EXPONENTIAL TIME DECAY ESTIMATES FOR THE LANDAU EQUATION ON TORUS KUNG-CHIEN WU 3 Abstract. We study the time decay estimates for the linearized Landau equation on 1 toruswhentheinitialperturbationisnotnecessarilysmooth. Ourresultrevealsthekinetic 0 2 and fluid aspects of the equation. We design a Picard-type iteration and Mixture lemma for constructing the increasingly regular kinetic like waves, they are carried by transport n equations and have exponential time decay rate. The fluid like waves are constructed as a J partofthelong-waveexpansioninthespectrumoftheFouriermodeforthespacevariable 4 and the time decay rate depends onthe size of the domain. The Mixture lemma playsan important role in this paper, this lemma is parallel to Boltzmann equation but the proof ] is more challenge. h p - h t a m 1. Introduction and Main Result [ 1.1. The Models. The generalized Landau equation reads 1 v 1 7 ∂ F +ξ ·∇ F = Q(F,F), 2  t x ε (1) 8   0 F(x,0,ξ) = F (x,ξ), . 0 1   0 with collision operator 3 1 : iv Q(F,G) = ∇ξ ·hZR3 Φ(ξ −ξ∗)[F(ξ∗)∇ξG(ξ)−G(ξ)∇ξ∗F(ξ∗)]dξ∗i. X r Here the positive constant ε is the Knudsen number, F(t,x,ξ) ≥ 0 is the spatially periodic a distribution function for the particles at time t ≥ 0, with spatial coordinates x ∈ T3, 1 the 3-dimensional torus with unit size of each side and microscopic velocity ξ ∈ R3. The positive semi-definite matrix Φ(ξ) has the general form Φ(ξ) = B(|ξ|)S(ξ), Date: December 11, 2013. 2010 Mathematics Subject Classification. 35Q20;82C40. Key words and phrases. Landauequation;Fluid-like wave;Kinetic-likewave;Maxwellianstates;Point- wise estimate. This paper is supervised by Cl´ement Mouhot in Cambridge. This work is supported by the Tsz-Tza Foundation in institute of mathematics, Academia Sinica, Taipei, Taiwan. 1 2 K.-C. WU where B(|ξ|) = |ξ|γ+2, γ ≥ −3, is a function depending on the nature of the interaction between the particles, and S(ξ) is the 3 by 3 matrix ξ ⊗ξ S(ξ) = I − . 3 |ξ|2 The original Landau collision operator for the Coulombian interaction corresponds to the case γ = −3. In order to remove the parameter ε from the equation, we introduce the new scaled variables: 1 1 x = x, t = t, ε ε after dropping the tilde, the equatioen (1) becomees ∂ F +ξ ·∇ F = Q(F,F), (x,t,ξ) ∈ T3 ×R+ ×R3 t x 1/ε (2)   F(0,x,ξ) = F (x,ξ), 0 where T3 means the 3-dimensional torus with size 1/ε of each side. The conservation of 1/ε mass, momentum, as well as energy, can be formulated as d 1,ξ,|ξ|2 F(t,x,ξ)dξdx = 0. dt ZT3 ZR3 n o 1/ε As in the Boltzmann equation, it is well know that Maxwellians are steady states to the Landau equation (2). We linearized the Landau equations (2) around a global Maxwellian M(ξ), 1 −|ξ|2 M(ξ) = exp , (2π)3/2 (cid:16) 2 (cid:17) which the standard perturbation f(x,t,ξ) to M as F = M +M1/2f . It is well-know that Q(M,M) = 0, by expanding Q(M +M1/2f,M +M1/2f) = 2Q(M,M1/2f)+Q(M1/2f,M1/2f), dropping the nonlinear term, we can define the linearized collision Landau operator L as (3) Lf = 2M−1/2Q(M,M1/2f). Similar to Boltzmann equation, we can decompose the linear collision operator L as diffu- sion part and compact part. Lf = Λf +Kf , the diffusion part e e Λf = M−1/2∇· (ϑM)∇(M−1/2f) , (cid:2) (cid:3) where the symmetric matrix ϑ(ξ) is e (4) ϑ(ξ) = Φ(ξ −ξ )M(ξ )dξ , Z ∗ ∗ ∗ R3 LANDAU EQUATION 3 and the compact part Kf = M−1/2(ξ)M−1/2(ξ )∇ ·∇ Z(ξ,ξ )f(ξ )dξ , Z ∗ ξ ξ∗ ∗ ∗ ∗ R3 e where Z(ξ,ξ ) = M(ξ)M(ξ )Φ(ξ −ξ ). ∗ ∗ ∗ The linearized Landau equation for f(t,x,ξ) now takes the form ∂ f +ξ ·∇ f = Lf , t x (5)   f(0,x,ξ) = I(x,ξ), here we define f(x,t,ξ) = GtI(x,ξ), i.e. Gt is the solution operator (Green function) ε ε of the linearized Landau equation (5). By assuming that initially F (x,ξ) has the same 0 total mass, momentum and total energy as the Maxwellian M, we can then rewrite the conservation laws as (6) 1,ξ,|ξ|2 M1/2I(t,x,ξ)dξdx = 0, ZT3 ZR3n o 1/ε this means the initial I satisfies the zero mean conditions. 1.2. Preliminaries. Before the presentation of the properties of the collision operator L, let us define some notations in this paper. For microscopic variable ξ, we shall use L2 to ξ denote the classical Hilbert space with norm 1/2 kfk = |f|2dξ , L2ξ (cid:16)ZR3 (cid:17) the Sobolev space of functions with all its s-th partial derivatives in L2 will be denoted by ξ Hs. The L2 inner product in R3 will be denoted by ·,· . For space variable x, we shall ξ ξ ξ use L2 to denote the classical Hilbert space with norm(cid:10) (cid:11) x 1 1/2 kfk = |f|2dx , L2x (cid:16)|T31/ε| ZT3 (cid:17) 1/ε the Sobolev space of functions with all its s-th partial derivatives in L2 will be denoted by x Hs. We denote the sup norm as x kfkL∞ = sup |f(x)|. x x∈T3 1/ε For notational of simplicity, we denote ξ s = (1 + |ξ|)s. In this paper, we need two projection operators, (cid:10) (cid:11) (i) For any h ∈ L2, we define the projection to the space {M1/2} as ξ p (h) = h,M1/2 M1/2(ξ). 0 ξ (cid:10) (cid:11) 4 K.-C. WU (ii) For any vector function u ∈ L2, we define the orthogonal projection to the vector ξ as ξ u·ξ P(ξ)u = ξ. |ξ|2 First, we list some important propositions about the diffusion operator Λ. Proposition 1. For any γ ≥ −3, we have e (i) Λf = ∇ · ϑ∇ f +∇ · ϑξ f −(ξ,ϑξ)f . ξ ξ ξ (cid:2) (cid:3) (cid:2) (cid:3) (ii) The spectrum of ϑ(eξ) consists of a simple eigenvalue λ (ξ) > 0 associated with the 1 eigenvector ξ, and a double eigenvalue λ (ξ) > 0 associated with eigenvector ξ⊥. Moreover, 2 there are constants c and c > 0 such that asymptotically, as |ξ| → ∞, we have 1 2 γ γ+2 λ (ξ) = c ξ , λ (ξ) = c ξ , 1 1 2 2 (cid:10) (cid:11) (cid:10) (cid:11) moreover (ξ,ϑξ) = λ (ξ)|ξ|2, 1 (u,ϑu) = λ (ξ)|P(ξ)u|2+λ (ξ) I −P(ξ) u 2, 1 2 3 (cid:12)(cid:2) (cid:3) (cid:12) (iii) For any |α| ≥ 1, (cid:12) (cid:12) |∂α(ϑξ)|+|∂αϑ| ≤ C ξ γ+2−|α|. ξ ξ α (cid:10) (cid:11) and |∂αλ (ξ)| ≤ C ξ γ+1−|α|, |∂αλ (ξ)| ≤ C ξ γ+2−|α|. ξ 1 α ξ 2 α (cid:10) (cid:11) (cid:10) (cid:11) (iv) (Coercivity estimate) There exists c ,ν > 0 such that 0 0 −Λ(f),f ≥ c k ξ γ/2+1fk2 +k ξ γ/2P(ξ)∇ fk2 ξ 0n L2ξ ξ L2ξ (cid:10) (cid:11) (cid:10) (cid:11) (cid:10) (cid:11) e + ξ γ/2+1 I −P(ξ) ∇ f 2 −ν M1/2,f 2. (cid:13)(cid:10) (cid:11) (cid:2) 3 (cid:3) ξ (cid:13)L2ξo 0(cid:10) (cid:11)ξ (cid:13) (cid:13) It is easy to see that if we want to get coercivity estimate, we need modify the decom- position of L. Now, we define L = Λ+K, where (7) Λf = Λf −ν p (f), Kf = Kf +ν p (f). 0 0 0 0 Under above definition, we caen rewrite the coercivityeestimate of Landau equation as −Λ(f),f ≥ c k ξ γ/2+1fk2 +k ξ γ/2P(ξ)∇ fk2 ξ 0n L2ξ ξ L2ξ (cid:10) (cid:11) (cid:10) (cid:11) (cid:10) (cid:11) + ξ γ/2+1 I −P(ξ) ∇ f 2 . (cid:13)(cid:10) (cid:11) (cid:2) 3 (cid:3) ξ (cid:13)L2ξo (cid:13) (cid:13) In order to estimate the Green function of the Landau equation in next section, we need to recall the spectrum Spec(εk), k ∈ Z3, of the operator −iπεξ ·k +L LANDAU EQUATION 5 Proposition 2. [3] For any γ ≥ −2, there exists δ > 0 and τ = τ(δ) > 0 such that (i) For any |εk| > δ, (8) Spec(εk) ⊂ {z ∈ C : Re(z) < −τ}. (ii) For any |εk| < δ, the spectrum within the region {z ∈ C : Re(z) > −τ} consisting of exactly five eigenvalues {σ (εk)}4 , j j=0 (9) Spec(εk)∩{z ∈ C : Re(z) > −τ} = {σ (εk)}4 , j j=0 and corresponding eigenvectors {e (εk)}4 , where j j=0 3 σ (εk) = a (i|εk|)n +O(εk)4, a > 0, j j,n j,2 X (10) n=1 3 e (εk) = e (i|εk|)n +O(εk)4, e ,e = δ . j j,n j k ξ jk Xn=1 (cid:10) (cid:11) (iii) 4 (11) e(−iπεξ·k+L)tf = Π f +χ eσj(−εk)t e (εk),f e (εk) δ {|εk|<δ} j ξ j Xj=0 (cid:10) (cid:11) where kΠ k = O(1)e−a(τ)t, a(τ) > 0, χ is the indicator function. δ L2 {·} ξ 1.3. Main Theorem and Method of Proof. In this paper, we decompose the solution of the linearized Landau equation as kinetic part, fluid part and remainder part. The kinetic aspect of the solution is described by the diffusive transport equation: ∂ g +ξ ·∇ g = Λg. t x The operator K is a smooth operator in ξ variable, we use this smoothness property to design a Picard-type iteration for constructing the increasingly regular kinetic like waves. The fluid behavior is studying by constructing the Green function with the Fourier series in space variable x: 1 Gt = eiπεk·x+(−iπεξ·k+L)t, ε |T3 | Xk∈Z 1/ε here the Green function is viewed as an operator for function of ξ. This prompts the analysis of the spectrum of the operator −iπεξ ·k+L. The spectrum includes five curves bifurcating from the origin. The origin is the multiple zero eigenvalues of L, the operator at k = 0. The kernel of L are the fluid variables. The fluid like waves are constructed from these curves near the origin. With the kinetic like waves and fluid like waves constructed, the rest of the solution is of sufficient smoothness and it has exponential time decay rate. 6 K.-C. WU Theorem 3. For any γ > −2, I ∈ L2 with compact support in x and satisfies the zero ξ mean conditions (6), then the solution GtI = Gt I +Gt I +Gt I , ε ε,K ε,F ε,R consists of the kinetic part Gt I: nonsmooth and time decay exponentially ε,K (12) kGt Ik = O(1)e−O(1)t; ε,K L2L2 x ξ the fluid part Gt I: the time decay rate depends on the size of the domain, i.e. there exist ε,F δ,δ ,C > 0 such that 0 0 (i) If ε > δ, (13) kGt Ik = 0, ε,F HxsL2ξ (ii) If δ < ε < δ, 0 (14) kGt Ik = O(1)e−O(1)ε2t, ε,F HxsL2ξ (iii) If 0 < ε < δ , 0 C (15) kGt Ik = O(1) 0 e−O(1)ε2t; ε,F HxsL2ξ (1+t)3/2 and the smooth remainder part Gt I: ε,R (16) kGt Ik = O(1)e−O(1)t. ε,R H2L2 x ξ The Mixture Lemma plays an important role to construct the kinetic like waves, it states that the mixture of the two operators S and K in Mt (see section 2) transports the j regularity in the microscopic velocity ξ to the regularity of the space time (x,t). This idea was introduced by Liu-Yu [10, 11, 12, 13] to construct the Green function of the Boltzmann equation. In Liu-Yu’s paper, the proof of Mixture lemma relies on the exact solution of the damped transport equations. In this paper, we introduce a differential operator to avoid constructing explicit solution, this operator commutes with free transport operator and can transports the microscopic velocity regularity to space regularity, this idea can also apply to Boltzmann equation to simplify Liu-Yu’s results. Actually, the Mixture Lemma is similar in spirit as the well-known Averaging Lemma, see [1, 4, 6]. These two lemmas have been introduced independently and used for different purposes. The spectrum analysis of the Landau equation was introduced by Degond-Lemou [2], this result parallel to the Boltzmann case done by Ellis-Pinsky [3], however, this paper is the first one give coercivity estimate of Landau equation (they called Poincare type inequality), and they also give some ideas to decompose the collection term as diffusion part and compact part, see (7). However, the Poincare type inequality is not enough in our analysis, we thank Mouhot [14] and Guo [5] give more shape coercivity estimates for the linearized Landau operator. LANDAU EQUATION 7 The analysis based on spectrum has been carried out by many authors. The exponential time decay rates for the Boltzmann equation with hard potentials on torus was firstly provided by Ukai [16]. The time-asymptotic nonlinear stability was obtained in [9, 17]. Using the Nishida approach, [8] obtained the time-asymptotic equivalent of Boltzmann solutions and Navier-Stokes solutions. These works yield the L2 theory, since the Fourier transform is isometric in L2. In this paper, we need to restrict γ > −2, there are two reasons about this restriction, first, we need spectrum gap analysis, but this only holds for γ ≥ −2, if γ < −2, the spectrum of the collision operator is not clear. The second reason is singular part of compact term K , the derivative of this operator need L2 norm bound, and the coefficient s ξ need bounded by the cut off distance, this property holds only when γ > −2, hence we need this restriction. The pointwise description of the one-dimensional linearized Boltzmann equation with hard sphere was firstly provided by Liu-Yu [10], the fluid like waves can be constructed by both complex and spectrum analysis, it reveals the dissipative behavior of the type of the Navier-Stokes equation as usually seems by the Chapman-Enskog expansion. The kinetic like waves can be constructed by Picard-type iteration and Mixture lemma. In this paper, we apply similar ideas to Landau equation on torus, we can also construct the kinetic and fluid like waves, which are both time decay exponentially. Moreover, the decay rate of the fluid like waves depends on the size of the domain. 1.4. Plain of the Paper. The rest of the paper is organized as follows. In section 2, we design Picard-type iteration and Mixture lemma for constructing the increasingly regular kinetic like waves. In section 3, we construct the Green function of the linearized Landau equation on torus, then use the long wave short wave decomposition and the spectrum analysis to obtain time decay rate, moreover, we improve the estimate of the fluid like waves. 2. Kinetic Part In this section, we will use the kinetic decomposition to construct the kinetic part, and apply Mixture lemma to show the tail term has enough regularity in x. 2.1. Picard iteration and Kinetic Decomposition. We rewrite the linearized Landau equation (5) as ∂ f +ξ ·∇ f −Λf = Kf , t x (17)   f(x,0,ξ) = I(x,ξ).  8 K.-C. WU The operator K has some regularizing effects with respect to the microscopic variable ξ. We decompose it as K = K +K , K ≡ K and K ≡ K : s r s s,D r r,D |ξ −ξ | K f = χ ∗ M−1/2(ξ)M−1/2(ξ )∇ ·∇ Z(ξ,ξ )f(ξ )dξ ,  s ZR3 (cid:16) D (cid:17) ∗ ξ ξ∗ ∗ ∗ ∗     (18)  Krf = Kf −Ksf ,   χ(r) = 1 for r ∈ [−1,1],      supp(χ) ⊂ [−2,2], χ ∈ C∞(R), χ ≥ 0.  c   Lemma 4. For γ > −2, we have the following smoothing properties about the operator K : (19) kK fk ≤ Dγ+2kfk , kK fk ≤ Ckfk . s H1 L2 r Hs L2 ξ ξ ξ ξ Proof. It is easy to see the regular part dominated by Guassian, so the estimate of regular part K is obvious. For singular part K , we introduce the Hardy-Littlewood maximum r s function: for any f ∈ L2 ξ M(f) = supr−3 |f(ξ )|dξ , Z ∗ ∗ r>0 |ξ−ξ∗|<r then ∂ K f can be bounded by Hardy-Littlewood maximum function M(f) ξ s |∂ K f(ξ)| ≤ C |ξ −ξ |(γ+2)−3|f(ξ )|dξ ξ s Z ∗ ∗ ∗ |ξ−ξ∗|<D ∞ = C |ξ −ξ |(γ+2)−3|f(ξ )|dξ Z ∗ ∗ ∗ Xn=0 2−(n+1)D<|ξ−ξ∗|<2−nD ∞ ≤ C (2−nD)γ+2(2−nD)−3 |f(ξ )|dξ Z ∗ ∗ Xn=0 |ξ−ξ∗|<2−nD ≤ CDγ+2M(f). By the Hardy-Littlewood-Wiener maximal theorem, we have kM(f)k ≤ Ckfk , L2 L2 ξ ξ and hence k∂ K fk ≤ CDγ+2kfk . ξ s L2 L2 ξ ξ (cid:3) We define the operator g = Stg and j = Otj as follows : 0 0 ∂ g +ξ ·∇ g −Λg = 0, t x (20)   g(x,0,ξ) = g (x,ξ), 0  LANDAU EQUATION 9 and ∂ j +ξ ·∇ j −Λj = K j, t x s (21)   j(x,0,ξ) = j (x,ξ). 0 We have the following standard estimates about diffusive transport equations: Lemma 5. (22) kStkL2L2 ≤ e−c0t, kOtkL2L2 ≤ e−c20t. x ξ x ξ Proof. Note that γ > −2, we have 1 d 1 (23) 2dtkgk2L2xL2ξ = |T31/ε| ZR3 ZT3 Λ(g)gdxdξ ≤ −c0kgk2L2xL2ξ , 1/ε this proves kStg0kL2xL2ξ ≤ e−c0tkg0k2L2xL2ξ. For Ot, we have 1 d 1 (24) 2dtkjk2L2xL2ξ = |T31/ε| ZR3 ZT3 Λ(j)jdxdξ ≤ (−c0 +Dγ+2)kjk2L2xL2ξ , 1/ε we get our result if we choose D sufficiently small. (cid:3) Now, we design a Picard type iteration, which treat the regular part K f as the source r term. The −1 order approximation of the linearized Landau equation (17) is the damped transport equation ∂ h(−1) +ξ ·∇ h(−1) −Λh(−1) = K h(−1), t x s (25)   h(−1)(x,0,ξ) = I(x,ξ). The difference f −h(−1) satisfies the equation ∂ (f −h(−1))+ξ ·∇ (f −h(−1))−Λ(f −h(−1)) = K(f −h(−1))+K h(−1), t x r (26)   (f −h(−1))(x,0,ξ) = 0, this means the zero order approximation h(0) is ∂ h(0) +ξ ·∇ h(0) −Λh(0) = K h(−1), t x r (27)   h(0)(x,0,ξ) = 0, for this process, we can define the kth order approximation h(k), k ≥ 1 ∂ h(k) +ξ ·∇ h(k) −Λh(k) = Kh(k−1), t x (28)   h(k)(x,0,ξ) = 0. This means the solutionf can be rewritten as f = h(−1) +h(0) +h(1) +···. 10 K.-C. WU The following lemma gives the L2L2 estimate of h(j). x ξ Lemma 6. For j ≥ −1, we have (29) kh(j)kL2L2 ≤ tj+1e−c20tkIkL2L2. x ξ x ξ Proof. This lemma canbeproven byinduction. The casej = −1 immediately fromLemma 5. Now, we prove the case j = 0. By Duhamel principle, t (30) h(0)(x,t,ξ) = St−s1K Os1I(·,s )ds . Z r 1 1 0 It is easy to see that kh(0)kL2L2 ≤ O(1)te−c20tkIkL2L2. x ξ x ξ Assume it holds for j, then by the compactness property of K, we have t kh(j+1)k = St−s(Kh(j))(·,s)ds L2xL2ξ (cid:13)(cid:13)Z0 (cid:13)(cid:13)L2xL2ξ (cid:13) t (cid:13) ≤ Z e−(t−s)e−c02ssj+1kIkL2xL2ξds 0 ≤ tj+2e−c02tkIkL2L2. x ξ (cid:3) Now, we can define the kinetic decomposition as 4 (31) GtI = h(j) +R = Gt I +R, ε ε,K X j=−1 then R satisfies the equation ∂ R+ξ ·∇ R = LR+Kh(4), t x (32)   R(x,0,ξ) = 0. By (32), the regularity of Rin x is equivalent to the regularity of h(4) in x, in order to make sure the tail term R has good regularity, we need to check h(4) has regularity H2. x To proceed, we need the Mixture lemma, define the Mixture operator as follow: t s1 s2j−1 (33) Mtf = ··· St−s1KSs1−s2KSs2−s3K ···Ss2j−1−s2jKSs2jf ds ···ds . j 0 Z Z Z 0 2j 1 0 0 0 Under this definition, we have t (34) h(4)(x,t,ξ) = Mt−s0K Os0I(·,s )ds . Z 2 r 0 0 0 Lemma 7. [7, 10](Mixture Lemma) For any f ∈ L2H2, j = 1,2, we have 0 x ξ (35) k∂xjMtjf0kL2xL2ξ ≤ tje−2c30tkf0kL2xHξj .