arXiv:1001.0048v2 [math.AP] 7 Jan 2010Nonlinear stability of periodic traveling wave solutions o f viscous conservation laws in dimensions one and two Mathew A. Johnson∗Kevin Zumbrun† November 12, 2018 Keywords : Periodic traveling waves; Bloch decomposition; modulate d waves. 2000 MR Subject Classification : 35B35. Abstract Extending results of Oh and Zumbrun in dimensions d≥3, we establish nonlin- ear stability and asymptotic behavior of spatially-periodic traveling- wave solutions of viscous systems of conservation laws in critical dimensions d= 1,2, under a natural set of spectral stability assumptions introduced by Schneider in th e setting of reaction diffusion equations. The key new steps in the analysis beyond that in d imensionsd≥3 are a refined Green function estimate separating off translation as the slowest decaying linear mode and a novel scheme for detecting cancellation at the leve l of the nonlinear iteration in the Duhamel representation of a modulated periodic wav e. 1 Introduction Nonclassical viscous conservation laws arising in multiph ase fluid and solid mechanics ex- hibit a rich variety of traveling wave phenomena, including homoclinic (pulse-type) and periodic solutions along with the standard heteroclinic (s hock, or front-type) solutions [GZ, Z6, OZ1, OZ2]. Here, we investigate stability of period ic traveling waves: specifi- cally, sufficient conditions for stability of the wave. Our ma in result, generalizing results of Oh and Zumbrun [OZ4] in dimensions d≥3, is to show that strong spectral stability in the sense of Schneider [S1, S2, S3] implies linearized and nonli nearL1∩HK→L∞bounded stability, for all dimensions d≥1, andasymptotic stability for dimensions d≥2. ∗Indiana University, Bloomington, IN 47405; matjohn@india na.edu: Research of M.J. was partially sup- ported by an NSF Postdoctoral Fellowship under NSF grant DMS -0902192. †Indiana University, Bloomington, IN 47405; kzumbrun@indi ana.edu: Research of K.Z. was partially supported under NSF grants no. DMS-0300487 and DMS-0801745 . 11 INTRODUCTION 2 More precisely, we show that small L1∩Hsperturbations of a planar periodic solution u(x,t)≡¯u(x1) (without loss of generality taken stationary) converge at Gaussian rate in Lp,p≥2 to a modulation (1.1) ¯ u(x1−ψ(x,t)), of the unperturbed wave, where x= (x1,˜x), ˜x= (x2,...,xd), andψis a scalar function whosex- andt-gradients likewise decay at least at Gaussian rate in all Lp,p≥2, but which itself decays more slowly by a factor t1/2; in particular, ψis merely bounded in L∞for dimensiond= 1. The one-dimensional study of spectral stability of spatial ly periodic traveling waves of systems of viscous conservation laws was initiated by Oh and Zumbrun [OZ1] in the “quasi- Hamiltonian” case that the traveling-wave equation posses ses an integral of motion, and in the general case by Serre[Se1]. An important contribution o f Serre was to point out a larger connection between the linearized dispersion relation (th e functionλ(ξ) relating spectra to wave number of the linearized operator about the wave) near z ero and the formal Whitham averaged system obtained by slow modulation, or WKB, approx imation. In [OZ3], this was extended to multi-dimensions, relating t he linearized dispersion rela- tion near zero to (1.2)∂tM+/summationdisplay j∂xjFj= 0, ∂t(ΩN)+∇x(ΩS) = 0, whereM∈Rndenotes the average over one period, Fjthe average of an associated flux, Ω =|∇xΨ| ∈R1the frequency, S=−Ψt/|∇xΨ| ∈R1the speeds, andN=∇xΨ/|∇xΨ| ∈ Rdthe normal νassociated with nearby periodic waves, with an additional c onstraint (1.3) curl (Ω N) = curl ∇xΨ≡0. As an immediate corollary, similarly as in [OZ1], [Se1] in th e one-dimensional case, this yieldedas anecessary condition formulti-dimensional sta bility hyperbolicityoftheaveraged system (1.2)–(1.3). The present study is informed by but does not directly rely on this observation relating Whitham averaging and spectral stability properties. Like wise, the Evans function tech- niquesusedin[Se1,OZ3]toestablishthisconnection play n oroleinouranalysis; indeed, the Evans function makes no appearance here. Rather, we rely on a direct Bloch-decomposition argument in the spirit of Schneider [S1, S2, S3], combining s harp linearized estimates with subtle cancellation in nonlinear source terms arising from the modulated wave approxima- tion. The analytical techniques used to realize this progra m are somewhat different from those of [S1, S2, S3], however, coming instead from the theor y of stability of viscous shock fronts through a line of investigation carried out in [OZ1, O Z2, OZ3, OZ4, HoZ]. In partic- ular, the nonsmooth dispersion relation at ξ= 0 typical for convection-diffusion equations1 INTRODUCTION 3 requires different treatment from that of [S1, S2, S3] in the re action diffusion case; see Re- mark 2.4. Moreover, we detect nonlinear cancellation in the physicalx-tdomain rather than the frequency domain as in [S1, S2, S3]. The main difference bet ween the present analysis and that of [OZ4] is the systematic incorporation of modulat ion approximation (1.1). 1.1 Equations and assumptions Consider a parabolic system of conservation laws (1.4) ut+/summationdisplay jfj(u)xj= ∆xu, u∈ U(open)∈Rn,fj∈Rn,x∈Rd,d≥1,t∈R+, and a periodic traveling wave solution (1.5) u= ¯u(x·ν−st), of periodX, satisfying the traveling-wave ODE ¯ u′′= (/summationtext jνjfj(¯u))′−s¯u′with boundary conditions ¯u(0) = ¯u(X) =:u0.Integrating, we obtain a first-order profile equation (1.6) ¯ u′=/summationdisplay jνjfj(¯u)−s¯u−q, where (u0,q,s,ν,X )≡constant. Without loss of generality take ν=e1,s= 0, so that ¯u= ¯u(x1) represents a stationary solution depending only on x1. Following [Se1, OZ3, OZ4], we assume: (H1)fj∈CK+1,K≥[d/2]+4. (H2) Themap H:R×U×R×Sd−1×Rn→Rntaking (X;a,s,ν,q)/mapsto→u(X;a,s,ν,q)−a is a submersion at point ( ¯X;¯u(0),0,e1,¯q), whereu(·;·) is the solution operator for (1.6). Conditions (H1)–(H2) imply that the set of periodic solutio ns in the vicinity of ¯ uform a smooth (n+d+1)-dimensional manifold {¯ua(x·ν(a)−α−s(a)t)}, withα∈R,a∈Rn+d. 1.1.1 Linearized equations Linearizing (1.4) about ¯ u(·), we obtain (1.7) vt=Lv:= ∆xv−/summationdisplay (Ajv)xj, where coefficients Aj:=Dfj(¯u) are now periodic functions of x1. Taking the Fourier transform in the transverse coordinate ˜ x= (x2,···,xd), we obtain (1.8)ˆvt=L˜ξˆv= ˆvx1,x1−(A1ˆv)x1−i/summationdisplay j/negationslash=1Ajξjˆv−/summationdisplay j/negationslash=1ξ2 jˆv, where˜ξ= (ξ2,···,ξd) is the transverse frequency vector.1 INTRODUCTION 4 1.1.2 Bloch–Fourier decomposition and stability conditions Following [G, S1, S2, S3], we define the family of operators (1.9) Lξ=e−iξ1x1L˜ξeiξ1x1 operating on the class of L2periodic functions on [0 ,X]; the (L2) spectrum of L˜ξis equal to the union of the spectra of all Lξwithξ1real with associated eigenfunctions (1.10) w(x1,˜ξ,λ) :=eiξ1x1q(x1,ξ1,˜ξ,λ), whereq, periodic, is an eigenfunction of Lξ. By continuity of spectrum, and discreteness of the spectrum of the elliptic operators Lξon the compact domain [0 ,X], we have that the spectra ofLξmay be described as the union of countably many continuous su rfacesλj(ξ). Without loss of generality taking X= 1, recall now the Bloch–Fourier representation (1.11) u(x) =/parenleftBig1 2π/parenrightBigd/integraldisplayπ −π/integraldisplay Rd−1eiξ·xˆu(ξ,x1)dξ1d˜ξ of anL2functionu, where ˆu(ξ,x1) :=/summationtext ke2πikx1ˆu(ξ1+ 2πk,˜ξ) are periodic functions of periodX= 1, ˆu(˜ξ) denoting with slight abuse of notation the Fourier transfo rm ofuin the full variable x. By Parseval’s identity, the Bloch–Fourier transform u(x)→ˆu(ξ,x1) is an isometry in L2: (1.12) /ba∇dblu/ba∇dblL2(x)=/ba∇dblˆu/ba∇dblL2(ξ;L2(x1)), whereL2(x1) is taken on [0 ,1] andL2(ξ) on [−π,π]×Rd−1. Moreover, it diagonalizes the periodic-coefficient operator L, yielding the inverse Bloch–Fourier transform representation (1.13) eLtu0=/parenleftBig1 2π/parenrightBigd/integraldisplayπ −π/integraldisplay Rd−1eiξ·xeLξtˆu0(ξ,x1)dξ1d˜ξ relating behavior of the linearized system to that of the dia gonal operators Lξ. Following [OZ4], weassumealongwith(H1)–(H2) the strong spectral stability conditions: (D1)σ(Lξ)⊂ {Reλ<0}forξ/ne}ationslash= 0. (D2) Reσ(Lξ)≤ −θ|ξ|2,θ>0, forξ∈Rdand|ξ|sufficiently small. (D3)λ= 0 is a semisimple eigenvalue of L0of multiplicity exactly n+1.1 For each fixed angle ˆξ:=ξ/|ξ|, expandLξ=L0+|ξ|L1+|ξ|2L2. By assumption (D3) and standard spectral perturbation theory, there exist n+1 smooth eigenvalues (1.14) λj(ξ) =−iaj(ξ)+o(|ξ|) 1The zero eigenspace of L0is at least ( n+1)-dimensional by the linearized existence theory and (H2 ), and hence n+ 1 is the minimal multiplicity; see [Se1, OZ3]. As noted in [O Z1, OZ3], minimal dimension of this zero eigenspace implies that ( M,NΩ) of (1.2) gives a nonsingular coordinatization of the fami ly of periodic traveling-wave solutions near ¯ u.1 INTRODUCTION 5 ofLξbifurcating from λ= 0 atξ= 0, where −iajare homogeneous degree one functions given by |ξ|times the eigenvalues of Π 0L1|KerL0, with Π 0the zero eigenprojection of L0. Conditions(D1)–(D3) areexactly thespectralassumptions of[S1,S2,S3], corresponding to “dissipativity” of the large-time behavior of the linear ized system. As in [OZ4], we make the further nondegeneracy hypothesis: (H3) The eigenvalues λ=−iaj(ξ)/|ξ|of Π0L1 KerL0are simple. The functions ajmay be seen to be the characteristics associated with the Whi tham av- eraged system (1.2)–(1.3) linearized about the values of M,S,N, Ω associated with the background wave ¯ u; see [OZ3, OZ4]. Thus, (D1) implies weak hyperbolicity of (1 .2)–(1.3) (reality ofaj), while (H1) corresponds to strict hyperbolicity. 1.2 Main results With these preliminaries, we can now state our main results. Theorem 1.1. Assuming (H1)–(H3) and (D1)–(D3), for some C >0andψ∈WK,∞(x,t), (1.15)|˜u−¯u(·−ψ)|Lp(t)≤C(1+t)−d 2(1−1/p)|˜u−¯u|L1∩HK|t=0, |˜u−¯u(·−ψ)|HK(t)≤C(1+t)−d 4|˜u−¯u|L1∩HK|t=0, |(ψt,ψx)|WK+1,p≤C(1+t)−d 2(1−1/p)|˜u−¯u|L1∩HK|t=0, and (1.16) |˜u−¯u|Lp(t),|ψ(t)|Lp≤C(1+t)−d 2(1−1 p)+1 2|˜u−¯u|L1∩HK|t=0 for allt≥0,p≥2,d= 1, for solutions ˜uof(1.4)with|˜u−¯u|L1∩HK|t=0sufficiently small. In particular, ¯uis nonlinearly bounded L1∩HK→L∞stable for dimension d= 1. Theorem 1.2. Assuming (H1)–(H3) and (D1)–(D3), for any ε >0, someC >0and ψ∈WK,∞(x,t), (1.17)|˜u−¯u(·−ψ)|Lp(t)≤C(1+t)−d 2(1−1/p)|˜u−¯u|L1∩HK|t=0, |˜u−¯u(·−ψ)|HK(t)≤C(1+t)−d 4|˜u−¯u|L1∩HK|t=0, |(ψt,ψx)|WK+1,p≤C(1+t)−d 2(1−1/p)+ε−1 2|˜u−¯u|L1∩HK|t=0, and (1.18)|˜u−¯u|Lp(t),|ψ(t)|Lp≤C(1+t)−d 2(1−1 p)+ε|˜u−¯u|L1∩HK|t=0, |˜u−¯u|HK(t),|ψ(t)|HK≤C(1+t)−d 4+ε|˜u−¯u|L1∩HK|t=0, for allt≥0,p≥2,d= 2, for solutions ˜uof(1.4)with|˜u−¯u|L1∩HK|t=0sufficiently small. In particular, ¯uis nonlinearly asymptotically L1∩HK→HKstable for dimension d= 2.1 INTRODUCTION 6 Remark 1.1. In Theorem 1.2, derivatives in x∈R2refer to total derivatives. Moreover, unless specified by an appropriate index, throughout this pa per derivatives in spatial variable xwill always refer to the total derivative of the function. In dimension one, Theorem 1.1 asserts only bounded L1∩HK→L∞stability, a very weak notion of stability. The absence of decay in perturbati on ˜u−¯uindicates the delicacy of the nonlinear analysis in this case. In particular, it is c rucial to separate off the slower- decaying modulated behavior (1.1) in order to close the nonl inear iteration argument. Remark 1.2. In dimension d= 1, it is straightforward to show that the results of Theorem 1.1 extend to all 1 ≤p≤ ∞using the pointwise techniques of [OZ2]; see Remark 3.3. Remark 1.3. The slow decay of |˜u−¯u|Lp(t)∼ |ψ(t)|Lpin (1.16) is due to nonlinear interactions; as shown in [OZ2, OZ4], the linearized decay r ate is faster by factor (1+ t)−1/2 (Proposition 2.1). In [OZ4], it was shown that for d≥3, where linear effects dominate behavior, (1.16) may be replaced by the stronger estimate |˜u−¯u|Lp(t),|ψ(t)|Lp≤C(1+ t)−d 2(1−1 p)|˜u−¯u|L1∩HK|t=0.These distinctions reflect fine details of both linearized es timates (Section 3) and nonlinear structure (Sections 4.1–4.2) tha t are not immediately apparent from the formal Whitham approximation (1.2)–(1.3). 1.3 Discussion and open problems Linearized stability under the same assumptions, with shar p rates of decay, was established ford= 1 [OZ2] and for d≥1 in [OZ4], along with nonlinear stability for d≥3. Theorem 1.1 completes this line of investigation by establishing no nlinear stability in the critical dimensions d= 1,2, a fundamental open problem cited in [OZ1, OZ4]. This gives a generalization of the work of [S1, S2, S3] for rea ction diffusion equations to the case of viscous conservation laws. Recall that the ana lysis of [S1, S2, S3] concerns also multiply periodic waves, i.e., waves that are either pe riodic or else constant in each coordinate direction. It is straightforward to verify that the methods of this paper apply essentially unchanged to this case, to give a corresponding stability result under the analog of (H1)–(H3), (D1)–(D3), as we intend to report further in a f uture work. Likewise, the extension from the semilinear parabolic case treated here t o the general quasilinear case is straightforward, following the treatment of [OZ4]. On the other hand, as noted in [OZ2], condition (D3) is in the c onservation law setting nongeneric, corresponding to the special “quasi-Hamilton ian” situation studied there; in particular, it implies that speed is to first order constant a mong the family of spatially pe- riodic traveling-wave solutions nearby ¯ u. In the generic case that (D3) is violated, behavior is essentially different [OZ1, OZ2], and perturbations decay more slowly at the linearized level. Nonlinear stability remains an interesting open pro blem in this setting. Our approach to stability in the critical dimensions d= 1,2, as suggested in [OZ4], is, loosely following the approach of [S1, S2, S3], to subtract o ut a slower-decaying part of the solution describedby anappropriatemodulation equation a ndshowthat theresidualdecays2 BASIC LINEARIZED STABILITY ESTIMATES 7 sufficiently rapidly to close a nonlinear iteration. It is wor th noting that the modulated approximation ¯ u(x1−ψ(x,t)) of (1.1) is not the full Ansatz (1.19) ¯ ua(Ψ(x,t)), Ψ(x,t) :=x1−ψ(x,t), associated with the Whitham averaged system (1.2)–(1.3) , where ¯ua isthemanifoldofperiodicsolutions near ¯ uintroducedbelow(H2), butonlythetranslational part not involving perturbations ain the profile. (See [OZ3] for the derivation of (1.2)– (1.3) and (1.19).) That is, we don’t need to separate out all v ariations along the manifold of periodic solutions, but only the special variations conn ected with translation invariance. The technical reason is an asymmetry in y-derivative estimates in the parts of the Green function associated with these various modes, something th at is not apparent without a detailed study of linearized behavior as carried out here. T his also makes sense formally, if one considers that (1.2) indicates that variables a,∇xΨ are roughly comparable, which would suggest, by the diffusive behavior Ψ >>∇xΨ, thatais neglible with respect to Ψ. However, note that in the case that (D3) holds, hence wave spe ed is stationary along the manifold of periodic solutions, the final equation of (1.2) d ecouples to (Ψ x)t= (ΩN)t= 0, and could be written as Ψ t= 0 in terms of Ψ alone. Hence, there is some ambiguity in this degenerate case which of Ψ, Ψ xis the primary variable, and in terms of linear behavior, the decay of variations aand Ψ are in fact comparable [OZ4]; in the generic case, aand Ψxare comparable at the linearized level [OZ2]. It would be very in teresting to better understand the connection between the Whitham averaged system (or suit able higher-order correction) and behavior at the nonlinear level, as explored at the linea r level in [OZ3, OZ4, JZ1, JZB]. 2 Basic linearized stability estimates We begin by recalling the basic linearized stability estima tes derived in [OZ4]. We will sharpen these afterward in Section 3. By standard spectral p erturbation theory [K], the total eigenprojection P(ξ) onto the eigenspace of Lξassociated with the eigenvalues λj(ξ), j= 1,...,n+1describedintheintroductioniswell-definedandanalyti cinξforξsufficiently small, since these (by discreteness of the spectra of Lξ) are separated at ξ= 0 from the rest of the spectrum of L0. Introducing a smooth cutoff function φ(ξ) that is identically one for|ξ| ≤εand identically zero for |ξ| ≥2ε,ε >0 sufficiently small, we split the solution operatorS(t) :=eLtinto low- and high-frequency parts (2.1) SI(t)u0:=/parenleftBig1 2π/parenrightBigd/integraldisplayπ −π/integraldisplay Rd−1eiξ·xφ(ξ)P(ξ)eLξtˆu0(ξ,x1)dξ1d˜ξ and (2.2) SII(t)u0:=/parenleftBig1 2π/parenrightBigd/integraldisplayπ −π/integraldisplay Rd−1eiξ·x/parenleftbig I−φP(ξ)/parenrightbig eLξtˆu0(ξ,x1)dξ1d˜ξ.2 BASIC LINEARIZED STABILITY ESTIMATES 8 2.1 High-frequency bounds By standard sectorial bounds [He, Pa] and spectral separati on ofλj(ξ) from the remaining spectra ofLξ, we have trivially the exponential decay bounds (2.3)/ba∇dbleLξt(I−φP(ξ))f/ba∇dblL2([0,X])≤Ce−θt/ba∇dblf/ba∇dblL2([0,X]), /ba∇dbleLξt(I−φP(ξ))∂l x1f/ba∇dblL2([0,X])≤Ct−l 2e−θt/ba∇dblf/ba∇dblL2([0,X]), /ba∇dbl∂l x1eLξt(I−φP(ξ))f/ba∇dblL2([0,X])≤Ct−l 2e−θt/ba∇dblf/ba∇dblL2([0,X]), forθ,C >0, and 0 ≤m≤K(Kas in (H1)). Together with (1.12), these give immediately the following estimates. Proposition 2.1 ([OZ4]).Under assumptions (H1)–(H3) and (D1)–(D2), for some θ, C >0, and allt>0,2≤p≤ ∞,0≤l≤K+1,0≤m≤K, (2.4)/ba∇dbl∂l xSII(t)f/ba∇dblL2(x),/ba∇dblSII(t)∂l xf/ba∇dblL2(x)≤Ct−l 2e−θt/ba∇dblf/ba∇dblL2(x), /ba∇dbl∂m xSII(t)f/ba∇dblLp(x),/ba∇dblSII(t)∂m xf/ba∇dblLp(x)≤Ct−d 2(1 2−1 p)−m 2e−θt/ba∇dblf/ba∇dblL2(x), where, again, derivatives in the variable x∈Rdrefer to total derivatives. Proof.The first inequalities follow immediately by (1.12) and (2.3 ). The second follows for x1derivatives in the case p=∞,m= 0 by Sobolev embedding from /ba∇dblSII(t)f/ba∇dblL∞(˜x;L2(x1))≤Ct−d−1 4e−θt/ba∇dblf/ba∇dblL2([0,X]) and /ba∇dbl∂x1SII(t)f/ba∇dblL∞(˜x;L2(x1))≤Ct−d−1 4−1 2e−θt/ba∇dblf/ba∇dblL2([0,X]), which follow by an application of (1.12) in the x1variable and the Hausdorff–Young in- equality /ba∇dblf/ba∇dblL∞(˜x)≤ /ba∇dblˆf/ba∇dblL1(˜ξ)in the variable ˜ x. The result for derivatives in x1and general 2≤p≤ ∞then follows by Lpinterpolation. Finally, the result for derivatives in ˜ xfollows from the inverse Fourier transform, equation (2.2), and the large|ξ|bound |eLtf|L2(x1)≤e−θ|˜ξ|2t|f|L2(x1),|ξ|sufficiently large , which easily follows from Parseval and the fact that Lξis a relatively compact perturbation of∂2 x−|ξ|2. Thus, by the above estimate we have /ba∇dbleLt∂˜xf/ba∇dblL2(x)≤C/ba∇dbleLξt|˜ξ|ˆf/ba∇dblL2(x1,ξ) ≤Csup/parenleftBig e−θ|˜ξ|2t|ξ|/parenrightBig /ba∇dblˆf/ba∇dblL2(x1,ξ) ≤Ct−1/2/ba∇dblf/ba∇dblL2(x). A similar argument applies for 1 ≤m≤K.2 BASIC LINEARIZED STABILITY ESTIMATES 9 2.2 Low-frequency bounds Denote by (2.5) GI(x,t;y) :=SI(t)δy(x) the Green kernel associated with SI, and (2.6) [ GI ξ(x1,t;y1)] :=φ(ξ)P(ξ)eLξt[δy1(x1)] the corresponding kernel appearing within the Bloch–Fouri er representation of GI, where the brackets on [ Gξ] and [δy] denote the periodic extensions of these functions onto the whole line. Then, we have the following descriptions of GI, [GI ξ], deriving from the spectral expansion (1.14) of Lξnearξ= 0. Proposition 2.2 ([OZ4]).Under assumptions (H1)–(H3) and (D1)–(D3), (2.7)[GI ξ(x1,t;y1)] =φ(ξ)n+1/summationdisplay j=1eλj(ξ)tqj(ξ,x1)˜qj(ξ,y1)∗, GI(x,t;y) =/parenleftBig1 2π/parenrightBigd/integraldisplay Rdeiξ·(x−y)[GI ξ(x1,t;y1)]dξ =/parenleftBig1 2π/parenrightBigd/integraldisplay Rdeiξ·(x−y)φ(ξ)n+1/summationdisplay j=1eλj(ξ)tqj(ξ,x1)˜qj(ξ,y1)∗dξ, where∗denotes matrix adjoint, or complex conjugate transpose, qj(ξ,·)and˜qj(ξ,·)are right and left eigenfunctions of Lξassociated with eigenvalues λj(ξ)defined in (1.14), normalized so that/an}b∇acketle{t˜qj,qj/an}b∇acket∇i}ht ≡1, whereλj/|ξ|is a smooth function of |ξ|andˆξ:=ξ/|ξ|andqjand˜qj are smooth functions of |ξ|,ˆξ:=ξ/|ξ|, andx1ory1, withℜλj(ξ)≤ −θ|ξ|2. Proof.Smooth dependence of λjand ofq, ˜qas functions in L2[0,X] follow from standard spectral perturbation theory [K] using the fact that λjsplit to first order in |ξ|asξis varied along rays through the origin, and that Lξvaries smoothly with angle ˆξ. Smoothness of qj, ˜qjinx1,y1then follow from the fact that they satisfy the eigenvalue eq uation forLξ, which has smooth, periodic coefficients. Likewise, (2.7)(i) is immediate from the spectral decomposition of elliptic operators on finite domains. Subs tituting (2.5) into (2.1) and computing (2.8) /hatwideδy(ξ,x1) =/summationdisplay ke2πikx1/hatwideδy(ξ+2πke1) =/summationdisplay ke2πikx1e−iξ·y−2πiky1=e−iξ·y[δy1(x1)], where the second and third equalities follow from the fact th at the Fourier transform either continuous or discrete of the delta-function is unity, we ob tain GI(x,t;y) =/parenleftBig1 2π/parenrightBigd/integraldisplayπ −π/integraldisplay Rd−1eiξ·xφP(ξ)eLξt/hatwideδy(ξ,x1)dξ =/parenleftBig1 2π/parenrightBigd/integraldisplayπ −π/integraldisplay Rd−1eiξ·(x−y)φP(ξ)eLξt[δy1(x1)]dξ,2 BASIC LINEARIZED STABILITY ESTIMATES 10 yielding (2.7)(ii) by (2.6)(i) and the fact that φis supported on [ −π,π]. Proposition 2.3 ([OZ4]).Under assumptions (H1)-(H3) and (D1)-(D3), (2.9) sup y/ba∇dblGI(·,t,;y)/ba∇dblLp(x),sup y/ba∇dbl∂x,yGI(·,t,;y)/ba∇dblLp(x)≤C(1+t)−d 2(1−1 p) for all2≤p≤ ∞,t≥0, whereC >0is independent of p. Proof.From representation (2.7)(ii) and ℜλj(ξ)≤ −θ|ξ|2, we obtain by the triangle in- equality (2.10) /ba∇dblGI/ba∇dblL∞(x,y)≤C/ba∇dble−θ|ξ|2tφ(ξ)/ba∇dblL1(ξ)≤C(1+t)−d 2, verifying the bounds for p=∞. Derivative bounds follow similarly, since derivatives fa lling onqjor ˜qjare harmless, whereas derivatives falling on eiξ·(x−y)bring down a factor of ξ, again harmless because of the cutoff function φ. To obtain bounds for p= 2, we note that (2.7)(ii) may be viewed itself as a Bloch– Fourier decomposition with respect to variable z:=x−y, withyappearing as a parameter. Recalling (1.12), we may thus estimate (2.11)sup y/ba∇dblGI(x,t;y)/ba∇dblL2(x)=/summationdisplay jsup y/ba∇dblφ(ξ)eλj(ξ)tqj(·,z1)˜q∗ j(·,y1)/ba∇dblL2(ξ;L2(z1∈[0,X])) ≤C/summationdisplay jsup y/ba∇dblφ(ξ)e−θ|ξ|2t/ba∇dblL2(ξ)/ba∇dblqj/ba∇dblL2(0,X)/ba∇dbl˜qj/ba∇dblL∞(0,X) ≤C(1+t)−d 4, where we have used in a crucial way the boundedness of ˜ qj; derivative bounds follow simi- larly. Finally, bounds for 2 ≤p≤ ∞follow byLp-interpolation. Remark 2.4. In obtaining the key L2-estimate, we have used in an essential way the periodic structure of qj, ˜qj. For, viewing GIas a general pseudodifferential expression rather than a Bloch–Fourier decomposition, we find that the s moothness of qj, ˜qjis not sufficient to apply standard L2→L2bounds of H¨ ormander, which require blowup in ξ derivatives at less than the critical rate |ξ|−1found here; see, e.g., [H] for further discussion. Nor do the weighted energy estimate techniques used in [S1, S 2, S3] apply here, as these also rely on the property of smoothness of λj,qj, ˜qjwith respect to ξat the origin ξ= 0. The lack of smoothness of the linearized dispersion relation at the origin is an essential technical difference separating the conservation law from the reaction diffusion case; see [OZ4] for further discussion. Remark 2.5. Underlying the above analysis, and also the technically rat her different approach of [OZ2], is the fundamental relation (2.12) G(x,t;y) =/parenleftBig1 2π/parenrightBigd/integraldisplayπ −π/integraldisplay Rd−1eiξ·(x−y)[Gξ(x1,t;y1)]dξ2 BASIC LINEARIZED STABILITY ESTIMATES 11 which, provided σ(Lξ) is semisimple, yields the simple formula G(x,t;y) =/parenleftBig1 2π/parenrightBigd/integraldisplayπ −π/integraldisplay Rd−1eiξ·(x−y)/summationdisplay jeλj(ξ)tqj(ξ,x1)˜qj(ξ,y1)∗dξ resembling that of the constant-coefficient case, where λjruns through the spectrum of Lξ. The basic idea in both cases is to separate off the principal pa rt of the series involving small λj(ξ) and estimate the remainder as a faster-decaying residual. Corollary 2.6 ([OZ4]).Under assumptions (H1)–(H3) and (D1)–(D3), for all p≥2,t≥0, (2.13) /ba∇dblSI(t)f/ba∇dblLp,/ba∇dbl∂xSI(t)f/ba∇dblLp,/ba∇dblSI(t)∂xf/ba∇dblLp≤C(1+t)−d 2(1−1 p)/ba∇dblf/ba∇dblL1. Proof.Immediate, from (2.9) and the triangle inequality, as, for e xample, /ba∇dblSI(t)f(·)/ba∇dblLp=/vextenddouble/vextenddouble/vextenddouble/integraldisplay RdGI(x,t;y)f(y)dy/vextenddouble/vextenddouble/vextenddouble Lp(x)≤/integraldisplay Rdsup y/ba∇dblGI(·,t;y)/ba∇dblLp|f(y)|dy. Proposition 2.1 ([OZ4]).Assuming (H1)-(H3), (D1)-(D3), for some C >0, allt≥0, p≥2,0≤l≤K, (2.14) /ba∇dblS(t)∂l xu0/ba∇dblLp≤Ct−l 2(1+t)−d 2(1 2−1 p)+l 2t−d 4−l 2/ba∇dblu0/ba∇dblL1∩L2. Proof.Immediate, from (2.4) and (2.13). 2.3 Additional estimates Lemma 2.7. Assuming (H1)–(H3), (D1)–(D3), for all t≥0,0≤l≤K, (2.15) /ba∇dbl∂l xSI(t)f/ba∇dblLp(x),/ba∇dblSI(t)∂l xf/ba∇dblLp(x)≤C(1+t)−d 2(1/2−1/p)/ba∇dblf/ba∇dblL2(x). Proof.From boundedness of the spectral projections Pj(ξ) =qj/an}b∇acketle{t˜qj,·/an}b∇acket∇i}htinL2[0,X] and their derivatives, another consequence of first-order splitting of eigenvalues λj(ξ) at the origin, we obtain boundedness of φ(ξ)P(ξ)eLξtand thus, by (1.12), the global bounds (2.16) /ba∇dbl∂l xSI(t)f/ba∇dblL2(x),/ba∇dblSI(t)∂l xf/ba∇dblL2(x)≤C/ba∇dblf/ba∇dblL2(x), for allt≥0, yielding the result for p= 2. Moreover, by boundedness of ˜ q,qin allLp(x1), we have |φ(ξ)P(ξ)eLξtˆf(ξ,·)|L∞(x1)≤Ce−θ|ξ|2t|P(ξ)ˆf(ξ,·)|L∞(x1)≤Ce−θ|ξ|2t|ˆf(ξ,·)|L2(x1),3 REFINED LINEARIZED ESTIMATES 12 C, θ>0, yielding by SIf=/parenleftBig 1 2π/parenrightBigd/integraltextπ −π/integraltext Rd−1eiξ·xφ(ξ)P(ξ)eLξtˆf(ξ,x1)dξ1d˜ξthe bound (2.17)/ba∇dblSI(t)f/ba∇dblL∞(x)≤/parenleftBig1 2π/parenrightBigd/integraldisplayπ −π/integraldisplay Rd−1|φ(ξ)P(ξ)eLξtˆf(ξ,·)|L∞(x1)dξ1d˜ξ ≤/parenleftBig1 2π/parenrightBigd/integraldisplayπ −π/integraldisplay Rd−1Cφ(ξ)e−θ|ξ|2t|ˆf(ξ,·)|L2(x1)dξ1d˜ξ ≤C|φ(ξ)e−θ|ξ|2t|L2(ξ)|ˆf|L2(ξ,x1) =C(1+t)−d 4/ba∇dblf/ba∇dblL2([0,X]), yielding the result for p=∞,l= 0. The result for p=∞, 1≤l≤Kfollows by a similar argument. The result for general 2 ≤p≤ ∞then follows by Lpinterpolation between p= 2 andp=∞. By Riesz–Thorin interpolation between (2.15) and (2.13), w e obtain the following, ap- parently sharp bounds between various LqandLp.2 Corollary 2.8. Assuming (H0)–(H3) and (D1)–(D3), for all 1≤q≤2≤p,t≥0, 0≤l≤K, (2.18) /ba∇dbl∂l xSI(t)f/ba∇dblLp,/ba∇dblSI(t)∂l xf/ba∇dblLp≤C(1+t)−d 2(1 q−1 p)/ba∇dblf/ba∇dblLq. Proposition 2.2. Assuming (H1)-(H3), (D1)-(D3), for some C >0, allt≥0,1≤q≤ 2≤p, and0≤l≤K, (2.19) /ba∇dblS(t)∂l xu0/ba∇dblLp≤C(1+t)−d 2(1 2−1 p)+l 2t−d 2(1 q−1 2)−l 2/ba∇dblu0/ba∇dblLq∩L2. Proof.Immediate, from (2.4) and (2.8). 3 Refined linearized estimates The bounds of Proposition 2.1 are sufficient to establish nonl inear stability and asymptotic behavior in dimensions d≥3, as shown in [OZ4]. However, they are not sufficient in the critical dimensions d= 1,2; see Remark 1, Section 7 of [OZ4]. Comparison with standard diffusive stability arguments as in [Z7] show that this is due t o the fact that the full solution operator |S(t)∂x|decays no faster than S(t), or, equivalently, Gyno faster than G. Following the basic strategy introduced in [ZH, Z1, MaZ2, Ma Z4] in the context of vis- cous shock waves, we now perform a refined linearized estimat e separating slower-decaying translational modes from a faster-decaying “good” part of t he solution operator. This will be used in Section 4 in combination with certain nonlinear ca ncellation estimates to show convergence to the modulated approximation (1.1) at a faste r rate sufficient to close the nonlinear iteration. The key to this decomposition is the following observation. 2The inclusion of general p≥2 in Lemma 2.7 repairs an omission in [OZ4], where the bounds ( 2.8) were stated but not used.3 REFINED LINEARIZED ESTIMATES 13 Lemma 3.1. Assuming (H1)–(H3), (D1)–(D3), let λj(ξ/|ξ|,ξ),qj(ξ/|ξ|,ξ,·),˜qj(ξ/|ξ|,ξ,·) denote the eigenvalues and associated right and left eigenf unctions of Lξ, withqj,˜qjsmooth functions of ξ/|ξ|and|ξ|as noted in Prop. 2.2. Then, without loss of generality, q1(ω,0,·)≡ ¯u′, while˜qj(ω,0,·)forj/ne}ationslash= 1are constant functions depending only on angle ω=ξ/|ξ|. Proof.Expanding Lξ=L0+|ξ|L1 ξ/|ξ|+|ξ|2L2 ξ/|ξ|as in the introduction, consider the con- tinuous family of spectral perturbation problems in |ξ|indexed by angle ω=ξ/|ξ|. Then, both facts follow by standard perturbation theory [K] using the observations that ¯ u′is in the right kernel of L0and constant functions care in the left kernel of L0, with /an}b∇acketle{tc,L1¯u′/an}b∇acket∇i}ht=/an}b∇acketle{tc,(ω1(2∂x1−A1)−/summationdisplay j/negationslash=1ωjAj))¯u′/an}b∇acket∇i}ht=/an}b∇acketle{tc,ω1∂2 x1¯u−/summationdisplay j/negationslash=1ωj∂x1fj(¯u)/an}b∇acket∇i}ht ≡0, where/an}b∇acketle{t·,·/an}b∇acket∇i}htdenotesL2(x1) inner product on the interval x1∈[0,X], that the dimension of kerL0by assumption is ( n+ 1), so that the orthogonal complement of ¯ u′in KerL0 is dimension nso exactly the set of constant functions, and that by (H3) the functions qj(ω,0,·) and ˜qj(ω,0) are right and left eigenfunctions of Π 0L1|kerL0(Π0as earlier denoting the zero eigenprojection associated with L0). Remark 3.2. The key observation of Lemma 3.1 can be motivated by the form o f the Whitham averaged system (1.2). For, recalling (Section 1.3 ) that (D3) implies that speed sis stationary to first order at ¯ ualong the manifold of nearby periodic solutions, we find that the last equation of (1.2) reduces to ( ∇xΨ)t= 0, i.e., the equation for the translational variation Ψ decouples from the equations for variations in o ther modes. This corresponds heuristically to the fact derived above that the translatio nal mode ¯u′(x1) decouples in the first-order eigenfunction expansion. Corollary 3.1. Under assumptions (H1)–(H3), (D1)–(D3), the Green functio nG(x,t;y) of(1.7)decomposes as G=E+˜G, (3.1) E= ¯u′(x)e(x,t;y), where, for some C >0, allt>0,1≤q≤2≤p≤ ∞,0≤j,k,l,j+l≤K,1≤r≤2, (3.2)/vextendsingle/vextendsingle/vextendsingle/integraldisplay+∞ −∞˜G(x,t;y)f(y)dy/vextendsingle/vextendsingle/vextendsingle Lp(x)≤C(1+t)−d 2(1/2−1/p)t−1 2(1/q−1/2)|f|Lq∩L2, /vextendsingle/vextendsingle/vextendsingle/integraldisplay+∞ −∞∂r y˜G(x,t;y)f(y)dy/vextendsingle/vextendsingle/vextendsingle Lp(x)≤C(1+t)−d 2(1/2−1/p)−1 2+r 2 ×t−d 2(1/q−1/2)−r 2|f|Lq∩L2, /vextendsingle/vextendsingle/vextendsingle/integraldisplay+∞ −∞∂r t˜G(x,t;y)f(y)dy/vextendsingle/vextendsingle/vextendsingle Lp(x)≤C(1+t)−d 2(1/2−1/p)−1 2+r ×t−d 2(1/q−1/2)−r|f|Lq∩L2.3 REFINED LINEARIZED ESTIMATES 14 (3.3)/vextendsingle/vextendsingle/vextendsingle/integraldisplay+∞ −∞∂j x∂k t∂l ye(x,t;y)f(y)dy/vextendsingle/vextendsingle/vextendsingle Lp≤(1+t)−d 2(1/q−1/p)−(j+k) 2|f|Lq. Moreover,e(x,t;y)≡0fort≤1. Proof.We first treat the simpler case q= 1. Recalling that (3.4) GI(x,t;y) =/parenleftBig1 2π/parenrightBigd/integraldisplay Rdeiξ·(x−y)φ(ξ)n+1/summationdisplay j=1eλj(ξ)tqj(ξ,x1)˜qj(ξ,y1)∗dξ, define (3.5) ˜e(x,t;y) =/parenleftBig1 2π/parenrightBigd/integraldisplay Rdeiξ·(x−y)φ(ξ)eλ1(ξ)t˜q1(ξ,y1)∗dξ, so that (3.6) GI(x,t;y)−¯u′(x1)˜e(x,t;y) =/parenleftBig1 2π/parenrightBigd/integraldisplay Rdeiξ·(x−y)φ(ξ)n+1/summationdisplay j=2eλj(ξ)tqj(ξ/|ξ|,0,x1)˜qj(ξ,y1)∗dξ +/parenleftBig1 2π/parenrightBigd/integraldisplay Rdn+1/summationdisplay j=1eiξ·(x−y)φ(ξ)eλj(ξ)tO(|ξ|)dξ. Noting, by Lemma 3.1, that ∂y˜q(ω,0,y)≡constant for j/ne}ationslash= 1, we have therefore (3.7)∂r y(GI(x,t;y)−¯u′(x1)˜e(x,t;y)) =/parenleftBig1 2π/parenrightBigd/integraldisplay Rdeiξ·(x−y)φ(ξ)n+1/summationdisplay j=1eλj(ξ)tO(|ξ|)dξ, which readily gives (3.8) |∂r y(GI(x,t;y)−¯u′(x1)˜e(x,t;y))|Lp≤C(1+t)−d 2(1−1/p)−1 2, p≥2, by the same argument used to prove (2.9), and similarly (3.9) |∂r t(GI(x,t;y)−¯u′(x1)˜e(x,t;y))|Lp≤c(1+t)−d 2(1−1/p)−1 2. These yield (3.2) by the triangle inequality. Defininge(x,t;y) :=χ(t)˜e(x,t;y), whereχisasmoothcutofffunctionsuchthat χ(t)≡1 fort≥2 andχ(t)≡0 fort≤1, and setting ˜G:=G−¯u′(x1)e(x,t;y), we readily obtain the estimates (3.2) by combining (3.9) with bound (2.4) on GII. Bounds (3.3) follow from (3.5) by the argument used to prove (2.9), together with the observ ation thatx- ort-derivatives bring down factors of |ξ|, followed again by an application of the triangle inequalit y. Thecases1 ≤q≤2followsimilarly, bytheargumentsusedtoprove(2.15)and (2.8).4 NONLINEAR STABILITY IN DIMENSION ONE 15 Remark 3.3. Despite their apparent complexity, the above bounds may be r ecognized as essentially just the standard diffusive bounds satisfied fo r the heat equation [Z7]. For dimensiond= 1, it may be shown using pointwise techniques as in [OZ2] tha t the bounds of Corollary 3.1 extend to all 1 ≤q≤p≤ ∞. Note the strong analogy between the Green function decompos ition of Corollary 3.1 and that of [MaZ3, Z4] in the viscous shock case. We pursue thi s analogy further in the nonlinearanalysisofthefollowingsections, combiningth e“instantaneous tracking” strategy of [ZH, Z1, Z4, Z7, MaZ2, MaZ4] with a type of cancellation est imate introduced in [HoZ]. 4 Nonlinear stability in dimension one For clarity, we carry out the nonlinear stability analysis i n detail in the most difficult, one-dimensional, case, indicating afterward by a few brief remarks the extension to d= 2. Hereafter, take x∈R1, dropping the indices on fjandxjand writing ut+f(u)x=uxx. 4.1 Nonlinear perturbation equations Given a solution ˜ u(x,t) of (1.4), define the nonlinear perturbation variable (4.1) v=u−¯u= ˜u(x+ψ(x,t))−¯u(x), where (4.2) u(x,t) := ˜u(x+ψ(x,t)) andψ:R×R→Ris to be chosen later. Lemma 4.1. Forv,uas in(4.1),(4.2), (4.3) ut+f(u)x−uxx= (∂t−L)¯u′(x1)ψ(x,t)+∂xR+(∂t+∂2 x)S, where R:=vψt+vψxx+(¯ux+vx)ψ2 x 1+ψx=O(|v|(|ψt|+|ψxx|)+/parenleftBig|¯ux|+|vx| 1−|ψx|/parenrightBig |ψx|2) and S:=−vψx=O(|v|(|ψx|). Proof.To begin, notice from the definition of uin (4.2) we have by a straightforward computation ut(x,t) = ˜ux(x+ψ(x,t),t)ψt(x,t)+ ˜ut(x+ψ,t) f(u(x,t))x=df(˜u(x+ψ(x,t),t))˜ux(x+ψ,t)·(1+ψx(x,t))4 NONLINEAR STABILITY IN DIMENSION ONE 16 and uxx(x,t) = (˜ux(x+ψ(x,t),t)·(1+ψx(x,t)))x = ˜uxx(x+ψ(x,t),t)·(1+ψx(x,t))+(˜ux(x+ψ(x,t),t)·ψx(x,t))x. Using the fact that ˜ ut+df(˜u)˜ux−˜uxx= 0, it follows that (4.4)ut+f(u)x−uxx= ˜uxψt+df(˜u)˜uxψx−˜uxxψx−(˜uxψx)x = ˜uxψt−˜utψx−(˜uxψx)x where it is understood that derivatives of ˜ uappearing on the righthand side are evaluated at (x+ψ(x,t),t). Moreover, by another direct calculation, using the fact t hatL(¯u′(x)) = 0 by translation invariance, we have (∂t−L)¯u′(x)ψ= ¯uxψt−¯utψx−(¯uxψx)x. Subtracting, and using the facts that, by differentiation of ( ¯u+v)(x,t) = ˜u(x+ψ,t), (4.5)¯ux+vx= ˜ux(1+ψx), ¯ut+vt= ˜ut+ ˜uxψt, so that (4.6)˜ux−¯ux−vx=−(¯ux+vx)ψx 1+ψx, ˜ut−¯ut−vt=−(¯ux+vx)ψt 1+ψx, we obtain ut+f(u)x−uxx= (∂t−L)¯u′(x)ψ+vxψt−vtψx−(vxψx)x+/parenleftBig (¯ux+vx)ψ2 x 1+ψx/parenrightBig x, yielding (4.3) by vxψt−vtψx= (vψt)x−(vψx)tand (vxψx)x= (vψx)xx−(vψxx)x. Corollary 4.2. The nonlinear residual vdefined in (4.1)satisfies (4.7) vt−Lv= (∂t−L)¯u′(x1)ψ−Qx+Rx+(∂t+∂2 x)S, where (4.8) Q:=f(˜u(x+ψ(x,t),t))−f(¯u(x))−df(¯u(x))v=O(|v|2), (4.9) R:=vψt+vψxx+(¯ux+vx)ψ2 x 1+ψx, and (4.10) S:=−vψx=O(|v|(|ψx|). Proof.Taylor expansion comparing (4.3) and ¯ ut+f(¯u)x−¯uxx= 0.4 NONLINEAR STABILITY IN DIMENSION ONE 17 4.2 Cancellation estimate Our strategy in writing (4.7) is motivated by the following b asic cancellation principle. Proposition 4.3 ([HoZ]).For anyf(y,s)∈Lp∩C2withf(y,0)≡0, there holds (4.11)/integraldisplayt 0/integraldisplay G(x,t−s;y)(∂s−Ly)f(y,s)dyds=f(x,t). Proof.Integrating the left hand side by parts, we obtain (4.12)/integraldisplay G(x,0;y)f(y,t)dy−/integraldisplay G(x,t;y)f(y,0)dy+/integraldisplayt 0/integraldisplay (∂t−Ly)∗G(x,t−s;y)f(y,s)dyds. Noting that, by duality, (∂t−Ly)∗G(x,t−s;y) =δ(x−y)δ(t−s), δ(·) here denoting the Dirac delta-distribution, we find that th e third term on the righthand side vanishes in (4.12), while, because G(x,0;y) =δ(x−y), the first term is simply f(x,t). The second term vanishes by f(y,0)≡0. Remark 4.1. Forψ=ψ(t), term (∂t−L)¯u′ψin (4.7) reduces to the term ˙ψ(t)¯u′(x) appearing in the shock wave case [ZH, Z1, Z4, Z7, MaZ2, MaZ4]. 4.3 Nonlinear damping estimate Proposition 4.2. Letv0∈HK(Kas in (H1)), and suppose that for 0≤t≤T, theHK norm ofvand theHK(x,t)norms ofψtandψxremain bounded by a sufficiently small constant. There are then constants θ1,2>0so that, for all 0≤t≤T, (4.13) |v(t)|2 HK≤Ce−θ1t|v(0)|2 HK+C/integraldisplayt 0e−θ2(t−s)/parenleftBig |v|2 L2+|(ψt,ψx)|2 HK(x,t)/parenrightBig (s)ds. Proof.Subtracting from the equation (4.4) for uthe equation for ¯ u, we may write the nonlinear perturbation equation as (4.14) vt+(df(¯u)v)x−vxx=Q(v)x+ ˜uxψt−˜utψx−(˜uxψx)x, where it is understood that derivatives of ˜ uappearing on the righthand side are evaluated at (x+ψ(x,t),t). Using (4.6) to replace ˜ uxand ˜utrespectively by ¯ ux+vx−(¯ux+vx)ψx 1+ψx and ¯ut+vt−(¯ux+vx)ψt 1+ψx, and moving the resulting vtψxterm to the lefthand side of (4.14), we obtain (4.15)(1+ψx)vt−vxx=−(df(¯u)v)x+Q(v)x+ ¯uxψt −((¯ux+vx)ψx)x+/parenleftBig (¯ux+vx)ψ2 x 1+ψx/parenrightBig x.4 NONLINEAR STABILITY IN DIMENSION ONE 18 Taking the L2inner product in xof/summationtextK j=0∂2j xv 1+ψxagainst (4.15), integrating by parts, and rearranging the resulting terms, we arrive at the inequalit y ∂t|v|2 HK(t)≤ −θ|∂K+1 xv|2 L2+C/parenleftBig |v|2 HK+|(ψt,ψx)|2 HK(x,t)/parenrightBig , for someθ >0,C >0, so long as |˜u|HKremains bounded, and |v|HKand|(ψt,ψx)|HK(x,t) remain sufficiently small. Using the Sobolev interpolation |v|2 HK≤ |∂K+1 xv|2 L2+˜C|v|2 L2for ˜C >0 sufficiently large, we obtain ∂t|v|2 HK(t)≤ −˜θ|v|2 HK+C/parenleftBig |v|2 L2+|(ψt,ψx)|2 HK(x,t)/parenrightBig from which (4.13) follows by Gronwall’s inequality. 4.4 Integral representation/ ψ-evolution scheme By Proposition 4.3, we have, applying Duhamel’s principle t o (4.7), (4.16)v(x,t) =/integraldisplay∞ −∞G(x,t;y)v0(y)dy +/integraldisplayt 0/integraldisplay∞ −∞G(x,t−s;y)(−Qy+Rx+St+Syy)(y,s)dyds+ψ(t)¯u′(x). Definingψimplicitly as (4.17)ψ(x,t) =−/integraldisplay∞ −∞e(x,t;y)u0(y)dy −/integraldisplayt 0/integraldisplay+∞ −∞e(x,t−s;y)(−Qy+Rx+St+Syy)(y,s)dyds, following [ZH, Z4, MaZ2, MaZ3], where eis defined as in (3.1), and substituting in (4.16) the decomposition G= ¯u′(x)e+˜Gof Corollary 3.1, we obtain the integral representation (4.18)v(x,t) =/integraldisplay∞ −∞˜G(x,t;y)v0(y)dy +/integraldisplayt 0/integraldisplay∞ −∞˜G(x,t−s;y)(−Qy+Rx+St+Syy)(y,s)dyds, and, differentiating (4.17) with respect to t, and recalling that e(x,s;y)≡0 fors≤1, (4.19)∂j t∂k xψ(x,t) =−/integraldisplay∞ −∞∂j t∂k xe(x,t;y)u0(y)dy −/integraldisplayt 0/integraldisplay+∞ −∞∂j t∂k xe(x,t−s;y)(−Qy+Rx+St+Syy)(y,s)dyds. Equations (4.18), (4.19) together form a complete system in the variables ( v,∂j tψ,∂k xψ), 0≤j≤1, 0≤k≤K, from the solution of which we may afterward recover the shif tψvia (4.17). From the original differential equation (4.7) togeth er with (4.19), we readily obtain short-time existence and continuity with respect to tof solutions ( v,ψt,ψx)∈HKby a standard contraction-mapping argument based on (4.13), (4 .17), and and (3.3).4 NONLINEAR STABILITY IN DIMENSION ONE 19 4.5 Nonlinear iteration Associated with the solution ( u,ψt,ψx) of integral system (4.18)–(4.19), define (4.20)ζ(t) := sup 0≤s≤t|(v,ψt,ψx)|HK(s)(1+s)1/4. Lemma 4.3. For allt≥0for whichζ(t)is finite, some C >0, andE0:=|u0|L1∩HK, (4.21) ζ(t)≤C(E0+ζ(t)2). Proof.By (4.9)–(4.10) and definition (4.20), (4.22) |(Q,R,S)|L1∩L∞≤ |(v,vx,ψt,ψx)|2 L2+|(v,vx,ψt,ψx)|2 L∞≤Cζ(t)2(1+t)−1 2, so long as |ψx| ≤ |ψx|HK≤ζ(t) remains small, and likewise (using the equation to bound t derivatives in terms of x-derivatives of up to two orders) (4.23) |(∂t+∂2 x)S|L1∩L∞≤ |(v,ψx)|2 H2+|(v,ψx)|2 W2,∞≤Cζ(t)2(1+t)−1 2. Applying Corollary 3.1 with q= 1,d= 1 to representations (4.18)–(4.19), we obtain for any 2≤p<∞ (4.24)|v(·,t)|Lp(x)≤C(1+t)−1 2(1−1/p)E0 +Cζ(t)2/integraldisplayt 0(1+t−s)−1 2(1/2−1/p)(t−s)−3 4(1+s)−1 2ds ≤C(E0+ζ(t)2)(1+t)−1 2(1−1/p) and (4.25) |(ψt,ψx)(·,t)|WK,p≤C(1+t)−1 2E0+Cζ(t)2/integraldisplayt 0(1+t−s)−1 2(1−1/p)−1/2(1+s)−1 2ds ≤C(E0+ζ(t)2)(1+t)−1 2(1−1/p). Using (4.13) and(4.24)–(4.25), we obtain |v(·,t)|HK(x)≤C(E0+ζ(t)2)(1+t)−1 4. Combining this with (4.25), p= 2, rearranging, and recalling definition (4.20), we obtain (4.3). Proof of Theorem 1.1. By short-time HKexistence theory, /ba∇dbl(v,ψt,ψx)/ba∇dblHKis continuous so long as it remains small, hence ηremains continuous so long as it remains small. By (4.3), therefore, it follows by continuous induction that η(t)≤2Cη0fort≥0, ifη0<1/4C, yielding by (4.20) the result (1.15) for p= 2. Applying (4.24)–(4.25), we obtain (1.15) for 2≤p≤p∗for anyp∗<∞, with uniform constant C. Takingp∗>4 and estimating |Q|L2,|R|L2,|S|L2(t)≤ |(v,ψt,ψx)|2 L4≤CE0(1+t)−3 45 NONLINEAR STABILITY IN DIMENSION TWO 20 in place of the weaker (4.22), then applying Corollary 3.1 wi thq= 2,d= 1, we obtain finally (1.15) for 2 ≤p≤ ∞, by a computation similar (4.24)–(4.25); we omit the detail s of this final bootstrap argument. Estimate (1.16) then follows using (3.3) with q=d= 1, by (4.26) |ψ(t)|Lp≤CE0+Cζ(t)2/integraldisplayt 0(1+t−s)−1 2(1−1/p)(1+s)−1 2ds≤C(1+t)1 2p(E0+ζ(t)2), together with the fact that ˜ u(x,t)−¯u(x) =v(x−ψ,t)+(¯u(x)−¯u(x−ψ),so that|˜u(·,t)−¯u| is controlled by the sum of |v|and|¯u(x)−¯u(x−ψ)| ∼ |ψ|. This yields stability for |u−¯u|L1∩HK|t=0sufficiently small, as described in the final line of the theore m. 5 Nonlinear stability in dimension two We now briefly sketch the extension to dimension d= 2. Given a solution ˜ u(x,t) of (1.4), define the nonlinear perturbation variable (5.1) v=u−¯u= ˜u(x1+ψ(x,t),x2,t)−¯u(x1), where (5.2) u(x,t) := ˜u(x1+ψ(x,t),t) andψ:Rd×R→Ris to be chosen later. Lemma 5.1. Forv,uas in(5.2), (5.3)ut+d/summationdisplay j=1fj(u)xj−d/summationdisplay j=1uxjxj= (∂t−L)¯u′(x1)ψ(x,t)+d/summationdisplay j=1∂xjRj+∂tS+T, where Rj=O((|v,ψt,ψx)||(v,vx,ψt,ψx)|), S:=−vψx1= (|v|(|ψx|), T:=O(|ψx|3+|(v,ψx)||ψxx|). Proof.Similarly as in the proof of Lemma 4.1, it follows by a straigh tforward computation Using the fact that ˜ ut+/summationtext jdfj(˜u)˜uxj−/summationtext j˜uxjxj= 0, it follows that (5.4)ut+/summationdisplay jdfj(u)uxj−/summationdisplay juxjxj= ˜ux1ψt−˜utψx1+/summationdisplay j/negationslash=1dfj(˜u)˜ux1ψxj −/summationdisplay j/negationslash=1˜uxjx1ψxj−/summationdisplay j(˜ux1ψxj)xj, where it is understood that derivatives of ˜ uappearing on the righthand side are evaluated at (x+ψ(x,t),t). Moreover, by another direct calculation, using the fact t hatL(¯u′(x1)) = 0 by translation invariance, we have (∂t−L)¯u′(x1)ψ= ¯ux1ψt−¯utψx1+/summationdisplay j/negationslash=1dfj(¯u)¯ux1ψxj−/summationdisplay j/negationslash=1¯uxjx1ψxj−/summationdisplay j(¯ux1ψxj)xj.5 NONLINEAR STABILITY IN DIMENSION TWO 21 Subtracting, and using (4.5) and (5.5)¯uxj+vxj= ˜uxj+ ˜ux1ψxj, ¯ut+vt= ˜ut+ ˜ux1ψt, so that (5.6)˜uxj−¯uxj−vxj=−(¯ux1+vx1)ψxj 1+ψx1, ˜ut−¯ut−vt=−(¯ux1+vx1)ψt 1+ψx1, we obtain ut+/summationdisplay jdfj(u)uxj−/summationdisplay juxjxj= (∂t−L)¯u′(x1)ψ+vx1ψt−vtψx1 +/summationdisplay j/negationslash=1(dfj(˜u)˜ux1−dfj(¯u)¯ux1)ψxj −/summationdisplay j/negationslash=1(˜uxjx1−¯uxjx1)ψxj−/summationdisplay j((˜ux1−¯ux1)ψxj)xj. Usingvx1ψt−vtψx1= (vψt)x1−(vψx1)t, dfj(˜u)˜ux1=f(u)x1−dfj(˜u)˜ux1ψx1=f(u)x1(1−ψx)−dfj(˜u)˜ux1ψ2 x1, and ˜uxjx1= (˜uxj)x1−˜uxjx1ψx1= (˜uxj)x1(1−ψx1)+ ˜uxjx1ψ2 x1,and rearranging, we obtain ut+/summationdisplay jdfj(u)uxj−/summationdisplay juxjxj= (∂t−L)¯u′(x1)ψ+(vψt)x1−(vψx1)t +/summationdisplay j/negationslash=1(fj(u)−fj(¯u))x1)ψxj −/summationdisplay j/negationslash=1f(u)x1ψx1ψxj−/summationdisplay j/negationslash=1dfj(˜u)˜ux1ψ2 x1ψxj −/summationdisplay j/negationslash=1(˜uxj−¯uxj)x1ψxj+/summationdisplay j/negationslash=1(˜uxj)x1ψx1ψxj +/summationdisplay j/negationslash=1˜uxjx1ψ2 x1ψxj −/summationdisplay j(vx1ψx1)xj−/summationdisplay j/parenleftBig (¯ux1+vx1)ψxjψx1 1+ψx1/parenrightBig xj. Noting that (fj(u)−fj(¯u))x1)ψxj= ((fj(u)−fj(¯u)ψxj)x1−(fj(u)−fj(¯u))ψxjx1,5 NONLINEAR STABILITY IN DIMENSION TWO 22 f(u)x1ψx1ψxj= (f(u)ψx1ψxj)x1−f(u)(ψx1ψxj)x1, and (˜uxj−¯uxj)x1ψxj= ((˜uxj−¯uxj)ψxj)x1−(˜uxj−¯uxj)ψxjx1, with|fj(u)−fj(¯u)|=O(|v|) and|˜uxj−¯uxj|=O(|v|),we obtain the result Proof of Theorem 1.2. The result of Lemma 5.1 is the only part of the analysis that di ffers essentially from that of the one-dimensional case. The canc ellation and nonlinear damping arguments go through exactly as before to yield the analogs o f Propositions 4.3 and (4.2). Likewise, we obtain a Duhamel representation analogous to ( 4.18)–(4.19), forming a closed system in variables ( v,ψx,ψt). To obtain the analog of Lemma 4.3, completing the proof of non linear stability, we can carry out a somewhat simpler argument than in the one-dimens ional case, using Corollary 3.1 withd= 2,q= 2 for all estimates, not only the final bootstrap argument, g iving in place of (4.24) the estimate (5.7) |v(·,t)|Lp(x)≤C(1+t)−(1−1/p)E0+Cζ(t)2/integraldisplayt 0(1+t−s)−(1/2−1/p)(t−s)−1 2(1+s)−1ds ≤C(E0+ζ(t)2)(1+t)−(1−1/p), (5.8)|(ψx,ψt)(·,t)|Lp(x)≤C(1+t)−(1−1/p)−1 2E0 +Cζ(t)2/integraldisplayt 0(1+t−s)−(1/2−1/p)(t−s)−1 2(1+s)−1ds ≤C(E0+ζ(t)2)(1+t)ε−(1−1/p)−1 2 for divergence-form source terms, and (5.9)|v(·,t)|Lp(x)≤Cζ(t)2/integraldisplayt 0(1+t−s)−(1/2−1/p)(1+s)−3 2ds ≤C(E0+ζ(t)2)(1+t)−(1−1/p), (5.10)|(ψx,ψt)(·,t)|Lp(x)≤C(1+t)−(1−1/p)−1 2E0 +Cζ(t)2/integraldisplayt 0(1+t−s)−(1/2−1/p)(t−s)−1 2(1+s)−3 2ds ≤C(E0+ζ(t)2)(1+t)ε−(1−1/p)−1 2 for faster-decaying nondivergence-form source terms. We omit the details, which are entirely similar to, but subst antially simpler than, those of the one-dimensional case.REFERENCES 23 References [G] R. Gardner, On the structure of the spectra of periodic traveling waves , J. Math. Pures Appl. 72 (1993), 415-439. [GZ] R. Gardner and K. Zumbrun, The Gap Lemma and geometric criteria for in- stability of viscous shock profiles , Comm. Pure Appl. Math. 51 (1998), no. 7, 797–85. [He] D. Henry, Geometric theory of semilinear parabolic equations , Lecture Notes in Mathematics, Springer–Verlag, Berlin (1981). [HoZ] D. Hoff and K. Zumbrun Asymptotic behavior of multidimensional scalar viscous shock fronts , Indiana Univ. Math. Journal, Vol. 49, No. 2 (2000). [H] I.L. Hwang, TheL2-boundedness of pseudodifferential operators, Trans. Amer. Math. Soc. 302 (1987) 55–76. [JZ1] M. Johnson and K. Zumbrun, Rigorous Justification of the Whitham Modulation Equations for the Generalized Korteweg-de Vries Equation, preprint (2009). [JZB] M. Johnson, K. Zumbrun, and J. Bronski, Bloch wave expansion vs. Whitham Modulation Equations for the Generalized Korteweg-de Vrie s Equation, inprepa- ration. [K] T. Kato, Perturbation theory for linear operators , Springer–Verlag, Berlin Hei- delberg (1985). [MaZ2] C. Mascia and K. Zumbrun, Stability of small-amplitude shock profiles of sym- metric hyperbolic-parabolic systems, Comm. Pure Appl. Math. 57 (2004), no. 7, 841–876. [MaZ3] C. Mascia and K. Zumbrun, Pointwise Green function bounds for shock profiles of systems with real viscosity. Arch. Ration. Mech. Anal. 169 (2003), no. 3, 177– 263. [MaZ4] C. Mascia and K. Zumbrun, Stability of large-amplitude viscous shock profiles of hyperbolic-parabolic systems, Arch. Ration. Mech. Anal. 172 (2004), no. 1, 93–131. [OZ1] M. Oh and K. Zumbrun, Stability of periodic solutions of viscous conservation laws with viscosity- 1. Analysis of the Evans function , Arch. Ration. Mech. Anal. 166 (2003), no. 2, 99–166. [OZ2] M. Oh and K. Zumbrun, Stability of periodic solutions of viscous conservation laws with viscosity- Pointwise bounds on the Green function , Arch.Ration. Mech. Anal. 166 (2003), no. 2, 167–196.REFERENCES 24 [OZ3] M. Oh, and K. Zumbrun, Low-frequency stability analysis of periodic traveling- wave solutions of viscous conservation laws in several dime nsions, Journal for Analysis and its Applications, 25 (2006), 1–21. [OZ4] M. Oh, and K. Zumbrun, Stability and asymptotic behavior of traveling-wave solutions of viscous conservation laws in several dimensio ns, to appear, Arch. Ration. Mech. Anal. [Pa] A. Pazy, Semigroups of linear operators and applications to partial differen- tial equations, Applied Mathematical Sciences, 44, Springer-Verlag, New Y ork- Berlin, (1983) viii+279 pp. ISBN: 0-387-90845-5. [S1] G.Schneider, Nonlinear diffusive stability of spatially periodic soluti ons– abstract theorem and higher space dimensions , Proceedings of the International Confer- ence on Asymptotics in Nonlinear Diffusive Systems (Sendai, 1 997), 159–167, Tohoku Math. Publ., 8, Tohoku Univ., Sendai, 1998. [S2] G. Schneider, Diffusive stability of spatial periodic solutions of the Swi ft- Hohenberg equation, (English. English summary) Comm. Math. Phys. 178 (1996), no. 3, 679–702. [S3] G. Schneider, Nonlinear stability of Taylor vortices in infinite cylinder s,Arch. Rat. Mech. Anal. 144 (1998) no. 2, 121–200. [Se1] D. Serre, Spectral stability of periodic solutions of viscous conser vation laws: Large wavelength analysis , Comm. Partial Differential Equations 30 (2005), no. 1-3, 259–282. [Z1] K. Zumbrun, Refined wave–tracking and stability of viscous Lax shocks , Methods Appl. Anal. 7 (2000) 747–768. [Z4] K. Zumbrun, Stability of large-amplitude shock waves of compressible N avier– Stokes equations, withanappendixbyHelge KristianJenssenandGregoryLyng, in Handbook of mathematical fluid dynamics. Vol. III, 311–53 3, North-Holland, Amsterdam, (2004). [Z6] K. Zumbrun, Dynamical stability of phase transitions in the p-system wi th viscosity-capillarity , SIAM J. Appl. Math. 60 (2000), 1913-1929. [Z7] K. Zumbrun, Instantaneous shock location and one-dimensional nonlinea r sta- bility of viscous shock waves, preprint (2009). [ZH] K. Zumbrun and P. Howard, Pointwise semigroup methods and stability of vis- cous shock waves . Indiana Mathematics Journal V47 (1998), 741–871; Errata, Indiana Univ. Math. J. 51 (2002), no. 4, 1017–1021.