Nonlinear instability of inhomogeneous steady states solutions to the HMF Model

(1)

HAL Id: hal-02048776

https://hal.archives-ouvertes.fr/hal-02048776v2

Submitted on 25 Mar 2019

HAL is a multi-disciplinary open access archive for the deposit and dissemination of sci- entific research documents, whether they are pub- lished or not. The documents may come from

L’archive ouverte pluridisciplinaire HAL, est destinée au dépôt et à la diffusion de documents scientifiques de niveau recherche, publiés ou non, émanant des établissements d’enseignement et de

Nonlinear instability of inhomogeneous steady states solutions to the HMF Model

Mohammed Lemou, Ana Maria Luz, Florian Méhats

To cite this version:

Mohammed Lemou, Ana Maria Luz, Florian Méhats. Nonlinear instability of inhomogeneous steady

states solutions to the HMF Model. Journal of Statistical Physics, Springer Verlag, 2020, 178 (3),

pp.645-665. �10.1007/s10955-019-02448-4�. �hal-02048776v2�

(2)

Nonlinear instability of inhomogeneous steady states solutions to the HMF Model

M. Lemou

^:

, A. M. Luz

^‹

, and F. M´ ehats

^;

:

Univ Rennes, CNRS, IRMAR, mohammed.lemou@univ-rennes1.fr

‹

IME, Universidade Federal Fluminense, analuz@id.uff.br

;

Univ Rennes, IRMAR, florian.mehats@univ-rennes1.fr

March 25, 2019

Abstract

In this work we prove the nonlinear instability of inhomogeneous steady states solutions to the Hamiltonian Mean Field (HMF) model. We first study the linear instability of this model under a simple criterion by adapting the techniques developed in [19]. In a second part, we extend to the inhomogeneous case some techniques developed in [14, 17, 18] and prove a nonlinear instability result under the same criterion.

1 Introduction

In this paper, we are interested in the nonlinear instability of inhomogeneous steady

states of the Hamiltonian Mean Field (HMF) system. The HMF system is a kinetic

model describing particles moving on a unit circle interacting via an infinite range

attractive cosine potential. This 1D model holds many qualitative properties of more

realistic long-range interacting systems as the Vlasov-Poisson model. The HMF model

has been the subject of many works in the physical community, for the study of non

equilibrium phase transitions [11, 26, 2, 24], of travelling clusters [6, 27] or of relaxation

processes [28, 3, 12]. The long-time validity of the N-particle approximation for the

HMF model has been investigated in [8, 9] and the Landau-damping phenomenon near

a spatially homogeneous state has been studied recently in [13]. The formal linear

stability of inhomogeneous steady states has been studied in [10, 23, 7]. In particular,

a simple criterion of linear stability has been derived in [23]. In [22], the authors of the

present paper have proved that, under the same criterion κ

0

ă 1 (see below for a precise

formulation), the inhomogeneous steady states of HMF that are nonincreasing functions

of the microscopic energy are nonlinearly stable. The aim of the present paper is to

show, in a certain sense, that this criterion is sharp: we show that if κ

0

ą 1, the HMF

model can develop instabilities, from both the linear and the nonlinear points of view.

(3)

In [15], Guo and Lin have derived a sufficient criterion for linear instability to 3D Vlasov-Poisson by extending an approach developped in [19] for BGK waves. Let us also mention that both works have adapted some techniques presented in [16] to prove the nonlinear instability of the 3D Vlasov-Poisson system. In the first part of this article we adapt these techniques and prove the linear instability of nonhomogeneous steady states to the HMF system. In [17], a nonlinear instability result for 1D Vlasov-Poisson equation was obtained for an initial data close to stationary homogeneous profiles that satisfy a Penrose instability criterion by using an approach developed in [14]. In [18], starting with the N- particles version of the HMF model, a nonlinear instability result is obtained for the corresponding Vlasov approximation by also considering a Penrose instability condition for stationary homogeneous profiles. In the second part of this article our aim is to prove the nonlinear instability of non-homogeneous steady states of HMF by adapting the techniques developed in [14, 17, 18].

In the HMF model, the distribution function of particles f p t, θ, v q solves the initial- valued problem

B

^t

f ` v B

θ

f ´ B

θ

φ

_f

B

^v

f “ 0, p t, θ, v q P R

_`

ˆ T ˆ R , (1.1) f p 0, θ, v q “ f

init

p θ, v q ě 0,

where T is the flat torus r 0, 2π s and where the self-consistent potential φ

_f

associated with a distribution function f is defined by

φ

_f

p θ q “ ´ ż

2π

0

ρ

_f

p θ

¹

q cos p θ ´ θ

¹

q dθ

¹

, ρ

_f

p θ q “ ż

R

f p θ, v q dv. (1.2) Introducing the so-called magnetization vector defined by

M

_f

“ ż

2π

0

ρ

_f

p θ q u p θ q dθ, with u p θ q “ p cos θ, sin θ q

^T

(1.3) we have

φ

_f

p θ q “ ´ M

_f

¨ u p θ q . (1.4) In this work will consider steady states of (1.1) of the form

f

0

p θ, v q “ F p e

0

p θ, v qq , (1.5) where F is a given nonnegative function and where the microscopic energy e

0

p θ, v q is given by

e

0

p θ, v q “ v

²

2 ` φ

0

p θ q with φ

0

“ φ

_f0

. (1.6) Without loss of generality, we assume that φ

0

p θ q “ ´ m

0

cos θ with m

0

ą 0. Here m

0

is the magnetization of the stationary state f

0

defined by m

0

“ ş

ρ

_f0

cos θdθ.

It is shown in [22] that (essentially) if F is decreasing then f

0

is nonlinearly stable

by the HMF flow (1.1) provided that the criterion κ

0

ă 1 is satisfied, where κ

0

is given

(4)

by

κ

0

“ ´ ż

2π

0

ż

_`8

´8

F

¹

p e

0

p θ, v qq

¨

˚ ˚

˚ ˝ ż

De0pθ,vq

p cos θ ´ cos θ

¹

qp e

0

p θ, v q ` m

0

cos θ

¹

q

^´¹^{²

dθ

¹

ż

De0pθ,vq

p e

0

p θ, v q ` m

0

cos θ

¹

q

^´¹^{²

dθ

¹

˛

‹ ‹

‹ ‚

2

dθdv, (1.7) with

D

_e

“ θ

¹

P T : m

0

cos θ

¹

ą ´ e ( .

In this paper, we explore situations where this criterion is not satisfied, i.e when κ

0

ą 1. Let us now state our two main results. The first one concerns the linearized HMF equation given by

B

^t

f “ Lf, (1.8)

where

Lf : “ ´ v B

^θ

f ` B

^θ

φ

0

B

^v

f ` B

^θ

φ

f

B

^v

f

0

. (1.9) Theorem 1.1 (Linear instability) . Let f

0

P L

¹

p T , R q be a stationary solution of (1.1) of the form (1.5), where F is a nonnegative C

¹

function on R such that F

¹

p e

0

p θ, v qq belongs to L

¹

p T , R q . Assume that κ

0

ą 1, where κ

0

is given by (1.7). Then there exists λ ą 0 and a non-zero f P L

¹

p T ˆ R q such that e

^λt

f is a nontrivial growing mode weak solution to the linearized HMF equation (1.8).

Our second result is the following nonlinear instability theorem.

Theorem 1.2 (Nonlinear instability) . Let f

0

be a stationary solution of (1.1) of the form (1.5), where F is a C

⁸

function on R , such that F p e q ą 0 for e ă e

_˚

, F p e q “ 0 for e ě e

_˚

, with e

_˚

ă m

0

and | F

¹

p e q| ď C | e

_˚

´ e |

^´^α

F p e q in the neighborhood of e

_˚

, for some α ě 1. Assume that κ

0

ą 1, where κ

0

is given by (1.7). Then f

0

is nonlinearly unstable in L

¹

p T ˆ R q , namely, there exists δ

0

ą 0 such that for any δ ą 0 there exists a nonnegative solution f p t q of (1.1) satisfying } f p 0 q ´ f

0

}

L¹

ď δ and

} f p t

δ

q ´ f

0

}

L¹

ě δ

0

, with t

_δ

“ O p| log δ |q as δ Ñ 0.

Remark 1.3. Note that in these two theorems we do not assume that the profile F is a decreasing function. Besides, the set of steady states satisfying the assumptions of these theorems is not empty, as proved in the Appendix (see Lemma A.1). Note also that the instability of Theorem 1.2 is not due to the usual orbital instability. Indeed the functional space of the pertubation can be restricted to the space of even functions in p θ, v q .

The outline of the paper is as follows: Sections 2 and 3 are respectively devoted to

the proofs of Theorem 1.1 and Theorem 1.2.

(5)

2 A linear instability result: proof of Theorem 1.1

The aim of this section is to prove Theorem 1.1. This proof will be done following the framework used by Lin for the study of periodic BGK waves in [19], which was generalized to the analysis of instabilities for the 3D Vlasov-Poisson system by Guo and Lin in [15]. We divide this proof into the three Lemmas 2.1, 2.2 and 2.3, respectively proved in Subsections 2.1, Subsection 2.2 and Subsection 2.3.

A growing mode of (1.8) is a solution of the form e

^λt

f , where f P L

¹

p T , R q is an unstable eigenfunction of L, i.e. a nonzero function satisfying Lf “ λf in the sense of distributions, with λ P R

˚

`

and with L defined by (1.9). Note that the equation Lf ´ λf “ 0 is invariant by translation: if f p θ, v q is an eigenfunction, then for all θ

0

, f p θ ` θ

0

, v q is also an eigenfunction. Since for all f P L

¹

we can find a θ

0

P T such that ş

ρ

_f

p θ ` θ

0

q sin θdθ “ 0, we can assume that our eigenfunction of L always satisfy ş ρ

_f

sin θdθ “ 0, i.e. φ

_f

“ ´ m cos θ with m “ ş

ρ

_f

cos θdθ.

Let us first define p Θ p s, θ, v q , V p s, θ, v qq as the solution of the characteristics problem

$ ’

&

’ %

dΘ p s, θ, v q

ds “ V p s, θ, v q dV p s, θ, v q

ds “ ´B

^θ

φ

0

p Θ p s, θ, v qq

(2.1)

with initial data Θ p 0, θ, v q “ θ, V p 0, θ, v q “ v. When there is no ambiguity, we denote simply Θ p s q “ Θ p s, θ, v q and V p s q “ V p s, θ, v q . Since φ

0

p θ q “ ´ m

0

cos θ, the solution p Θ, V q is globally defined and belongs to C

8

p R ˆ T ˆ R q . Note that the energy e

0

p Θ p s q , V p s qq “

^V^p2^s^q²

` φ

0

p Θ p s qq does not depend on s.

We shall reduce the existence of a growing mode of (1.8) to the existence of a zero of the following function, defined for all λ P R

^˚

`

: G p λ q “ 1 `

ż

2π

0

ż

R

F

¹

ˆ v

²

2 ´ m

0

cos θ

˙

cos

²

θdθdv

´ ż

2π

0

ż

R

F

¹

ˆ v

²

2 ´ m

0

cos θ

˙ ˆż

0

´8

λe

^λs

cos Θ p s, θ, v q ds

˙

cos θdθdv. (2.2) Lemma 2.1. Let f

0

P L

¹

p T , R q be a stationary solution of (1.1) of the form (1.5), where F is a C

¹

function on R such that F

¹

p e

0

p θ, v qq belongs to L

¹

p T , R q . Then the function G defined by (2.2) is well-defined and continuous on R

^˚

`

. Moreover, there exists a growing mode e

^λt

f solution to (1.8) associated with the eigenvalue λ ą 0 if and only if G p λ q “ 0.

An unstable eigenfunction f of L is defined by f p θ, v q “ ´ F

¹

p e

0

p θ, v qq cos θ ` F

¹

p e

0

p θ, v qq

ż

0

´8

λe

^λs

cos p Θ p s, θ, v qq ds. (2.3) Lemma 2.2. Under the assumptions of Lemma 2.1, the function G defined by (2.2) satisfies

lim

λÑ0^`

G p λ q “ 1 ´ κ

0

, (2.4)

(6)

where κ

0

is defined by (1.7).

Lemma 2.3. Under the assumptions of Lemma 2.1, the function G defined by (2.2) satisfies

λ

lim

Ñ`8

G p λ q “ 1. (2.5)

Proof of Theorem 1.1. From these three lemmas, it is clear that if κ

0

ą 1, we have

λ

lim

Ñ0^`

G p λ q ă 0 and lim

λÑ`8

G p λ q ą 0 so by continuity of G, there exists λ ą 0 such that G p λ q “ 0. This means that there exists a growing mode to (1.8) and this proves Theorem 1.1.

2.1 First properties of the function G p λ q: proof of Lemma 2.1 In this subsection, we prove Lemma 2.1. Let λ P R

˚

`

. Since, by assumption, the function F

¹

p e

0

p θ, v qq belongs to L

¹

p T ˆ R q , and since

@p θ, v q ˇ ˇ ˇ ˇ

ż

0

´8

λe

^λs

cos p Θ p s qq ds ˇ ˇ ˇ ˇ ď

ż

0

´8

λe

^λs

ds “ 1, both functions

F

¹

p e

0

p θ, v qq cos θ and F

¹

p e

0

p θ, v qq ż

0

´8

λe

^λs

cos p Θ p s, θ, v qq ds

belong to L

¹

p T ˆ R q , so the function f defined by (2.3) also belongs to L

¹

p T ˆ R q . Hence, by integrating with respect to cos θdθdv, we deduce that G p λ q is well-defined by (2.2).

The continuity of G on R

^˚

`

stems from dominated convergence.

Consider now a (nonzero) growing mode e

^λt

f of (1.8) associated to an eigenvalue λ ą 0. Let us prove that G p λ q “ 0. From Lf “ λf and (2.1), we get, in the sense of distributions,

d ds

´

e

^λs

f p Θ p s q , V p s qq ¯

“ e

^λs

φ

¹_f

p Θ p s qq V p s q F

¹

p e

0

p Θ, V qq . Integrating this equation from ´ R to 0, we get, for almost all p θ, v q and all R,

f p θ, v q “ e

^´^λR

f p Θ p´ R q , V p´ R qq ` F

¹

p e

0

q ż

0

´R

e

^λs

φ

¹_f

p Θ p s qq V p s q ds,

where we recall that e

0

p Θ, V q “ e

0

p θ, v q . We multiply by a test function ψ p θ, v q P C

₀⁸

p T ˆ R q and integrate with respect to p θ, v q ,

ż

2π

0

ż

R

f p θ, v q ψ p θ, v q dθdv “ e

^´^λR

ż

2π

0

ż

R

f p θ, v q ψ p Θ p R q , V p R qq dθdv

` ż

0

´R

ż

2π

0

ż

R

e

^λs

F

¹

p e

0

q φ

¹_f

p Θ p s qq V p s q ψ p θ, v q dsdθdv.

(7)

In the first integral of the right-hand side, we have performed the change of variable p θ, v q “ p Θ p R, θ

¹

, v

¹

q , V p R, θ

¹

, v

¹

qq . In the second integral, we remark that | φ

¹_f

p Θ p s qq | ď } f }

L¹

and, the support of ψ being compact, v is bounded. Hence, by

^v₂²

´ m

0

cos θ “

V²

2

´ m

0

cos Θ, V p s q is bounded. Therefore, by dominated convergence (using that F

¹

p e

0

q P L

¹

), as R Ñ 8 , we get

ż

2π

0

ż

R

f p θ, v q ψ p θ, v q dθdv “ ż

0

´8

ż

2π

0

ż

R

e

^λs

F

¹

p e

0

q φ

¹_f

p Θ p s qq V p s q ψ p θ, v q dsdθdv i.e.

f p θ, v q “ F

¹

p e

0

q ż

0

´8

e

^λs

φ

¹_f

p Θ p s qq V p s q ds “ F

¹

p e

0

q ż

0

´8

e

^λs

d

ds p φ

_f

p Θ p s qqq ds

“ F

¹

p e

0

q φ

f

p θ q ´ F

¹

p e

0

q ż

0

´8

λe

^λs

φ

f

p Θ p s qq ds almost everywhere. Recall that φ

f

p θ q “ ´ m cos θ, with m “ ş

2π

0

ρ

f

p θ q cos θdθ. Then we can rewrite this expression of f as

f p θ, v q “ ´ mF

¹

p e

0

q cos θ ´ mF

¹

p e

0

q ż

0

´8

λe

^λs

cos Θ p s q ds.

Integrating both sides of this equation with respect to cos θdθdv we get m “

ż

2π

0

ρ

_f

p θ q cos θdθ “ ´ m ż

2π

0

ż

R

F

¹

p e

0

q cos

²

θdθdv

` m ż

2π

0

ż

R

F

¹

p e

0

q ˆż

0

´8

λe

^λs

cos Θ p s q ds

˙

cos θdθdv, i.e. mG p λ q “ 0. It is clear that m ‰ 0, otherwise f “ 0 a.e.. Finally, we get G p λ q “ 0.

Reciprocally, assume that G p λ q “ 0 for some λ ą 0. Let f be given by (2.3). We have proved above that this function belongs to L

¹

p T ˆ R q . Moreover, since Θ p s, ´ θ, ´ v q “ Θ p s, θ, v q , we have ş

ρ

_f

sin θdθ “ 0. Multiplying (2.3) by cos θ and integrating with respect to θ and v, and using that G p λ q “ 0, we get ş

ρ

f

cos θdθ “ 1, so f is not the zero function and we have φ

_f

p θ q “ ´ cos θ.

We now check that the function f given by (2.3) is an eigenfunction of L associated with λ. From (2.3), we get

f p Θ p t q , V p t qq “ ´ F

¹

p e

0

q φ

_f

p Θ p t qq ` F

¹

p e

0

q ż

0

´8

λe

^λs

φ

_f

p Θ p s, Θ p t q , V p t qqq ds.

(8)

Note that Θ p s, Θ p t q , V p t qq “ Θ p s ` t, θ, v q . Therefore f p Θ p t q , V p t qq “ ´ F

¹

p e

0

q φ

_f

p Θ p t qq ` F

¹

p e

0

q

ż

0

´8

λe

^λs

φ

_f

p Θ p s ` t qq ds

“ ´ F

¹

p e

0

q φ

_f

p Θ p t qq ` F

¹

p e

0

q e

^´^λt

ż

t

´8

λe

^λs

φ

_f

p Θ p s qq ds

“ F

¹

p e

0

q e

^´^λt

ż

t

´8

e

^λs

φ

¹_f

p Θ p s qq V p s q ds, where we integrated by parts. Then

e

^λt

f p Θ p t q , V p t qq “ F

¹

p e

0

q ż

t

´8

e

^λs

φ

¹_f

p Θ p s qq V p s q ds

“ ż

t

´8

e

^λs

φ

¹_f

p Θ p s qq B

^v

f

0

p Θ p s q , V p s qq ds.

Differentiating both sides with respect to t, we obtain in the sense of distributions that for all t P R ,

e

^λt

`

λf p Θ p t q , V p t qq ` V p t qB

^θ

f p Θ p t q , V p t qq ´ φ

¹₀

p Θ p t qqB

^v

f p Θ p t q , V p t qq ˘

“ e

^λt

φ

¹_f

p Θ p t qq B

^v

f

0

p Θ p t q , V p t qq . By writing this equation at t “ 0, we get Lf “ λf: f is an unstable eigenfunction of L.

This ends the proof of Lemma 2.1.

2.2 Limiting behavior of G p λ q near λ “ 0: proof of Lemma 2.2 To study lim

λÑ0^`

G p λ q we need to analyze the limit of the function g

_λ

p θ, v q “

ż

0

´8

λe

^λs

cos Θ p s, θ, v q ds (2.6) as λ Ñ 0. We provide this result in the next lemma, where we also recall some well- known facts on the solution of the characteristics equations (2.1), which are nothing but the pendulum equations.

Lemma 2.4. Let p θ, v q P T ˆ R and e

0

“

^v2²

´ m

0

cos θ. Consider the solution p Θ p s, θ, v q , V p s, θ, v qq to the characteristics equations (2.1). Then the following holds true.

(i) If e

0

ą m

0

then, for all s P R , we have

Θ p s ` T

_e0

q “ Θ p s q ` 2π, V p s ` T

_e0

q “ V p s q , (2.7) with

T

_e0

“ ż

2π

0

dθ

¹

a 2 p e

0

` m

0

cos θ

¹

q ą 0. (2.8)

(9)

(ii) If ´ m

0

ă e

0

ă m

0

then Θ and V are periodic with period given by T

_e0

“ 4

ż

θm0

0

a dθ

2 p e

0

` m

0

cos θ q “ 4

? m

0

ż

π{2

0

b dθ

1 ´

^m2m⁰^`0^e⁰

sin

²

θ ą 0, (2.9) where θ

_m0

“ arccos p´

m^e⁰0

q .

(iii) We have

λ

lim

Ñ0^`

g

_λ

p θ, v q “

$ ’

’ ’

’ &

’ ’

% 1 T

e0

ż

2π

0

cos θ

¹

a 2 p e

0

` m

0

cos θ

¹

q dθ

¹

if e

0

ą m

0

, 4

T

_e0

ż

θm0

0

cos θ

¹

a 2 p e

0

` m

0

cos θ

¹

q dθ

¹

if ´ m

0

ă e

0

ă m

0

. (2.10) Proof. (i) Let e

0

ą m

0

. Without loss of generality, since Θ p s, ´ θ, ´ v q “ Θ p s, θ, v q and V p s, ´ θ, ´ v q “ V p s, θ, v q , we can only treat the case v ą 0. As we have

V p s q

²

2 “ e

0

` m

0

cos Θ p s q ě e

0

´ m

0

ą 0,

V p s q does not vanish and remains positive. Hence Θ p s q is the solution of the following autonomous equation

Θ 9 p s q “ V p s q “ a

2 p e

0

` m

0

cos Θ p s qq (2.11) and is strictly increasing with Θ p s q Ñ `8 as s Ñ `8 . Let T

_e0

be the unique time such that Θ p T

e0

q “ θ ` 2π. By Cauchy-Lipschitz’s theorem, we have Θ p s ` T

e0

q ´ 2π “ Θ p s q and (2.7) holds. Defining

P p τ q “ ż

τ

0

dθ

¹

a 2 p e

0

` m

0

cos θ

¹

q ,

the solution of (2.11) satisfies P p Θ p s qq ´ P p θ q “ s. Therefore, we have T

_e0

“ P p θ ` 2π q ´ P p θ q , from which we get (2.8).

(ii) Let ´ m

0

ă e

0

ă m

0

. In this case, Θ p s q will oscillate between the two values θ

_m0

“ arccos p´

_m^e⁰0

q and ´ θ

_m0

with a period T

_e0

given by (2.9). On the half-periods where Θ is increasing, we also have (2.11). We skip the details of the proof, which is classical.

(iii) We remark that cos Θ p s ` kT

_e0

q “ cos Θ p s q for all s P R and k P Z . Indeed, by (i),

for e

0

ą m

0

we have Θ p s ` kT

_e0

q “ Θ p s q ` 2πk and, by (ii), for ´ m

0

ă e

0

ă m

0

we have

(10)

Θ p s ` kT

_e0

q “ Θ p s q . Hence, we compute from (2.6) g

_λ

p θ, v q “

`8

ÿ

k“0

ż

_pk`1qTe0

kTe0

λe

^´^λs

cos Θ p´ s q ds

“

`8

ÿ

k“0

ż

Te0

0

λe

^´^λs^´^kλT^e⁰

cos Θ p´ s q ds

“

˜

_`8

ÿ

k“0

e

^´^kλT^e0

¸ ż

Te0

0

λe

^´^λs

cos Θ p´ s q ds

“ λ

1 ´ e

^´^λT^e⁰

ż

Te0

0

e

^´^λs

cos Θ p´ s q ds.

Therefore, clearly, lim

λÑ0^`

g

λ

p θ, v q “ 1 T

_e0

ż

Te0

0

cos Θ p´ s q ds “ 1 T

_e0

ż

Te0

0

cos Θ p s q ds.

If e

0

ą m

0

, we perform the change of variable θ

¹

“ Θ p s q which is strictly increasing from r 0, T

_e0

s to r θ, θ ` 2π s . Using (2.11), we obtain

lim

λÑ0^`

g

_λ

p θ, v q “ 1 T

_e0

ż

θ`2π

θ

cos θ

¹

a 2 p e

0

` m

0

cos θ

¹

q dθ

¹

“ 1 T

_e0

ż

2π

0

cos θ

¹

a 2 p e

0

` m

0

cos θ

¹

q dθ

¹

. If ´ m

0

ă e

0

ă m

0

, we can always choose a time t

0

such that Θ p t

0

q “ ´ θ

m0

, Θ p t

0

` T

_e0

{ 2 q “ θ

_m0

, Θ p s q is strictly increasing on r t

0

, t

0

` T

_e0

{ 2 s and such that Θ p s q “ Θ p 2t

0

` T

_e0

´ s q for s P r t

0

` T

_e0

{ 2, t

0

` T

_e0

s . We have

λ

lim

Ñ0^`

g

_λ

p θ, v q “ 1 T

_e0

ż

t0`Te0{2

t0

cos Θ p s q ds ` 1 T

_e0

ż

t0`Te0

t0`Te0{2

cos Θ p s q ds

“ 1 T

e0

ż

t0`Te0{2

t0

cos Θ p s q ds ` 1 T

e0

ż

t0`Te0

t0`Te0{2

cos Θ p 2t

0

` T

e0

´ s q ds

“ 2 T

_e0

ż

t0`Te0{2

t0

cos Θ p s q ds

“ 2 T

e0

ż

θm0

´θm0

cos θ

¹

a 2 p e

0

` m

0

cos θ

¹

q dθ

¹

“ 4 T

e0

ż

θm0

0

cos θ

¹

a 2 p e

0

` m

0

cos θ

¹

q dθ

¹

, where, on the time interval r t

0

, t

0

` T

_e0

{ 2 s , we performed the change of variable θ

¹

“ Θ p s q .

Proof of Lemma 2.2. Now we come back to the definition (2.2) of G

_λ

, which reads G p λ q “ 1 `

ż

2π

0

ż

R

F

¹

ˆ v

²

2 ´ m

0

cos θ

˙

cos

²

θdθdv

´ ż

2π

0

ż

R

F

¹

ˆ v

²

2 ´ m

0

cos θ

˙

g

λ

p θ, v q cos θdθdv. (2.12)

(11)

We remark that | g

_λ

p θ, v q| ď 1 and recall that the function F

¹

´

v²

2

´ m

0

cos θ ¯

belongs to L

¹

p T ˆ R q . Therefore, we can pass to the limit in the second integral by dominated convergence and deduce from Lemma 2.4 (iii) (note that the set tp θ, v q : e

0

p θ, v q ď ´ m

0

u is of measure zero) that

λ

lim

Ñ0^`

G p λ q “ 1 ` ż

2π

0

ż

R

F

¹

p e

0

q cos

²

θdθdv

´ ĳ

e0pθ,vqąm0

F

¹

p e

0

q

˜ 1 T

e

ż

2π

0

cos θ

¹

a 2 p e

0

` m

0

cos θ

¹

q dθ

¹

¸

cos θdθdv

´

ĳ

´m0ăe0pθ,vqăm0

F

¹

p e

0

q

˜ 4 T

e

ż

θm0

0

cos θ

¹

a 2 p e

0

` m

0

cos θ

¹

q dθ

¹

¸

cos θdθdv

“ 1 ` ż

2π

0

ż

R

F

¹

p e

0

q cos

²

θdθdv

´ ż

2π

0

ż

R

F

¹

p e

0

q

¨

˚ ˚

˚ ˝ ż

De0

cos θ

¹

p e

0

` m

0

cos θ

¹

q

^´¹^{²

dθ

¹

ż

De0

p e

0

` m

0

cos θ

¹

q

^´¹^{²

dθ

¹

˛

‹ ‹

‹ ‚ cos θdθdv

“ 1 ` ż

2π

0

ż

R

F

¹

p e

0

q cos p θ q

²

dvdθ ´ ż

2π

0

ż

R

F

¹

p e

0

q p Π

m0

p cos θ qq

²

dθdv, with

p Π

m0

h qp e q “ ż

De

p e ` m

0

cos θ q

^´¹^{²

h p θ q dθ ż

De

p e ` m

0

cos θ q

^´¹^{²

dθ

, (2.13)

for all function h p θ q and

D

_e

“ θ

¹

P T : m

0

cos θ

¹

ą ´ e ( .

Here the operator Π

_m0

is a variant of the operator Π given by (3.8) in [21], this operator should be understood as the “projector”onto the functions which depend only on the microscopic energy e

0

p θ, v q . A projector of this type is also mentioned in the work by Guo and Lin [15].

Now we remark that straightforward calculations give

1 ´ κ

0

“ 1 ` ż

2π

0

ż

R

F

¹

p e

0

q ´

cos

²

θ ´ 2 cos θΠ

m0

p cos θ q ` p Π

m0

p cos θ qq

²

¯ dθdv,

“ 1 ` ż

2π

ż

F

¹

p e

0

q cos

²

θdθdv ´ ĳ

F

¹

p e

0

q p Π

m0

p cos θ qq

²

dθdv,

(12)

where

´ κ

0

: “ ż

2π

0

ż

R

F

¹

p e

0

q

¨

˚ ˚

˝ ż

De

p cos θ ´ cos θ

¹

qp e

0

` m

0

cos θ

¹

q

^´¹^{²

dθ

¹

ż

De

p e

0

` m

0

cos θ

¹

q

^´¹^{²

dθ

¹

˛

‹ ‹

‚

2

dθdv.

This calculation uses that Π

m0

is a projector. We finally get (2.4) and the proof of Lemma 2.2 is complete.

2.3 Limiting behavior of G p λ q as λ Ñ 8: proof of Lemma 2.3

In this subsection, we prove Lemma 2.3. An integration by parts in (2.6) yields g

_λ

p θ, v q “ cos θ `

ż

0

´8

e

^λs

V p s, θ, v q sin Θ p s, θ, v q ds.

The velocity can be bounded independently of s thanks to the conservation of the energy,

| V p s q| “ `

v

²

` 2m

0

cos θ ´ 2m

0

cos Θ p s q ˘

1{2

ď `

v

²

` 4m

0

˘

1{2

. Thus

ˇ ˇ ˇ ˇ

ż

0

´8

e

^λs

sin Θ p s, θ, v q V p s, θ, v q ds ˇ ˇ ˇ ˇ ď `

v

²

` 4m

0

˘

1{2

ż

0

´8

e

^λs

ds “

` v

²

` 4m

0

˘

1{2

λ and, for all p θ, v q ,

λÑ`8

lim g

_λ

p θ, v q “ cos θ.

Using again that | g

λ

p θ, v q| ď 1 and that F

¹

p e

0

p θ, v qq belongs to L

¹

, we deduce directly (2.5) from (2.12) and from dominated convergence. The proof of Lemma 2.3 is complete.

3 A nonlinear instability result: proof of Theorem 1.2

We start by an analysis of the linearized HMF operator L around the inhomogeneous equilibrium state f

0

, where L is given by (1.9). We write

L “ L

0

` K, (3.1)

where

L

0

f “ ´ v B

^θ

f ´ E

_f0

B

^v

f, Kf “ ´ E

_f

B

^v

f

0

, E

_f

“ ´B

^θ

φ

_f

. (3.2)

(13)

3.1 Estimates on the semigroup e

^tL

Let us state some useful properties of the operator L

0

given by (3.2). Since φ

0

p θ q “

´ m

0

cos θ is smooth, the characteristics equations (2.1) admit a unique solution Θ p s, θ, v q , V p s, θ, v q , which is globally defined and C

8

in the variables p s, θ, v q . Moreover, this solution has bounded derivatives with respect to θ and v, locally in time. Let k P N . For any f in the Sobolev space W

^k,1

p T ˆ R q , the function

e

^tL⁰

f p t, θ, v q : “ f p Θ p´ t, θ, v q , V p´ t, θ, v qq , @ t ě 0, (3.3) belongs to C

⁰

p R , W

^k,1

p T ˆ R qq and is clearly a solution to B

^t

g “ L

0

g with initial data f . This means that the semigroup e

^tL⁰

generated by the operator L

0

is strongly continuous on W

^k,1

p T ˆ R q .

Our aim is to apply the abstract results in [25] concerning perturbation theory of linear operators. Hence, we need to prove the following estimate. For all β ą 0, there exists a positive constant M

_β,k

such that

} e

^tL⁰

f }

W^k,1

ď M

_β,k

e

^tβ

} f }

W^k,1

@ f P W

^k,1

p T ˆ R q , @ t ě 0, (3.4) where M

β,k

depends on β and k. In fact this estimate will be proved for a subclass of functions f . From the assumptions of Theorem 1.2 on F , there exists e

_˚

ă m

0

such that support of F is p´8 , e

_˚

s . This means that the support of f

0

is contained in Ω

0

, where Ω

0

is the smooth open set

Ω

0

“

"

p θ, v q : v

²

2 ´ m

0

cos θ ă e

_˚

* . We then introduce the functional space

E

_k

“ !

f P W

^k,1

p T ˆ R q : Supp p f q Ă Ω

0

) , and claim that for all f P E

_k

, we have e

^tL⁰

f P E

_k

and, for all β ą 0,

} e

^tL⁰

f }

W^k,1

ď M

_β,k

e

^tβ

} f }

W^k,1

, @ f P E

_k

, @ t ě 0, (3.5) M

_β,k

being a positive constant depending on β and k.

Assume for the moment that estimate (3.5) holds true. From the assumptions of Theorem 1.2, one deduces that B

^v

f

0

P E

_k

. It is then easy to check that e

^tL⁰

K is a compact operator on E

_k

, for all t P R , and the map t ÞÑ e

^tL⁰

K P L p E

_k

q is continuous on R . Hence, K is L

0

-smoothing in the sense of [25] (page 707). Assumptions of Theorem 1.1 in [25] are therefore satisfied, which implies that L generates a strongly continuous semigroup e

^tL

. Now, from Theorem 1.2 in [25], for all β ą 0, any point of the spectrum σ p L q lying in the half plane Re z ą β is an isolated eigenvalue with finite algebraic multiplicity. Furthermore, the set σ p L q X t Re z ą β u is finite.

The assumptions of Theorem 1.2 clearly imply those of Theorem 1.1. Hence L admits at least one eigenvalue λ P R

˚

`

associated with an eigenfunction f r P L

¹

p T ˆ R q . We claim

(14)

that, in fact, f r P E

_k

which will be proved below. This means that the set of eigenvalues of L on E

_k

with positive real part is not empty, and we therefore can choose an eigenvalue γ with positive maximal real part. Finally, we apply Theorem 1.3 in [25] and get that, for all β ą Re γ, there exists a positive constant M

_β,k

such that

} e

^tL

f }

W^k,1

ď M

β,k

e

^tβ

} f }

W^k,1

@ f P E

_k

, @ t ě 0. (3.6) Proof of (3.5) and of the claim f r P E

_k

. Let f P E

_k

. From (3.3), we clearly have

} e

^tL⁰

f }

L¹

“ } f }

L¹

, @ f P L

¹

, @ t ě 0.

Moreover, we know from the analysis of the characteristics problem (2.1) performed in Section 2, that by conservation of the energy, for all p θ, v q P Ω

0

, we have p Θ p t, θ, v q , V p t, θ, v qq P Ω

0

. Thus

Supp p e

^tL⁰

f q Ă Ω

0

.

Let k ě 1. By (3.3), to get an estimate of e

^tL⁰

f in W

^k,1

p T ˆ R q , it is sufficient to estimate Θ and V in W

^k,⁸

p T ˆ R q for p θ, v q P Ω

0

. Recall that, since e

_˚

ă m

0

, Θ and V are periodic functions with period T

_e0

. Moreover, (2.9) shows that T

_e0

is a C

8

function of e

0

on r´ m

0

, e

_˚

s , which means that it is also a C

⁸

function of p θ, v q . Note also that

?2πm0

ď T

_e0

ď T

_e_˚

. Define now the following 1-periodic functions

Θ r p s, θ, v q “ Θ p sT

e0

, θ, v q , V r p s, θ, v q “ V p sT

e0

, θ, v q satisfying

dr Θ

ds “ T

_e0

V , r d V r

ds “ ´ m

0

T

_e0

sin Θ. r Applying Gronwall Lemma, one gets

|B

^rs

B

θ^j

B

v^ℓ

Θ r | ` |B

s^r

B

^jθ

B

v^ℓ

V r | ď C

_k

e

^C^k^s

, @ s ě 0, @p θ, v q P Ω

0

, for r ` j ` ℓ ď k.

The period of Θ and r V r being independent of p θ, v q , the functions B

θ^j

B

v^ℓ

Θ and r B

^jθ

B

v^ℓ

V r are also 1-periodic and therefore

|B

^rs

B

θ^j

B

^ℓv

Θ r | ` |B

^rs

B

θ^j

B

v^ℓ

V r | ď C

_k¹

“ C

_k

e

^C^k

, @ s ě 0, @p θ, v q P Ω

0

, for r ` j ` ℓ ď k.

Coming back to Θ and V , we deduce

|B

_θ^j

B

v^ℓ

Θ | ` |B

_θ^j

B

^ℓv

V | ď C

_k

p 1 ` s

^k

q , @ s ě 0, @p θ, v q P Ω

0

, for j ` ℓ ď k. (3.7) Using this estimate and (3.3), we finally get (3.5).

Let us finally prove that f r P E

_k

. By Lemma 2.1, the function f r is given by f r p θ, v q “ ´ F

¹

p e

0

p θ, v qq cos θ ` F

¹

p e

0

p θ, v qq

ż

0

´8

λe

^λs

cos p Θ p s, θ, v qq ds.

(15)

Hence, the support of F

¹

p e

0

p θ, v qq being in Ω

0

, the support on f r will also be contained in Ω

0

. Moreover, by using (3.7), we obtain that, for some C

_k

ą 0, we have

@ j ` ℓ ď k, @p θ, v q P T ˆ R , ˇ ˇ ˇ ˇ B

θ^j

B

^ℓv

ż

0

´8

λe

^λs

cos p Θ p s, θ, v qq ds ˇ ˇ ˇ ˇ ď C

_k

.

This is sufficient to deduce from F P C

8

that f r P E

_k

. 3.2 An iterative scheme

In this part, we prove Theorem 1.2 by following the strategy developed by Grenier in [14], which has been also used in [17, 18] to analyse instabilities for homogeneous steady states of Vlasov-Poisson models. Let N ě 1 be an integer to be fixed later. According to the previous subsection, we can consider an eigenvalue γ of L on E

_N

with maximal real part, Re γ ą 0. Let g P E

_N

be an associated eigenfunction. With no loss of generality, we may assume that } Re g }

L¹

“ 1. Let

f

1

p t, θ, v q “ Re `

e

^γt

g p θ, v q ˘

χ

_δ

p e

0

p θ, v qq , (3.8) with e

0

p θ, v q “

^v2²

` φ

0

p θ q and where 0 ď χ

_δ

p e q ď 1 is a smooth real-valued truncation function to be defined further, in order to ensure the positivity of f

0

` δf

1

p 0 q . Note that f

1

is almost a growing mode solution to the linearized HMF model (1.8) since we have

pB

^t

´ L q f

1

“ Re p e

^γt

R r

δ

q , where

R r

_δ

“ `

´ E

_p₁_´_χ_δ_q_g

` p 1 ´ χ

_δ

q E

g

˘

B

^v

f

0

(3.9)

will be small. We now construct an approximate solution f

_app^N

to the HMF model (1.1) of the form

f

_app^N

“ f

0

` ÿ

N

k“1

δ

^k

f

k

, (3.10)

for sufficiently small δ ą 0, in which f

_k

(k ě 2) solves inductively the linear problem pB

^t

´ L q f

k

`

k

ÿ

´1 j“1

E

fj

B

^v

f

k´j

“ 0 (3.11) with f

_k

p 0 q “ 0. Then f

_app^N

approximately solves the HMF model (1.1) in the sense that

B

^t

f

_app^N

` v B

^θ

f

_app^N

´ B

^θ

φ

_f^N

app

B

^v

f

_app^N

“ R

N

` δ Re p e

^γt

R r q , (3.12) where the remainder term R

_N

is given by

R

_N

“ ÿ

1ďj,ℓďN;j`ℓěN`1

δ

^j^`^ℓ

E

_f_j

B

v

f

_ℓ

. (3.13)

(16)

Step 1. Estimate of f

_k

. We claim that f

_k

P E

_N

´k`1

and, for all 1 ď k ď N ,

} f

k

}

W^N´k`¹^,¹

ď C

k

e

^ktRe^γ

. (3.14) We proceed by induction. From (3.8), this estimate is a consequence, for k “ 1, of

} gχ

_δ

}

W^N,¹

ď C

1

, (3.15)

which is proved below in Step 5. Let k ě 2. We have f

k

p t q “ ´

ż

t

0

e

^L^p^t^´^s^q

k

ÿ

´1 j“1

E

fj

p s qB

^v

f

k´j

p s q ds.

Therefore, for Re γ ă β ă 2 Re γ, } f

_k

}

W^N´k`¹^,¹

ď

k

ÿ

´1 j“1

ż

t

0

› ›

› e

^L^p^t^´^s^q

`

E

_f_j

p s qB

v

f

_k_´_j

p s q ˘›› ›

W^N´k`^1,1

ds ď M

β,N´k`1

k

ÿ

´1 j“1

ż

t

0

e

^β^p^t^´^s^q

› ›E

_f_j

p s q › ›

W^N´k`^1,8

}B

^v

f

k´j

p s q}

W^N´k`¹^,¹

ds ď M

_β,N_´_k_`1

k

ÿ

´1

j“1

ż

t

0

e

^β^p^t^´^s^q

} f

_j

p s q}

L¹

} f

_k_´_j

p s q}

_W^N´k`^2,1

ds since k ´ j ď k ´ 1,

ď M

β,N´k`1

˜

_k_´₁

ÿ

j“1

C

j

C

k´j

¸ ż

t 0

e

^β^p^t^´^s^q

e

^ks^Re^γ

ds

ď M

_β,N_´_k_`₁

k Re γ ´ β

˜

_k_´₁

ÿ

j“1

C

j

C

_k_´_j

¸

e

^kt^Re^γ

,

where we used (3.6) and the recursive assumption. This ends the proof of (3.14).

Step 2. Estimates of f

_app^N

´ f

0

and R

_N

. The parameter δ and the time t will be such that

δe

^t^Re^γ

ď min ˆ 1

2 , 1 2K

_N

˙

, K

_N

“ max

1ďkďN

C

_k

. (3.16) Hence, from (3.14) we obtain

} f

_app^N

´ f

0

}

W¹^,¹

ď ÿ

N

k“1

δ

^k

C

k

e

^kt^Re^γ

ď K

N

δe

^t^Re^γ

1 ´ δe

^t^Re^γ

ď 1 and

} R

_N

}

L¹

ď

`8

ÿ

k“N`1

δ

^k

e

^kt^Re^γ

ÿ

1ďj,ℓďN;j`ℓ“k

C

_j

C

_ℓ

ď C r

_N

`

δe

^t^Re^γ

˘

N`1

.

(17)

Step 3. Estimate of f ´ f

_app^N

. Let f p t q be the solution of (1.1) with initial data f

0

` δ Re gχ

_δ

and let h “ f ´ f

_app^N

. Note that the positivity of f p t q is ensured by f

0

` δ Re gχ

_δ

ě 0 and that we have

} f p 0 q ´ f

0

}

L¹

ď δ.

The function h satisfies the following equation B

t

h ` v B

θ

h ` E

_f

B

v

h “ ´

E

_fN

app

´ E

_f

¯

B

v

f

_app^N

´ R

_N

´ δ Re p e

^γt

R r

_δ

q

with h p 0 q “ 0. To get a L

¹

-estimate of h, we multiply this equation by sign p h q and integrate in p θ, v q . We get

d

dt } h }

L¹

ď › › › E

_fN

app

´ E

_f

› › ›

_L₈

› › B

v

f

_app^N

› ›

_L1

` } R

_N

}

L¹

` δe

^t^Re^γ

} R r

_δ

}

L¹

ď } h }

L¹

› › B

^v

f

_app^N

› ›

L¹

` } R

_N

}

L¹

` δe

^t^Re^γ

} R r

_δ

}

L¹

. From Step 2 we have › › B

^v

f

_app^N

› ›

L¹

ď }B

^v

f

0

}

L¹

` 1, which implies that } h p t q}

L¹

ď

ż

t

0

e

^p^t^´^s^qp}B^v^f⁰^}^L¹^`¹^q

´

} R

_N

p s q}

L¹

` δe

^s^Re^γ

} R r

_δ

}

L¹

¯ ds.

Again from Step 2, we then get } h p t q}

L¹

ď

ż

t

0

e

^p^t^´^s^qp}B^v^f⁰^}^L¹^`¹^q

´ C r

N

`

δe

^s^Re^γ

˘

^N`1

` δe

^s^Re^γ

} R r

δ

}

L¹

¯ ds We now fix N as follows (with the notation t ¨ u for the integer function)

N : “

Z }B

v

f

0

}

L¹

` 1 Re γ

^

` 1 ě 1 and claim that χ may be chosen such that

} R r

_δ

}

L¹

ď `

δe

^s^Re^γ

˘

N

, (3.17)

see Step 5 for the proof. This yields

} f ´ f

_app^N

}

L¹

p t q ď C q

_N

`

δe

^t^Re^γ

˘

N`1

(3.18) with C q

_N

“

¹3 Re^`^C^r^Nγ

.

Step 4. End of the proof. Since Re g is not zero, we can choose a real valued function ϕ p θ, v q in L

⁸

such that } ϕ }

^L⁸

=1 and

Re z

_g

ą 0 with z

_g

“ ż

2π

0

ż

R

gϕdθdv.

(18)

Denoting

z

_g,δ

“ ż

2π

0

ż

R

gχ

_δ

ϕdθdv, we have

ĳ

f

1

ϕdθdv “ e

^t^Re^γ

Re `

e

^it^Im^γ

z

_g,δ

˘ ě e

^t^Re^γ

Re `

e

^it^Im^γ

z

_g

˘

´ e

^t^Re^γ

| z

_g

´ z

_g,δ

| ě e

^t^Re^γ

Re `

e

^it^Im^γ

z

_g

˘

´ e

^t^Re^γ

} g p 1 ´ χ

_δ

q}

L¹

We claim that

δ

lim

Ñ0

}p 1 ´ χ

_δ

q g }

L¹

“ 0, (3.19) which again will be proved in Step 5. In order to end the proof of Theorem 1.2, we estimate from below, using (3.18) and (3.14),

} f ´ f

0

}

L¹

ě ĳ

p f ´ f

0

q ϕdθdv “ ĳ

p f

_app^N

´ f

0

q ϕdθdv ` ĳ

p f ´ f

_app^N

q ϕdθdv ě δ

ĳ

f

1

ϕdθdv ´ ÿ

N

k“2

δ

^k

} f

_k

}

L¹

´ C q

_N

`

δe

^t^Re^γ

˘

N`1

ě δ ĳ

f

1

ϕdθdv ´ ÿ

N

k“2

C

_k

`

δe

^t^Re^γ

˘

k

´ C q

N

`

δe

^t^Re^γ

˘

N`1

ě δ ĳ

f

1

ϕdθdv ´ 2K

N

`

δe

^t^Re^γ

˘

²

´ C q

N

`

δe

^t^Re^γ

˘

N`1

ě δe

^t^Re^γ

´ Re `

e

^it^Im^γ

z

g

˘

´ }p 1 ´ χ

δ

q g }

L¹

´ 2K

N

δe

^t^Re^γ

´ C q

N

`

δe

^t^Re^γ

˘

N

¯ Assume for a while that

Re `

e

^it^Im^γ

z

g

˘

ě Re z

_g

2 . (3.20)

We have

} f ´ f

0

}

L¹

ě δe

^t^Re^γ

Re z

g

2 ˜

1 ´ 2 }p 1 ´ χ

_δ

q g }

L¹

Re z

_g

´ 4K

N

δe

^t^Re^γ

Re z

_g

´ 2 C q

N

`

δe

^t^Re^γ

˘

N

Re z

_g

¸

Let δ

0

ą 0 be such that 32K

_N

p Re z

_g

q

²

δ

0

` 2 C q

_N

8

^N

p Re z

_g

q

^N^`¹

δ

^N₀

ď 1

4 and 8δ

0

Re z

_g

ď min ˆ 1

2 , 1 2K

N

˙

(note that N ě 1) and consider times t such that 4δ

0

Re z

_g

ď δe

^t^Re^γ

ď 8δ

0

Re z

_g

. (3.21)

(19)

Owing to (3.19), we also choose δ small enough such that 2 }p 1 ´ χ

_δ

q g }

L¹

Re z

_g

ď 1 4 . We conclude from these inequalities that

} f ´ f

0

}

L¹

ě δ

0

and that (3.16) is satisfied.

To end the proof, it remains to fix the time t

_δ

and to choose the truncation function χ

_δ

. Let us show that, for δ small enough, there exists a time t

_δ

satisfying both (3.20) and (3.21). If Im γ “ 0, then (3.20) is clearly satisfied since Re z

_g

ą 0: a suitable t

_δ

is then

t

_δ

“ 1 Re γ log

ˆ 6δ

0

δ Re z

_g

˙ .

Assume now that Im γ ‰ 0. For δ small enough, the size of the interval of times t satisfying (3.21) becomes larger than

_|_Im^2π_γ_|

. This means that it is possible to find a time t

δ

in this interval satisfying (3.20).

Step 5. Choice of χ

_δ

. For all δ ą 0, we have to fix the function χ

_δ

P C

8

p R q such that (3.15), (3.17), (3.19) are satisfied and such that f

0

` δf

1

p 0 q ě 0. First of all, proceeding as in the proof of Lemma 2.1, we obtain that g takes the form

g p θ, v q “ ´ mF

¹

p e

0

q cos θ ´ mF

¹

p e

0

q ż

0

´8

γe

^γs

cos Θ p s q ds, (3.22) with m “ ş

2π

0

ρ

_f

p θ q cos θdθ. Hence,

| g p θ, v q| ď | m | ˆ

1 ` | γ | Re γ

˙

| F

¹

p e

0

q| . (3.23) The assumptions on F and F

¹

in Theorem 1.2 imply that

@ e ă e

_˚

, | F

¹

p e q| ď C p e

_˚

´ e q

^´^α

F p e q (3.24) with α ě 1. Since F p e q ą 0 for e ă e

_˚

, the local assumption becomes global. Let χ be a C

8

function such that 0 ď χ ď 1 and

$ &

%

χ p t q “ 0 for t ď 0, χ p t q ď 2t

^α

for t ě 0, χ p t q “ 1 for t ě 1 and let

χ

δ

p e q “ χ

ˆ e

_˚

´ e δ

¹^{p^2α^q

˙

. (3.25)

(20)

From

}p 1 ´ χ

_δ

q g }

L¹

ď } g

¹_e_˚_´_δ¹{p2αqăe0pθ,vqăe_˚

}

L¹

and dominated convergence, we clearly have (3.19). By (3.23) and (3.24), we have, for all p θ, v q P T ˆ R ,

δ | Re g p θ, v q χ

_δ

p e

0

p θ, v qq| ď δC | e

_˚

´ e

0

|

^´^α

F p e

0

q | e

_˚

´ e

0

|

^α

δ

¹^{²

“ Cδ

¹^{²

f

0

p θ, v q , so for δ small enough, we have f

0

` δf

1

p 0 q ě 0.

By differentiating (3.22) and using (3.7), we get

@ j ` ℓ ď N, |B

^jθ

B

^ℓv

g p θ, v q| ď C max

kďN`1

F

^p^k^q

p e

0

q ď C | e

_˚

´ e

0

|

^N^´¹

,

where we used Taylor formulas and the fact that F P C

8

with F p e q “ 0 for e ě e

_˚

. Besides, from (3.25), we obtain (if δ ď 1)

@ 1 ď j ` ℓ ď N, |B

^jθ

B

^ℓv

χ

_δ

p e

0

p θ, v qq| ď Cδ

^´^N^{p^2α^q¹_e_˚_´_δ¹{p2αqăe0pθ,vqăe_˚

. Therefore

} gχ

_δ

}

W^N,¹

ď } g }

L¹

` Cδ

^´^N^{p^2α^q

ż

2π

0

ż

R

| e

_˚

´ e

0

p θ, v q|

^N^´¹¹e˚´δ¹^{p^2αqăe0pθ,vqăe˚

dθdv ď } g }

L¹

` Cδ

^´^N^{p^2α^q

ż

e˚

e˚´δ¹^{p^2αq

p e

_˚

´ e q

^N^´¹

˜ 4

ż

θm0

0

a dθ

2 p e ` m

0

cos θ q

¸ de

“ } g }

L¹

` Cδ

^´^N^{p^2α^q

ż

e_˚

e˚´δ¹^{p^2αq

p e

_˚

´ e q

^N^´¹

T

e

de,

where θ

m0

“ arccos p´

m^e0

q and T

e

is given by (2.9). Now we recall that for e ď e

_˚

we have T

e

ď T

e_˚

. This yields

} gχ

_δ

}

W^N,1

ď } g }

L¹

` CT

_e_˚

δ

^´^N^{p^2α^q

ż

e_˚

e˚´δ¹^{p^2αq

p e

_˚

´ e q

^N^´¹

de “ } g }

L¹

` CT

e_˚

N . We have proved (3.15).

By (3.9), we have

} R r

_δ

}

L¹

ď C p}p 1 ´ χ

_δ

q g }

L¹

` }p 1 ´ χ

_δ

qB

^v

f

0

}

L¹

q ď C ´

} g

¹_e_˚_´_δ¹{p2αqăe0pθ,vqăe˚

}

L¹

` }B

^v

f

0¹e˚´δ¹^{p²^αqăe0pθ,vqăe˚

}

L¹

¯ , so by dominated convergence,

δ

lim

Ñ0

} R r

_δ

}

L¹

“ 0.

We now choose δ small enough such that } R

_δ

}

L¹

ď

ˆ 4δ

0

Re z

_g

˙

N

.

From (3.21), we obtain (3.17), which ends the proof of Theorem 1.2.

(21)

A Appendix. Existence of unstable steady states

In this section, we prove that the set of steady states satisfying the assumptions of Theorems 1.1 and 1.2 is not empty. More precisely, we prove the following

Lemma A.1. Let m ą 0. There exist m ą 0, e

_˚

ă m and there exists a nonincreasing function F , C

⁸

on R , such that F p e q ą 0 for e ă e

_˚

, F p e q “ 0 for e ě e

_˚

and

| F

¹

p e q| ď C | e

_˚

´ e |

^´^α

F p e q in the neighborhood of e

_˚

, for some α ě 1, and such that the function f p θ, v q “ F p

^v2²

´ m cos θ q is a steady state solution to the HMF model (1.1) and such that κ p m, F q ą 1, where κ p m, F q is given by

κ p m, F q “ ż

2π

0

ż

_`8

´8

ˇ ˇF

¹

p e p θ, v qq ˇ ˇ

¨

˚ ˚

˚ ˝ ż

Depθ,vq

p cos θ ´ cos θ

¹

qp e p θ, v q ` m cos θ

¹

q

^´¹^{²

dθ

¹

ż

Depθ,vq

p e p θ, v q ` m cos θ

¹

q

^´¹^{²

dθ

¹

˛

‹ ‹

‹ ‚

2

dθdv,

with

e p θ, v q “ v

²

2 ´ m cos θ, D

_e

“ θ

¹

P T : m cos θ

¹

ą ´ e ( .

Proof. Let m ą 0 and F a nonincreasing C

8

function on R supported in p´8 , m q , which is not identically zero on p´ m, m q . We first observe that f p θ, v q “ F p

^v2²