N Stability in the Energy Space of the Sum of Peakons for the Degasperis-Procesi Equation Andr´e Kabakouala L.M.P.T., U.F.R Sciences et Techniques, Universit´e de Tours, Parc Grandmont, 6 37200 Tours, France. 1 0 2 [email protected] n a J Abstract 8 The Degasperis-Procesi equation possesses well-known peaked solitary waves that are called 2 peakons. Their stability has been established by Lin and Liu in [5]. In this paper, we localize ] theproof(insomesuitablesensedetailed inSection3)ofthestability ofasinglepeakon. Thanksto P this, we extend theresult of stability to thesum of N peakons traveling to theright with respective A speeds c1,...,cN,such that thedifference between consecutivelocations of peakons is large enough. . h t a m [ 1 Introduction 1 v The Degasperis-Procesi(DP) equation 4 7 u u +4uu =3u u +uu , (t,x) R R (1.1) 6 t− txx x x xx xxx ∈ ∗+× 7 is completely integrable (see [1]) and possesses, among others, the following invariants 0 . 1 E(u)= yv and F(u)= u3, (1.2) 0 R R Z Z 6 1 where y = (1−∂x2)u and v = (4−∂x2)−1u. Substituting u by 4v−vxx in (1.2) and using integration by : parts (we suppose that u( )=v( )=v ( )=0), the conservation laws can be rewritten as v ±∞ ±∞ x ±∞ i X E(u)= 4v2+5v2+v2 and F(u)= v3 +12vv2 48v2v +64v3 . (1.3) r ZR x xx ZR − xx xx− xx a (cid:0) (cid:1) (cid:0) (cid:1) One can see that the conservationlaw E() is equivalent to 2 . Indeed, using integration by parts · k·kL2(R) u 2 = u2 = (4v v )2 = 16v2+8v2+v2 4E(u), (1.4) k kL2(R) R R − xx R x xx ≤ Z Z Z (cid:0) (cid:1) and applying Plancherel-Parsevalidentity 1+ω2 E(u)= yv = u(ω)2 u(ω)2 = u 2 , (1.5) R R 4+ω2| | ≤ R| | k kL2(R) Z Z Z where u denotes the Fourier transform of u. In thebsequel we wbill denote u = E(u). (1.6) b k kH p 1 Notethat,byreversingtheoperator(1 ∂2)()in(1.1),theDPequationcanberewritteninconservation − x · form as 1 3 ut+ 2∂xu2+ 2(1−∂x2)−1∂xu2 =0, (t,x)∈R∗+×R. (1.7) The DP equation possesses solitary waves called peakons (see Fig. 1a) and defined by u(t,x)=ϕc(x−ct)=cϕ(x−ct)=ce−|x−ct|, c∈R∗, (t,x)∈R∗+×R, (1.8) but they are not smooth since ϕ / C1(R) (see Fig. 1b). The peakons are only global weak solutions of c (1.7). It means, for any smooth te∈st function φ C (R R), it holds ∞ + ∈ × +∞ ϕ (x ct)φ (t,x)dtdx+ 1 +∞ ϕ2(x ct)φ (t,x)dtdx Z0 ZR c − t 2Z0 ZR c − x + 3 +∞ (1 ∂2) 1ϕ2(x ct)φ (t,x)dtdx+ ϕ (x)φ(0,x)dx=0. 2Z0 ZR − x − c − x ZR c The goal of our work is to prove that ordered trains of peakons are stable under small perturbations in the energy space (equivalent to L2). H Definition 1.1 (Stability). Let c>0 be given. The peakon ϕ is said stable in , if for all ε>0, there c H exists δ >0 such that if u ϕ δ, (1.9) 0 c k − kH ≤ then for all t 0, there exists ξ(t) such that ≥ u(t, ) ϕ ( ξ(t)) ε, (1.10) c k · − ·− kH ≤ where u(t) is the solution to (1.1) emanating from u . 0 Lin and Liu proved in [5] the stability of a single peakon under the additional condition that (1 ∂2)u +(R). Using this result and the general strategy introduced by Martel, Merle and Tsai in [7−] x 0 ∈M for the generalized Korteweg-de Vries (gKdV) equation and adapted by El Dika and Molinet in [3] and [2] for the Camassa-Holm (CH) equation, we prove here the stability of the sum of N peakons for the DP equation. Before stating the main result we introduce the function space where will live our class of solutions to the equation. For I a finite or infinite time interval of R , we denote by (I) the function space 1 + X (I)= u C I;H1(R) L I;W1,1(R) , u L (I;BV(R)) . (1.11) ∞ x ∞ X ∈ ∩ ∈ The main result of th(cid:8)e presen(cid:0)t paper is(cid:1)the fol(cid:0)lowing theor(cid:1)em. (cid:9) Theorem 1.1 (Stability of the Sum of N Peakons). Let be given N velocities c ,...,c such that 1 N 0 < c < ...< c . Let u ([0,T[), with 0 < T + , be a solution of the DP equation. There exist 1 N ∈ X ≤ ∞ C >0, L >0 and ε >0 only depending on the speeds (c )N , such that if 0 0 i i=1 y =(1 ∂2)u +(R) (1.12) 0 − x 0 ∈M and N u ϕ ( z0) ε2, with 0<ε<ε , (1.13) (cid:13) 0− ci ·− i (cid:13) ≤ 0 (cid:13)(cid:13) Xi=1 (cid:13)(cid:13)H for some z0,...,z0 satisfyin(cid:13)g (cid:13) 1 N (cid:13) (cid:13) z0 <...<z0 and z0 z0 L, with L>L >0, i=2,...,N, (1.14) 1 N i − i−1 ≥ 0 1W1,1(R) is the space of L1(R) functions with derivatives inL1(R) and BV(R) is the space of function with bounded variation. 2 then there exist ξ1(t),...,ξN(t) such that 1 1 N u(t) ϕ ( ξi(t)) C(√ε+L 1/8), t [0,T[ (1.15) (cid:13) − ci ·− 1 (cid:13) ≤ − ∀ ∈ (cid:13)(cid:13) Xi=1 (cid:13)(cid:13)H (cid:13) (cid:13) and (cid:13) (cid:13) L ξ1i(t)−ξ1i−1(t)> 2, ∀t∈[0,T[, i=2,...,N, (1.16) where ξ1(t),...,ξN(t) are defined in Subsection 4.1. 1 1 2 Preliminaries Inthissection,webrieflyrecalltheglobalwell-posednessresultsfortheDPequationanditsconsequences (see [4] and [6] for details). Theorem 2.1 (Global Weak Solution; See [4] and [6]). Assume that u L2(R) with y =(1 ∂2)u +(R). Then the DP equation has a unique global weak solution u 0(∈R ) such that0 − x 0 ∈ + M ∈X y(t, )=(1 ∂2)u(t, ) +(R), t R . (2.1) · − x · ∈M ∀ ∈ + Moreover E() and F() are conserved by the flow. · · Remark 2.1 (Control of L Norm by L2 Norm). From (2.1), it holds ∞ u(x)= e−x x ex′y(x)dx + ex +∞e x′y(x)dx ′ ′ − ′ ′ 2 2 Z−∞ Zx and u (x)= e−x x ex′y(x)dx + ex +∞e x′y(x)dx, x ′ ′ − ′ ′ − 2 2 Z−∞ Zx which lead to u (x) u(x), x R. (2.2) x | |≤ ∀ ∈ Then,usingthe Sobolevembedding ofH1(R)into L (R)and(2.2), weinferthatthere existsaconstant ∞ C >0 such that S u L∞(R) CS u H1(R) 2CS u L2(R). (2.3) k k ≤ k k ≤ k k Lemma 2.1 (Positivity; See [6]). Let u H1(R) with y =(1 ∂2)u +(R). If k 1, then we have ∈ − x ∈M 1 ≥ (k ∂ )u(x) 0, x R. (2.4) 1 x ± ≥ ∀ ∈ Lemma2.2(Positivity;See[6]). Letw(x)=(k ∂ )u(x). Assumethatu H1(R)withy =(1 ∂2)u +(R). If k 1 and k 2, then we have 1± x ∈ − x ∈ 1 2 M ≥ ≥ (k ∂ )(4 ∂2) 1w(x) 0, x R. (2.5) 2± x − x − ≥ ∀ ∈ 3 Stability of a single peakon TheproofofLinandLiuin[5]isnotentirelysuitableforourwork,becauseitinvolvesalllocalextremaof thefunction v =(4 ∂2) 1uonR,andthusis notlocal. Forourwork,wehaveto localizethe estimates. − x − Therefore, we need to modify a little the proof of Lin and Liu. We do this first for a single peakon. 3 Theorem 3.1 (Stability of Peakons). Let u ([0,T[), with 0 < T + , be a solution of the DP ∈ X ≤ ∞ equation and ϕ bethe peakon defined in (1.8), traveling to theright at the speedc>0. There exist C >0 c and ε >0 only depending on the speed c>0, such that if 0 y =(1 ∂2)u +(R) (3.1) 0 − x 0 ∈M and u ϕ ε2, with 0<ε<ε , (3.2) 0 c 0 k − kH ≤ then u(t, ) ϕ ( ξ (t)) C√ε, t [0,T[, (3.3) c 1 k · − ·− kH ≤ ∀ ∈ where ξ (t) R is any point where the function v(t, )=(4 ∂2) 1u(t, ) attains its maximum. 1 ∈ · − x − · To prove this theorem we first need the following lemma that enables to control the distance of E(u) and F(u) to respectively E(ϕ ) and F(ϕ ). c c Lemma 3.1 (Control of Distances Between Energies). Let u H1(R) with y =(1 ∂2)u +(R). If ∈ − x ∈M u ϕ ε2, then c k − kH ≤ E(u) E(ϕ ) O(ε2) (3.4) c | − |≤ and F(u) F(ϕ ) O(ε2), (3.5) c | − |≤ where O() only depends on the speed c. · Proof. For the first estimate, applying triangular inequality, and using that u ϕ ε2 and c ϕ =c/√3, we have k − kH ≤ c k kH E(u) E(ϕ ) = u ϕ ( u + ϕ ) c c c | − | |k kH−k kH| k kH k kH u ϕ ( u ϕ +2 ϕ ) c c c ≤k − kH k − kH k kH 2c ε2 ε2+ ≤ √3 (cid:18) (cid:19) O(ε2). ≤ Forthe secondestimate, applyingthe Ho¨lderinequality,andusing that u ϕ ε2 and(2.3), we c k − kH ≤ have F(u) F(ϕ ) u3 ϕ3 | − c |≤ R − c Z (cid:12) (cid:12) (cid:12)u ϕ ((cid:12)u2+uϕ +ϕ2) ≤ R| − c| c c Z 1/2 ≤ku−ϕckL2(R) R(u2+uϕc+ϕ2c)2 (cid:18)Z (cid:19) 1/2 =ku−ϕckL2(R) R(u4+2u3ϕc+3u2ϕ2c +2uϕ3c +ϕ4c) (cid:18)Z (cid:19) u ϕc L2(R) ≤k − k 1/2 8 1 · 4CS2kuk4L2(R)+4cCSkuk2L2(R)+3c2kuk2L2(R)+ 3c3CSkukL2(R)+ 2c4 (cid:18) (cid:19) O(ε2), ≤ where we also use that the L2 norm of u is bounded and the following measures of peakon: 2 1 kϕckL∞(R) =c, kϕckL3(R) = 3 3c and kϕckL4(R) = √42c. r 4 This proves the lemma. (cid:3) Now, to proveTheorem 3.1, by the conservationof E(), F() and the continuity of the map t u(t) from [0,T[ to H1(R) ֒ (since L2 and u L2(R)· u· H1(R)), it suffices to prove that f7→or any function u H1(R) sa→tisfHying y =(H1 ≃∂2)u k+k(R), (3≤.4)kankd (3.5), if ∈ − x ∈M inf u ϕ ( ξ) ε1/4, (3.6) c ξ Rk − ·− kH ≤ ∈ then u ϕ ( ξ ) C√ε, (3.7) c 1 k − ·− kH ≤ where ξ R is any point where the function v =(4 ∂2) 1u attains its maximum. 1 ∈ − x − WedividetheproofofTheorem3.1intoasequenceoflemmas. Inthesequel,wewillneedtointroduce the following smooth-peakons defined for all x R by: ∈ c c ρc(x)=cρ(x)=(4−∂x2)−1ϕc(x)= 3e−|x|− 6e−2|x|. (3.8) One can check that ρ H3(R) ֒ C2(R) (by the Sobolev embedding) since ϕ H1(R). Indeed, we c c ∈ → ∈ have (1+ω2)3 ρ 2 = ϕ (ω)2 (1+ω2)ϕ (ω)2 ϕ 2 =2c2. (3.9) k ckH3(R) R (4+ω2)2| c | ≤ R | c | ≤k ckH1(R) Z Z Moreover,ρ is a positive even function which decays to 0 at infinity, and admits a single maximum c/6 c b b at point 0 (see Fig. 1a-1c). Lemma 3.2 (Uniform Estimates). Let u H1(R) with y = (1 ∂2)u +(R), and ξ R. If ∈ − x ∈ M ∈ u ϕ ( ξ) ε1/4, then c k − ·− kH ≤ u ϕc( ξ) L∞(R) O(ε1/8) (3.10) k − ·− k ≤ and v ρc( ξ) L∞(R) O(ε1/4), (3.11) k − ·− k ≤ where v =(4 ∂2) 1u and ρ is defined in (3.8). − x − c Proof. For the second estimate, applying the Ho¨lder inequality and using assumption, we get for all x R, ∈ 1 v(x) ρ (x ξ) e 2x′ u(x x) ϕ [(x x) ξ] dx c − | | ′ c ′ ′ | − − |≤ 4 R | − − − − | Z 1/2 1/2 1 e 4x′ dx u(x) ϕ (x ξ)2dx − | | ′ ′ c ′ ′ ≤ 4 R R| − − | (cid:18)Z (cid:19) (cid:18)Z (cid:19) 1 u ϕ ( ξ) c ≤ 2√2k − ·− kH O(ε1/4). ≤ For the first estimate, note that the assumption y =(1 ∂2)u 0 implies that u=(1 ∂2) 1y 0 and satisfies (2.2). Then, applying triangular inequality, a−ndxusin≥g that ϕ = ϕ on R −andx(2−.3),≥we | ′c| c have u ϕc( ξ) H1(R) u H1(R)+ ϕc H1(R) k − ·− k ≤k k k k 2 u L2(R)+2 ϕc L2(R) ≤ k k k k 2 u ϕc( ξ) L2(R)+4 ϕc L2(R) ≤ k − ·− k k k O(ε1/4)+O(1). ≤ 5 Now, applying the Gagliardo-Nirenberginequality and using assumption, we obtain ku−ϕc(·−ξ)kL∞(R) ≤CGku−ϕc(·−ξ)k1L/22(R)ku−ϕc(·−ξ)k1H/12(R) O(ε1/8) O(ε1/8)+O(1) ≤ O(ε1/8).(cid:16) (cid:17) ≤ This proves the lemma. (cid:3) 1 0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0 −10 −8 −6 −4 −2 0 2 4 6 8 10 (a)ϕ(x)andρ(x)profiles. 1 2 0.8 1.8 0.6 1.6 0.4 1.4 0.2 1.2 0 1 −0.2 0.8 −0.4 0.6 −0.6 0.4 −0.8 0.2 −1 0 −10 −8 −6 −4 −2 0 2 4 6 8 10 −10 −8 −6 −4 −2 0 2 4 6 8 10 (b)ϕ′(x)andρ′(x)profiles. (c)(1−∂x2)ϕ(x)=2δ0 and(1−∂x2)ρ(x)=(1/2)e−2|x| profiles Figure 1: Variation of peakon and smooth-peakon at initial time with the speed c=1. Lemma 3.3 (Quadratic Identity; See [5]). For any u L2(R) and ξ R, it holds ∈ ∈ c E(u) E(ϕ )= u ϕ ( ξ) 2 +4c v(ξ) , (3.12) c c − k − ·− kH − 6 (cid:16) (cid:17) where v =(4−∂x2)−1u and c/6=ρc(0)=maxx∈Rρc(x−ξ). 6 Sketch of proof. The prooffollows by direct computation, with the aidof two integrationby parts, and using that (1 ∂2)ϕ ( ξ)=2cδ , where δ denotes the Dirac mass applied at point ξ. (cid:3) − x c ·− ξ ξ Letu H1(R)withy =(1 ∂2)u +(R),andassumethatthereexistsξ Rsuchthat(3.6)holds for some ∈ξ R. We consider n−owxthe∈inMterval in which the peakon (respectivel∈y the smooth-peakon) is ∈ concentrated, and we will decompose this interval according to the variation of v = (4 ∂2) 1u in the − x − following way: we set c c α=sup x<ξ, v(x)= and β =inf x>ξ, v(x)= . (3.13) 2400 2400 n o n o According to Lemma 3.2, we know that v is close to ρ ( ξ) in L norm with ρ (0)=c/6. Therefore v c ∞ c ·− musthaveatleastonelocalmaximumon[α,β]. Assumethaton[α,β]thefunctionv admitsk+1points (ξ )k+1 with local maximal values for some integer k 0, where ξ is the first local maximum point and j j=1 ≥ 1 ξ the last local maximum point2. Then between ξ and ξ , the function v admits k points (η )k k+1 1 k+1 j j=1 with local minimal values. We rename α=η and β =η so that it holds 0 k+1 η <ξ <η <...<ξ <η <ξ <η <...<η <ξ <η . (3.14) 0 1 1 j j j+1 j+1 k k+1 k+1 Let M =v(ξ ), j =1,...,k+1, and m =v(η ), j =1,...,k. (3.15) j j j j By construction v (x) 0, x [η ,ξ ], j =1,...,k (3.16) x j 1 j ≥ ∀ ∈ − and v (x) 0, x [ξ ,η ], j =1,...,k+1. (3.17) x j j ≤ ∀ ∈ We claim that c v(x) , x R [η ,η ], (3.18) 0 k+1 ≤ 300 ∀ ∈ \ c u(x) , x R [η ,η ], (3.19) 0 k+1 ≤ 300 ∀ ∈ \ and there exits C >0 such that 0 [η ,η ] [ξ C ,ξ+C ]. (3.20) 0 k+1 0 0 ⊂ − Indeed, for some 0<ε 1 fixed, using (3.11) we have ≪ c c ρ (η ξ)=v(η )+O(ε1/4)= +O(ε1/4) . c 0 0 − 2400 ≤ 350 Please note that, we abuse notation by writing that the difference between v and ρ ( ξ) is equal to c ·− O(ε1/4). Therefore, using that ρ ( ξ) is increasing on ] ,ξ], it holds for all x ] ,η [, c 0 ·− −∞ ∈ −∞ c c v(x)=ρ (x ξ)+O(ε1/4) +O(ε1/4) . c − ≤ 350 ≤ 300 Proceeding in the same way for x ]η ,+ [, we obtain (3.18). k+1 One can remark that for all x ∈R, ∞ ∈ c c ϕc(x) 6ρc(x)=ce−|x| 6 e−|x| e−2|x| = ce−|x|+ce−2|x| 0. (3.21) − − 3 − 6 − ≤ (cid:16) (cid:17) Thus, combining (3.10), (3.21) and proceeding as for the estimate (3.18), we infer (3.19). Finally, from (3.11) we have c c ρ (η ξ)=v(η )+O(ε1/4)= +O(ε1/4) . c 0 0 − 2400 ≥ 3000 2Inthecaseofaninfinitecountable numberoflocalmaximalvalues,theproofisexactlythesame. 7 Therefore, since ρ = (c/3)e (c/6)e 2 and that x (1/3)e x (1/6)e 2x is a positive even c −|·| − |·| −| | − | | function decreasing to 0 on R (−see Fig. 1a), there exists7→a universal co−nstant C > 0 such that (3.20) + 0 holds. We now are ready to establish the connection between the conservation laws. Please note that, we will change the order of the extrema of v =(4 ∂2) 1u while keeping the same notations as in (3.15). − x − Lemma 3.4 (Connection Between E() and the Local Extrema of v). Let u H1(R) and v = (4 ∂2) 1u H3(R). Define the function g·by ∈ − x − ∈ 2v+v 3v , x<ξ , xx x 1 − 2v+v +3v , ξ <x<η , g(x)= xx x j j j =1,...,k. (3.22) 2v+vxx−3vx, ηj <x<ξj+1, 2v+v +3v , x>ξ , xx x k+1 Then it holds k k g2(x)dx=E(u) 12 M2 m2 . (3.23) R − j+1− j Z j=0 j=1 X X Proof. We have g2(x)dx= ξ1 g2(x)dx+ k ξj+1g2(x)dx+ +∞g2(x)dx. (3.24) ZR Z−∞ Xj=1Zξj Zξk+1 For j =1,...,k, ξj+1 ηj ξj+1 g2(x)(x)dx = (2v+v +3v )2+ (2v+v 3v )2 xx x xx x − Zξj Zξj Zηj =J +I. Let us compute I, ξj+1 I = 4v2+v2 +9v2+4vv 12vv 6v v xx x xx− x− x xx Zηj (cid:0) (cid:1) ξj+1 ξj+1 ξj+1 ξj+1 = 4v2+v2 +9v2 +4 vv 12 vv 6 v v xx x xx− x− x xx Zηj Zηj Zηj Zηj (cid:0) (cid:1) ξj+1 = 4v2+v2 +9v2 +I +I +I . xx x 1 2 3 Zηj (cid:0) (cid:1) Applying integration by parts and using that v (ξ )=v (η )=0, we get x j x j ξj+1 ξj+1 ξj+1 I = 4 v2, I = 6 ∂ (v2)= 6v2(ξ )+6v2(η ) and I = 3 ∂ (v2)=0. 1 − x 2 − x − j+1 j 3 − x x Zηj Zηj Zηj Therefore ξj+1 I = 4v2+5v2+v2 6v2(ξ )+6v2(η ). (3.25) x xx − j+1 j Zηj (cid:0) (cid:1) Similar computations lead to ηj ξ1 ξ1 J = 4v2+5v2+v2 6v2(ξ )+6v2(η ), g2(x)dx= 4v2+5v2+v2 6v2(ξ ) (3.26) x xx − j j x xx − 1 Zξj (cid:0) (cid:1) Z−∞ Z−∞(cid:0) (cid:1) 8 and + + ∞g2(x)dx= ∞ 4v2+5v2+v2 6v2(ξ ). (3.27) x xx − k+1 Zξk+1 Zξk+1 (cid:0) (cid:1) Adding I and J, and summing over j 1,...,k , we obtain ∈{ } ξk+1 ξk+1 k k k g2(x)dx= 4v2+5v2+v2 6 v2(ξ ) 6 v2(ξ )+12 v2(η ). (3.28) x xx − j+1 − j j Zξ1 Zξ1 j=1 j=1 j=1 (cid:0) (cid:1) X X X The lemma follows by combining (3.24) and (3.26)-(3.28). (cid:3) Lemma 3.5 (Connection Between F() and the Local Extrema of v). Let u H1(R) and v = (4 ∂2) 1u H3(R). Define the function h·by ∈ − x − ∈ v 6v +16v, x<ξ , xx x 1 − − v +6v +16v, ξ <x<η , h(x)= − xx x j j j =1,...,k. (3.29) −vxx−6vx+16v, ηj <x<ξj+1, v +6v +16v, x>ξ , xx x k+1 − Then it holds k k h(x)g2(x)dx=F(u) 144 M3 m3 . (3.30) R − j+1− j Z j=0 j=1 X X Proof. We have h(x)g2(x)dx= ξ1 h(x)g2(x)dx+ k ξj+1h(x)g2(x)dx+ +∞h(x)g2(x)dx. (3.31) ZR Z−∞ Xj=1Zξj Zξk+1 For j =1,...,k, ξj+1 ηj h(x)g2(x)dx= ( v 6v +16v)(2v+v 3v )2 xx x xx x − − − Zξj Zξj ξj+1 + ( v +6v +16v)(2v+v +3v )2 xx x xx x − Zηj =J +I. Let us compute I, ξj+1 ξj+1 ξj+1 I = v3 +12vv2 +64v3+60v2v 54 v3+27 v2v − xx xx xx − x x xx Zηj Zηj Zηj (cid:0) (cid:1) ξj+1 ξj+1 ξj+1 108 vv v 216 v2v +216 vv2 − x xx− x x Zηj Zηj Zηj ξj+1 = v3 +12vv2 +64v3+60v2v − xx xx xx Zηj (cid:0) (cid:1) ξj+1 54 v3+I +I +I +I . − x 1 2 3 4 Zηj Applying integration by parts and using that v (ξ )=v (η )=0, we get x j x j ξj+1 ξj+1 ξj+1 I =9 ∂ (v3)=0, I =54 v3, I = 72 ∂ (v3)= 72v3(ξ )+72v3(η ) 1 x x 2 x 3 − x − j+1 j Zηj Zηj Zηj 9 and ξj+1 ξj+1 I =108 ∂ (v2)v = 108 v2v . 4 x x xx − Zηj Zηj Therefore ξj+1 I = v3 +12vv2 +64v3+60v2v 72v3(ξ )+72v3(η ). (3.32) − xx xx xx − j+1 j Zηj (cid:0) (cid:1) Similar computations lead to ηj J = v3+12vv2 +64v3+60v2v 72v3(ξ )+72v3(η ), (3.33) − x xx xx − j j Zξj (cid:0) (cid:1) ξ1 ξ1 h(x)g2(x)dx= v3+12vv2 +64v3+60v2v 72v3(ξ ) (3.34) − x xx xx − 1 Z−∞ Z−∞(cid:0) (cid:1) and + + ∞h(x)g2(x)dx= ∞ v3+12vv2 +64v3+60v2v 72v3(ξ ). (3.35) − x xx xx − k+1 Zξk+1 Zξk+1 (cid:0) (cid:1) Adding (3.25) and (3.33), and summing over j 1,...,k , we obtain ∈{ } ξk+1 ξk+1 k h(x)g2(x)dx= v3+12vv2 +64v3+60v2v 72 v3(ξ ) − x xx xx − j+1 Zξ1 Zξ1 j=1 (cid:0) (cid:1) X k k 72 v3(ξ )+72 v3(η ). (3.36) j j − j=1 j=1 X X The lemma follows by combining (3.31) and (3.34)-(3.36). (cid:3) Lemma 3.6 (Connection Between E() and F()). Let u H1(R), with y = (1 ∂2)u +(R), that satisfies (3.6) for some ξ R. Assume·that v =(·4 ∂2) 1u∈satisfies (3.13)-(3.20)−, wxith lo∈caMl extrema on ∈ − x − [η ,η ] arranged in decreasing order in the following way: 0 k+1 M M ... M 0, m m ... m 0, M m , j =1,...,k. (3.37) 1 2 k+1 1 2 k j+1 j ≥ ≥ ≥ ≥ ≥ ≥ ≥ ≥ ≥ There exists ε >0 only depending on the speed c, such that if 0<ε<ε , then it holds 0 0 1 1 M3 E(u)M + F(u) 0. (3.38) 1 − 4 1 72 ≤ Proof. The key is to show that h 18M on R. Note that by (3.6) we know that 18M c/4. We 1 1 ≤ ≥ rewrite the function h as v 6v +16v, x<η , xx x 0 − − ∂2+3∂ +2 v 3v +18v, η <x<ξ , − x x − x 0 1 h(x)= −−(cid:0)(cid:0)∂∂xx22+−33∂∂xx++22(cid:1)(cid:1)vv−+33vvxx++1188vv,, ηξjj <<xx<<ηξjj+,1, j =1,...,k. First, one can remarktha−−t(cid:0)v(cid:0)f∂oxxrx2+a−ll63xv∂xx++R12,6(cid:1)(cid:1)vv,+x3v>xη+k+118,v, ξk+1 <x<ηk+1, ∈ v(x)= e−2x x e2x′u(x)dx + e2x +∞e 2x′u(x)dx ′ ′ − ′ ′ 4 4 Z−∞ Zx 10