• Aucun résultat trouvé

Analytic continuation of the resolvent of the Laplacian and the dynamical zeta function

N/A
N/A
Protected

Academic year: 2023

Partager "Analytic continuation of the resolvent of the Laplacian and the dynamical zeta function"

Copied!
6
0
0

Texte intégral

(1)

Partial Differential Equations

Analytic continuation of the resolvent of the Laplacian and the dynamical zeta function

Vesselin Petkov

a

, Latchezar Stoyanov

b

aUniversité Bordeaux I, Institut de mathématiques, 351, cours de la Libération, 33405 Talence, France bUniversity of Western Australia, School of Mathematics and Statistics, Perth, WA 6009, Australia

Received 24 September 2007; accepted 8 October 2007 Available online 7 November 2007

Presented by Gilles Lebeau

Abstract

Lets0<0 be the abscissa of absolute convergence of the dynamical zeta function Z(s)for several disjoint strictly convex compact obstaclesKi⊂RN, i=1, . . . , κ0, κ03 and letRχ(z)=χ (Dz2)1χ , χC0 (RN), be the cut-off resolvent of the Dirichlet Laplacian−DinΩ=RN\κ0

i=1Ki. We prove that there existsσ2< s0such thatZ(s)is analytic for(s) σ2 and the cut-off resolventRχ(z)has an analytic continuation for(z) <−iσ2, |(z)|C.To cite this article: V. Petkov, L. Stoyanov, C. R. Acad. Sci. Paris, Ser. I 345 (2007).

©2007 Académie des sciences. Published by Elsevier Masson SAS. All rights reserved.

Résumé

Prolongement analytique de la résolvante du Laplacien et de la fonction zeta dynamique.Soits0<0 l’abscisse de conver- gence absolue de la fonciton zeta dynamiqueZ(s)pour des obstacles compacts, disjoints et strictement convexesKi⊂RN, i= 1, . . . , κ0, κ03 et soitRχ(z)=χ (−Dz2)1χ , χC0(RN), la résolvante tronquée du Laplacien de DirichletDdans Ω=RN\κ0

i=1Ki. On prouve qu’il existeσ2< s0tel queZ(s)est analytique pour(s)σ2et la résolvante tronquéeRχ(z) admet un prolongement analytique pour(z) <−iσ2, |(z)|C.Pour citer cet article : V. Petkov, L. Stoyanov, C. R. Acad. Sci.

Paris, Ser. I 345 (2007).

©2007 Académie des sciences. Published by Elsevier Masson SAS. All rights reserved.

Version française abrégée

SoitK=K1K2∪· · ·∪Kk0,Ki⊂RN, N2,sont des domaines compacts, disjoints et strictement convexes ayant des frontières Γi =∂Ki et κ03. SoitΩ =RN\K etΓ =∂K. On suppose que K satisfait la condition suivante :

(H) Pour chaque coupleKi, Kj de différentes composantes connexes deKl’enveloppe convexe deKiKj n’a pas de points communs avec les autres composantes connexes deK.

E-mail addresses:petkov@math.u-bordeaux1.fr (V. Petkov), stoyanov@maths.uwa.edu.au (L. Stoyanov).

1631-073X/$ – see front matter ©2007 Académie des sciences. Published by Elsevier Masson SAS. All rights reserved.

doi:10.1016/j.crma.2007.10.019

(2)

Étant donné un rayon périodique réfléchissantγΩ avecmγ réflexions, soitTγ la période primitive (longueur) de γ et soit Pγ l’application de Poincaré linéaire associée à γ (cf. [8]). Notons par λi,γ, i =1, . . . , N−1, les valeurs propres dePγ telles que|λi,γ|>1 et désignons parP l’ensemble de rayons primitifs périodiques. Soitδγ=

12log(λ1,γ· · ·λN1,γ),rγ=0 simγ est pair etrγ=1 simγ est impair. On considère la fonction zeta dynamique Z(s)=

m=1

1 m

γ∈P

(−1)mrγem(sTγ+δγ).

On voit facilement qu’il existe une abscisse de convergence absolues0∈Rtelle que pour(s) > s0la sérieZ(s) est absolument convergente. D’autre part, en utilisant la dymanique symbolique et les résultats de [7], on conclut queZ(s)est méromorphe pour(s) > s0a, a >0 (cf. [4]). En suivant les résultats récents (cf. [9] pourN=2 et [10] pourN=3 sous certaines conditions) on sait qu’il existe 0< < atel que la fonction zeta dynamiqueZ(s) admet un prolongement analytique pour(s) > s0. On considère maintenant pour(s) <0 la résolvante tronquée Rχ(z)=χ (Dz2)1χ:L2(Ω)L2(Ω), oùχC0(RN), χ=1 surKet−Dest le Laplacien de Dirirchlet dansΩ=RN\K. La résolvanteRχ(z)admet un prolongement mémororphe dansCpourNimpair et dansC\{iR+} pourN pair avec des pôleszj, (zj) >0. On se propose d’étudier la liaison entre les prolongements analytiques de Z(s)etRχ(z). Le cass0>0 est plus facile car on sait que pour−is0(z) <0 la résolvante tronquéeRχ(z)est analytique [6]. Dans la suite on suppose ques0<0. Sous l’hypothèse s0<0, Ikawa [3] a démontré que pour tout >0 il existeC>0 tel queRχ(z)est analytique pour(z) <i(s0+),|(z)|C. Un résultat similaire pour un problème du contrôle a été établi par Burq [1]. La fonction zeta dynamiqueZ(s)est liée aux périodes des rayons périodiques et formellementZ(s)ne contient pas une information sur la dynamique de rayons dans un voisinage de l’ensemble ‘non-wandering’ (captif). Dans cette Note on examine le cas(s) < s0 en exploitant les propriétés spectrales de l’opérateur de RuelleLs(cf. Section 2). Notre résultat principal est le suivant :

Théorème 0.1.Soits0<0. Supposons que l’opérateur de RuelleLssatisfait les estimations(6). Alors il existeσ2< s0 tel queZ(s)est analytique pour(s) > σ2et la résolvante tronquéeRχ(z)est analytique pour

(z) <−iσ2, (s)C.

Les estimations (6) sont un analogue aux estimations de Dolgopyat [2]. Ces estimaitons ont été démontrées pour N=2 dans [9] et pour N 3 sous certaines conditions dans [10]. On espère que (6) sont valables pour N3 sans aucune restriction. Notons qu’il y a quelques ans, Ikawa [5] a annoncé un résultat concernant le prolongement analytique deRχ(z)dans un domaine−iD, où

D=

s∈C: (s) > s0(s)α,(s)C,0< α <1

en imposant des conditions fortes sur le comportement de la fonction proprewde l’opérateur de Ruelle associée à la valeur propre maximale et un prolongement analytique deZ(s)dansD. De plus, il suppose qu’on ait l’estimation

|Z(s)||s|1, 0< <1, s∈D. A notre connaissance la preuve de ce résultat n’a pas été publiée ailleurs.

1. Introduction

LetK be a subset ofRN, N2, of the formK=K1K2∪ · · · ∪Kk0, whereKi are compact strictly convex disjoint domains inRNwithCboundariesΓi=∂Ki andk03. SetΩ=RN\KandΓ =∂K. We assume thatK satisfies the following (no-eclipse) condition:

(H) For every pairKi,Kj of different connected components ofKthe convex hull ofKiKjhas no common points with any other connected component ofK.

With this condition, the billiard flowφt defined on the cosphere bundleS(Ω)in the standard way is called an open billiard flow. Given a periodic reflecting rayγΩ withmγ reflections, denote byTγ the primitive period (length) ofγ and byPγ the linear Poincaré map associated toγ(see [8]). Letλi,γ, i=1, . . . , N−1, be the eigenvalues ofPγ

(3)

with|λi,γ|>1 and letP be the set of primitive periodic rays. ForγP setδγ = −12log(λ1,γ· · ·λN1,γ),rγ=0 if mγ is even andrγ =1 ifmγ is odd. Consider the dynamical zeta function

Z(s)= m=1

1 m

γ∈P

(−1)mrγem(sTγ+δγ).

It is easy to show that there exists an abscissa of absolute convergences0∈Rsuch that for(s) > s0the seriesZ(s)is absolutely convergent. On the other hand, using symbolic dynamics and the results of [7], we deduce thatZ(s)is mero- morphic for(s) > s0a, a >0 (see [4]). According to some recent results (see [9] forN=2 and [10] forN3 un- der some additional conditions) there exists 0< < aso that the dynamical zeta functionZ(s)admits an analytic con- tinuation for(s)s0. For(z) <0 consider the cut-off resolventRχ(z)=χ (Dz2)1χ:L2(Ω)L2(Ω), whereχC0(RN), χ =1 onKand−Dis the Dirichlet Laplacian inΩ=RN\K. The cut-off resolventRχ(z) has a meromorphic continuation inCforN odd and inC\ {iR+}forN even with poleszj such that(zj) >0. The analytic properties and the estimates ofRχ(z)play a crucial role in many problems related to the local energy decay, distribution of the resonances etc. We study the link between the analytic continuations ofZ(s)andRχ(z). The case s0>0 is much easier, since we know that for−is0(z) <0 the cut-off resolventRχ(z)is analytic (see [6]).

In the followingwe assume thats0<0. The problem is to examine the link between the analyticity ofZ(s)for (s) > s0and the behavior ofRχ(z)for 0(z) <is0. Assumings0<0, Ikawa [3] proved that for every >0 there existsC>0 so that the cut-off resolventRχ(z)is analytic for(z) <i(s0+),|(z)|C. A similar result for a control problem has been established by Burq [1]. The proofs in [3] and [1] are based on the construction of an asymptotic solutionUM(x, s;k)with boundary datam(x;k)=eikϕ(x)h(x), k∈R, k1, whereϕis a phase function (∇ϕ =1) andhC(Γ )has a small support. More precisely,UM(·,s;k)isC(Ω)-valued holomorphic function¯ inD0= {s∈C: (s) > s0}, and we have

(s2)UM(·,s;k)=0 forxΩ,(s) > s0, (1)

UM(·,s;k)L2(Ω) if(s) >0, (2)

UM(x, s;k)=m(x, k)+rM(x, s;k) onΓ, (3)

where, forrM(x, s;k)andsD0, |s+ik|1, we have the estimates rM(·,s;k)

Cp(Γ )CpkM+p

ϕCM2+p+2(Γ )+1 hCM2+p+2(Γ ),p∈N. (4) The functionUM(x, s;k)is given by a finite sum of series having the form

n=0

|j|=n+2,jn+2=l

M

q=0

ej(x) 2q ν=0

aj,q,ν(x, s;k)(s+ik)ν (ik)q, (5)

wherej=(j1, . . . , jm), ji ∈ {1, . . . , κ0}are configurations,|j| =m, ϕj(x)are phase functions and the amplitudes aj,q,ν(x, s;k)depend ons∈Candk∈R. The main difficulty is to establish the summability of these series and to obtain for(s) > s0suitableCpestimates of their traces onΓ. The absolute convergence ofZ(s)makes it possible to establish the absolute convergence of the series in (5) and to get crude estimates leading to (1)–(4). The dynamical zeta functionZ(s)is related to the periods of periodic rays and formally fromZ(s)we get no information about the dynamics of all rays in aneighborhood of the non-wandering(trapped)set. In this Note we study the case(s) < s0 by means of the Ruelle operatorLs (see Section 2 for the definition). Our main result is the following:

Theorem 1.1.Lets0<0. Assume that for the Ruelle operatorLs the estimates(6)hold. Then there existsσ2< s0

such thatZ(s)is analytic for(s) > σ2and the cut-off resolventRχ(z)is analytic for (z) <−iσ2, (s)C.

The estimates (6) are analogous to Dolgopyat’s estimates in [2]. For open billiard flows (6) have been established in [9] forN=2 and under some conditions in [10] forN3. We expect that (6) hold forN3 without any restrictions.

(4)

Several years ago, Ikawa [5] announced a result concerning an analytic continuation ofRχ(z)in a domain−iD, where

D=

s∈C: (s) > s0(s)α,(z)C,0< α <1

assuming some strong conditions on the behavior of the eigenfunctionwof the corresponding Ruelle operator re- lated to its maximal eigenvalue as well as an analytic continuation of Z(s) in D combined with an estimate

|Z(s)||s|1, 0< <1, s∈D. To our best knowledge a proof of the above result of Ikawa has not been published anywhere.

2. Ruelle operator Introduce the spaces

ΣA=

(. . . , ηm, . . . , η1, η0, η1, . . . , ηm, . . .): 1ηjκ0, ηj∈N, ηj=ηj+1for allj∈Z , ΣA+=

0, η1, . . . , ηm, . . .): 1ηjκ0, ηj∈N, ηj=ηj+1for allj1 . We define the operatorσ:ΣA+ΣA+by(σ ξ )i=ξi+1, i∈N. GivenξΣA, let

. . . , P2(ξ ), P1(ξ ), P0(ξ ), P1(ξ ), P2(ξ ), . . .

be the successive reflection points of the unique billiard trajectory in the exterior ofKsuch thatPj(ξ )Kξj for all j∈Z. Setf (ξ )= P0(ξ )P1(ξ ). Following [4], one constructs a sequence{ϕξ,j}j=−∞ of phase functions such that for eachj,ϕξ,j is defined and smooth in a neighborhoodUξ,j of the segment[Pj(ξ ), Pj+1(ξ )]inΩ and

(i) ∇ϕξ,j(Pj(ξ ))=PPj+1j+1(ξ )(ξ )PPjj(ξ )(ξ ), (ii) ϕξ,j =ϕξ,j+1onΓξj+1Uξ,jUξ,j+1,

(iii) for eachxUξ,j the surfaceCξ,j(x)= {yUξ,j:ϕξ,j(y)=ϕξ,j(x)}is strictly convex with respect to its normal field∇ϕξ,j. For anyyUξ,j denote byGξ,j(y)theGauss curvatureofCξ,j(x)aty.

Now defineg:ΣA→Rby g(ξ )= 1

N−1 lnGξ,0(P1(ξ )) Gξ,0(P0(ξ )).

By Sinai’s Lemma, there existf ,˜ g˜depending on future coordinates only andχ1, χ2such that f (ξ )= ˜f (ξ )+χ1(ξ )χ1(σ ξ ), g(ξ )= ˜g(ξ )+χ2(ξ )χ2(σ ξ ), ξΣA.

Settingr(ξ, s)˜ = −sf (ξ )˜ + ˜g(ξ )+iπ, we define theRuelle transfer operator Ls:C(ΣA+)C(ΣA+)byLsu(ξ )=

σ η=ξer(η,s)˜ u(η) for any continuous (complex-valued) function uon ΣA+ and any ξΣA+. For our analysis the Dolgopyat type estimates [2] for the norms of Lns play a crucial role. Following the results in [9,10] there exist constantsC >0, σ0< s0and 0< ρ <1 so that fors=τ +it withτ σ0andn=p[log|t|] +l, p∈N, 0l [log|t|] −1, we have

Lnsp[log|t|]elPr(τf˜+ ˜g), |t|t0, (6) Pr(G)being the topological pressure of the functionG(see [7]).

3. Idea of the proof of Theorem 1.1

Fixl∈ {1, . . . , κ0}. Given a phase functionϕ(x)and an amplitudeh(x)C(Γ ), we wish to construct an asymp- totic solutionUM(x, s;k)which is a holomorphic function for(s)σ2andUM has properties similar to (1)–(4).

The first approximation ofUMis an infinite sum w0,l(x,−is)=

n=0

|j|=n+2.jn+2=l

uj(x,−is),

(5)

whereuj(x,is)=(−1)m1ej(x)aj(x)is related to a configurationj= {j1, . . . , jm}by using successive phase functionsϕj(x)and amplitudes aj(x)determined by the transport equation (see [5]). To justify the convergence of this series, we need to compare the general term with a suitable composition of operators related to the dynamics. Let μ=0=1, μ1, . . .)ΣA+. It follows from [3] that there exists a unique pointy(μ)Γ1such that the rayγ (y, ϕ) issued from a pointy(μ)in direction∇ϕ(y(μ))follows the configurationμ. LetQ0(μ)=y(μ), Q1(μ), . . . ,be the consecutive reflection points of this ray. Definefj+(μ)= Qj(μ)Qj+1(μ), and

gj+(μ)= 1

N−1lnGϕμ,j(Qj+1(μ)) Gϕμ,j(Qj(μ)) <0,

whereGϕμ,j(y)denotes theGauss curvatureof the surfaceCϕμ,j(x)= {z: ϕ01,...,μj)(z)=ϕ01,...,μj)(x)}aty. We define an extensione:ΣA+ΣA. Fors∈CandξΣA+withξ0=1, following [5], set

φ+(ξ, s)= n=0

s f

σne(ξ )fn+(ξ ) +

g

σne(ξ )gn+(ξ ) .

Formally, we defineφ+(ξ, s)=0 whenξ0=1, thus obtaining a functionφ+:ΣA+×C→C. Setχ (ξ, s)= −1(ξ )+ χ2(ξ )and for anys∈Cdefine the operatorGs:C(ΣA+)C(ΣA+)by

Gsv(ξ )=

σ η=ξ, η0=1

eφ+(η,s)+χ (e(η),s)sf (η)˜ + ˜g(η)v(η), vC(ΣA+), ξΣA+.

Fix an arbitraryl∈ {1, . . . , κ0}and an arbitrary pointx0Γl. Define the functionφ(x0; ·,·):ΣA×C→C (depending onlas well) as follows. First, setφ(x0;η, s)=0 ifη0=l. Next, assume thatηΣA satisfiesη0=l.

There exists a unique billiard trajectory inΩwith successive reflection pointsP˜j(x0;η)∂Kηj (−∞< j0) such thatx0= ˜P1(x0;η)+tϕη(P˜1(x0;η))for somet >0. In general the segment[ ˜P1(x0;η), x0]may intersect the interior ofKl. If this is the case, set againφ(x0;η, s)=0. Otherwise, denoteP˜0(x0;η)=x0and for anyj <0 set

fj(x0;η)=P˜j+1(x0;η)− ˜Pj(x0;η), gj(x0;η)= 1

N−1 lnGη,j(P˜j+1(x0;η)) Gη,j(P˜j(x0;η)) , and defineφ(x0;η, s)= −s−∞

j=−1[f (σj(η))fj(x0;η)] +−∞

j=−1[g(σj(η))gj(x0;η)]. Next, similarly to [5], introduce the operatorMn,s(x0):C(ΣA+)C(ΣA+)by

Mn,s(x0)v (ξ )=

σ η=ξ

eφ(x0;σn+1e(η),s)χ (σn+1e(η),s)sf (η)˜ + ˜g(η)v(η), vC(ΣA+), ξΣA+. Introduce the functionvs(ξ )=esϕ(Q0(ξ ))h(Q0(ξ ))ifξ0=1 andvs(ξ )=0 otherwise and define the norms

fΓ ,p=max

xΓ max

a(1),...,a(p)TxΓ

(Da(1)· · ·Da(p)f )(x), fΓ ,(p)= max

0jpfΓ ,j

wherea(j ) =1 for allj =1, . . . , p.

Theorem 3.1.There exist global constantsC >0, c >0, a∈(0,1]andθ(0,1)depending only onKsuch that for any choice ofl∈ {1, . . . , κ0}the following holds:For any integersp1andn1, anyξΣA+withξ0=land any s∈Cwith(s)s0awe have

LnsMn,s(·)Gsv˜s (ξ )

|j|=n+2,jn+2=l

uj(·,is) Γ ,p

C(θ+ca)neC[|(s)|(1+ϕΓ ,0)+∇ϕΓ ,(1)] p

j=0

|s|∇ϕΓ ,j+ ∇ϕΓ ,j+1 j+1

hΓ ,pj. (7)

(6)

A similar estimate holds forp=0; in this case the sum in the right-hand side of (7) has to be replaced by[(|s| +

ϕΓ ,(1))hΓ ,0+ hΓ ,(1)]. Applying the casep=0, we reduce the convergence ofw0,lto the summability of the series

n=0Lns. On the other hand, forτσ0,|t|2 the estimates (6) yield

n=0

Lns C 1−ρ[log|t|]

[log|t|]−1 j=0

ejPr(τf˜+ ˜g)C1max

log|t|,|t|Pr(τf˜+ ˜g) .

Moreover, forσ0sufficiently close tos0there exists 0< β <1 such thatLnsMn,sGΓ ,0C|t|1+β and we conclude that w0,l(x,−iτ +t )Γl,0B|t|1+β. Exploiting the case p1, we obtain similar estimates for w0,l(x,−iτ + t )Γl,p, p1 and we get the first approximation. Repeating this procedure, we complete the construction ofUM. References

[1] N. Burq, Controle de l’équation des plaques en présence d’obstacles strictement convexes, Mém. Soc. Math. France 55 (1993) 126.

[2] D. Dolgopyat, On decay of correlations in Anosov flows, Ann. Math. 147 (1998) 357–390.

[3] M. Ikawa, Decay of solutions of the wave equation in the exterior of several convex bodies, Ann. Inst. Fourier 2 (1988) 113–146.

[4] M. Ikawa, Singular perturbations of symbolic flows and the poles of the zeta function, Osaka J. Math. 27 (1990) 281–300.

[5] M. Ikawa, On zeta function and scattering poles for several convex bodies, in: Conf. EDP, Saint-Jean de Monts, SMF, 1994.

[6] M. Ikawa, On scattering by several convex bodies, J. Korean Math. Soc. 37 (2000) 991–1005.

[7] W. Parry, M. Pollicott, Zeta functions and the periodic orbit structure of hyperbolic dynamics, Astérisque 187–188 (1990).

[8] V. Petkov, L. Stoyanov, Geometry of Reflecting Rays and Inverse Spectral Problems, John Wiley & Sons, 1992.

[9] L. Stoyanov, Spectrum of the Ruelle operator and exponential decay of correlations for open billiard flows, Amer. J. Math. 123 (2001) 715–759.

[10] L. Stoyanov, Spectra of Ruelle transfer operators for contact flows on basic sets, Preprint, 2007.

Références

Documents relatifs

Since the sequence {F^} converges uniformly on C^ to F, we conclude that given N&gt;o, there exists n^eZ,, n^&gt;_m^ such that if n and n' are integers not less than n^ then

the scattering kernel. Using this, certain information on the behavior of the modified resolvent of the Laplacian and the distribution of poles of the scattering

( S ^ S ) of section 6 below), proved the acyclicity except in degree 0 of this complex in the smooth case, and related the characteristic polynomial of Frobenius acting on Ho to

of this fact we shall prove a new omega result for E2(T), the remainder term in the asymptotic formula for the fourth power mean of the Riemann zeta-function.. We start

over the rationals, and such a formula is still lacking except for curves, abelian varieties, rational surfaces, varieties with « only the algebraic coho- mology

In this section, we recall some results from number theory, and relate the formula above for the zeta function of monodromy at oo to a formula estimating the

The Newton polyhedron of f and its associated toroidal modification We first recall some of the terminology and basic properties of the Newton polyhedron of an

By Poincaré duality (loc. cit.) the pairing in the second line is nondegenerate, and therefore the pairing in the first line is too... PROOF: The first equality is