HAL Id: hal-02047472
https://hal.archives-ouvertes.fr/hal-02047472
Preprint submitted on 24 Feb 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 teaching and research institutions in France or abroad, or from public or private research centers.
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 recherche français ou étrangers, des laboratoires publics ou privés.
Asymptotic behavior for the Vlasov-Poisson equations with strong external curved magnetic field. Part II :
general initial conditions
Mihai Bostan
To cite this version:
Mihai Bostan. Asymptotic behavior for the Vlasov-Poisson equations with strong external curved
magnetic field. Part II : general initial conditions. 2019. �hal-02047472�
Asymptotic behavior for the Vlasov-Poisson
equations with strong external curved magnetic field.
Part II : general initial conditions
Miha¨ı BOSTAN ∗ (February 4, 2019)
Abstract
We discuss the asymptotic behavior of the Vlasov-Poisson system in the framework of the magnetic confinement, that is, under a strong external mag- netic field. We concentrate on curved three dimensional magnetic fields. We derive second order approximations, when the magnetic field becomes large, for general initial particle densities.
Keywords: Vlasov-Poisson system, averaging, homogenization.
AMS classification: 35Q75, 78A35, 82D10.
1 Introduction
The asymptotic analysis of the transport of charged particles under strong magnetic fields is a very important topic in plasma physics [7, 8, 9, 10, 14, 15, 11, 12, 13, 1, 2, 3, 4, 5]. It is related to real life applications, such that the energy production through magnetic confinement. When the particle velocities are small with respect to the light speed, the evolution of the particle density f = f(t, x, v) is described by the Vlasov-Poisson system
∂
tf
ε+ v ·∇
xf
ε+ q
m {E[f
ε(t)](x) + v ∧ B
ε(x)}·∇
vf
ε= 0, (t, x, v) ∈ R
+× R
3× R
3. (1) E[f
ε(t)] = −∇
xΦ[f
ε(t)], Φ[f
ε(t)](x) = q
4π
0Z
R3
Z
R3
f
ε(t, x
0, v
0)
|x − x
0| dv
0dx
0where ε > 0 is a small parameter, entering the strong external non vanishing magnetic field
B
ε(x) = B
ε(x)e(x), B
ε(x) = B(x)
ε , |e(x)| = 1, x ∈ R
3.
∗
Aix Marseille Universit´ e, CNRS, Centrale Marseille, I2M, Marseille France, Centre de
Math´ ematiques et Informatique, UMR 7373, 39 rue Fr´ ed´ eric Joliot Curie, 13453 Marseille Cedex
13 France. E-mail : mihai.bostan@univ-amu.fr.
The potential Φ[f
ε] satisfies the Poisson equation
−∆
xΦ[f
ε(t)] = q
0Z
R3
f
ε(t, x, v) dv, (t, x) ∈ R
+× R
3whose fundamental solution is z →
4π|z|1, z ∈ R
3\ {0}. Here
0represents the electric permittivity. For any particle density f = f(x, v), the notation E[f] stands for the Poisson electric field
E[f ](x) = q 4π
0Z
R3
Z
R3
f (x
0, v
0) x − x
0|x − x
0|
3dv
0dx
0(2) and ρ[f ], j[f] are the charge and current densities respectively
ρ[f ] = q Z
R3
f (·, v) dv, j[f ] = q Z
R3
f (·, v)v dv.
The above system is supplemented by the initial condition f
ε(0, x, v) = f
in(x, v), (x, v) ∈ R
3× R
3.
In [6] a regular reformulation (when ε & 0) of the Vlasov-Poisson system has been derived, in the three dimensional setting, for well prepared initial particle densities. In this work we extend the previous analysis to general initial particle densities. Consider- ing general initial conditions leads to fast oscillations in time. In order to describe the asymptotic behavior (when ε & 0), we need to introduce a fast time variable s = t/ε.
The analysis follows closely that in [6] and the arguments rely on averaging along the flow of a vector field. As a fast time variable has been introduced, we need to consider the extended phase space (s, x, v) for averaging functions and vector fields.
Our paper is organized as follows. The average operators on the extended phase space and main properties are discussed in Section 2. The regular reformulation of the Vlasov-Poisson problem with strong external magnetic field is derived in Section 3 and revisited in the last Section 4.
2 Average operators and main properties
As in [6], we introduce the relative velocity with respect to the electric cross field drift
˜
v = v − ε E
ε(t, x) ∧ e(x) B (x) .
Accordingly, at any time t ∈ [0, T ], we consider the new particle density f ˜
ε(t, x, v) = ˜ f
εt, x, v ˜ + ε E[f
ε(t)](x) ∧ e(x) B(x)
, (x, ˜ v) ∈ R
3× R
3. The particle densities f
ε, f ˜
εhave the same charge density
ρ[ ˜ f
ε(t)] = q Z
R3
f ˜
ε(t, ·, ˜ v) d˜ v = q Z
R3
f
ε(t, ·, v) dv = ρ[f
ε(t)], t ∈ [0, T ]
implying that the Poisson electric fields corresponding to f
ε, f ˜
εcoincide E[f
ε(t)] = E[ ˜ f
ε(t)], t ∈ [0, T ].
Therefore we can use the same notation E
ε(t) for denoting them. We assume that the magnetic field satisfies
B
0:= inf
x∈R3
|B(x)| > 0 or equivalently ω
0:= inf
x∈R3
|ω
c(x)| > 0.
The new particle densities ( ˜ f
ε)
ε>0verify
∂
tf ˜
ε+
˜
v + ε E
ε∧ e B
· ∇
xf ˜
ε− ε
∂
tE
ε∧ e B + ∂
xE
ε∧ e
B ˜ v + ε E
ε∧ e B
· ∇
˜vf ˜
ε+ h ω
cε v ˜ ∧ e + q
m (E
ε· e) e i
· ∇
˜vf ˜
ε= 0, (t, x, v ˜ ) ∈ [0, T ] × R
3× R
3(3) f ˜
ε(0, x, ˜ v) = f
inx, v ˜ + ε E[f
in](x) ∧ e(x) B(x)
, (x, ˜ v) ∈ R
3× R
3. As in [6], thanks to the continuity equation
∂
tρ[f
ε] + div
xj[f
ε] = 0
we obtain the following representation for the time derivative of the electric field E
ε, in terms of the particle density ˜ f
ε∂
tE[f
ε] = − 1 4π
0div
xZ
R3
x − x
0|x − x
0|
3⊗
j [ ˜ f
ε(t)](x
0) + ερ[ ˜ f
ε(t)](x
0) E
ε(t, x
0) ∧ e(x
0) B(x
0)
dx
0. We introduce the new Larmor center ˜ x = x + ε
˜v∧e(x)ωc(x)
, which is a second order ap- proximation of the Larmor center x + ε
v∧e(x)ωc(x)
. We decompose the transport field in the Vlasov equation in such a way that ˜ x remains invariant with respect to the fast dynamics. We will distinguish between the orthogonal and parallel directions, taking as reference direction the magnetic line passing through the new Larmor center ˜ x, that is e(˜ x) (which is left invariant with respect to the fast dynamics)
˜
v = [˜ v − (˜ v · e(˜ x))e(˜ x)] + (˜ v · e(˜ x))e(˜ x).
Finally the Vlasov equation (3) writes
∂
tf ˜
ε+c
ε[ ˜ f
ε(t)]·∇
x,˜vf ˜
ε+εa
ε[ ˜ f
ε(t)]·∇
x,˜vf ˜
ε+ b
εε ·∇
x,˜vf ˜
ε= 0, (t, x, ˜ v) ∈ [0, T ]× R
3× R
3(4) where the autonomous vector field
bεε· ∇
x,˜vis given by
b
εε · ∇
x,˜v= [˜ v − (˜ v · e(˜ x)) e(˜ x) + εA
εx(x, v)] ˜ · ∇
x+ ω
c(x)
ε (˜ v ∧ e(˜ x)) · ∇
v˜and for any particle density ˜ f , a
ε[ ˜ f ] · ∇
x,˜v, c
ε[ ˜ f] · ∇
x,˜vstand for the vector fields a
ε[ ˜ f] · ∇
x,˜v= E[ ˜ f] ∧ e
B − A
εx!
· ∇
x+
"
−∂
xE[ ˜ f ] ∧ e B
!
˜
v + ε E [ ˜ f ] ∧ e B
!
+ 1
4π
0B div
xZ
R3
x − x
0|x − x
0|
3⊗ j [ ˜ f] + ερ[ ˜ f] E[ ˜ f] ∧ e B
!
(x
0) dx
0∧ e(x)
#
· ∇
v˜c
ε[ ˜ f ] · ∇
x,˜v= (˜ v · e(˜ x)) e(˜ x) · ∇
x+
ω
c(x)˜ v ∧ e(x) − e(˜ x)
ε + q
m (E[ ˜ f] · e(x)) e(x)
· ∇
˜v= q
m (E[ ˜ f ] · e(x)) e(x) − ω
cv ˜ ∧ Z
10
∂
xe
x + εs ˜ v ∧ e(x) ω
c(x)
v ˜ ∧ e(x) ω
c(x) ds
· ∇
v˜+ (˜ v · e(˜ x)) e(˜ x) · ∇
x.
The vector field A
εx(x, ˜ v) · ∇
xwill be determined by imposing that the Larmor center
˜
x is left invariant by the fast dynamics b
ε· ∇
x,˜vx + ε v ˜ ∧ e(x) ω
c(x)
= 0 that is
I
3+ ε∂
x˜ v ∧ e ω
cA
εx(x, v) = ˜ −∂
x˜ v ∧ e ω
c[˜ v−(˜ v·e(˜ x)) e(˜ x)]− e(˜ x) − e(x)
ε ∧(˜ v∧e(˜ x)).
The method employed in [6] applies as well when the initial particle density is not well prepared. In this case we deal with two time scales: the slow time variable t and the fast time variable s = t/ε. We need to average in the extended phase space(s, x, v ˜ ).
We say that a function u = u(s, x, ˜ v) is S = S(x, v) periodic with respect to ˜ s iff u(s + S(x, v), x, ˜ v) = ˜ u(s, x, v), ˜ (s, x, v) ˜ ∈ R × R
3× R
3.
Similarly, we say that a function u = u(s, x, v) is ˜ S
ε= S
ε(x, v) periodic with respect ˜ to s iff
u(s + S
ε(x, ˜ v), x, v) = ˜ u(s, x, ˜ v), (s, x, v) ˜ ∈ R × R
3× R
3.
With the notations in [6] Propositions 3.1, 3.2, we observe that if u is S periodic with respect to s, therefore the function (s, x, ˜ v) → u(Λ
ε(s; x, ˜ v), T
ε(x, v)) is ˜ S
εperiodic with respect to s. For establishing that, notice that
Λ
ε(s + S
ε(x, ˜ v); x, v ˜ ) = Λ
ε(s; x, ˜ v) + S(T
ε(x, v)). ˜ Indeed, we have, thanks to Proposition 3.2 [6]
Λ
ε(s + S
ε(x, v ˜ ); x, v) = ˜
Z
s+Sε(x,˜v) 0λ
ε( X
ε(σ; x, v), ˜ V ˜
ε(σ; x, ˜ v)) dσ
= Z
s0
λ
ε( X
ε(σ; x, v), ˜ V ˜
ε(σ; x, ˜ v)) dσ +
Z
Sε(x,˜v) 0λ
ε(( X
ε, V ˜
ε)(τ ; ( X
ε, V ˜
ε)(s; x, ˜ v))) dτ
= Λ
ε(s; x, v) + ˜
Z
Sε((Xε,V˜ε)(s;x,˜v)) 0λ
ε(( X
ε, V ˜
ε)(τ; ( X
ε, V ˜
ε)(s; x, ˜ v))) dτ
= Λ
ε(s; x, v) + Λ ˜
ε(S
ε(( X
ε, V ˜
ε)(s; x, ˜ v)); ( X
ε, V ˜
ε)(s; x, ˜ v))
= Λ
ε(s; x, v) + ˜ S(T
ε(( X
ε, V ˜
ε)(s; x, ˜ v)))
= Λ
ε(s; x, v) + ˜ S(( X , V ˜ )(Λ
ε(s; x, v); ˜ T
ε(x, ˜ v)))
= Λ
ε(s; x, v) + ˜ S(T
ε(x, v)). ˜
It is easily seen that
u(Λ
ε(s + S
ε(x, v ˜ ); x, v), T ˜
ε(x, v)) = ˜ u(Λ
ε(s; x, ˜ v) + S(T
ε(x, v)), T ˜
ε(x, ˜ v))
= u(Λ
ε(s; x, ˜ v), T
ε(x, ˜ v))
saying that the function (s, x, v) ˜ → u(Λ
ε(s; x, v ˜ ), T
ε(x, v)) is ˜ S
εperiodic with respect to s.
Observe that for any (s, x, ˜ v) ∈ R × R
3× R
3, the characteristics of ∂
s+ b · ∇
x,˜v, ∂
s+ b
ε· ∇
x,˜vissued from (s, x, v) are ˜
(s + σ, X (σ; x, ˜ v), V ˜ (σ; x, v ˜ )), (s + σ, X
ε(σ; x, v), ˜ V ˜
ε(σ; x, v ˜ ))
respectively. We define the average operators for continuous S periodic, S
εperiodic functions by
hui (s, x, v) = ˜ 1 S(x, v) ˜
Z
S(x,˜v) 0u(s + σ, X (σ; x, ˜ v), V ˜ (σ; x, ˜ v)) dσ, (s, x, v ˜ ) ∈ R × R
3× R
3hui
ε(s, x, ˜ v) = 1
S
ε(x, v) ˜
Z
Sε(x,˜v) 0u(s+σ, X
ε(σ; x, v), ˜ V ˜
ε(σ; x, ˜ v)) dσ, (s, x, v) ˜ ∈ R × R
3× R
3. Notice that the above operators extend the corresponding average operators defined in Proposition 3.1 [6] for continuous functions, not depending on s. As in Proposition 3.2 [6], we establish a relation between the average operators h·i , h·i
ε. We will work under the hypothesis ∇
xω
c= 0, implying that S(x, ˜ v) = S
ε(x, v ˜ ) = 2π/ω
c, λ
ε(x, ˜ v) = 1, Λ
ε(s; x, ˜ v) = s, (s, x, v) ˜ ∈ R × R
3× R
3.
Proposition 2.1
Let u ∈ C( R × R
3× R
3) be a S periodic function with respect to s such that supp u ⊂ {(s, x, v) ˜ ∈ R × R
3× R
3: |˜ v| ≤ R} for some R > 0. For any ε > 0 satisfying εRk∂
xek
L∞/|ω
c| < 1 we have
hu(·, T
ε)i
ε= hui (·, T
ε).
Proof.
It is enough to consider (s, x, ˜ v) ∈ R × R
3× R
3such that |˜ v| ≤ R. In that case we have, cf. Proposition 3.2 [6]
hu(·, T
ε)i
ε(s, x, ˜ v) = 1 S
Z
S 0u(s + σ, T
ε(( X
ε, V ˜
ε)(σ; x, v))) dσ ˜
= 1 S
Z
S 0u(s + σ, ( X , V ˜ )(σ; T
ε(x, ˜ v))) dσ
= hui (s, T
ε(x, v)). ˜
We also need to adapt the result in Proposition 3.3 [6] for S periodic functions.
Proposition 2.2
Let z ∈ C( R × R
3× R
3) be a S periodic function of zero average hzi (s, x, ˜ v) = 1
S Z
S0
z(s + σ, x, V ˜ (σ; x, ˜ v)) dσ = 0, (s, x, v) ˜ ∈ R × R
3× R
3.
1. There is a unique continuous S periodic function u of zero average whose deriva- tive along the flow of ∂
s+ b · ∇
x,˜vis z
(∂
s+ b · ∇
x,˜v)u = z, hui = 0.
If z is bounded, so is u and
kuk
C(R×B(Rx)×B(Rv˜))≤ S
2 kzk
C(R×B(Rx)×B(Rv˜))for any R
x, R
˜v> 0. If supp z ⊂ R × B (R
x) × B(R
v˜), then supp u ⊂ R × B(R
x) × B(R
˜v).
2. If z is of class C
1, then so is u and we have for any R
x, R
v˜> 0 k∇
v˜uk
C(R×B(Rx)×B(Rv˜))≤ S √
3k∇
˜vzk
C(R×B(Rx)×B(R˜v))k∇
xuk
C(R×B(Rx)×B(R˜v))≤ C k∇
xzk
C(R×B(Rx)×B(Rv˜))+ R
v˜k∇
˜vzk
C(R×B(Rx)×B(R˜v))k∂
suk
C(R×B(Rx)×B(Rv˜))≤ kzk
C(R×B(Rx)×B(R˜v))+ 2 √
3R
v˜k∇
˜vzk
C(R×B(Rx)×B(R˜v))for some constant C depending on k∂
xek
L∞and S.
Proof.
1. Take
u(s, x, ˜ v) = 1 S
Z
S 0(σ − S)z(s + σ, ( X , V ˜ )(σ; x, v)) dσ, ˜ (s, x, v) ˜ ∈ R × R
3× R
3. 2. Use the vector fields (c
i· ∇
x,˜v)
1≤i≤6, see Proposition 3.1 [6], which are in involution with ∂
s+ b · ∇
x,˜v, since ∇
xω
c= 0.
3 The limit model and convergence result
We are ready to investigate the limit model in (4) as ε & 0. In this case we intend to capture the fast oscillations due to the operator ∂
t+
bεε· ∇
x,˜v. We are looking for a development whose dominant term belongs to the kernel of ∂
t+
bεε· ∇
x,˜v. It is easily seen that for any function u ∈ ker(∂
s+ b · ∇
x,˜v), we have u(·, T
ε) ∈ ker(∂
s+ b
ε· ∇
x,˜v), since
u(s + σ, T
ε(( X
ε, V ˜
ε)(σ; x, v))) = ˜ u(s + σ, ( X , V ˜ )(σ; T
ε(x, v))) = ˜ u(s, T
ε(x, v)). ˜ Similarly, for any S periodic function of zero average hui = 0, the S periodic function u(·, T
ε) has zero average
hu(·, T
ε)i
ε= hui (·, T
ε) = 0.
The previous discussion suggests to consider the Ansatz
f ˜
ε(t) = ˜ f
ε(t, t/ε) ◦ T
ε+ ε f ˜
ε1(t, t/ε) ◦ T
ε+ ε
2f ˜
ε2(t, t/ε) ◦ T
ε+ . . . (5) where (∂
s+ b · ∇
x,˜v) ˜ f
ε= 0, D
f ˜
ε1E
= 0. As in [6], at the leading order, the particle
density ˜ f
εhas no fluctuation (with respect to the extended average operator), and the
averages at the orders O(ε
0), O(ε) combine together in ˜ f
ε(t, t/ε) ◦ T
ε. Notice also that the constraint (∂
s+b ·∇
x,˜v) ˜ f
ε= 0 is equivalent to ˜ f
ε(t, s, x, v) = ˜ ˜ F
ε(t, ( X , V ˜ )(−s; x, ˜ v)), for some function ˜ F
ε(t). We are looking for a closure determining ˜ f
ε, f ˜
ε1. The error estimate will require to introduce the second order correction ε
2f ˜
ε2(t, t/ε)◦T
ε. Plugging the Ansatz (5) in (4) we obtain
1
ε ∂
sf ˜
ε(t, s) ◦ T
ε+ ∂
tf ˜
ε(t, s) ◦ T
ε+ ∂
sf ˜
ε1(t, s) ◦ T
ε+ ε∂
tf ˜
ε1(t, s) ◦ T
ε+ ε∂
sf ˜
ε2(t, s) ◦ T
ε+ . . . + c
ε[ ˜ f
ε(t, s) ◦ T
ε+ ε f ˜
ε1(t, s) ◦ T
ε+ . . .] · ∇[( ˜ f
ε+ ε f ˜
ε1+ . . .)(t, s) ◦ T
ε] + εa
ε[( ˜ f
ε+ . . .)(t, s) ◦ T
ε] · ∇[( ˜ f
ε+ . . .)(t, s) ◦ T
ε]
+ b
εε · ∇[( ˜ f
ε+ ε f ˜
ε1+ ε
2f ˜
ε2+ . . .)(t, s) ◦ T
ε] = 0.
By construction we have (∂
s+ b
ε· ∇)( ˜ f
ε◦ T
ε) = 0, and therefore we deduce
∂
tf ˜
ε(t, s) ◦ T
ε+ (∂
s+ b
ε· ∇)(( ˜ f
ε1+ ε f ˜
ε2)(t, s) ◦ T
ε) + ε∂
tf ˜
ε1(t, s) ◦ T
ε(6) + c
ε[( ˜ f
ε+ ε f ˜
ε1)(t, s) ◦ T
ε] · ∇[( ˜ f
ε+ ε f ˜
ε1)(t, s) ◦ T
ε]
+ εa
ε[ ˜ f
ε(t, s) ◦ T
ε] · ∇[ ˜ f
ε(t, s) ◦ T
ε] = O(ε
2).
We will take the (extended) average of (6) by discarding all second order contributions.
Obviously we have
D
∂
tf ˜
ε(t, ·) ◦ T
εE
ε
= ∂
tf ˜
ε(t, ·) ◦ T
εD
(∂
s+ b
ε· ∇
x,˜v)( ˜ f
ε1+ ε f ˜
ε2)(t, ·) ◦ T
ε) E
ε
= 0 D
∂
tf ˜
ε1(t, ·) ◦ T
εE
ε
= D
∂
tf ˜
ε1(t, ·) E
◦ T
ε= ∂
tD
f ˜
ε1(t, ·) E
◦ T
ε= 0 and
εa
ε[ ˜ f
ε(t, s) ◦ T
ε] · ∇
x,˜v[ ˜ f
ε(t, s) ◦ T
ε] = ε(a[ ˜ f
ε(t, s)] · ∇
x,˜vf ˜
ε(t, s)) ◦ T
ε+ O(ε
2) which implies, cf. Proposition 2.1
D
εa
ε[ ˜ f
ε(t, ·) ◦ T
ε] · ∇
x,˜v[ ˜ f
ε(t, ·) ◦ T
ε] E
ε
= ε D
(a[ ˜ f
ε(t, ·)] · ∇
x,˜vf ˜
ε(t, ·)) ◦ T
εE
ε
+ O(ε
2)
= ε D
a[ ˜ f
ε(t, ·)] · ∇
x,˜vf ˜
ε(t, ·) E
◦ T
ε+ O(ε
2).
We concentrate now on the term corresponding to the vector field c
ε· ∇
x,˜vc
ε[( ˜ f
ε+ ε f ˜
ε1)(t, s) ◦ T
ε] · [( ˜ f
ε+ ε f ˜
ε1)(t, s) ◦ T
ε] = c
ε[ ˜ f
ε(t, s) ◦ T
ε] · ∇( ˜ f
ε(t, s) ◦ T
ε)
+ ε(c
ε[ ˜ f
ε(t, s)] · ∇ f ˜
ε1(t, s)) ◦ T
ε+ ε q
m [(E[ ˜ f
ε1(t, s)] · e)(e · ∇
˜vf ˜
ε(t, s)] ◦ T
ε+ O(ε
2)
= (c
0[ ˜ f
ε(t, s)] · ∇ f ˜
ε(t, s)) ◦ T
ε+ ε(c
1[ ˜ f
ε(t, s)] · ∇ f ˜
ε(t, s)) ◦ T
ε+ ε(c
0[ ˜ f
ε(t, s)] · ∇ f ˜
ε1(t, s)) ◦ T
ε+ ε q
m [(E[ ˜ f
ε1(t, s)] · e)(e · ∇
v˜f ˜
ε)] ◦ T
ε+ O(ε
2) where
c
0[ ˜ f ] · ∇
x,˜v= (˜ v · e)e · ∇
x+ q
m (E[ ˜ f ] · e)e · ∇
˜v− [˜ v ∧ ∂
xe(˜ v ∧ e)] · ∇
˜v.
We claim that the average along the flow of ∂
s+ b · ∇
x,˜vof (E[ ˜ f
ε1(t, s)] · e)e · ∇
˜vf ˜
ε(t, s) vanishes. Indeed, as e · ∇
˜vis in involution with respect to ∂
s+ b · ∇
x,˜v, we have
(∂
s+ b · ∇
x,˜v)(e · ∇
v˜f ˜
ε(t)) = e · ∇
v˜((∂
s+ b · ∇
x,˜v) ˜ f
ε(t)) = 0 implying that
D
(E[ ˜ f
ε1(t, ·)] · e)(e · ∇
˜vf ˜
ε(t)) E
= D
E[ ˜ f
ε1(t)] · e E
e · ∇
v˜f ˜
ε(t) = 0 because the charge density of ˜ f
ε1(t) has zero average
D
ρ[ ˜ f
ε1(t)] E
(s, x) = q S
Z
S 0Z
R3
f ˜
ε1(t, s + σ, x, ˜ v) d˜ vdσ
= q S
Z
S 0Z
R3
f ˜
ε1(t, s + σ, x, V ˜ (σ; x, ˜ v)) d˜ vdσ
= q Z
R3
D f ˜
ε1(t) E
(s, x, v) d˜ ˜ v = 0.
Therefore, thanks to Proposition 2.1 we obtain D
[(E[ ˜ f
ε1(t)] · e)(e · ∇
˜vf ˜
ε(t))] ◦ T
εE
ε
= D
(E[ ˜ f
ε1(t)] · e)(e · ∇
˜vf ˜
ε(t)) E
◦ T
ε= 0 and thus
D
c
ε[( ˜ f
ε(t, ·) + ε f ˜
ε1(t, ·)) ◦ T
ε] · ∇[( ˜ f
ε(t, ·) + ε f ˜
ε1(t, ·)) ◦ T
ε] E
ε
= D
c
0[ ˜ f
ε(t, ·)] · ∇ f ˜
ε(t, ·) E
◦ T
ε+ ε D
c
1[ ˜ f
ε(t, ·)] · ∇ f ˜
ε(t, ·) E
◦ T
ε+ ε D
c
0[ ˜ f
ε(t, ·)] · ∇ f ˜
ε1(t, ·) E
◦ T
ε+ O(ε
2).
The previous computations lead to the following model for the particle density ˜ f
ε∂
tf ˜
ε(t, s) + D
c
0[ ˜ f
ε(t, ·)] · ∇ f ˜
ε(t, ·) E + ε D
(a[ ˜ f
ε(t, ·)] + c
1[ ˜ f
ε(t, ·)]) · ∇ f ˜
ε(t, ·) E
(7) + ε D
c
0[ ˜ f
ε(t, ·)] · ∇ f ˜
ε1(t, ·) E
= 0, (t, s, x, v) ˜ ∈ [0, T ] × R × R
3× R
3together with the constraint (∂
s+ b · ∇
x,˜v) ˜ f
ε= 0. The equation for the fluctuation ˜ f
ε1comes by comparing (6) with respect to (7). Indeed, we have
(∂
s+ b
ε· ∇)[( ˜ f
ε1+ ε f ˜
ε2)(t, s) ◦ T
ε] = ∂
s( ˜ f
ε1+ ε f ˜
ε2) ◦ T
ε+ ∂T
εb
ε| {z }
b◦Tε
·[∇( ˜ f
ε1+ ε f ˜
ε2)] ◦ T
ε= [(∂
s+ b · ∇)( ˜ f
ε1+ ε f ˜
ε2)] ◦ T
εand therefore (6) also writes
∂
tf ˜
ε(t, s) + (∂
s+ b · ∇)( ˜ f
ε1+ ε f ˜
ε2) + ε∂
tf ˜
ε1+ c
0[ ˜ f
ε(t, s)] · ∇ f ˜
ε(t, s) (8) + ε(a[ ˜ f
ε(t, s)] + c
1[ ˜ f
ε(t, s)]) · ∇ f ˜
ε(t, s) + εc
0[ ˜ f
ε(t, s)] · ∇ f ˜
ε1(t, s)
+ ε q
m (E[ ˜ f
ε1(t, s)]·)(e · ∇
˜vf ˜
ε(t, s)) = O(ε
2).
Taking the difference between (8), (7) yields c
0[ ˜ f
ε(t, s)] · ∇ f ˜
ε(t, s) − D
c
0[ ˜ f
ε(t, ·)] · ∇ f ˜
ε(t, ·) E
+ ε(a[ ˜ f
ε(t, s)] + c
1[ ˜ f
ε(t, s)]) · ∇ f ˜
ε(t, s)
− ε D
(a[ ˜ f
ε(t, ·)] + c
1[ ˜ f
ε(t, ·)]) · ∇ f ˜
ε(t, ·) E + εc
0[ ˜ f
ε(t, s)] · ∇ f ˜
ε1(t, s) − ε
D
c
0[ ˜ f
ε(t, ·)] · ∇ f ˜
ε1(t, ·) E + ε q
m (E[ ˜ f
ε1(t, s)] · e)(e · ∇
v˜f ˜
ε(t, s)) + (∂
s+ b · ∇)( ˜ f
ε1+ ε f ˜
ε2) + ε∂
tf ˜
ε1= O(ε
2).
The above equality suggests to determine the fluctuation ˜ f
ε1by c
0[ ˜ f
ε(t, s)] · ∇ f ˜
ε(t, s) − D
c
0[ ˜ f
ε(t, ·)] · ∇ f ˜
ε(t, ·) E
+ (∂
s+ b · ∇) ˜ f
ε1(t, s) = 0, D f ˜
ε1E
= 0 (9) and to consider the corrector ˜ f
ε2such that
(a[ ˜ f
ε(t, s)] + c
1[ ˜ f
ε(t, s)]) · ∇ f ˜
ε(t, s) − D
(a[ ˜ f
ε(t, ·)] + c
1[ ˜ f
ε(t, ·)]) · ∇ f ˜
ε(t, ·) E + c
0[ ˜ f
ε(t, s)] · ∇ f ˜
ε1(t, s) − D
c
0[ ˜ f
ε(t, ·)] · ∇ f ˜
ε1(t, ·) E + q
m (E[ ˜ f
ε1(t, s)] · e)(e · ∇
v˜f ˜
ε(t, s)) + ∂
tf ˜
ε1+ (∂
s+ b · ∇) ˜ f
ε2= 0, D f ˜
ε2E
= 0.
The well posedness of (7), (9) is stated in Section 4, see Theorem 4.1. As in Theorem 1.2 [6], we can establish the following error estimate. The proof details are left to the reader.
Theorem 3.1
Let B = Be ∈ C
b4( R
3) be a smooth magnetic field such that ∇
xB = 0, div
xB = 0.
Consider a non negative, smooth, compactly supported initial particle density G ˜ ∈ C
c3( R
3× R
3) and g(s, x, ˜ v) = ˜ ˜ G(x, V ˜ (−s; x, v ˜ )), (s, x, v) ˜ ∈ R × R
3× R
3. We denote by (f
ε)
0<ε≤1the solutions of the Vlasov-Poisson equations with external magnetic field (1), (2) on [0, T ], 0 < T < T (f
in), cf. Theorem 2.1 [6], corresponding to the initial condition
f
ε(0, x, v) = (˜ g + ε˜ g
1) 0, x + ε v ∧ (x)
ω
c, v − ε E[ ˜ G] ∧ e(x) B
!
, (x, v) ∈ R
3× R
3where
c
0[˜ g] · ∇˜ g − hc
0[˜ g] · ∇˜ gi + (∂
s+ b · ∇)˜ g
1= 0,
˜ g
1= 0.
For ε ∈]0, ε
T] small enough, we consider the solution f ˜
ε= ˜ f
ε(t, s, x, ˜ v) on [0, T ] of the problem
∂
tf ˜
ε(t, s) + D
c
0[ ˜ f
ε(t, ·)] · ∇ f ˜
ε(t, ·) E + ε D
(a[ ˜ f
ε(t, ·)] + c
1[ ˜ f
ε(t, ·)]) · ∇ f ˜
ε(t, ·) E + ε D
c
0[ ˜ f
ε(t, ·)] · ∇ f ˜
ε1(t, ·) E
= 0, (t, s, x, v) ˜ ∈ [0, T ] × R × R
3× R
3c
0[ ˜ f
ε(t, s)] · ∇ f ˜
ε(t, s) − D
c
0[ ˜ f
ε(t, ·)] · ∇ f ˜
ε(t, ·) E
+ (∂
s+ b · ∇) ˜ f
ε1(t, s) = 0, D f ˜
ε1E
= 0
corresponding to the initial condition
f ˜
ε(0, s, x, ˜ v) = ˜ G(x, V ˜ (−s; x, v)) = ˜ ˜ g(s, x, ˜ v), (s, x, v) ˜ ∈ R × R
3× R
3. Therefore there is a constant C
Tsuch that for any 0 < ε ≤ ε
Tsup
t∈[0,T]
Z
R3
Z
R3
f
ε(t, x, v) − ( ˜ f
ε+ ε f ˜
ε1) t, t/ε, x + ε v ∧ e
ω
c, v − ε E[ ˜ f
ε(t, t/ε)] ∧ e B
!
2
dvdx
1/2
≤ C
Tε
2.
4 Equivalent formulation of the limit model
We determine now the equivalent formulation for (7), (9) by computing the average of the vector fields entering this model. Most of the computations has been performed in [6], where the formulae for hai · ∇, hc
0i · ∇, hc
1i · ∇ are detailed, and we only need to complete them by treating the extra terms due to the presence of the fast time variable.
Proposition 4.1
Assume that e ∈ C
2( R
3), ∇
xω
c= 0, div
xe = 0 and let us consider f ˜
ε∈ C
2([0, T ] × R × R
3× R
3). Then f ˜
εsolves
∂
tf ˜
ε+ D
c
0[ ˜ f
ε] · ∇ f ˜
εE + ε D
(a[ ˜ f
ε] + c
1[ ˜ f
ε]) · ∇ f ˜
εE + ε D
c
0[ ˜ f
ε] · ∇ f ˜
ε1E
= 0, (∂
s+ b · ∇) ˜ f
ε= 0 (10) c
0[ ˜ f
ε] · ∇ f ˜
ε− D
c
0[ ˜ f
ε] · ∇ f ˜
εE
+ (∂
s+ b · ∇) ˜ f
ε1= 0, D f ˜
ε1E
= 0 (11)
iff f ˜
εsatisfies f ˜
ε(t, s, x, v ˜ ) = ˜ F
ε(t, ( X , V ˜ )(−s; x, v)), where ˜
∂
tF ˜
ε+ D
c
0[ ˜ F
ε] E
· ∇ F ˜
ε+ ε D a[ ˜ F
ε] E
+ D
c
1[ ˜ F
ε] E
· ∇ F ˜
ε+ εD[ ˜ F
ε] · ∇ F ˜
ε= 0 (12)
D[ ˜ F ] · ∇
X,V˜=
"
j[ ˜ F ] ∧ e(X) 3
0B +
R
R3
N (e(X), X − X
0, e(X
0))j[ ˜ F ](X
0) dX
08π
0B ∧ e(X)
#
· ∇
V˜+
"
q
m e ⊗ e E
"
(e · rot
Xe) ( ˜ V · e) ω
cF ˜
#
− E
"
V ˜ ∧ e
ω
c· ∇
XF ˜
#!
+ d
V˜(X, V ˜ )
#
· ∇
V˜d
V˜(X, V ˜ ) =
*
(c
0· ∇
x,˜v)
3
X
k=1
[(A
k)
−scos(ksω
c) + (B
k)
−ssin(ksω
c)]
+
(0, X, V ˜ )
f ˜
ε1(t, s, x, v) = ˜
3
X
k=1
[A
k(( X , V ˜ )(−s; x, v)) cos(ksω ˜
c) + B
k(( X , V ˜ )(−s; x, v)) sin(ksω ˜
c)]
· (∇
V˜F ˜
ε(t))(( X , V ˜ )(−s; x, v)) ˜ (13)
where N (e, z, e
0) = (I
3−e⊗e)K(z)(I
3−e
0⊗e
0)−M [e]K(z)M [e
0], K (z) = (I
3− 3z ⊗ z/|z|
2) /|z|
3A
k(X, V ˜ ) = 1
kπ Z
S0
F(s, X, V ˜ ) sin(ksω
c) ds, B
k(X, V ˜ ) = − 1 kπ
Z
S 0F (s, X, V ˜ ) cos(ksω
c) ds for k ∈ {1, 2, 3} and
F (s, X, V ˜ ) = ∂
x,˜vV ˜ (−s; ( X , V ˜ )(s; X, V ˜ ))c
0(( X , V ˜ )(s; X, V ˜ )) c
0(x, ˜ v) · ∇
x,˜v= (˜ v · e)e · ∇
x− [˜ v ∧ ∂
xe(˜ v ∧ e)] · ∇
˜v. Proof.
Clearly the constraint (∂
s+ b · ∇) ˜ f
ε= 0 writes ˜ f
ε(t, s, x, v ˜ ) = ˜ F
ε(t, ( X , V ˜ )(−s; x, v)) for ˜ some function ˜ F
ε= ˜ F
ε(t, X, V ˜ ) ∈ C
2([0, T ]× R
3× R
3). We need to compute the averages D
c
0[ ˜ f
ε] · ∇ f ˜
εE , D
(a[ ˜ f
ε] + c
1[ ˜ f
ε]) · ∇ f ˜
εE , D
c
0[ ˜ f
ε] · ∇ f ˜
ε1E
along the flow of ∂
s+b ·∇
x,˜vand to invert the operator ∂
s+ b · ∇
x,˜von zero average functions, in order to solve (11).
Recall that for any particle density ˜ f , the vector field a[ ˜ f] · ∇
x,˜vwrites a[ ˜ f] · ∇
x,˜v= E[ ˜ f] ∧ e
B − A
x(x, ˜ v)
!
· ∇
x− ∂
xE[ ˜ f] ∧ e B
!
˜ v · ∇
v˜+ 1
4π
0B
div
xZ
R3
x − x
0|x − x
0|
3⊗ j[ ˜ f](x
0) dx
0∧ e(x)
· ∇
˜v.
We have ρ[ ˜ f
ε(t, s)] = ρ[ ˜ F
ε(t)] implying that E[ ˜ f
ε(t, s)] = E[ ˜ F
ε(t)] and j[ ˜ f
ε(t, s)](x) = R(−sω
c, e(x))j [ ˜ F
ε(t)](x)
and therefore
a[ ˜ f
ε(t, s)] · ∇
x,˜v= a[ ˜ F
ε(t)] · ∇
x,˜v+ a
s[ ˜ F
ε(t)] · ∇
v˜where
a
s[ ˜ F
ε]·∇
˜v= 1 4π
0B
div
xZ
R3
x − x
0|x − x
0|
3⊗ [R(−sω
c, e(x
0)) − I
3] j[ ˜ F
ε](x
0) dx
0∧ e(x)
·∇
˜v.
Thanks to the equality ˜ f
ε(t, s, x, v) = ˜ ˜ f
ε(t, s + σ, ( X , V ˜ )(σ; x, v)) we have ˜
∇
x,˜vf ˜
ε(t, s, x, v) = ˜
t∂( X , V ˜ )(σ; x, v ˜ )(∇ f ˜
ε)(t, s + σ, ( X , V ˜ )(σ; x, v)) ˜ implying that
(∇ f ˜
ε)(t, s + σ, ( X , V ˜ )(σ; x, ˜ v)) =
t∂ ( X , V ˜ )(−σ; ( X , V ˜ )(σ; x, ˜ v))∇ f ˜
ε(t, s, x, ˜ v) and
(a[ ˜ f
ε] · ∇ f ˜
ε)(t, s + σ, ( X , V ˜ )(σ; x, ˜ v))
= ∂( X , V ˜ )(−σ; ( X , V ˜ )(σ; x, v))a[ ˜ ˜ F
ε(t)](( X , V ˜ )(σ; x, v)) ˜ · ∇ f ˜
ε(t, s, x, v) ˜
+ R(σω
c, e(x))a
s+σ[ ˜ F
ε(t)](x) · ∇
˜vf ˜
ε(t, s, x, v). ˜
Averaging with respect to σ, we obtain cf. Proposition 5.1 [6]
D
a[ ˜ f
ε] · ∇ f ˜
εE
(t, s, x, ˜ v) = 1 S
Z
S 0∂( X , V ˜ )(−σ; ( X , V ˜ )(σ; x, v))a[ ˜ ˜ F
ε(t)](( X , V ˜ )(σ; x, v)) dσ ˜
· ∇
x,˜vf ˜
ε(t, s, x, v ˜ ) + 1 S
Z
S 0R(σω
c, e(x))a
s+σ[ ˜ F
ε(t)](x) dσ · ∇
˜vf ˜
ε(t, s, x, ˜ v)
= D
a[ ˜ F
ε(t)] E
(x, v ˜ ) · ∇
x,˜vf ˜
ε(t, s, x, ˜ v) + 1 S
Z
S 0R(σω
c, e(x))a
s+σ[ ˜ F
ε(t)](x) dσ · ∇
˜vf ˜
ε(t, s, x, ˜ v)
= ∂( X , V ˜ )(−s; x, v) ˜ D
a[ ˜ F
ε(t)] E
(x, ˜ v) · (∇
X,V˜F ˜
ε)(t, ( X , V ˜ )(−s; x, v ˜ )) + 1
S Z
S0
R((s + σ)ω
c, e(x))a
s+σ[ ˜ F
ε(t)](x) dσ · (∇
V˜F ˜
ε)(t, ( X , V ˜ )(−s; x, v)) ˜
= D
a[ ˜ F
ε(t)] E
(( X , V ˜ )(−s; x, ˜ v)) · (∇
X,V˜F ˜
ε)(t, ( X , V ˜ )(−s; x, v)) ˜ + 1
S Z
S0
R(σω
c, e(x))a
σ[ ˜ F
ε(t)](x) dσ · (∇
V˜F ˜
ε)(t, ( X , V ˜ )(−s; x, ˜ v))
where hai · ∇ has been computed in Proposition 5.4 [6]. Notice that in the last equality we have used the involution of D
a[ ˜ F
ε(t)] E
· ∇ with respect to the vector field b · ∇, cf.
Proposition 5.1 [6]. For the last average, observe that 1
S Z
S0
R(σω
c, e(x))a
σ[ ˜ F
ε(t)](x) dσ
= 1
4π
0BS Z
S0
R(σω
c, e(x))
div
xZ
R3
x − x
0|x − x
0|
3⊗ R(−σω
c, e(x
0))j [ ˜ F
ε(t)](x
0) dx
0∧ e(x)
dσ
= 1
4π
0BS Z
S0
[cos(σω
c)(I
3− e(x) ⊗ e(x)) + sin(σω
c)M[e(x)] + e(x) ⊗ e(x)]
div
xZ
R3
x − x
0|x − x
0|
3⊗ [cos(σω
c)(I
3− e(x
0) ⊗ e(x
0)) − sin(σω
c)M[e(x
0)] + e(x
0) ⊗ e(x
0)]j[ ˜ F
ε(t)] dx
0∧ e(x) dσ
= 1
4π
0B
I
3− e ⊗ e 2
div
xZ
R3
x − x
0|x − x
0|
3⊗ [I
3− e(x
0) ⊗ e(x
0)]j[ ˜ F
ε(t)](x
0) dx
0∧ e(x)
− 1
4π
0B M [e]
2
div
xZ
R3
x − x
0|x − x
0|
3⊗ M [e(x
0)]j[ ˜ F
ε(t)](x
0) dx
0∧ e(x)
= − M [e]
8π
0B lim
δ&0
Z
|x−x0|>δ
[I
3− e(x) ⊗ e(x)](x − x
0)
|x − x
0|
3div{[I
3− e(x
0) ⊗ e(x
0)]j[ ˜ F
ε(t)](x
0)} dx
0+ M [e]
8π
0B lim
δ&0
Z
|x−x0|>δ
M [e(x)](x − x
0)
|x − x
0|
3div{M [e(x
0)]j[ ˜ F
ε(t)](x
0)} dx
0= − M[e(x)]
8π
0B lim
δ&0
Z
|x−x0|>δ
N (e(x), x − x
0, e(x
0))j[ ˜ F
ε(t)](x
0) dx
0− M [e(x)]
8π
0B lim
δ&0
Z
|x−x0|=δ
{[I
3− e(x) ⊗ e(x)] x − x
0|x − x
0|
3⊗ [I
3− e(x
0) ⊗ e(x
0)] x − x
0|x − x
0| + M [e(x)] x − x
0|x − x
0|
3⊗ M [e(x
0)] x − x
0|x − x
0| }j[ ˜ F
ε(t)](x
0) dσ(x
0)
where K(z) = (I
3− 3
|z|z⊗
|z|z)/|z|
3, z ∈ R
3\ {0}, N (e, z, e
0) = (I
3− e ⊗ e)K (z)(I
3− e
0⊗ e
0) − M [e]K(z)M[e
0], e, z, e
0∈ R
3\ {0}, |e| = |e
0| = 1. Performing the change of variable x
0= x + δz in the last integral yields
Z
|x−x0|=δ
{(I
3− e(x) ⊗ e(x)) x − x
0|x − x
0|
3⊗ (I
3− e(x
0) ⊗ e(x
0)) x − x
0|x − x
0| + M [e(x)] x − x
0|x − x
0|
3⊗ M [e(x
0)] x − x
0|x − x
0| }j[ ˜ F
ε(t)](x
0)dσ(x
0)
= Z
|z|=1
{(I
3− e(x) ⊗ e(x))z ⊗ (I
3− e(x + δz) ⊗ e(x + δz))z + M [e(x)]z ⊗ M [e(x + δz)]z}j [ ˜ F
ε(t)](x + δz)dσ(z)
δ&0
→ Z
|z|=1
{(I
3− e(x) ⊗ e(x))z ⊗ (I
3− e(x) ⊗ e(x))z + M [e(x)]z ⊗ M [e(x)]z}dσ(z) j[ ˜ F
ε(t)](x)
= Z
|z|=1
|z ∧ e(x)|
2(I
3− e(x) ⊗ e(x))dσ j[ ˜ F
ε(t)](x)
= 8π
3 (I
3− e(x) ⊗ e(x)) j[ ˜ F
ε(t)](x).
Therefore the average of a[ ˜ f
ε] · ∇
x,˜vf ˜
εwrites D
a[ ˜ f
ε] · ∇ f ˜
εE
(t, s, x, ˜ v) = D
a[ ˜ F
ε(t)] E
· ∇
X,V˜F ˜
ε(t)
(( X , V ˜ )(−s; x, v)) ˜ (14)
−
e(x)
8π
0B ∧ lim
δ&0
Z
|x−x0|>δ
N (e(x), x − x
0, e(x
0))j[ ˜ F
ε(t)](x
0) dx
0· (∇
V˜F ˜
ε(t))(( X , V ˜ )(−s; x, v)) ˜ + j[ ˜ F
ε(t)](x) ∧ e(x)
3
0B · (∇
V˜F ˜
ε(t))(( X , V ˜ )(−s; x, v)). ˜ We inquire about the convergence, when δ & 0 of Z
|x−x0|>δ
N (e(x), x − x
0, e(x
0))j [ ˜ F
ε](x
0) dx
0= Z
δ<|x−x0|<R
N (e(x), x − x
0, e(x
0))(j [ ˜ F
ε](x
0) − j[ ˜ F
ε](x)) dx
0+
Z
δ<|x−x0|<R
N(e(x), x − x
0, e(x
0)) dx
0j[ ˜ F
ε](x) for R large enough. We are done if we establish the convergence of R
1
{δ<|x−x0|<R}N (e(x), x−
x
0, e(x
0)) dx
0when δ & 0. But we can write Z
δ<|x−x0|<R
N (e(x), x − x
0, e(x
0)) dx
0= Z
δ<|x−x0|<R
(N (e(x), x − x
0, e(x
0)) − N (e(x), x − x
0, e(x))) dx
0+
Z
δ<|x−x0|<R
N(e(x), x − x
0, e(x)) dx
0and therefore it is enough to prove the convergence, as δ & 0, of R
1
{δ<|x−x0|<R}N (e(x), x−
x
0, e(x)) dx
0. Actually, for any r > 0 we have R
|z|=r
N (e(x), z, e(x)) dσ(z) = 0. This is a consequence of the fact that K (z) has zero trace for any z ∈ R
3\ {0}. Indeed, for any ξ ∈ R
3we have
K (z) : (I
3− e ⊗ e)ξ ⊗ (I
3− e ⊗ e)ξ + K(z) : M[e]ξ ⊗ M [e]ξ + |ξ ∧ e|
2K (z) : e ⊗ e
= |ξ ∧ e|
2traceK(z) = 0, z ∈ R
3\ {0}
implying that
[N (e, z, e) + (K(z)e · e)(I
3− e ⊗ e)] : ξ ⊗ ξ = 0, ξ ∈ R
3.
As the matrix N (e, z, e) + (K (z)e · e)(I
3− e ⊗ e) is symmetric, we obtain N (e, z, e) =
−(K(z)e · e)(I
3− e ⊗ e). By direct computation we obtain R
|z|=r
(K(z)e · e) dσ(z) = 0, r > 0. Similarly we have
D
c
0[ ˜ f
ε] · ∇
x,˜vf ˜
εE
(t, s, x, v) = ˜ D
c
0[ ˜ F
ε(t)]
E
· ∇
X,V˜F ˜
ε(t)
(( X , V ˜ )(−s; x, v)) ˜ (15) where hc
0i · ∇ has been computed in Proposition 5.5 [6]. In order to treat the average of c
1[ ˜ f
ε] · ∇
x,˜vf ˜
εwe need to compute E[(˜ v ∧ e) · ∇
xf ˜
ε] and therefore the charge density
q Z
R3
(˜ v ∧ e) · ∇
xf ˜
ε(t, s, x, v) d˜ ˜ v = q div
xZ
R3
f ˜
ε(t, s, x, v)(˜ ˜ v ∧ e) d˜ v
− q Z
R3
f ˜
ε(t, s, x, v)div ˜
x(˜ v ∧ e) d˜ v
= div
x(j[ ˜ f
ε(t, s)] ∧ e) + j [ ˜ f
ε(t, s)] · rot
xe
= e · rot
x(R(−sω
c, e(x))j[ ˜ F
ε(t)]).
We obtain the following formula for the vector field c
1· ∇
c
1[ ˜ f
ε(t, s)] · ∇
x,˜v= c
1[ ˜ F
ε(t)] · ∇
x,˜v+ c
1s[ ˜ F
ε(t)] · ∇
˜vwhere
c
1s[ ˜ F
ε] · ∇
˜v= 1 B
E[e · rot
x((R(−sω
c, e(x)) − I
3)˜ v F ˜
ε)] · e e · ∇
v˜and therefore
D
c
1[ ˜ f
ε] · ∇ f ˜
εE
(t, s, x, v) = ˜ D
c
1[ ˜ F
ε(t)] E
· ∇
X,V˜F ˜
ε(t)
(( X , V ˜ )(−s; x, v)) ˜ + 1
S Z
S0