HAL Id: hal-00529940
https://hal.archives-ouvertes.fr/hal-00529940
Preprint submitted on 27 Oct 2010
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.
Bounded Solutions of the Boltzmann Equation in the Whole Space
Radjesvarane Alexandre, Yoshinori Morimoto, Seiji Ukai, Chao-Jiang Xu, Tong Yang
To cite this version:
Radjesvarane Alexandre, Yoshinori Morimoto, Seiji Ukai, Chao-Jiang Xu, Tong Yang. Bounded So-
lutions of the Boltzmann Equation in the Whole Space. 2010. �hal-00529940�
IN THE WHOLE SPACE
R. ALEXANDRE, Y. MORIMOTO, S. UKAI, C.-J. XU, AND T. YANG
ABSTRACT. We construct bounded classical solutions of the Boltzmann equation in the whole space without specifying any limit behaviors at the spatial infinity and without as- suming the smallness condition on initial data. More precisely, we show that if the initial data is non-negative and belongs to a uniformly local Sobolev space in the space variable with Maxwellian type decay property in the velocity variable, then the Cauchy problem of the Boltzmann equation possesses a unique non-negative local solution in the same func- tion space, both for the cutoff and non-cutoff collision cross section with mild singularity.
The known solutions such as solutions on the torus (space periodic solutions) and in the vacuum (solutions vanishing at the spatial infinity), and solutions in the whole space having a limit equilibrium state at the spatial infinity are included in our category.
1. I
NTRODUCTIONConsider the Boltzmann equation,
(1.1) f
t+ v · ∇
xf = Q( f , f ),
where f = f (t ,x, v) is the density distribution function of particles with position x ∈ R
3and velocity v ∈ R
3at time t. The right hand side of (1.1) is given by the Boltzmann bilinear collision operator
Q(g, f ) =
ZR3 Z
S2
B(v − v
∗, σ )
g(v
′∗) f (v
′) −g(v
∗) f (v) d σ dv
∗,
which is well-defined for suitable functions f and g specified later. Notice that the collision operator Q(· , ·) acts only on the velocity variable v ∈ R
3. In the following discussion, we will use the σ −representation, that is, for σ ∈ S
2,
v
′= v +v
∗2 + |v− v
∗|
2 σ , v
′∗= v + v
∗2 − |v − v
∗| 2 σ , which give the relations between the pre- and post- collisional velocities.
It is well known that the Boltzmann equation is a fundamental equation in statistical physics. For the mathematical theories on this equation, we refer the readers to [10, 11, 12, 13, 25], and the references therein also for the physics background.
In addition to the special bilinear structure of the collision operator, the cross-section B(v − v
∗, σ ) is a function of only |v − v
∗| and θ where
cos θ = v − v
∗|v − v
∗| , σ , 0 ≤ θ ≤ π 2 .
B varies with different physical assumptions on the particle interactions and it plays an important role in the well-posedness theory for the Boltzmann equation. In fact, except for
2000 Mathematics Subject Classification. 35A01, 35A02, 35A09, 35S05, 76P05, 82C40.
Key words and phrases. Boltzmann equation, non-cutoff cross section, local existence, locally uniform Sobolev space, spatial behavior at infinity, pseudo-differential calculus.
1
the hard sphere model, for most of the other molecular interaction potentials such as the inverse power laws, the cross section B(v− v
∗, σ ) has a non-integrable angular singularity.
For example, if the interaction potential obeys the inverse power law r
−(p−1)for 2 < p < ∞, where r denotes the distance between two interacting molecules, the cross-section behaves like
B(|v − v
∗|, σ ) ∼ |v − v
∗|
γθ
−2−2s, with
−3 < γ = p − 5
p − 1 < 1, 0 < s = 1 p −1 < 1 .
As usual, the hard and soft potentials correspond to p > 5 and 2 < p < 5 respectively, and the Maxwellian potential corresponds to p = 5.
The main consequence of the non-integrable singularity of B at θ = 0 is that it makes the collision operator Q behave like a pseudo differential operator, as pointed out by many authors, e.g. [2, 18, 20, 23]. To avoid this difficulty, Grad [13] introduced an assumption to cutoff this singularity. This was a substantial step made in the study of the Boltzmann equation (1.1) and is now called Grad’s angular cutoff assumption.
One of the main issues in the study of (1.1) is the existence theory of the solutions. Many authors have developed various methods for constructing local and global solutions for different situations. Among them, the Cauchy problem has been studied most extensively for both cutoff and non-cutoff cases.
An essential observation here is that so far, all solutions for the Cauchy problem have been constructed so as to satisfy one of the following three spatial behaviors at infin- ity; x-periodic solutions (solutions on the torus, [13, 14, 21]), solutions approaching an equilibrium ([4, 5, 6, 7, 15, 19, 22]) and solutions approaching 0 (solutions near vacuum, [1, 3, 9, 12]). Notice that the solutions constructed in [16] are also solutions approaching an global equilibrium.
However, it is natural to wonder if there are any other solutions behaving differently at x-infinity. In fact the aim of the present paper is to show that the admissible limit be- haviors are not restricted to the above three behaviors. More precisely, we show that the Cauchy problem admits a very large solution space which includes not only the solutions of the above three types but also the solutions having no specific limit behaviors such that almost periodic solutions and perturbative solutions of arbitrary bounded functions which are not necessarily equilibrium state. This will be done for both cutoff and non-cutoff cases without the smallness condition on initial data.
The method developed in this paper works for local existence theory. The global exis- tence in the same solution space is a big open issue and is our current subject. Also the present method works for the Landau equation but since the extension is straightforward, the detail is omitted.
Our assumption on the cross section is as follows. For the non-cutoff case we assume as usual that B takes the form
B(v − v
∗, σ ) = Φ(|v − v
∗|)b(cos θ ), (1.2)
in which it contains a kinetic part
Φ(|v − v
∗|) = Φ
γ(|v − v
∗|) = |v − v
∗|
γ, and a factor related to the collision angle with singularity,
b(cos θ ) ≈ K θ
−2−2swhen θ → 0+,
for some constant K > 0. For the cutoff case we assume that b takes the form (2.2) or is bounded by it.
In order to introduce our working function spaces, set
∂
βα= ∂
xα∂
vβ, α , β ∈ N
3. Let φ
1= φ
1(x) be a smooth cutoff function
φ
1∈ C
0∞( R
3), 0 ≤ φ
1(x) ≤ 1, φ
1(x) =
1, |x| ≤ 1, 0, |x| ≥ 2.
(1.3)
Our function space is the uniformly local Sobolev space with respect to the space vari- able and the usual Sobolev space with respect to the velocity variable with weight. More precisely, let k, ℓ ∈ N and W
ℓ= (1 +|v|
2)
ℓ/2be a weight function. We define
H
ulk,ℓ( R
6) = {g | kgk
2Hulk,ℓ(R6)
(1.4)
= ∑
|α+β|≤k
sup
a∈R3 Z
R6
| φ
1(x− a)W
ℓ∂
βαg(x, v)|
2dxdv < +∞}.
We will set H
ulk( R
6) = H
ulk,0( R
6).
The uniformly local Sobolev space was first introduced by Kato in [17] as a space of functions of x variable, and was used to develop the local existence theory on the quasi- linear symmetric hyperbolic systems without specifying the limit behavior at infinity.
This space could be defined also by the cutoff function φ
R(x) = φ (x/R) for any R > 0, but the choice of R is not a matter. Indeed let R > 1. Then we see that
kgk
Hulk,ℓ(R6)≤ ∑
|α+β|≤k
sup
a∈R3 Z
R6
| φ
R(x − a)W
ℓ∂
βαg(x, v)|
2dxdv
≤ ∑
|α+β|≤k
∑
j∈Z3,|j|≤R
sup
a∈R3 Z
R6
| φ
1(x − a − j)W
ℓ∂
βαg(x,v)|
2dxdv
≤ CR
3kgk
Hulk,ℓ(R6)
.
The case 0 < R < 1 can be proved similarly. In the sequel, therefore, we fix R = 1.
This space shares many important properties with the usual Sobolev space such as the Sobolev embedding and hence it is contained in the space of bounded functions if k > 3.
An important difference from the usual Sobolev space is that no limit property is specified at x-infinity for the space (1.4).
We shall consider the solutions satisfying the Maxwellian type exponential decay in the velocity variable. More precisely, set hvi = (1 + |v|
2)
1/2. For k ∈ N , our function space of initial data will be
E
k0
( R
6) = n
g ∈ D
′( R
6x,v); ∃ ρ
0> 0 s.t. e
ρ0<v>2g ∈ H
ulk( R
6x,v) o , while the function space of solutions will be, for T > 0,
E
k([0, T ] × R
6x,v) = n
f ∈ C
0([0, T ]; D
′( R
6x,v)); ∃ ρ > 0 s.t. e
ρhvi2f ∈ C
0([0, T ]; H
ulk( R
6x,v)) o
. Our main result is stated as follows.
Theorem 1.1. Assume that the cross section B takes the form (1.2) with 0 < s < 1/2,
γ > −3/2 and 2s + γ < 1. If the initial data f
0is non-negative and belongs to the function
space E
k00
( R
6) for some k
0∈ N , k
0≥ 4, then, there exists T
∗> 0 such that the Cauchy problem
(1.5)
f
t+v · ∇
xf = Q( f , f ), f |
t=0= f
0,
admits a non-negative unique solution in the function space E
k0([0,T
∗] × R
6).
Remark 1.2. For the cutoff case, if γ > −3/2, the same theorem and the same proof are valid because our assumption is that the cross-section b is given by (2.2) or is bounded by it. In the sequel, therefore, we consider the non-cutoff case only.
Before closing this section we give some comparison of this paper and our recent paper [3]. First, we shall compare the existence results. Both papers solve the same modified Cauchy problem (2.1). Thus, we shall compare Theorem 2.1 of this paper and Theorem 4.1 of [3].
The solution space in [3] is the usual weighted Sobolev space H
ℓk( R
6x,v) , so that Theorem 4.1 of [3] gives solutions vanishing at x-infinity (solutions near vacuum). And it is easy to see that even if the space is replaced by H
ℓk( T
3x
× R
3v
), the proof of [3] is still valid and gives rise to x-periodic solutions (solutions on torus). The same space was used also in [14]. On the other hand, it is clear that the space H
ulk,ℓ( R
6) defined by (1.4), the locally uniform Sobolev space with respect to x-variables, contains, as its subset, not only the spaces H
ℓk( R
6) and H
ℓk( T
3× R
3) but also the set of functions having the form G = µ + µ
1/2g. If µ is a global Maxwellian, then we have well-known perturbative solutions of equilibrium as in [5, 6, 7, 22], but more generally G can be any bounded functions. Hence, Theorem 2.1 gives, for example, almost periodic solutions, solutions having different limits at x- infinity like shock profile solutions which attain different equilibrium at the right and left infinity, and bounded solution behaving in more general way at x-infinity. Thus the present paper extends extensively the function space of admissible solutions. This is an essential difference between the two papers.
Another big difference is the collision cross section. The present paper deals with the original kinetic factor Φ(z) = |z|
γwithout regularizing the singularity at z = 0 whereas [3] considers only a regularized one of the form Φ(z) = (1 + |v|
2)
γ/2. This regulariza- tion simplifies drastically the estimates of collision operator Q. For example the proof of Proposition 4.4 with non-regularized kinetic factor is far more subtle than (2.1.2) of [3].
Also the case γ < 0 should be handled separately from the case γ ≥ 0 if the kinetic factor is not regularized, and the range of admissible values of γ is restricted in this paper.
The rest of this paper is organized as follows. In the next section we first rewrite the Cauchy problem (1.5) by the one involving the wight function of the time-dependent Maxwellian type and approximate it by introducing the cutoff cross section. After es- tablishing the upper bounds of the cutoff collision operator, we introduce linear iterative Cauchy problems and show that the iterative solutions converge to the solutions to the cut- off Cauchy problem. In Section 3 we derive the a priori estimates satisfied by the solutions to the cutoff Cauchy problem uniformly with respect to the cutoff parameter. The estimates thus obtained are enough to conclude the local existence and hence to lead to Theorem 1.1.
The last section is devoted to the proof of Lemma 3.3 which is essential for the uniform estimate established in Section 3.
2. C
ONSTRUCTION OFA
PPROXIMATES
OLUTIONSAs will be seen later (Lemma 2.2 and Theorem 4.4), the non-linear collision operator
induces a weight loss, which implies that it cannot be Lipschitz continuous so that the usual
iteration procedure is not valid for constructing local solutions. This difficulty can be over- come, however, by introducing weight functions in v of time-dependent Maxwellian type, developed previously in [3, 23, 24]. Indeed it compensates the weight loss by producing an extra gain term of one order higher weight in the velocity variable at the expense of the loss of the decay order of the time dependent Maxwellian-type weight.
2.1. Modified Cauchy Problem. More precisely, we set, for any κ , ρ > 0, T
0= ρ /(2 κ ),
and put
µ
κ(t) = µ (t, v) = e
−(ρ−κt)(1+|v|2), and
f = µ
κ(t)g, Γ
t(g,g) = µ
κ(t)
−1Q( µ
κ(t)g, µ
κ(t)g) for t ∈ [0, T
0]. Then the Cauchy problem (1.5) is reduced to
(2.1)
g
t+ v · ∇
xg + κ (1 + |v|
2)g = Γ
t(g,g), g|
t=0= g
0.
Define
M
k,ℓ(]0, T [× R
6)) = {g | kgk
Mk,ℓ(]0,T[×R6))= ∑
|α+β|≤k
sup
a∈R3 Z
]0,T[×R6
| φ
0(x − a)W
ℓ∂
βαg(x, v)|
2dtdxdv < +∞}.
Our existence theorem can be stated as follows
Theorem 2.1. Assume that 0 < s < 1/2, γ > −3/2 and 2s + γ < 1. Let κ , ρ > 0 and let g
0∈ H
ulk,ℓ( R
6), g
0≥ 0 for some k ≥ 4 and ℓ ≥ 3. Then there exists T
∗∈]0, T
0] such that the Cauchy problem (2.1) admits a unique non-negative solution satisfying
g ∈ C
0([0, T
∗]; H
ulk,ℓ( R
6))
\M
k,ℓ+1(]0,T
∗[× R
6)).
The strategy of proof is in the same spirit as in [3]. That is, first, approximate the non- cutoff cross-section by a family of cutoff cross-sections and construct the corresponding solutions by a sequence of iterative linear equations. Then the existence of solutions to these approximate linear equations and by obtaining a uniform estimate on these solutions with respect to the cutoff parameter in the uniformly local Sobolev space, the compactness argument will lead to the convergence of the approximate solutions to the desired solution for the original problem.
2.2. Cutoff Approximation. Recall that the cross-section takes the form (1.2). For 0 <
ε < < 1, we approximate (cutoff) the cross-section by b
ε(cos θ ) =
b(cos θ ), if | θ | ≥ 2 ε , b(cos ε ), if | θ | ≤ 2 ε . (2.2)
Denote the corresponding cutoff cross-section by B
ε= Φ(v−v
∗)b
ε(cos θ ) and the collision operator by Γ
tε(g, g). We shall establish a upper weighted estimate on the cutoff collision operator in the uniformly local Sobolev space H
ulk,ℓ( R
6).
Lemma 2.2. Let γ > −3/2.Then for any ε > 0, k ≥ 4, l ≥ 0, and for any U,V belonging to H
ulk,ℓ( R
6), it holds that
Γ
tε(U, V ) ∈ H
ulk,ℓ( R
6)
with
(2.3) kΓ
tε(U, V )k
Hk,ℓul(R6)
≤ CkUk
Hulk,ℓ+γ+(R6)
kV k
Hulk,ℓ+γ+(R6)
, 0 ≤ t ≤ T
0, for some C > 0 depending on ε , k, ℓ as well as ρ , κ .
Proof. First, for simplicity of notations, denote µ
κ(t) by µ (t) without any confusion. By using the collisional energy conservation,
|v
′∗|
2+ |v
′|
2= |v
∗|
2+|v|
2,
we have µ
∗(t ) = µ
−1(t) µ
∗′(t) µ
′(t ). Then for suitable functions U,V , it holds that Γ
tε(U, V )(v)
= µ
−1(t, v)
ZZR3v
∗×S2σ
B
ε(v − v
∗, σ ) µ
∗′(t)U
∗′µ
′(t)V
′− µ
∗(t )U
∗µ (t)V dv
∗d σ
=
ZZR3 v∗×S2σ
B
ε(v − v
∗, σ ) µ
∗(t) U
∗′V
′−U
∗V dv
∗d σ (2.4)
= T
ε(U, V, µ (t)).
Then we have by the Leibniz formula in the x variable and by the translation invariance property in the v variable that for any α , β ∈ N
3,
∂
βαΓ
tε(U, V ) = ∑
α1+α2=α;
β1+β2+β3=β
C
α1,α2,β1,β2,β3T
ε( ∂
xα1∂
vβ1U, ∂
xα2∂
vβ2V, ∂
vβ3µ (t )).
Next, recall the cutoff function φ
1in (1.3) and set φ
2(x) = φ
1(x/2), that is, φ
2∈ C
0∞( R
3), 0 ≤ φ
2(x) ≤ 1, φ
2(x) =
1, |x| ≤ 2, 0, |x| ≥ 4.
Since φ
1(x −a) = φ
1(x− a) φ
2(x − a) holds, we see that for a ∈ R
3, φ
1(x − a) ∂
βαΓ
tε(U, V )
=
α∑
1+α2=α;
β1+β2+β3=β
C
α1,α2,β1,β2,β3T
ε( φ
1(x −a) ∂
βα11
U, φ
2(x− a) ∂
βα22
V, ∂
vβ3µ (t)).
To prove (2.6), put
g
1= φ
1(x − a) ∂
βα11
U, h
2= φ
2(x − a) ∂
βα22
V, µ
3(t) = ∂
vβ3µ (t ), T
ε(g
1,h
2, µ
3(t)) = T
ε+− T
ε−.
Throughout this section, we often use the estimates
µ (t, v), | µ
3(t )| = | ∂
vβ3µ (t, v)| ≤ C
ρ,ke
−ρhvi2/4, t ∈ [0,T
0], v ∈ R
3. We compute T
ε+as follows.
|W
ℓT
ε+| ≤ C
ZZ|v − v
∗|
γ| µ
3(t ,v
∗)| W
ℓ(W
ℓ)
′∗(W
ℓ)
′|(W
ℓg
1)
′∗||(W
ℓh
2)
′|dv
∗d σ
≤ C h
ZZ|v − v
∗|
2γ| µ
3(t, v
∗)|
2dv
∗d σ i
1/2× h
ZZ|(W
ℓg
1)
′∗(W
ℓh
2)
′|
2dv
∗d σ i
1/2≤ C h
ZZW
2γ|(W
ℓg
1)
′∗(W
ℓh
2)
′|
2dv
∗d σ i
1/2,
where we have used
W
l(W
l)
′∗(W
l)
′≤ 1
which comes from the energy conservation, and an elementary inequality
ZZ|v−v
∗|
γ| µ
3(t ,v
∗)|dv
∗d σ (2.5)
≤ C
ZZ|v− v
∗|
γe
−ρ|v∗|2/4dv
∗≤ Chvi
γ, γ > −3, with some ρ > 0. Since the change of variables
(v,v
∗, σ ) → (v
′, v
′∗, σ
′), σ
′= (v − v
∗)/|v − v
∗|, has a unit Jacobian, and since W
γ≤ (W
γ+)
′∗(W
γ+)
′holds, we get
kW
ℓT
ε+k
2L2(R6)≤ C
ZZZZ|(W
ℓ+γ+g
1)
′∗(W
ℓ+γ+h
2)
′|
2dv
∗d σ dvdx
≤ C
ZkW
ℓ+γ+g
1k
2L2(R3v)
kW
ℓ+γ+h
2k
2L2(R3 v)dx.
If | α
1+ β
1| ≤ 2, by virtue of the Sobolev embedding and by the assumption k ≥ 4, we have kW
ℓT
ε+k
L2(R6)≤ CkW
ℓ+γ+g
1k
L∞(R3x;L2(R3v))kW
ℓ+γ+h
2k
L2(R6x,v)≤ Ck φ
1(x − a)W
ℓ+γ+∂
βα11
Uk
H2x(L2v)
k φ
2(x − a)W
ℓ+γ+∂
βα22
V k
L2(R6 x,v). Taking the supremum of both sides with respect to a ∈ R
3and since φ
1and φ
2define the equivalent norms, we have
k T
ε+k
H0,ℓul(R6)
≤ CkUk
Hulk,ℓ+γ+(R6)
kV k
Hulk,ℓ+γ+(R6)
.
The computation is similar when | α
2+ β
2| > 2 for which | α
1+ β
1| ≤ k − | α
2+ β
2| ≤ k− 3.
Also, the the estimate of T
ε−can be done similarly but more straightforwardly. This completes the proof of the lemma. In the below we will use the following estimates which
comes directly from the above proof.
Lemma 2.3. Let γ > −3/2.Then for any ε > 0, k ≥ 4, l ≥ 0, and for any U,V belonging to H
ulk,ℓ( R
6), it holds that
k φ
1(x− a)W
ℓ∂
βαΓ
t,±ε(U, V )k
L2(R6)(2.6)
≤ C ∑ k φ
1(x − a)W
ℓ+γ+∂
βα11U k
H2x(L2v)
k φ
2(x − a)W
ℓ+γ+∂
βα22
V k + k φ
1(x − a)W
ℓ+γ+∂
βα11Uk
L2(R6)k φ
2(x− a)W
ℓ+γ+∂
βα22V k
H2x(L2v)
.
for some C > 0 depending on ε , k, ℓ as well as ρ , κ .
We now study the following Cauchy problem for the cutoff Boltzmann equation
(2.7)
g
t+v · ∇
xg + κ hvi
2g = Γ
tε(g, g), g|
t=0= g
0,
for which we shall obtain uniform estimates in weighted Sobolev spaces. We first prove
the existence of weak solutions.
Theorem 2.4. Assume that −3/2 < γ ≤ 1. Let k ≥ 4, l ≥ 0, ε > 0 and D
0> 0. Then, there exists T
ε∈]0, T
0] such that for any initial data g
0satisfying
(2.8) g
0∈ H
ulk,ℓ( R
6), g
0≥ 0, kg
0k
Hk,ℓul(R6)
≤ D
0, the Cauchy problem (2.7) admits a unique solution g
εhaving the property
g
ε∈ C
0(]0,T
ε[; H
ulk,ℓ( R
6)), g ≥ 0, kg
εk
L∞(]0,Tε[; Hulk,ℓ(R6))
≤ 2D
0. Moreover, this solution enjoys a moment gain in the sense that
g
ε∈ M
k,ℓ+1(]0, T
ε[×( R
6)).
Remark 2.5. (1) Notice that we do not assume g
0∈ H
ulk,ℓ+1( R
6). The moment gain will be essentially used below to control the weight loss of the collision operator shown in Lemma 2.2.
(2) The regularity of g
εwith respect to t variable follows directly from the equation (2.7).
Proof of Theorem 2.4. We prove the existence of non-negative solutions by successive ap- proximation that preserves the non-negativity, which is defined by using the usual splitting of the collision operator (2.4) into the gain (+) and loss (-) terms,
Γ
t,+ε(g, h) =
ZZR3v
∗×S2σ
B
ε(v − v
∗, σ ) µ
∗(t)g
′∗h
′dv
∗d σ , Γ
t,−ε(g, h) = hL
ε(g),
L
ε(g) =
ZZR3 v∗×S2σ
B
ε(v − v
∗, σ ) µ (t, v
∗)g
∗dv
∗d σ .
Evidently, Lemma 2.2 applies to Γ
t,±ε, and in view of (2.5), the linear operator L
εsatisfies
| ∂
βαL
ε(g)(t, x,v)| ≤ C ∑
β1+β2=β Z
|v− v
∗|
γ|( ∂
β1µ )(t, v
∗)| |( ∂
βα2g)(t, x, v
∗)|dv
∗≤ C h
Z|v − v
∗|
2γe
−ρ|v∗|2/4dv
∗i
1/2 β∑
2≤βh
Z|( ∂
βα2g)(t, x, v
∗)|
2dv
∗i
1/2(2.9)
≤ Chvi
γk ∂
xαgk
H|β|(R3v)
, t ∈ [0, T
0], for a constant C > 0 depending on ε .
We now define a sequence of approximate solutions {g
n}
n∈Nby
(2.10)
g
0= g
0;
∂
tg
n+1+ v · ∇
xg
n+1+ κ h|v|i
2g
n+1= Γ
t,+ε(g
n, g
n) − Γ
t,−ε(g
n, g
n+1), g
n+1|
t=0= g
0.
Actually, in view of (2.9) we can consider the mild form g
n+1(t, x, v) = e
−κh|v|i2t−Vn(t,0)g
0(x− tv, v) (2.11)
+
Z t0
e
−κh|v|i2(t−s)−Vn(t,s)Γ
s,+ε(g
n, g
n)(s, x − (t −s)v , v)ds, where
V
n(t, s) =
Z ts
L
ε(g
n)(s, x− (t − s)v, v)ds.
We shall first notice that (2.11) defines well an approximate sequence of solutions. In fact, it follows from Lemma 2.2 that if
g
0∈ H
ulk,ℓ( R
6), g
0≥ 0, (2.12)
g
n∈ L
∞(]0, T [; H
ulk,ℓ( R
6)), g
n≥ 0,
for any T ∈]0, T
0[, then the mild form (2.11) determines g
n+1in the function class (2.13) g
n+1∈ L
∞(]0, T [; H
ulk,ℓ−γ+( R
6)), g
n+1≥ 0,
and solves (2.10). Thus g
n+1exists and is non-negative. However, it appears to have a loss of weight in the velocity variable. We shall now show that the term κ hvi
2g
n+1in (2.10) not only recovers this weight loss but also creates higher moments. To see this, introduce the space and norm defined by
Y = L
∞(]0, T [; H
ulk,ℓ( R
6)) ∩ M
k,ℓ+1(]0,T [× R
6),
||g||
Y2= kgk
2L∞(]0,T[; Hulk,ℓ(R6))
+ κ kgk
2Mk,ℓ+1(]0,T[×R6).
Lemma 2.6. Assume that −3/2 < γ ≤ 1 and let k ≥ 4, l ≥ 0, ε > 0. Then, there exist positive numbers C
1,C
2such that if ρ > 0, κ > 0 and if g
0and g
nsatisfy (2.12) with some T ≤ T
0, the function g
n+1given by (2.11) enjoys the properties
g
n+1∈ Y
||g
n+1||
Y≤ e
C1KnTkg
0k
2Hulk,ℓ(R6)
+ C
2κ ||g
n||
4L4(]0,T[; Hulk,ℓ(R6)), where K
nis a positive constant depending on kg
nk
L∞(]0,T[; Hk,ℓul(R6))
and κ . Proof. Put
h
nℓ= φ
1(x − a)W
ℓ∂
βαg
n. (2.14)
Differentiate (2.10) with respect to x, v and multiply by φ
1(x − a) to deduce
∂
th
n+1ℓ+ v · ∇
xh
n+1ℓ+ κ hvi
2h
n+1ℓ= G
+1− G
−1+ G
2+ G
3, G
+1= φ
1(x− a)W
ℓ∂
βαΓ
t,+ε(g
n, g
n),
G
−1= φ
1(x− a)W
ℓ∂
βαΓ
t,−ε(g
n, g
n+1), G
2= −W
ℓ[ φ
1(x − a) ∂
β, v · ∇
x] ∂
αg
n+1, G
3= − κ W
ℓ∑
|β′|=1,2
C
β′φ
1(x − a)( ∂
β′hvi
2) ∂
β−βα ′g
n+1. Let χ
j∈ C
0∞( R
3), j ∈ N , be the cutoff functions
χ
j(v) =
1 , |v| ≤ j , 0 , |v| ≥ j + 1 . Let S
N(D
x) be a mollifier definded by the Fourier multiplier
S
N( ξ ) = 2
−NS(2
−Nξ ), S( ξ ) ∈ S ( R
3), S( ξ ) = 1 (| ξ | ≤ 1), = 0 (| ξ | ≥ 1).
We remark that (2.13) does not necessarily imply h
n+1ℓ+1(t ) ∈ L
2( R
6), but χ
jh
n+1ℓ+1(t ) does for all j ∈ N . Hence, we can use χ
2jS
2N(D
x)h
n+1ℓas a test function to get
1 2
d
dt kS
N(D
x) χ
jh
n+1ℓk
2+ κ kS
N(D
x) χ
jh
n+1ℓ+1k
2(2.15)
= (G
+1− G
−1+G
2+G
3, S
N(D
x)
2χ
2jh
n+1ℓ).
Here and in what follows, the norm k k and inner product ( , ) are those of L
2( R
6x,v) unless otherwise stated. We shall evaluate the inner products on the right hand side. Observe that Lemma 2.3 gives, for t ∈ [0,T ], since γ ≤ 1 is assumed, and putting ˜h
nℓ= φ
2(x−a)W
ℓ∂
βαg
n,
(G
+1, S
2Nχ
2jh
n+1ℓ) =
(S
Nχ
jW
−1G
+1, S
Nχ
jh
n+1ℓ+1)
≤ CkW
−1G
+1k kS
Nχ
jh
n+1ℓ+1k
≤ Ckh
nℓ−1+γ+k k˜h
nℓ−1+γ+k kS
Nχ
jh
n+1ℓ+1k
≤ Ckg
nk
2Hulk,ℓ−1+γ+(R6)
kS
Nχ
jh
n+1ℓ+1k
≤ C
κ kg
nk
4Hulk,ℓ(R6)+ κ
4 kS
Nχ
jh
n+1ℓ+1k
2,
(G
−1, S
2Nχ
2jh
n+1ℓ) =
(S
Nχ
jW
−1G
−1, S
Nχ
jh
n+1ℓ+1)
≤ CkW
−1G
−1k kS
Nχ
jh
n+1ℓ+1k
≤ Ckh
nℓ−1+γ+k k˜h
n+1ℓ−1+γ+k kS
Nχ
jW
ℓ+1h
n+1ℓ+1k
≤ Ckg
nk
Hulk,ℓ−1+γ+(R6)
kg
n+1k
Hulk,ℓ−1+γ+(R6)
kS
Nχ
jW
ℓ+1h
n+1ℓ+1k
≤ C
κ kg
nk
2Hulk,ℓ(R6)kg
n+1k
2Hulk,ℓ(R6)
+ κ
4 kS
Nχ
jW
l+1h
n+1ℓ+1k
2. Here C is a positive constants independent of κ , and we have used
kh
nℓk ≤ Ckg
nk
Hk,ℓ ul(R6).
and similarly for ˜h
nℓ. The estimation on the remaining two inner products are more straight- forward and can be given as follows. With some abuse of notation,
|G
2| ≤ |v|W
ℓ|∇
xφ
1(x − a))| | φ
2(x − a) ∂
βαg
n+1| + φ
1(x −a)W
ℓη
β| ∂
β−1α+1g
n+1| where η
β= 0 for β = 0 and = 1 for | β | ≥ 1,while
|G
3| ≤ κ CW
ℓ∑
|β′|=1,2
φ
1(x − a)(|v| + 1)| ∂
βα−β′g
n+1|.
Therefore we get
(G
2+ G
3,S
2Nχ
2jh
n+1ℓ) =
W
−1(G
2+ G
3), S
2Nχ
2jW
1h
n+1ℓ)
≤ CkW
−1(G
2+ G
3)k kS
Nχ
jh
n+1l+1k
≤ C(1 + κ )kg
n+1k
Hk,ℓul(R6)
kS
Nχ
jh
n+1l+1k
≤ C
′(1 + κ )
2κ kg
n+1k
2Hulk,ℓ(R6)+ κ
2 kS
Nχ
jh
n+1ℓ+1k
2.
The constants C
′are independent of ε and κ .
Putting together all the estimates obtained above in (2.15) yields d
dt kS
Nχ
jh
n+1ℓk
2+ κ kS
Nχ
jh
n+1ℓ+1k
2≤ C
′′κ ((1 + κ )
2+ kg
nk
2Hulk,ℓ(R6)
)kg
n+1k
2Hulk,ℓ(R6)
+ C
κ kg
nk
4Hlk(R6). The constants C,C
′′are independent of ε and κ . Integrate this over [0,t ] to deduce
kS
Nχ
jh
n+1ℓ(t)k
2+ κ
Z t0
kS
Nχ
jh
n+1ℓ+1( τ )k
2d τ
≤ kS
Nχ
jh
n+1ℓ(0)k
2+C
1K
n Zt0
kg
n+1( τ )k
2Hulk,ℓ(R6)
d τ + C
2κ
Z t
0
kg
n( τ )k
4Hulk,ℓ(R6)
d τ . where
K
n= 1 κ
kg
nk
2L∞(]0,T[;Hulk,ℓ(R6))
+ (1 + κ )
2,
and C
1> 0 is a constant independent of ε , κ while C
2is independent of κ but depends on ε . It is easy to see that we can now take the limit N → ∞ and j → ∞, which results in
kh
n+1ℓ(t )k
2+ κ
Z t0
kh
n+1ℓ+1( τ )k
2d τ
≤ kh
n+1ℓ(0)k
2+C
1K
n Z t0
kg
n+1( τ )k
2Hulk,ℓ(R6)
d τ + C
2κ
Z t 0
kg
n( τ )k
4Hulk,ℓ(R6)
d τ . Sum up this for | α + β | ≤ k (see (2.14)) and take the speremum of the left hand side with respect to a ∈ R
3. Knowing that the right hand side is independent of α , β and also of a ∈ R
3, we have
kg
n+1(t)k
2Hulk,ℓ(R6)
+ κ kg
n+1k
2Mk,ℓ+1(]0,t[×R6)≤ kg
0k
2Hulk,ℓ(R6)
+C
1K
n Z t0
kg
n+1( τ )k
2Hulk,ℓ(R6)
d τ + C
2κ
Z t
0
kg
n( τ )k
4Hulk,ℓ(R6)
d τ . The Gronwall inequality then gives
kg
n+1(t)k
2Hulk,ℓ(R6)
+ κ kg
n+1k
2Mk,ℓ+1(]0,t[×R6)(2.16)
≤ e
C1Kntkg
0k
2Hulk,ℓ(R6)
+ C
2κ
Z t 0
e
C1Kn(t−τ)kg
n( τ )k
4Hulk,ℓ(R6)
d τ . for all t ∈ [0, T ].
Now the proof of Lemma 2.6 is completed.
We are now ready to prove the convergence of {g
n}
n∈N. Fix κ > 0. Let D
0, g
0be as in Theorem 2.4 and introduce an induction hypothesis
(2.17) kg
nk
L∞(]0,T[; Hk,ℓul(R6))
≤ 2D
0.
for some T ∈]0, T
0]. Notice that the factor 2 can be any number > 1.
(2.17) is true for n = 0 due to (2.8). Suppose that this is true for some n > 0. We shall determine T independent of n. A possible choice is given by
e
C1K0T= 2, 2
4C
2κ T D
20= 1 where K
0= 1
κ (2D
0+ (1 + κ )
2) or
T = min log 2
C
1K
0, κ 2
4C
2D
20.
(2.18)
In fact, (2.16) and (2.17) imply that g
n+1∈ Y and
||g
n+1||
2Hulk,ℓ(R6)
≤ e
C1K0Tkg
0k
2Hulk,ℓ(R6)
+ C
2κ T ||g
n||
4L∞(]0,T[; Hulk,ℓ(R6))≤ e
C1K0TD
20+ C
2κ T 2
4D
40≤ 4D
20.
That is, the induction hypothesis (2.17) is fulfilled for n + 1, and hence holds for all n.
For the convergence, set w
n= g
n(t) − g
n−1(t), for which (2.10) leads to
∂
tw
n+1+v · ∇
xw
n+1+ κ h|v|i
2w
n+1= Γ
t,+ε(w
n, g
n)
+Γ
t,+ε(g
n−1, w
n) − Γ
t,−ε(w
n, g
n+1) − Γ
t,−ε(g
n−1, w
n+1), w
n+1|
t=0= 0.
By the same computation as in (2.15), we get
||w
n+1||
Y2≤ 1
2 C
2e
C1K0T1 κ T
n kg
n+1k
2L∞(]0,T[; Hulk,ℓ(R6))
+ kg
nk
2L∞(]0,T[; Hulk,ℓ(R6))
+ kg
n−1k
2L∞(]0,T[; Hulk,ℓ(R6))
o kw
nk
2L∞(]0,T[; Hulk,ℓ(R6))
, with the same constants C
1,C
2and K
0as above. Then, (2.17) and (2.18) give
||g
n+1− g
n||
2Y≤ 2
4C
2D
20κ
−1T kg
n− g
n−1k
2L∞(]0,T[; Hulk,ℓ(R6))
. Finally, choose T smaller if necessary so that
2
4C
2D
20κ
−1T ≤ 1 4 . Then, we have proved that for any n ≥ 1,
(2.19) ||g
n+1− g
n||
Y≤ 1
2 ||g
n− g
n−1||
Y. Consequently, {g
n} is a convergence sequence in Y , and the limit
g
ε∈ Y,
is therefore a non-negative solution of the Cauchy problem (2.7). The estimate (2.19) also implies the uniqueness of solutions.
By means of the mild form (2.11), it can be proved also that for each n, g
n∈ C
0([0, T ]; H
ulk,ℓ( R
6))
and hence so is the limit g
ε. The non-negativity of g
εfollows because g
n≥ 0. Now the proof of Theorem 2.4 is completed.
3. U
NIFORME
STIMATEWe now prove the convergence of approximation sequence {g
ε} as ε → 0. The first step is to prove the uniform boundedness of this approximation sequence. Below, the constant C are various constants independent of ε > 0.
Theorem 3.1. Assume that 0 < s < 1, γ > −3/2, γ + 2s < 1. Let g
0∈ H
ulk,ℓ( R
6), g
0≥ 0 for some k ≥ 4, l ≥ 3. Then there exists T
∗∈]0, T
0] depending on kg
0k
Hk,ℓul(R6)