ¥ 3¡α P THEOREM D. If u0 H (R), s 2 , α (0, 1), then there exist a positive time T = T(}u0}Hs ) ¡ 0 and a unique solution of the IVP (0.1) such that (a) u P C([¡T, T] : Hs(R)), 1 8 (b) Bxu P L ([¡T, T] : L (R)), (Strichartz), 7. PROPAGATION OF REGULARITY 35
(c) Smoothing effect: for R ¡ 0, » » § § § § ! T R § α+1 §2 § α+1 §2 § r+ 2 § § r+ 2 § §BxDx u§ + §HBxDx u§ dx dt ¤ C ¡T ¡R