Lecture 7

Non-symmetric perturbation with polynomial tail

Renjun DUAN

The Chinese University of Hong Kong

Course: Perturbation theory of the Boltzmann equation

23th Summer School on Nonlinear PDEs
16-20 July 2025, Capital Normal University

1/61
Outline

§7.1 Main Results

§7.2 Preliminaries

§7.3 Local in time existence

§7.4 Hard potential case

§7.5 Soft potential case

2/61
§7.1 Main Results

3/61
Let F = F (t, x, v ) with t ≥ 0, x ∈ R3 , v ∈ R3 satisfies
∂t F + v · ∇x F = Q(F , F ), F (0, x, v ) = F0 (x, v ), (1)
Z Z
B(v − u, θ) G (u ′ )F (v ′ ) − G (u)F (v ) dωdu
 
Q(G , F )(v ) =
R3 S2
:= Q+ (G , F )(v ) − Q− (G , F )(v ), (2)

v ′ = v − [(v − u) · ω] ω, u ′ = u + [(v − u) · ω] ω,
B(v − u, θ) = |v − u|γ b(θ), (3)
− 3 < γ ≤ 1, 0 ≤ b(θ) ≤ C | cos θ|. (4)
We call hard potentials in case of 0 ≤ γ ≤ 1 and soft potentials in case of
−3 < γ < 0.

4/61
Set
µ = µ(v ) := (2π)−3/2 exp −|v |2 /2 .

F =µ+g (5)
for a new unknown function g = g (t, x, v ). Substituting (5) into (1),
∂t g + v · ∇x g + Lg = Q(g , g ), g (0, x, v ) = g0 (x, v ) := F0 (x, v ) − µ(v ), (6)
where
Lg = −[Q(µ, g ) + Q(g , µ)].
Moreover, L can be split into
L = ν − K,
with
Z Z
ν(v ) = B(v − u, θ)µ(u)dωdu ∼ (1 + |v |)γ , (7)
R3 S2
and
Z Z
B(v − u, θ) µ(u ′ )f (v ′ ) + µ(v ′ )f (u ′ ) − µ(v )f (u) dωdu.
 
Kf (v ) = (8)
R3 S2
By integrating
Z t
g (t, x, v ) =e −ν(v )t g0 (x − vt, v ) + e −ν(v )(t−s) (Kg )(s, x − v (t − s), v ) ds
0
Z t
+ e −ν(v )(t−s) Q(g , g )(s, x − v (t − s), v ) ds. (9)
0

5/61
Given a function f = f (x, v ), the Lqv Lpx norm for 1 ≤ p, q ≤ ∞ is defined by
(Z Z q )1
q
p
∥f ∥Lqv Lpx := |f (x, v )|p dx dv ,
R3 R3

with the usual convention for the L∞ norm in case p = ∞ or q = ∞. For brevity,
we write ∥f ∥Lpx,v = ∥f ∥Lpv Lpx for 1 ≤ p ≤ ∞, in particular, ∥f ∥L∞
x,v
= ∥f ∥L∞ ∞ and
v Lx
∥f ∥L2 = ∥f ∥L2 L2 . Furthermore, we define a polynominal velocity weight function
x,v v x

wk = wk (v ) = (1 + |v |)k .
For a function f = f (x, v ) ∈ L∞ ∞ 1
v (Lx ∩ Lx ), the velocity weighted Xj,k norm is
defined as
∥f ∥Xj,k = ∥wk f ∥L∞
x,v
+ ∥wj f ∥L∞ L2 +∥wj f ∥L∞ L1 . (10)
v x v x

Similarly, for a function f = f (x, v ) ∈ L∞ ∞

v (Lx ∩ L2x ), the velocity weighted Yj,k
norm is defined as
∥f ∥Yj,k = ∥wk f ∥L∞
x,v
+ ∥wj f ∥L∞ L2 . (11)
v x

6/61
Theorem 1.1 (Hard potentials)
Let 0 ≤ γ ≤ 1, then there is k0 > 0 large enough such that for any k ≥ j ≥ k0 , the
following holds true. There are ϵ0 > 0 and C > 0 such that if it holds that
F0 (x, v ) = µ(v ) + g0 (x, v ) ≥ 0 with g0 ∈ L∞ ∞ 1
v (Lx ∩ Lx ) satisfying

∥g0 ∥Xj,k ≤ ϵ0 , (12)

then the Cauchy problem on the Boltzmann equation (1) or (6) admits a unique
global mild solution F (t, x, v ) = µ(v ) + g (t, x, v ) ≥ 0 in the sense of (9) with
g ∈ L∞ (0, ∞; L∞ ∞ 2
v (Lx ∩ Lx )) satisfying the estimate
3
∥g (t)∥Yj,k ≤ C (1 + t)− 4 ∥g0 ∥Xj,k , (13)
for all t ≥ 0.

7/61
Theorem 1.2 (Soft potentials)
Let −3 < γ < 0 and 0 < ϵ ≤ 12 , then there is k0 > 0 large enough such that for any
k ≥ j ≥ k0 , the following holds true. There are ϵ0 and C > 0 such that if
F0 (x, v ) = µ(v ) + g0 (x, v ) ≥ 0 with g0 ∈ L∞ ∞ 1
v (Lx ∩ Lx ) satisfying
∥g0 ∥Xj+|γ|,k+|γ| ≤ ϵ0 , (14)

then the Cauchy problem on the Boltzmann equation (1) or (6) admits a unique
global mild solution F (t, x, v ) = µ(v ) + g (t, x, v ) ≥ 0 in the sense of (9) with
g ∈ L∞ (0, ∞; L∞ ∞ 2
v (Lx ∩ Lx )) satisfying the estimate
3
∥g (t)∥Yj,k ≤ C (1 + t)− 4 +ϵ ∥g0 ∥Xj+|γ|,k+|γ| , (15)
for all t ≥ 0. Note that C may depend on ϵ such that C tends to ∞ as ϵ → 0.

8/61
§7.2 Preliminaries

9/61
Lemma 2.1
Let −3 < γ ≤ 1. For any k > max{3, 3 + γ}, there is a constant Ck > 0 depending
only on k such that
wk (v ) − |u′ |2
Z Z
C ν(v )
|v − u|γ e 2 dωdu ≤
γ+3
ν(v ) + Ck , ∀ v ∈ R3 , (16)
R 3 S2 wk (v ′ ) k 4 (1 + |v |)2
and
Z Z
wk (v )
|v − u|γ dωdu ≤ Ck ν(v ), ∀ v ∈ R3 , (17)
R3 S2 wk (v ′ )wk (u ′ )
where C > 0 is a universal constant independent of k. Morever, when 0 ≤ γ ≤ 1, the
bound in (17) can be refined as the one similar to (16), that is,
Z Z
wk (v ) C ν(v )
|v − u|γ dωdu ≤ ν(v ) + Ck , ∀ v ∈ R3 . (18)
R3 S2 wk (v ′ )wk (u ′ ) k (1 + |v |)2

Ref.: C. Cao, Cutoff Boltzmann equation with polynomial perturbation near

Maxwellian. J. Funct. Anal. 283 (2022), 179–184.

10/61
A simple example:
Proposition 2.2
For any γ > −3, k > max{3, 3 + γ} we have
Z
c
|v − v∗ |γ ⟨v∗ ⟩−k dv∗ ≤ ⟨v ⟩γ + Ck ⟨v ⟩γ−2 .
R3 k

Proof: Case: 0 ≤ γ ≤ 1: we have

Z Z
|v − v∗ |γ ⟨v∗ ⟩−k dv∗ ≤ C (|v |γ + |v∗ |γ )⟨v∗ ⟩−k dv∗ .
R3 R3
It is easily seen that
Z
c
|v∗ |γ ⟨v∗ ⟩−k dv∗ ≤ Ck ≤ ⟨v ⟩γ + Ck ⟨v ⟩γ−2 .
R3 k
Using the beta function we have
Z Z +∞ 1 c
⟨v ⟩γ ⟨v∗ ⟩−k dv∗ ≤ C ⟨v ⟩γ r2 dr ≤ 3/2 ⟨v ⟩γ ,
R3 0 (1 + r 2 )k/2 k
so the case γ ∈ [0, 1] is thus proved.

11/61
Case γ ∈ (−3, 0) :
If |v | ≤ 12 , then |v∗ | + 12 ≤ 1 + |v − v∗ |, so
Z Z Z
|v −v∗ |γ ⟨v∗ ⟩−k dv∗ = |v∗ |γ ⟨v −v∗ ⟩−k dv∗ ≤ Ck |v∗ |γ ⟨v∗ ⟩−k dv∗ ≤ Ck ≤ Ck ⟨v ⟩γ−2 .
R3 R3 R3

If |v | > 21 , then split the integral into two regions |v − v∗ | > ⟨v ⟩/4 and
|v − v∗ | ≤ ⟨v ⟩/4.
For the first region
Z Z
|v − v∗ |γ ⟨v∗ ⟩−k dv∗ ≤ C ⟨v ⟩γ ⟨v∗ ⟩−k dv∗
⟨v ⟩
|v −v∗ |> 4 R3
Z ∞ 1 c
≤ C ⟨v ⟩γ r2 dr ≤ 3/2 ⟨v ⟩γ .
0 (1 + r 2 )k/2 k
For the second region, |v | > 1/2 and |v − v∗ | ≤ ⟨v ⟩/4 imply |v∗ | ≥ ⟨v ⟩/8, hence
Z Z
|v −v∗ |γ ⟨v∗ ⟩−k dv∗ ≤ Ck ⟨v ⟩−k |v −v∗ |γ dv∗ ≤ Ck ⟨v ⟩−k+γ+3
⟨v ⟩ ⟨v ⟩
|v −v∗ |≤ 4 |v −v∗ |≤ 4
c
≤ ⟨v ⟩γ + Ck ⟨v ⟩γ−2 ,
k
so the proposition is thus proved.

12/61
Lemma 2.3
Let −3 < γ ≤ 1. For any k > max{3, 3 + γ}, there is a constant Ck > 0 depending
only k such that for any M > 0,
!
C Ck
χ{|v |>M} wk (v )|Kf (v )| ≤ γ+3
+ ν(v )∥wk f ∥L∞ , ∀ v ∈ R3 , (19)
k 4 M2 v

where C > 0 is a universal constant independent of k and M.

13/61
In the symmetric framework, the linear operator L defined as
1 √ 1 √ 1 √
Lf := √ K( µf ) = − √ Q(µ, µf ) − √ Q( µf , µ), (20)
µ µ µ
is self-adjoint and non-negative definite on L2v .
Lemma 2.4
Let −3 < γ ≤ 1. There is a kernel function k(·, ·) with k(v , η) = k(η, v ) such that
Z
Kf (v ) = k(v , η)f (η) dη,
R3
with
|v |2 |η|2 |v −η|2
||v |2 −|η|2 |2
C −
|k(v , η)| ≤ C |v − η|γ e − 4 e− 4 + 3−γ
e− 8 e 8|v −η|2 .
|v − η| 2

Moreover, for any k ∈ R, there is a constant Ck such that

wk (v ) |v −η|2
Z
|k(v , η)| e 20 dη ≤ Ck (1 + |v |)−1 , ∀ v ∈ R3 . (21)
R3 wk (η)

14/61
Decomposition of K :

A cut-off smooth function χm = χm (τ ) with a parameter 0 ≤ m ≤ 1 such that

0 ≤ χm ≤ 1, and χm (τ ) = 1 for τ ≤ m and χm (τ ) = 0 for τ ≥ 2m. One can split K
into two parts K = K m + K c with
Z Z p
K m f (v ) = B(v − u, θ)χm (|v − u|) µ(u)
R3 S2
p 
µ(u ′ )f (v ′ ) + µ(v ′ )f (u ′ ) − µ(v )f (u) dωdu,
p p
· (22)

and K c = K − K m .

15/61
Lemma 2.5
Let −3 < γ < 0 and 0 < m ≤ 1. For the part K m , there is a constant C independent
of m such that
|v |2
|K m f (v )| ≤ Cmγ+3 e − 10 ∥f ∥L∞
v
, ∀ v ∈ R3 . (23)
For the part K c , there is a kernel function l(·, ·) with l(v , η) = l(η, v ) such that
Z
K c f (v ) = l(v , η)f (η)dη, (24)
R3
with
|v |2 |η|2 |v −η|2
||v |2 −|η|2 |2
Cm −
|l(v , η)| ≤ C |v − η|γ e − 4 e− 4 + 3−γ
e− 16 e 16|v −η|2 .
|v − η| 2

Moreover, for l(v , η), the following estimates also hold true. For any k ∈ R, there is a
constant Cm,k such that
Z
wk (v )
|l(v , η)| dη ≤ Cm,k (1 + |v |)−1 , (25)
R3 wk (η)
wk (v ) − |η|2 |v |2
Z
|l(v , η)| e 20 dη ≤ Cm,k e − 100 ,
R3 wk (η)
wk (v ) |v −η|2
Z
ν(v )
|l(v , η)| e 20 dη ≤ Cm,k , (26)
R3 wk (η) (1 + |v |)2
for all v ∈ R3 .

16/61
Lemma 2.6
Let −3 < γ ≤ 1 and r > 0. There is a constant Cr > 0 such that in case 0 ≤ γ ≤ 1, it
holds
Z t
e −ν(v )(t−s) ν(v )(1 + s)−r ds ≤ Cr (1 + t)−r , ∀ t ≥ 0, (27)
0
and in case −3 < γ < 0 and r ̸= 1, it holds
Z t
e −ν(v )(t−s) ν(v )(1 + s)−r ds ≤ Cr (1 + t)− min{r ,1} , ∀ t ≥ 0. (28)
0

17/61
Lemma 2.7
Let r > 0 and 0 < q ≤ r , then there is a constant Cr ,q > 0 such that for any t ≥ 0,

−q
Z t Cr ,q (1 + t) ,
 r > 1,
(1 + t − s)−q (1 + s)−r ds ≤ Cr ,q (1 + t)−q log(1 + t) r = 1, (29)
0 C (1 + t)−q−r +1

r ,q r < 1.
Furthermore, let r > 0 and λ > 0, then there is a constant Cr ,λ > 0 such that for any
t ≥ 0,
Z t
e −λ(t−s) (1 + s)−r ds ≤ Cr ,λ (1 + t)−r . (30)
0

18/61
§7.3 Local in time existence

19/61
We define the function space
XT ,k,j := {g (t, x, v ) ∈ L∞ (0, T ; L∞ ∞ 2
x,v ∩ Lv Lx ) :
sup ∥wk g (t)∥L∞ + sup wj g (t) L∞ 2 < ∞}.
0≤t≤T x,v
0≤t≤T v Lx

Theorem 3.1
Assume (3) and (4). There is k0 > 0 large enough such that for any k ≥ j ≥ k0 , the
following holds true. There are ϵ1 > 0 and T∗ > 0 such that if
F0 (x, v ) := µ(v ) + g0 (x, v ) ≥ 0 with
∥wk g0 ∥L∞
x,v
< ϵ1 , (31)

then the Cauchy problem on the Boltzmann equation (1) or (6) admits a unique mild
solution F (t, x, v ) = µ(v ) + g (t, x, v ) ≥ 0 with (t, x, v ) ∈ [0, T∗ ] × R3x × R3v in the
sense of (9) such that g ∈ XT∗ ,k,j satisfies the following estimate:


sup ∥wk g (t)∥L∞ + sup wj g (t) L∞ L2 ≤ 2 ∥wk g0 ∥L∞ + wj g0 L∞ L2 . (32)
x,v v x x,v v x
0≤t≤T∗ 0≤t≤T∗

20/61
Proof: We first rewrite (6) as
∂t g + v · ∇x g + νg + Q− (g , g ) = Kg + Q+ (g , g ),
g (0, x, v ) = g0 (x, v ) = F0 (x, v ) − µ(v ),
where ν, K, Q− and Q+ are defined in (7), (8) and (2). Then we construct the
approximation sequence {g n }∞
n=0 as

∂t g n+1 + v · ∇x g n+1 + νg n+1 + Q− (g n , g n+1 ) = Kg n + Q+ (g n , g n ),

with g n+1 (0, x, v ) = g0 (x, v ) and g 0 (t, x, v ) = 0. It is noted that
Rt n
g n+1 (t, x, v ) =e − 0 ζ (τ,x−v (t−τ ),v )dτ g0 (x − vt, v )
Z t
− st ζ n (τ,x−v (t−τ ),v )dτ
R
+ e (Kg n )(s, x − v (t − s), v )ds
0
Z t Rt n
+ e− s ζ (τ,x−v (t−τ ),v )dτ Q+ (g n , g n )(s, x − v (t − s), v )ds, (33)
0
where
Z Z
ζ n (τ, y , v ) = B(v − u, θ) [µ(u) + g n (τ, y , u)] dωdu.
R3 S2

Defining F n = µ + g n , we have that (33) is equivalent to

Rt n
F n+1 (t, x, v ) =e − 0 ζ (τ,x−v (t−τ ),v )dτ F0 (x − vt, v )
Z t
− st ζ n (τ,x−v (t−τ ),v )dτ
R
+ e Q+ (F n , F n )(s, x − v (t − s), v )ds.
0
Then it is direct to see that the nonnegativity F n ≥ 0 follows by induction.
21/61
We prove the bound (32) for the approximation sequence {g n }∞
n=0 in case of soft
and hard potentials separately. We can show that there is T∗ > 0 such that for any
n ≥ 0,
sup ∥wk g n (s)∥L∞
x,v
≤ 2∥wk g0 ∥L∞
x,v
, (34)
0≤s≤T∗

sup ∥wj g n (s)∥L∞ L2 ≤ 2∥wj g0 ∥L∞ L2 . (35)

v x v x
0≤s≤T∗

We prove it by induction.

22/61
 Soft potentials −3 < γ < 0 : It is obvious to hold for n = 0, since g 0 ≡ 0. We
assume (34) and R R(35) for n ≥ 0. By (33) and the fact that
ζ n (τ, y , v ) = R3 S2 B(v − u, θ)F n (τ, y , u)dωdu ≥ 0, a direct calculation shows that
Z t
wk (v )g n+1 (t, x, v ) ≤ |wk (v )g0 (x − vt, v )| + |wk (v )(Kg n )(s, x − v (t − s), v )| ds
0
Z t
n n
+ |wk (v )Q+ (g , g )(s, x − v (t − s), v )| ds.
0
Applying the definitions of K in (8) and Q+ in (2) yields
wk (v )g n+1 (t, x, v )
≤ |wk (v )g0 (x − vt, v )|
Z t Z Z
+ C ∥wk g n (s)∥L∞x,v
|v − u|γ
0 R3 S2
wk (v ) − |u′ |2 wk (v ) − |v ′ |2 wk (v ) − |v |2

× e 2 + e 2 + e 2 dωduds
wk (v ′ ) wk (u ′ ) wk (u)
Z tZ Z
wk (v )
+ C ∥wk g n (s)∥2L∞ |v − u|γ dωduds. (36)
x,v
0 R3 S2 wk (v ′ )wk (u ′ )

23/61
Then it follows from (16), (17) and (36) that for 0 ≤ t ≤ T∗ and k ≥ k0 > 3,
wk (v )g n+1 (t, x, v )
Z t
ν(v ) ∥wk g n (s)∥L∞ + ∥wk g n (s)∥2L∞ ds


≤∥wk g0 ∥L∞
x,v
+ C k x,v x,v
0
∥wk g n (s)∥L∞ ∥wk g n (s)∥2L∞ .


≤∥wk g0 ∥L∞
x,v
+ C k T∗ sup x,v
+ sup (37)
x,v
0≤s≤T∗ 0≤s≤T∗

Using the assumption (34), one gets

wk (v )g n+1 (t, x, v ) ≤ ∥wk g0 ∥L∞ + ∥wk g0 ∥2L∞ .


x,v
+ Ck T∗ ∥wk g0 ∥L∞
x,v
(38)
x,v

Similarly, for k ≥ j ≥ k0 > 3,

wj g n+1 (t) L∞ 2
v Lx
Z t
≤ wj g0 L∞ 2 + wj (Kg n )(s) L∞ 2 ds
v Lx 0 v Lx
Z t
+ wj Q+ (g n , g n )(s) L∞ 2 ds
0 v Lx

∥wj g n (s)∥L∞ L2 1 + ∥wk g n (s)∥L∞

≤∥wj g0 ∥L∞ L2 + Cj T∗ sup sup x,v
v x v x
0≤s≤T∗ 0≤s≤T∗


≤∥wj g0 ∥L∞ L2 + Cj T∗ ∥wj g0 ∥L∞ L2 1 + ∥wk g0 ∥L∞
x,v
. (39)
v x v x
Let
1
T∗ = > 0,
6(Ck + Cj )(1 + ∥wk g0 ∥L∞
x,v
)
then it follows from (38) and (39) that (34) and (35) are satisfied for n + 1. Hence,
(34) and (35) are proved for all n ≥ 0.
24/61
Next, we prove that the constructed sequence {g n }∞ n=0 is a Cauchy sequence in the
wk -weighted norm of L∞ x,v . By taking the difference and using similar arguments as
in Li1 , one can obtain that
sup wk g n+2 (s) − wk g n+1 (s) L∞
0≤s≤T x,v

≤ CT∗ (1 + ∥wk g0 ∥L∞

x,v
) sup wk g n+1 − wk g n L∞
0≤s≤T∗ x,v

C
≤ sup wk g n+1 − wk g n L∞
,
6Ck 0≤s≤T∗ x,v

for some C > 0. Then we can choose Ck to be large enough such that
1
sup wk g n+2 (s) − wk g n+1 (s) L∞
≤ sup wk g n+1 − wk g n L∞
. (40)
0≤s≤T∗ x,v 2 0≤s≤T∗ x,v

One may refer to Section 3 in Li and Appendix in DHWY2 for more details. Hence,
we have proved that {g n }∞ n=0 is a Cauchy sequence in the wk -weighted norm of
L∞
x,v . We take the limit n → ∞ to obtain a local mild solution g ∈ XT∗ ,k,j satisfying
(32) in terms of the uniform estimates (34) and (35). Recall F n ≥ 0 for any n ≥ 0,
so the limiting function F = µ + g is also nonnegative. For the uniqueness, let
g1 , g2 ∈ XT∗ ,k,j be two solutions to (6) satisfying (32). Then, following the similar
arguments for deducing (40), it is direct to prove g1 = g2 . Thus, the case of soft
potentials is proved.
1
Z.-G. Li, Large Amplitude Solutions in Lpv L∞ ∞
T Lx to the Boltzmann Equation for Soft
Potentials. SIAM J. Math. Anal. 54 (2022), no. 4, 4163–4197.
2
R.-J. Duan, F.-M. Huang, Y. Wang and T. Yang, Global well-posedness of the Boltzmann
equation with large amplitude initial data. Arch. Rational. Mech. Anal. 225 (2017), 375–424.
25/61
 Hard potentials 0 ≤ γ ≤ 1 : We need to modify the proof above since the term
ν(v ) in (38) is unbounded when v is large. As before, for some T∗ > 0 which will
be determined later, we prove (34) and (35) for any n ≥ 0 by induction. For the
case that n = 0, the above estimates hold naturally since g 0 ≡ 0. We assume (34)
and (35) for n ≥ 0. Here, due to the difficulty caused by the nonlinear term, we
make an additional assumption that ∥wk g0 ∥L∞ x,v
< ϵ1 . Recalling the definition of ζ
that ζ n (τ, y , v ) = R3 S2 B(v − u, θ) [µ(u) + g n (τ, y , u)] dωdu, in order to overcome
R R
the unboundedness of ν(v ), we start to deduce that ζ behaves like ν(v ) up to a
small positive time. We consider 0 ≤ τ ≤ T∗ . A direct calculation shows that
Z Z
ζ n (τ, y , v ) = B(v − u, θ) [µ(u) + g n (τ, y , u)] dωdu
R3 S2
Z Z
≥ ν(v ) − C ∥wk g n (τ )∥L∞
x,v
|v − u|γ (1 + |u|)−k dωdu. (41)
R3 S2
For k ≥ k0 > 3, it follows from (41) that
ζ n (τ, y , v ) ≥ ν(v )(1 − Ck sup ∥wk g n (s)∥L∞
x,v
),
0≤s≤T∗

where Ck depends only on k. Recalling sup0≤s≤T∗ ∥wk g n (s)∥L∞

x,v
≤ 2∥wk g0 ∥L∞
x,v
, we
have
ζ n (τ, y , v ) ≥ ν(v )(1 − 2Ck ∥wk g0 ∥L∞
x,v
).

26/61
Defining
1
ϵ1 = min{1, },
4Ck
and
1
T1 = ,
6Ck (1 + ∥wk f0 ∥L∞
x,v
)
then for ∥wk g0 ∥L∞
x,v
< ϵ1 and 0 ≤ τ ≤ t ≤ T∗ ≤ T1 , one gets

ζ n (τ, y , v ) ≥ C ν(v ). (42)

From (33) and (42), for 0 ≤ t ≤ T∗ ≤ T1 , it follows that
wk (v )g n+1 (t, x, v ) ≤ |wk (v )g0 (x − vt, v )|
Z t
+ e −C ν(v )(t−s) |wk (v )(Kg n )(s, x − v (t − s), v )| ds
0
Z t
+ e −C ν(v )(t−s) |wk (v )Q+ (g n , g n )(s, x − v (t − s), v )| ds.
0
Using a similar argument as in (36), we have from (16) and (18) that
wk (v )g n+1 (t, x, v ) ≤∥wk g0 ∥L∞
x,v
Z t !
C
+ e −C ν(v )(t−s) ν(v )ds T∗ Ck + γ+3
0 k 4

∥wk g n (s)∥L∞ ∥wk g n (s)∥2L∞ .

× sup x,v
+ sup
x,v
0≤s≤T∗ 0≤s≤T∗

27/61
 Rt
By the fact that e −C ν(v )(t−s) ν(v )ds is bounded, we apply (34) to obtain
0
!
n+1 C1
+ ∥wk g0 ∥2L∞ ,


wk (v )g (t, x, v ) ≤∥wk g0 ∥L∞ x,v
+ T∗ Ck + γ+3 ∥wk g0 ∥L∞
x,v x,v
k 4
(43)
where C1 is a generic constant. Similarly as (39) and (43), using (16), (18), (34)
and (35), one has
!
C1
∥wj g n+1 (t)∥L∞ L2 ≤∥wj g0 ∥L∞ L2 + T∗ Cj + γ+3 ∥wj g0 ∥L∞ L2 (1 + ∥wk g0 ∥L∞
x,v
).
v x v x v x
j 4
(44)
Based on the estimates above, we choose k0 to be large enough such that
2
k ≥ j ≥ k0 > (4C1 ) γ+3 to obtain
C1 1 C1 1
γ+3
< , γ+3
< ,
k 4 4 j 4 4

which, together with our requirement ∥wk g0 ∥L∞ x,v

< ϵ1 ≤ 1, yields
!
C1 1
+ ∥wk g0 ∥2L∞ ≤ 2(T∗ Ck + )∥wk g0 ∥L∞


T∗ Ck + γ+3 ∥wk g0 ∥L∞
x,v x,v
, (45)
k 4
x,v 4
and
!
C1 1
T∗ Ck + γ+3
∥wj g0 ∥L∞ L2 (1 + ∥wk g0 ∥L∞
x,v
) ≤ 2(T∗ Ck + )∥wj g0 ∥L∞ L2 . (46)
j 4
v x 4 v x

28/61
We are ready to define
1
T∗ = min{1, T1 , }. (47)
8(Ck + Cj )(1 + ∥wk g0 ∥L∞
x,v
)
It then holds that
sup ∥wk g n+1 (s)∥L∞
x,v
≤ 2∥wk g0 ∥L∞
x,v
0≤s≤T∗
and
sup ∥wj g n+1 (s)∥L∞ L2 ≤ 2∥wj g0 ∥L∞ L2
v x v x
0≤s≤T∗

from (43), (44), (45) and (46), which give the bounds (34) and (35) for all n ≥ 0 by
induction.

29/61
Next, we prove that {g n }∞ n=0 is a Cauchy sequence in the velocity weighted Lx,v
∞

norm. Due to the growth of ν(v ) in large velocities, we choose k0 large enough
such that k − 2γ ≥ k0 − 2γ > 0 and consider the sequence {wk−2γ g n }∞ n=0 instead of
{wk g n }∞
n=0 . Recalling (33) and taking the difference, it is straightforward to get

|wk−2γ (v )(g n+2 −g n+1 )(t, x, v )| ≤ I0 (t, x, v )+I1 (t, x, v )+I2 (t, x, v )+I3 (t, x, v ), (48)
where
R t n+1 Rt n
I0 (t, x, v ) = e − 0 ζ (τ,x−v (t−τ ),v )dτ
− e− 0 ζ (τ,x−v (t−τ ),v )dτ ·|wk−2γ (v )g0 (x−vt, v )|,

Z t R t n+1 Rt n
I1 (t, x, v ) = e− s ζ (τ,x−v (t−τ ),v )dτ
− e− s ζ (τ,x−v (t−τ ),v )dτ
0
· wk−2γ (v )|Kg n+1 (s, x − v (t − s), v )|
+wk−2γ (v )|Q+ (g n+1 , g n+1 )(s, x − v (t − s), v )| ds,

Z t Rt n
e− 0 ζ (τ,x−v (t−τ ),v )dτ Kg n+1 − Kg n (s, x − v (t − s), v ) ds,


I2 (t, x, v ) = wk−2γ (v )
0
and
Z t Rt n
I3 (t, x, v ) = e− 0 ζ (τ,x−v (t−τ ),v )dτ wk−2γ (v )
0
Q+ (g n+1 , g n+1 ) − Q+ (g n , g n ) (s, x − v (t − s), v ) ds.


·

30/61
Using the fact that |e −a − e −b | ≤ |a − b| for any a, b ≥ 0, we have
R t n+1 Rt n
e− s ζ (τ,x−v (t−τ ),v )dτ
− e− s ζ (τ,x−v (t−τ ),v )dτ

Z t Z Z
B(v − u, θ) g n+1 (τ, y , u) − g n (τ, y , u) dωdu dτ
 
≤
0 R3 S2
Z t Z
≤C sup ∥wk−2γ g n+1 (s) − wk−2γ g n (s)∥L∞
x,v
|v − u|γ w2γ−k (u)dudτ.
0≤s≤T∗ 0 R3
(49)
We let k0 > 6 so as to deduce
Z
|v − u|γ w2γ−k (u)du ≤ C ν(v ).
R3
Then combining the above inequality with (49), one gets
R t n+1 Rt n
e− s ζ (τ,x−v (t−τ ),v )dτ
− e− s ζ (τ,x−v (t−τ ),v )dτ

≤ CT∗ ν(v ) sup ∥wk−2γ g n+1 (s) − wk−2γ g n (s)∥L∞

x,v
. (50)
0≤s≤T∗

As a special case that s = 0 in the left hand side of (50), it is direct to see that
I0 (t, x, v ) ≤ CT∗ sup ∥wk−2γ g n+1 (s) − wk−2γ g n (s)∥L∞
x,v
0≤s≤T∗

· |wk−2γ (v )ν(v )g0 (x − vt, v )|

≤ CT∗ sup ∥wk−2γ g n+1 (s) − wk−2γ g n (s)∥L∞
x,v
∥wk g0 ∥L∞
x,v
. (51)
0≤s≤T∗

31/61
For I1 (t, x, v ), from (50) and our choice that T∗ ≤ 1 in (47), we have that

I1 (t, x, v ) ≤ C sup ∥wk−2γ g n+1 (s) − wk−2γ g n (s)∥L∞

x,v
0≤s≤T∗
Z t 
wk (v )
· |Kg n+1 (s, x − v (t − s), v )|
0 ν(v )

wk (v )
+ |Q+ (g n+1 , g n+1 )(s, x − v (t − s), v )| ds.
ν(v )
Using similar arguments for obtaining (36), (37) and (38), it further holds by (16),
(17) and (34) that

I1 (t, x, v ) ≤ T∗ Ck sup ∥wk−2γ g n+1 (s) − wk−2γ g n (s)∥L∞

x,v
0≤s≤T∗

+ ∥wk g0 (s)∥2L∞ .


· ∥wk g0 (s)∥L∞
x,v
(52)
x,v

A similar argument to derive (43) shows that

I2 (t, x, v ) ≤ T∗ Ck sup ∥wk−2γ g n+1 (s) − wk−2γ g n (s)∥L∞
x,v
. (53)
0≤s≤T∗

32/61
Using the fact that
Q+ (g n+1 , g n+1 ) = Q+ (g n+1 − g n , g n ) + Q+ (g n+1 , g n+1 − g n ),
one gets from (42) and the definition of Q+ in (2) that
I3 (t, x, v )
Z t
≤ e −C ν(v )(t−s) wk−2γ (v )(|Q+ (g n+1 − g n , g n )| + |Q+ (g n+1 , g n+1 − g n )|)ds
0
≤ C ( sup ∥wk g n (s)∥L∞
x,v
+ sup ∥wk g n+1 (s)∥L∞
x,v
)
0≤s≤T∗ 0≤s≤T∗

· sup ∥wk−2γ (g n+1 − g n )(s)∥L∞

x,v
0≤s≤T∗
Z t Z Z
wk−2γ (v )
· e −C ν(v )(t−s) |v − u|γ dωduds.
0 R3 S2 wk−2γ (v ′ )wk−2γ (u ′ )
By our choice that k0 > 6, we get k − 2γ > 3 + γ. Then it follows from (18) and
(34) that
!
C
I3 (t, x, v ) ≤ C T∗ Ck + γ+3 ∥wk g0 (s)∥L∞
x,v
sup ∥wk−2γ g n+1 (s)−wk−2γ g n (s)∥L∞
x,v
.
k 4 0≤s≤T∗
(54)

33/61
Combining (48), (51), (52), (53) and (54), and using the condition (31) with ϵ ≤ 1,
one has
|wk−2γ (v )(g n+2 − g n+1 )(t, x, v )|
!
C
≤ T∗ Ck + γ+3 sup ∥wk−2γ g n+1 (s) − wk−2γ g n (s)∥L∞
x,v
,
k 4 0≤s≤T∗

where Ck depends only on k and C is independent of k and j. We may choose C1

in (43) so large that
|wk−2γ (v )(g n+2 − g n+1 )(t, x, v )|
!
C1
≤ T∗ Ck + γ+3 sup ∥wk−2γ g n+1 (s) − wk−2γ g n (s)∥L∞
x,v
.
k 4 0≤s≤T∗

2
Then by our choice that k0 > (4C1 ) γ+3 , we can deduce from the definition of T∗ in
(47) that for 0 ≤ t ≤ T∗ ,
1
∥wk−2γ g n+2 (t) − wk−2γ g n+1 (t)∥L∞
x,v
≤ sup ∥wk−2γ g n+1 (s) − wk−2γ g n (s)∥L∞
x,v
.
2 0≤s≤T∗
(55)
Hence, {wk−2γ g n }∞ ∞
n=0 is a Cauchy sequence in LT∗ ,x,v . We take the limit to obtain
a local mild solution. Recalling F n ≥ 0 for any n ≥ 0, so the limiting function
F = µ + g is also nonnegative. Also (32) follows by letting n → ∞ in (34) and (35).
For the uniqueness, let g1 , g2 ∈ XT∗ ,k,j be two solutions to (6) satisfying (31).
Then, following the similar arguments for deriving (55), it is straightforward to
prove g1 = g2 . We then conclude the proof of Theorem 3.1.

34/61
§7.4 Hard potential case

35/61
Caflisch’s decomposition:

First resolve the problem (6) into a coupling system of g1 = g1 (t, x, v ) and
g2 = g2 (t, x, v ) where g1 and g2 satisfy
√ √
∂t g1 + v · ∇x g1 + νg1 = Ks g1 + Q(g1 + µg2 , g1 + µg2 ), (56)
∂t g2 + v · ∇x g2 + Lg2 = Kb g1 , (57)
with
g1 (0, x, v ) = g10 (x, v ) = F0 (x, v ) − µ(v ), g2 (0, x, v ) = g20 (x, v ) = 0. (58)
The linear operators Ks and Kb above are respectively defined by
Ks g1 (t, x, v ) := χ{|v |≥M} Kg1 (t, x, v ), (59)
and
Kb g1 (t, x, v ) := χ{|v |<M} µ−1/2 (v )Kg1 (t, x, v ), (60)
where for any set E , χ{v ∈E } = 1 if v ∈ E and χ{v ∈E } = 0 otherwise. Note
√
K = Ks + µKb .

36/61
Step 1.Estimates on g1
Lemma 4.1
Let 0 ≤ γ ≤ 1, then there is k0 > 0 large enough such that for any k ≥ k0 , there is
√
M > 0 for the decomposition K = Ks + µKb such that the following estimates hold:
3
∥wk g1 (t)∥L∞
x,v
≤ C (1 + t)− 2 ∥wk g10 ∥L∞
x,v
3 3
+ Ck (1 + t)− 2 sup ∥(1 + s) 4 wk g (s)∥2L∞ , (61)
x,v
0≤s≤t

3
∥wk g1 (t)∥L∞ L2 ≤ C (1 + t)− 2 ∥wk g10 ∥L∞ L2
v x v x
(
3 3
+ Ck (1 + t)− 2 sup ∥(1 + s) 4 wk g (s)∥2L∞
x,v
0≤s≤t
)
3
+ sup ∥(1 + s) 4 wk g (s)∥2L∞ L2 , (62)
0≤s≤t v x

and
3
∥wk g1 (t)∥L∞ L1 ≤ C (1 + t)− 2 ∥wk g10 ∥L∞ L1
v x v x
3 3
+ Ck (1 + t)− 2 sup ∥(1 + s) 4 wk g (s)∥2L∞ L2 , (63)
0≤s≤t v x

where Ck is a constant which may depend on k.

37/61
Key:
Linear term Ks g1 := χ{|v |≥M} Kg1 (t, x, v ) can be small for M large enough.
Proof of Lemma 4.1: We only show (61) in L∞ x,v -norm. We integrate along the
characteristic in (56) to get
Z t
g1 (t, x, v ) =e −ν(v )t g10 (x − vt, v ) + e −ν(v )(t−s) Ks g1 (s, x − v (t − s), v )ds
0
Z t
−ν(v )(t−s)
+ e Q(g , g )(s, x − v (t − s), v )ds. (64)
0
We now start to prove the first estimate (61). We directly obtain from (64) that
|wk (v )g1 (t, v )| ≤ Ce −λt ∥wk g10 ∥L∞
x,v
+ I4 (t, x, v ) + I5 (t, x, v ), (65)
with
Z t
I4 (t, x, v ) := e −ν(v )(t−s) |wk (v )Ks g1 (s, x − v (t − s), v )| ds,
0
Z t
I5 (t, x, v ) := e −ν(v )(t−s) |wk (v )Q(g , g )(s, x − v (t − s), v )| ds,
0
where λ > 0 is a constant.

38/61
We choose k0 > 3 + γ such that Lemma 2.1 and Lemma 2.3 hold for k ≥ k0 .
Taking the L∞
x norm and using the definition of Ks in (59), we have
Z t
∥I4 (t, v )∥L∞
x
≤ e −ν(v )(t−s) ∥χ{|v |>M} wk (v )Kg1 (t, v )∥L∞
x
ds.
0
It follows from (19) that
!
3
Z t C Ck
−ν(v )(t−s) − 23
∥I4 (t, v )∥L∞
x
≤ C sup ∥(1+s) wk g1 (s)∥L∞
2
x,v
e ν(v )(1+s) γ+3
+ 2 ds,
0≤s≤t 0 k 2 M

which combining with (27) yields

!
1 Ck 3 3
∥I4 (t, v )∥L∞
x
≤C2 γ+3
+ 2 (1 + t)− 2 sup ∥(1 + s) 2 wk g1 (s)∥L∞
x,v
, (66)
k 2 M 0≤s≤t

2
for some constant C2 . We choose k0 large enough, for instance, k0 > (4C2 ) γ+3 , so
that for any k ≥ k0 ,
C2 1
γ+3
≤ . (67)
k 2 4
Then we let
p
M=2 C2 Ck (68)
to get from (66) and (67) that
1 3 3
∥I4 (t, v )∥L∞
x
≤ (1 + t)− 2 sup ∥(1 + s) 2 wk g1 (s)∥L∞
x,v
. (69)
2 0≤s≤t

39/61
For I5 , by the definition of Q in (2), one gets
Z t
|I5 (t, x, v )| ≤ e −ν(v )(t−s) |wk (v )Q(g , g )(s, x − v (t − s), v )| ds
0
3
Z t 3
≤ C sup ∥(1 + s) 4 wk g (s)∥2L∞ e −ν(v )(t−s) (1 + s)− 2
x,v
0≤s≤t 0
Z Z  
wk (v ) 1
× |v − u|γ + dωduds. (70)
R3 S2 wk (u ′ )wk (v ′ ) wk (u)
R |v −u|γ
Noticing the fact that R3 wk (u)
dωdu ≤ Ck ν(v ) for k ≥ k0 > 3 + γ, we deduce by
(17) and (27) that
3
Z t 3
|I5 (t, x, v )| ≤ Ck sup ∥(1 + s) 4 wk g (s)∥2L∞ e −ν(v )(t−s) ν(v )(1 + s)− 2 ds
x,v
0≤s≤t 0
3 3
≤ Ck (1 + t)− 2 sup ∥(1 + s) 4 wk g (s)∥2L∞ . (71)
x,v
0≤s≤t

Collecting (65), (69) and (71), we obtain

1 3 3
∥wk g1 (t)∥L∞
x,v
≤Ce −λt ∥wk g10 ∥L∞
x,v
+ (1 + t)− 2 sup ∥(1 + s) 2 wk g1 (s)∥L∞
x,v
2 0≤s≤t
3 3
+ Ck (1 + t)− 2 sup ∥(1 + s) 4 wk g (s)∥2L∞ , (72)
x,v
0≤s≤t

which yields
3 3
sup ∥(1 + s) 2 wk g1 (s)∥L∞
x,v
≤C ∥wk g10 ∥L∞
x,v
+ Ck sup ∥(1 + s) 4 wk g (s)∥2L∞ . (73)
x,v
0≤s≤t 0≤s≤t

Then (61) holds from the above inequality.

40/61
Step 2.Time decay in the symmetric case

We consider the corresponding linear homogeneous problem

∂t f + v · ∇x f + Lf = 0, f (0, x, v ) = f0 (x, v ). (74)

Proposition 4.2
Let 0 ≤ γ ≤ 1. Let f be the solution to the problem (74), then it holds
3
∥f (t)∥L2 ≤ C (1 + t)− 4 (∥f0 ∥L2 + ∥f0 ∥L2 L1 ), (75)
x,v x,v v x

for any t ≥ 0.

Remark:
The proof can be based on either the semigroup result or the Fourier energy
method.

41/61
Lemma 4.3
Let 0 ≤ γ ≤ 1 and k > 3 + γ. Let f be the solution to the problem (74), then it holds
3
∥wk f (t)∥L∞
x,v
≤ Ck (1 + t)− 4 (∥wk f0 ∥L∞
x,v
+ ∥f0 ∥L2 + ∥f0 ∥L2 L1 ), (76)
x,v v x

− 43
∥wk f (t)∥L∞ L2 ≤ Ck (1 + t) (∥wk f0 ∥L∞ L2 + ∥f0 ∥L2 L1 ), (77)
v x v x v x

for any t ≥ 0.

Remark:
L2x,v → L∞ 2 ∞ 2
x,v and Lx,v → Lv Lx interplay.

42/61
Proof of Lemma 4.3: Prove (76) only. Recall the definition of the function
k = k(v , η) in Lemma 2.4. Define
wk (v )
kwk (v , η) = k(v , η) (78)
wk (η)
Z
Kwk f (v ) = kwk (v , η)f (η)dη.
R3
We can rewrite the equation (74) as
∂t wk f + v · ∇x wk f + νwk f − Kwk wk f = 0.
Then
wk (v )f (t, x, v ) = e −ν(v )t wk (v )f0 (x − vt, v )
Z t Z
+ e −ν(v )(t−s) kwk (v , η)(wk f )(s, x − v (t − s), η)dηds.
0 R3
We further have
wk (v )f (t, x, v )
= e −ν(v )t wk (v )f0 (x − vt, v )
Z t Z
+ e −ν(v )(t−s) kwk (v , η)e −ν(η)s wk (η)f0 (x − v (t − s), η)dηds
0 R3
Z t Z Z s
+ e −ν(v )(t−s) kwk (v , η) e −ν(η)(s−s1 )
0 R3 0
Z
× kwk (η, ξ)wk (ξ)f (s1 , x1 − η(s − s1 ), ξ)dξds1 dηds, (79)
R3
where x1 = x − v (t − s).
43/61
For the L∞ x,v estimate (76), by the fact that ν(v ) ≥ ν0 for some constant ν0 > 0,
the first term on the right hand side above is directly bounded by e −ν0 t ∥wk f0 ∥L∞ x,v
.
For the second term, it holds from (21) that
Z t Z
e −ν(v )(t−s) kwk (v , η)e −ν(η)s wk (η)f0 (x − v (t − s), η)dηds
0 R3
Z
≤ Cte −ν0 t ∥wk f0 ∥L∞x,v
kwk (v , η)dη
R3
ν
− 20 t
≤ Ck e ∥wk f0 ∥L∞
x,v
.
Then we have
|wk (v )f (t, x, v )| ≤ Ck e −λ0 t ∥wk f0 ∥L∞
x,v
+ I6 (t, x, v ), (80)
where λ0 > 0 is a constant and
Z t Z Z
I6 (t, x, v ) = e −ν(v )(t−s) |kwk (v , η)kwk (η, ξ)|
0 R3 R3
Z s
× e −ν(η)(s−s1 ) |wk (ξ)f (s1 , x1 − η(s − s1 ), ξ)| dξds1 dηds. (81)
0
We split I6 (t, x, v ) into four cases:

44/61
Case 1. |v | ≥ N.
Ck 3 3
I6 (t, x, v ) ≤ (1 + t)− 4 sup ∥(1 + s) 4 wk f (s)∥L∞
x,v
. (82)
N 0≤s≤t

Case 2. |v | ≤ N, |η| ≥ 2N or |η| ≤ 2N, |ξ| ≥ 3N. In this case, we have either
|v − η| ≥ N or |η − ξ| ≥ N. Then,
Z t
N2 3
I6 (t, x, v ) ≤ Ce − 20 sup ∥(1 + s) 4 wk f (s)∥L∞
x,v
e −ν0 (t−s)
0≤s≤t 0
|v −η|2 |η−ξ|2
Z Z
· kwk (v , η)e 20 kwk (η, ξ)e 20
R3 R3
Z s 3
· e −ν0 (s−s1 ) (1 + s1 )− 4 dξds1 dηds
0
Ck 3 3
≤ (1 + t)− 4 sup ∥(1 + s) 4 wk f (s)∥L∞
x,v
. (83)
N 0≤s≤t

Case 3. s − s1 ≤ λ ≪ 1. It follows from the fact s − λ ≤ s1 ≤ s that

3
I6 (t, x, v ) ≤ C sup ∥(1 + s) 4 wk f (s)∥L∞
x,v
0≤s≤t
Z t Z Z
× e −ν0 (t−s) |kwk (v , η)kwk (η, ξ)|
0 R3 R3
Z s
3
· e −ν0 (s−s1 ) (1 + s1 )− 4 dξds1 dηds
s−λ
3 3
≤ Ck λ(1 + t)− 4 sup ∥(1 + s) 4 wk f (s)∥L∞
x,v
. (84)
0≤s≤t

45/61
Case 4. |v | ≤ N, |η| ≤ 2N, |ξ| ≤ 3N, s − s1 ≥ λ. This case needs to be treated
more carefully. By the property of k (21), we can approximate kwk by a smooth
function kN with compact support such that
Z
Ck
sup |kwk (v , η) − kN (v , η)| dη ≤ . (85)
|v |≤3N |η|≤3N N
We can split I6
3
Z t
I6 (t, x, v ) ≤ C sup ∥(1 + s) 4 wk f (s)∥L∞
x,v
e −ν0 (t−s)
0≤s≤t 0
Z Z Z s−λ 3
× |kwk (v , η) − kN (v , η)| |kwk (η, ξ)| e −ν0 (s−s1 ) (1 + s1 )− 4 ds1 dηdξds
R3 R3 0
3
Z t
+ C sup ∥(1 + s) 4 wk f (s)∥L∞
x,v
e −ν0 (t−s)
0≤s≤t 0
Z Z Z s−λ 3
× |kN (v , η)| |kwk (η, ξ) − kN (η, ξ)| e −ν0 (s−s1 ) (1 + s1 )− 4 ds1 dηdξds
R3 R3 0
Z t ZZ
−ν(v )(t−s)
+ e |kN (v , η)kN (η, ξ)|
0 |η|≤2N,|ξ|≤3N
Z s−λ
× e −ν(η)(s−s1 ) |(wk f )(s1 , x1 − η(s − s1 ), ξ)| ds1 dηdξds.
0

46/61
Combining the above inequality and (85), we get
Ck 3 3
I6 (t, x, v ) ≤ (1 + t)− 4 sup ∥(1 + s) 4 wk f (s)∥L∞
x,v
+ I61 (t, x, v ), (86)
N 0≤s≤t

where
Z t Z s−λ
I61 (t, x, v ) = Ck,N e −ν0 (t−s) e −ν0 (s−s1 )
0 0
ZZ
× |f (s1 , x1 − η(s − s1 ), ξ)| dηdξds1 ds.
|η|≤2N,|ξ|≤3N

Using Cauchy-Schwarz’s inequality and change of variable y = x1 − η(s − s1 ), we

have
Z t Z s−λ
I61 (t, x, v ) ≤ Ck,N e −ν0 (t−s) e −ν0 (s−s1 )
0 0
!1
ZZ 2
2
× |f (s1 , x1 − η(s − s1 ), ξ)| dηdξ ds1 ds
|η|≤2N,|ξ|≤3N
!1
Z t Z s−λ ZZ 2
−ν0 (t−s) −ν0 (s−s1 ) 2
≤ Ck,N,λ e e |f (s1 , y , ξ)| dydξ ds1 d
0 0 |η|≤2N,|ξ|≤3N
Z t Z s−λ
≤ Ck,N,λ e −ν0 (t−s) e −ν0 (s−s1 ) ∥f (s1 )∥L2 ds1 ds. (87)
x,v
0 0

47/61
Then it follows from (30), (75) and (87) that
Z t Z s−λ
3
I61 (t, x, v ) ≤ Ck,N,λ e −ν0 (t−s) e −ν0 (s−s1 ) (1 + s1 )− 4 (∥f0 ∥L2 + ∥f0 ∥L2 L1 )ds1 ds
x,v v x
0 0
− 43
≤ Ck,N,λ (1 + t) (∥f0 ∥L2 + ∥f0 ∥L2 L1 ). (88)
x,v v x

For Case 4, it holds by (86) and (88) that

Ck 3 3
I6 (t, x, v ) ≤ (1 + t)− 4 sup ∥(1 + s) 4 wk f (s)∥L∞
x,v
N 0≤s≤t
3
+ Ck,N,λ (1 + t)− 4 (∥f0 ∥L2 + ∥f0 ∥L2 L1 ). (89)
x,v v x

Collecting four cases (82), (83), (84) and (89), one has from (80) that
∥wk f (t)∥L∞
x,v
≤ Ck e −λ0 t ∥wk f0 ∥L∞
x,v

Ck 3 3
+ Ck λ (1 + t)− 4 sup ∥(1 + s) 4 wk f (s)∥L∞


+ x,v
N 0≤s≤t
3
+ Ck,N,λ (1 + t)− 4 (∥f0 ∥L2 + ∥f0 ∥L2 L1 ).
x,v v x

Then (76) holds by choosing N large and λ small.

48/61
Step 3. Estimate on g2
Lemma 4.4
Let 0 ≤ γ ≤ 1 and k > 3 + γ. Let the operator Kb be defined in (60) for the
Caflisch’s decomposition. Then, for any 1 ≤ p ≤ ∞ and j ≥ 0, it holds that
∥wj Kb f ∥L∞ Lpx ≤ Cj,k ∥wk f ∥L∞ Lpx . (90)
v v

49/61
Recall
Z t
g2 (t) = U(t − s)Kb g1 (s) ds. (91)
0

Lemma 4.5
Let 0 ≤ γ ≤ 1, then there is k0 > 0 large enough such that for any k ≥ k0 , there is
√
M > 0 for the decomposition K = Ks + µKb such that the following estimates hold:
3
n o
∥wk g2 (t)∥L∞
x,v
≤ Ck (1 + t)− 4 ∥wk g10 ∥L∞
x,v
+ ∥wk g10 ∥L∞ L2 + ∥wk g10 ∥L∞ L1
v x v x
(
3 3
+ Ck (1 + t)− 4 sup ∥(1 + s) 4 wk g (s)∥2L∞
x,v
0≤s≤t
)
3
+ sup ∥(1 + s) 4 wk g (s)∥2L∞ L2 , (92)
0≤s≤t v x

and
3
n o
∥wk g2 (t)∥L∞ L2 ≤ Ck (1 + t)− 4 ∥wk g10 ∥L∞ L2 + ∥wk g10 ∥L∞ L1
v x v x v x
(
3 3
+ Ck (1 + t)− 4 sup ∥(1 + s) 4 wk g (s)∥2L∞
x,v
0≤s≤t
)
3
+ sup ∥(1 + s) 4 wk g (s)∥2L∞ L2 . (93)
0≤s≤t v x

50/61
Proof of Lemma 4.5: Prove (92) only. We use (91) and (76) to get
Z t
|wk g2 (t)| ≤ |wk U(t − s)(Kb g1 )(s)|ds
0
Z t 3
≤ Ck (1 + t − s)− 4 (∥wk Kb g1 (s)∥L∞
x,v
+ ∥Kb g1 (s)∥L2 + ∥Kb g1 (s)∥L2 L1 )ds.
x,v v x
0
Noticing the fact that ∥f ∥L2 ≤ C ∥wk f ∥L∞ for k ≥ k0 > 3, we have
Z t
3
|wk g2 (t)| ≤ Ck (1 + t − s)− 4 (∥wk Kb g1 (s)∥L∞
x,v
+ ∥wk Kb g1 (s)∥L∞ L2 + ∥wk Kb g1 (s)∥L∞ L1 )d
v x v x
0
Then it follows from (90) that
Z t
3
|wk g2 (t)| ≤ Ck (1 + t − s)− 4 (∥wk g1 (s)∥L∞
x,v
+ ∥wk g1 (s)∥L∞ L2 + ∥wk g1 (s)∥L∞ L1 )ds.
v x v x
0
Recalling our estimate for g1 (61), (62) and (63), it holds that
Z t
3 3
(1 + t − s)− 4 (1 + s)− 2 ∥wk g10 ∥L∞


|wk g2 (t)| ≤ Ck x,v
+ ∥wk g10 ∥L∞ L2 + ∥wk g10 ∥L∞ L1 ds
v x v x
0
Z t
− 43 − 32
+ Ck (1 + t − s) (1 + s)
0
( )
3 3
× sup ∥(1 + s) 4 wk g (s)∥2L∞ + sup ∥(1 + s) 4 wk g (s)∥2L∞ L2 ds.
x,v v x
0≤s≤t 0≤s≤t
(94)
Thus, (92) follows by (29) and (94).

51/61
Proof of Theorem 1.1 in Hard Potential Case:
√
Recall our definition for g that g = g1 + µg2 . Then it is straightforward to see
that for k ≥ j ≥ k0 ,
√
∥wk g (t)∥L∞
x,v
≤ ∥wk g1 (t)∥L∞
x,v
+ ∥wk µg2 (t)∥L∞
x,v

≤ ∥wk g1 (t)∥ L∞
x,v
+ Cj,k ∥wj g2 (t)∥L∞
x,v
.

We have from (61), (92) and the condition j ≤ k that

∥wk g (t)∥L∞
x,v
3
n o
≤ Cj,k (1 + t)− 4 ∥wk g10 ∥L∞
x,v
+ ∥wj g10 ∥L∞ L2 + ∥wj g10 ∥L∞ L1
v x v x
( )
3 3 3
+ Cj,k (1 + t)− 4 sup ∥(1 + s) 4 wk g (s)∥2L∞ + sup ∥(1 + s) 4 wj g (s)∥2L∞ L2 .
x,v v x
0≤s≤t 0≤s≤t
(95)
Due to the L∞ 2
v Lx norm in (95), we need to estimate
3
sup0≤s≤t ∥(1 + s) 4 wj g (s)∥2L∞ L2 . By (62), (93) and the condition j ≤ k, one has
v x

∥wj g (t)∥L∞ L2
v x
3
n o
≤ Cj (1 + t)− 4 ∥wj g10 ∥L∞ L2 + ∥wj g10 ∥L∞ L1
v x v x
( )
3 3 3
+ Cj (1 + t)− 4 sup ∥(1 + s) 4 wk g (s)∥2L∞ + sup ∥(1 + s) 4 wj g (s)∥2L∞ L2 .
x,v v x
0≤s≤t 0≤s≤t
(96)
52/61
Combining (95) and (96), we obtain
3 3
sup ∥(1 + s) 4 wk g (s)∥L∞
x,v
+ sup ∥(1 + s) 4 wj g (s)∥L∞ L2
v x
0≤s≤t 0≤s≤t
n o
≤ Cj,k ∥wk g10 ∥L∞
x,v
+ ∥wj g10 ∥L∞ L2 + ∥wj g10 ∥L∞ L1
v x v x
( )
3 3
+ Cj,k sup ∥(1 + s) 4 wk g (s)∥2L∞ + sup ∥(1 + s) 4 wj g (s)∥2L∞ L2 . (97)
x,v v x
0≤s≤t 0≤s≤t

Recall our definitions of ∥ · ∥Xj,k in (10) and ∥ · ∥Yj,k in (11). By the local-in-time
existence together with the continuity argument, from (97), (13) follows by (12) for
a small constant ϵ0 which depends on j and k. Hence, the global existence is
established and the proof of Theorem 1.1 is complete.

53/61
§7.5 Soft potential case

54/61
Step 1. Estimates on g1
Lemma 5.1
Let −3 < γ < 0 and 0 < ϵ ≤ 12 . There is k0 > 0 large enough such that for any
√
k ≥ k0 , there is a constant M > 0 for the decomposition K = Ks + µKb such that
the following estimates hold:
∥wk g1 (t)∥L∞
x,v
≤Cϵ (1 + t)−1+ϵ ∥wk+|γ| g10 ∥L∞
x,v
3
+ Cϵ,k (1 + t)−1+ϵ sup ∥(1 + s) 4 −ϵ wk g (s)∥2L∞ ,
x,v
0≤s≤t
(98)
∥wk g1 (t)∥L∞ L2 ≤ Cϵ (1 + t)−1+ϵ ∥wk+|γ| g10 ∥L∞ L2
v x v x
(
3
+Cϵ,k (1 + t)−1+ϵ sup ∥(1 + s) 4 −ϵ wk g (s)∥2L∞
x,v
0≤s≤t
)
3
+ sup ∥(1 + s) 4 −ϵ wk g (s)∥2L∞ L2 , (99)
0≤s≤t v x

∥wk g1 (t)∥L∞ L1 ≤ Cϵ (1 + t)−1+ϵ ∥wk+|γ| g10 ∥L∞ L1

v x v x
3
+ Cϵ,k (1 + t)−1+ϵ sup ∥(1 + s) 4 −ϵ wk g (s)∥2L∞ L2 ,
0≤s≤t v x

(100)
where Cϵ depends only on ϵ and Cϵ,k depends only on ϵ and k.

55/61
Step 2. Time decay in the symmetric case
Proposition 5.2
Let −3 < γ < 0. Let f be the solution to the problem (74), then it holds
3
∥f (t)∥L2 ≤ C (1 + t)− 4 (∥ν −1 f0 ∥L2 + ∥ν −1 f0 ∥L2 L1 ), (101)
x,v x,v v x

for any t ≥ 0.

56/61
With this proposition, we are able to obtain the L∞
v decay properties of the solution
to (74).
Lemma 5.3
Let −3 < γ < 0 and k > 3. Let f be the solution to the problem (74), then it holds
3
∥wk f (t)∥L∞
x,v
≤ Ck (1 + t)− 4 (∥wk+|γ| f0 ∥L∞
x,v
+ ∥ν −1 f0 ∥L2 + ∥ν −1 f0 ∥L2 L1 ), (102)
x,v v x

− 34
∥wk f (t)∥L∞ L2 ≤ Ck (1 + t) (∥wk+|γ| f0 ∥L∞ L2 + ∥ν −1 f0 ∥L2 L1 ), (103)
v x v x v x

for any t ≥ 0.

Lemma 5.4
Let −3 < γ < 0 and k > 3. Let the operator Kb be defined in (60) with the constant
M = M(ϵ, k). Then, for any 1 ≤ p ≤ ∞ and j ≥ 0, it holds that
∥wj Kb f ∥L∞ Lpx ≤ Cϵ,j,k ∥wk f ∥L∞ Lpx , (104)
v v

where Cϵ,j,k depends only on ϵ, j and k.

57/61
Step 3. Estimate on g2
Lemma 5.5
Let −3 < γ < 0 and 0 < ϵ ≤ 12 . There is k0 > 0 large enough such that for any
√
k ≥ k0 , there is a constant M > 0 for the decomposition K = Ks + µKb such that
the following estimates hold:
∥wk g2 (t)∥L∞
x,v
3
n o
≤ Cϵ,k (1 + t)− 4 +ϵ ∥wk+|γ| g10 ∥L∞
x,v
+ ∥wk+|γ| g10 ∥L∞ L2 + ∥wk+|γ| g10 ∥L∞ L1
v x v x
(
3 3
+ Cϵ,k (1 + t)− 4 +ϵ sup ∥(1 + s) 4 −ϵ wk g (s)∥2L∞
x,v
0≤s≤t
)
3
+ sup ∥(1 + s) 4 −ϵ wk g (s)∥2L∞ L2 , (105)
0≤s≤t v x

3
n o
∥wk g2 (t)∥L∞ L2 ≤ Cϵ,k (1 + t)− 4 +ϵ ∥wk+|γ| g10 ∥L∞ L2 + ∥wk+|γ| g10 ∥L∞ L1
v x v x v x
(
3 3
+ Cϵ,k (1 + t)− 4 +ϵ sup ∥(1 + s) 4 −ϵ wk g (s)∥2L∞
x,v
0≤s≤t
)
3
+ sup ∥(1 + s) 4 −ϵ wk g (s)∥2L∞ L2 .
0≤s≤t v x

(106)

58/61
Proof of Theorem 1.2 in Soft potential case:
√
Recall that g = g1 + µg2 . For k ≥ j ≥ k0 , a direct calculation shows that
√
∥wk g (t)∥L∞
x,v
≤ ∥wk g1 (t)∥L∞
x,v
+ ∥wk µg2 (t)∥L∞ x,v
≤ ∥wk g1 (t)∥L∞
x,v
+ Cj,k ∥wj g2 (t)∥L∞
x,v
.

By (98), (105) and the condition j ≤ k, it holds that

∥wk g (t)∥L∞
x,v
3
n o
≤ Cϵ,j,k (1 + t)− 4 +ϵ ∥wk+|γ| g10 ∥L∞
x,v
+ ∥wj+|γ| g10 ∥L∞ L2 + ∥wj+|γ| g10 ∥L∞ L1
v x v x
( )
3 3 3
+Cϵ,j,k (1 + t)− 4 +ϵ sup ∥(1 + s) 4 −ϵ wk g (s)∥2L∞ + sup ∥(1 + s) 4 −ϵ wj g (s)∥2L∞ L2 .
x,v v x
0≤s≤t 0≤s≤t
(107)
Also we have
√
∥wj g (t)∥L∞ L2 ≤ ∥wj g1 (t)∥L∞ L2 + ∥wj µg2 (t)∥L∞ L2 ≤ ∥wj g1 (t)∥L∞ L2 + ∥wj g2 (t)∥L∞ L2 .
v x v x v x v x v x

Then it follows from (99), (106) and the condition j ≤ k that

∥wj g (t)∥L∞ L2
v x
3
n o
≤ Cϵ,j (1 + t)− 4 +ϵ ∥wj+|γ| g10 ∥L∞ L2 + ∥wj+|γ| g10 ∥L∞ L1
v x v x
( )
3 3 3
+Cϵ,j (1 + t)− 4 +ϵ sup ∥(1 + s) 4 −ϵ wk g (s)∥2L∞ + sup ∥(1 + s) 4 −ϵ wj g (s)∥2L∞ L2 .
x,v v x
0≤s≤t 0≤s≤t
(108)

59/61
Combining (107) and (108), we obtain
3 3
sup ∥(1 + s) 4 −ϵ wk g (s)∥L∞
x,v
+ sup ∥(1 + s) 4 −ϵ wj g (s)∥L∞ L2
v x
0≤s≤t 0≤s≤t
n o
≤ Cϵ,j,k ∥wk+|γ| g10 ∥L∞
x,v
+ ∥wj+|γ| g10 ∥L∞ L2 + ∥wj+|γ| g10 ∥L∞ L1
v x v x
( )
3 3
+ Cϵ,j,k sup ∥(1 + s) 4 −ϵ wk g (s)∥2L∞ + sup ∥(1 + s) 4 −ϵ wj g (s)∥2L∞ L2 .
x,v v x
0≤s≤t 0≤s≤t
(109)
Recall our definitions of ∥ · ∥Xj,k in (10) and ∥ · ∥Yj,k in (11). By the local-in-time
existence together with the continuity argument, from (109), (15) follows by (14)
for a small constant ϵ0 which depends on ϵ, j and k. Hence, the global solution is
established and the proof of Theorem 1.2 is complete.

60/61
End of Lecture 7

61/61