Lecture 2

Initial Value Problem on the Spatially

Homogeneous Equation

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

1/43
 Spatially homogeneous Boltzmann equation:

f = f (t, ξ), t ≥ 0, ξ ∈ R3

satisfies
(
∂t f = Q(f , f ), t > 0, ξ ∈ R3
(IVP)
f (0, ξ) = f0 (ξ), ξ ∈ R3 .

Hard sphere model:

Z Z
Q(f , f )(ξ) = dξ∗ dω |(ξ − ξ∗ ) · ω|(f ′ f∗′ − ff∗ )
R3 (ξ−ξ∗ )·ω≥0

f ′ = f (ξ ′ ), f∗′ = f (ξ∗′ ), f∗ = f (ξ∗ )

ω-representation:

ξ ′ = ξ − (ξ − ξ∗ ) · ω ω, ξ∗′ = ξ + (ξ − ξ∗ ) · ω ω.
2/43
Outline

§2.1 Existence for truncated equation

§2.2 Removing truncation

§2.3 Uniqueness and Regularity

§2.4 Pointwise bounds

3/43
2.1 Existence for truncated equation

4/43
The truncated (ignore large relative velocities) equation:
(
∂t f M = Q M (f M , f M )
(1.1)
f M (0, ·) = f0

The Boltzmann collision operator with cutoff is

Z
1
M
Q (f , g ) = χM (|ξ − ξ∗ |)(f ′ g∗′ + g ′ f∗′ − fg∗ − gf∗ )
2 (ξ−ξ∗ )·ω≥0
(ξ − ξ∗ ) · ω dω dξ∗

where the cut-off function χM is defined by

(
1 r ≤M
χM (r ) = .
0 otherwise

5/43
It can be checked that ∥Q M (f , f )∥L1 ≤ CM∥f ∥2L1 and

∥Q M (f , f ) − Q M (g , g )∥L1
=∥Q M (f + g , f − g )∥L1
≤CM∥f + g ∥L1 ∥f − g ∥L1 .

As a consequence, there exists a local solution

f M ∈ C 1 ([0, T ]; L1 )

to problem (1.1) for T > 0 which is small enough.

6/43
Question: f0 ≥ 0 =⇒ f M ≥ 0, ∀t > 0?
Z Z
Note: f0 dξ = 1 =⇒ f M (t, ξ) dξ = 1, 0 ≤ ∀t ≤ T .

Consider the equation (µ > 0 to be chosen)


∂t g + µg = ΓM (g )  

g (0, ·) = f0 ≥ 0

Z Good if µ ≫ 1.

M M
Γ (g ) = Q (g ) + µg g (ξ) dξ 



Use the iterations

 Z t
g n = e −µt f +

e −µ(t−s) ΓM (g n−1 ) ds
0
0
 0
 g =0

and we have for sufficiently large µ,

ΓM positive =⇒ g ≥ 0.

7/43
 ▶ g n ≥ g n−1 ≥ · · · ≥ g 1 ≥ g 0 = 0
Z Z t Z 2
▶ g n dξ = e −µt + µ e −µ(t−s) g n−1 dξ ds
0
Z Z
∴ if g n−1 dξ ≤ 1 then g n dξ ≤ 1.
Z
=⇒ ∃g such that 0 ≤ g = lim g n with g dξ ≤ 1
n→∞

In the end, we show Z

gdξ = 1.

Then, the problem on g is the same as on the one on f M . By

uniqueness (f → Q M (f , f ) is Lip),

f M = g ≥ 0.

8/43
2.2 Removing truncation

9/43
Recall: Weak compactness in L1 .

Let {fn }n≥1 ⊂ L1 (Rd ). Then the following (i) and (ii) are
equivalent:
(i) {fn } is contained in a weakly sequentially compact set of
L1 (Rd ).
(iia) {fn } is bounded in L1 (Rd ).
(iib) ∀ε > 0, ∃δ > 0 such that ∀E ⊂ Rd (E measurable) with
λ(E ) < δ, Z
sup |fn | dx ≤ ε.
n E
Z
d
(iic) ∀ε > 0, ∃K compact, K ⊂ R such that sup |fn | dx ≤ ε.
n Rd \K

Remark: Equiintegrability (iib)+(iic)

Ref.: Dunford-Schwartz, Linear Operators, I. (1963)

10/43
(iia):

Z
f M (ξ, t) dξ = 1, f M (ξ, t) ≥ 0.
R3

11/43
(iic): Z Z
M 1 2E
f (ξ, t) dξ ≤ 2 ξ 2 f M (ξ, t) dξ ≤
|ξ|>R R R2
where Z
1
E = E (f0 ) = ξ 2 f0 (ξ) dξ
2
from the conservation of energy.

12/43
 R
(iib): For any A with |A|R= A dξ < ε, fix a > 1 (TBD) and recall
the H-theorem: H(f ) = f ln f dξ, H(f M ) ≤ H(f0 ),
Z Z
1
f M (t, ξ) dξ ≤ f M (t, ξ) ln f M (t, ξ)χ{f M (t)>a} dξ + aε
A ln a A
Z
1
≤ f M (t, ξ) ln f M (t, ξ)χ{f M (t)>1} dξ + aε
ln a
Z
H(f0 ) 1
≤ + aε − f M (t, ξ) ln f M (t, ξ)χ{f M (t)<1} dξ
ln a ln a
Z
H(f0 ) 1
≤ + aε − f M (t, ξ) ln f M (t, ξ)χ{e −|ξ|2 <f M (t)<1} dξ
ln a ln a
Z
1
− f M (t, ξ) ln f M (t, ξ)χ{f M (t)≤e −|ξ|2 } dξ
ln a
Z 
|ξ|2
Z
H(f0 ) 1
≤ + aε + f M (t, ξ)|ξ|2 dξ + C e − 2 dξ
ln a ln a
1 1
(because x 2 | ln x| is bounded if x < 1, and after choosing a = ε− 2 )
2H(f0 ) √ C (1 + E )
≤− + ε− .
ln ε ln ε

13/43
Dunford-Pettis theorem ⇒ there exists f ∈ L1 ([0, T ] × R3 ) such that
w
fM ⇀f.

It is obvious to see:
▶ f ≥ 0,
Z Z
▶ |ξ| f (t, ξ) dξ ≤ |ξ|2 f0 (ξ) dξ, a.e. t ∈ [0, T ].
2
R3

14/43
Question: Q n (f n , f n ) ≡ Q Mn (f n , f n ) → Q(f , f ) ?

A major problem with weak convergence is that nonlinear

functions are in general NOT weakly continuous.

Typical Example:
w
sin(nx)1[0,1] (x) ⇀ 0 in L1 (R)
w 1
sin2 (nx)1[0,1] (x) ⇀ 1[0,1] (x) in L1 (R)
2
by Riemann-Lebesgue Lemma.

15/43
Notation:
Z
s
∥f ∥1,s = (1 + |ξ|2 ) 2 |f (ξ)| dξ,

L1s = {f : ∥f ∥1,s < ∞}.

Lemma (Povzner inequality) Suppose s ≥ 2, 0 ≤ f , g ∈ L1s . Then

Z
s
(1 + |ξ|2 ) 2 Q(f , g ) dξ ≤ C (s){∥f ∥1,s ∥g ∥1,2 + ∥g ∥1,s ∥f ∥1,2 }.

(To be proved later, also true for Q M .)

16/43
Then we have
Z t
∥f M (t)∥1,s ≤ ∥f (0)∥1,s + C (s) ∥f M (τ )∥1,s ∥f M (τ )∥1,2 dτ (Povzner)
0

∥f M (t)∥1,s ≤ ∥f (0)∥1,s e C (s)Et (Gronwall’s inequality)

and

∥Q+M (f , f )∥1,2 = ∥fR M (f )∥1,2

Z
≤ C (1 + |ξ|2 )f (ξ)(|ξ|∥f ∥1 + ∥f ∥1,2 ) dξ

≤ C {∥f ∥1,3 ∥f ∥1 + ∥f ∥21,2 }

Z
since R M (f )(ξ) ≤ C |ξ − ξ∗ |f (ξ∗ ) dξ∗ ≤ C (|ξ|∥f ∥1 + ∥f ∥1,2 ).

Therefore, on condition that f0 ∈ L14 :

Z t
∥f M (t) − f M (r )∥1,2 ≤ ∥Q M (f M (τ ), f M (τ ))∥1,2 dτ ≤ C (T )|t − r |.
r

17/43
Equicontinuity ⇒ f ∈ C ([0, T ]; L12 )

∴ up to a subsequence
w
f n (t) ⇀ f (t) in L1 (R3 ), ∀t ∈ [0, T ]

and
Z Z
φ(ξ, ξ∗ )f n (ξ)f n (ξ∗ ) dξ dξ∗ → φ(ξ, ξ∗ )f (ξ)f (ξ∗ ) dξ dξ∗

for t ∈ [0, T ], φ ∈ L∞ (R3 × R3 ).

18/43
Apply this to gain term: For φ ∈ L∞ (R3 ),
Z
n
[Q+ (f n ) − Q+ (f )](ξ)φ(ξ) dξ
Z Z
≤ dξ dξ∗ dω (ξ − ξ∗ ) · ω
(ξ−ξ∗ )·ω≥0

× χA (|ξ − ξ∗ |)φ(ξ ′ ){f n (ξ)f n (ξ∗ ) − f (ξ)f (ξ∗ )} (→ 0 as n → ∞)

Z
+ C ∥φ∥∞ (|ξ| + |ξ∗ |)f n (ξ)f n (ξ∗ ) dξ dξ∗ (→ 0 as A → ∞)
|ξ|> A2
Z !
+ (|ξ| + |ξ∗ |)f (ξ)f (ξ∗ ) dξ dξ∗ (→ 0 as A → ∞).
|ξ|> A2

Note: it is also true for loss term.

19/43
Proposition 2.1 Z
If 0 ≤ f0 ∈ L13 and its entropy f0 ln f0 dξ is finite, then
∃f ∈ C ([0, T ]; L12 ) (T > 0: arbitrary) such that
Z t
f (t) = f0 + Q(f (s), f (s)) ds. (2.1)
0

And also
▶ 0 ≤ f (t) ∈ L13 , ∀t finite,
Z  Z 
▶ f (t, ξ) dξ = 1 = f0 (ξ) dξ , ∀t,
R3 R3
Z Z
▶ |ξ|2 f (t, ξ) dξ = |ξ|2 f0 (ξ) dξ, ∀t.
R3 R3

Remark: Although the weak compactness argument gives only the

energy inequality, multiplying (2.1) by |ξ|2 and taking integration
recovers the conservation of energy.

20/43
Proof of Povzner inequality:

For positive numbers a < b and any s ≥ 2, γ ∈ [0, 1]:

(a + b)s ≤ as + b s + (2s − 2)b s−1 a ≤ as + b s + (2s − 2)b s−γ aγ ,

and hence for positive a and b:

as + b s ≤ (a + b)s ≤ as + b s + C (b s−γ aγ + as−γ b γ ).

As a consequence, for γ ∈ [0, 2]:

s s s s
(1 + |ξ ′ |2 ) 2 + (1 + |ξ∗′ |2 ) 2 − (1 + |ξ|2 ) 2 − (1 + |ξ∗ |2 ) 2
h γ s−γ s−γ γ
i
≤ Cs (1 + |ξ|2 ) 2 (1 + |ξ∗ |2 ) 2 + (1 + |ξ|2 ) 2 (1 + |ξ∗ |2 ) 2 . (2.2)

21/43
 Z
s
∴ (1 + |ξ|2 ) 2 Q(f , g ) dξ
Z
1
= f (ξ∗ )g (ξ)(ξ − ξ∗ ) · ω
2 (ξ−ξ∗ )·ω≥0
s s s s 
(1 + |ξ ′ |2 ) 2 + (1 + |ξ∗′ |2 ) 2 − (1 + |ξ|2 ) 2 − (1 + |ξ∗ |2 ) 2 dξ∗ dω dξ.


By using the obvious inequality:

1 1
|ξ − ξ∗ | ≤ (1 + |ξ|2 ) 2 (1 + |ξ∗ |2 ) 2

and choosing γ = 1 in (2.2), we easily get

Z
s
(1 + |ξ|2 ) 2 Q(f , g ) dξ ≤ C (s){∥f ∥1,s ∥g ∥1,2 + ∥g ∥1,s ∥f ∥1,2 } .

22/43
2.3 Uniqueness and Regularity

23/43
 Theorem 3.1
Let 0 ≤ f0 ∈ L14 and f0 ln f0 < ∞. Then there exists a unique
R

0 ≤ f ∈ C 1 ([0, T ]; L12 ) satisfying

(
∂t f = Q(f , f )
.
f (0, ·) = f0

Remark:
• Due to Povzner, f0 ∈ L13 is enough to give uniqueness.
• For ∂t f ∈ C ([0, T ]; L12 ), we need f0 ∈ L14 .

24/43
Proof: Note

∥Q+ (f , g )∥1,s ≤ C (s){∥f ∥1,s+1 ∥g ∥1 + ∥g ∥1,s+1 ∥f ∥1 },

∥fR(g ) + gR(f )∥1,s ≤ C (s){∥f ∥1,s+1 ∥g ∥1 + ∥g ∥1,s+1 ∥f ∥1 }.

So that
Z t
∥f (t) − f (r )∥1,s ≤ ∥Q(f (τ ), f (τ ))∥1,s dτ ≤ C (s, T )|t − r |
|r {z }
if f0 ∈L1s+1

where we need s = 3 later.

25/43
Furthermore,
f (t + h) − f (t)
− Q(f (t), f (t))
h 1,2
Z t+h
≤h−1 ∥Q(f (τ ) + f (t), f (τ ) − f (t))∥1,2 dτ
t
Z t+h
−1
≤Ch ∥f (τ ) + f (t)∥1,3 ∥f (τ ) − f (t)∥1,3 dτ
t
≤Ch.

Therefore, ∂t f = Q(f , f ) ∈ L12 and

f ∈ C 1 ([0, T ]; L12 ).

26/43
Uniqueness: Assume there are two solutions f , g ∈ C 1 ([0, T ]; L12 )
with f |t=0 = f0 = g |t=0 :
d
∥f (t) − g (t)∥1,2
dt
Z
= (1 + |ξ|2 ) sgn(f (t) − g (t))Q(f (t) + g (t), f (t) − g (t)) dξ
ZZZ
= (1 + |ξ|2 ) sgn(f − g ) (ξ − ξ∗ ) · ω
 ′
(f∗ + g∗′ )(f ′ − g ′ ) + (f ′ + g ′ )(f∗′ − g∗′ )
− (f∗ + g∗ )(f − g ) − (f + g )(f∗ − g∗ ) dω dξ∗ dξ
ZZZ
≤ (1 + |ξ|2 ) (ξ − ξ∗ ) · ω
 ′
(f∗ + g∗′ )|f ′ − g ′ | + (f ′ + g ′ )|f∗′ − g∗′ |
−(f∗ + g∗ )|f − g | + (f + g )|f∗ − g∗ | dω dξ∗ dξ (3.1)
≤C ∥f (t) − g (t)∥1,2 ∥f (t) + g (t)∥1,3 . (3.2)

Exercise 3.2
Derive the inequality from (3.1) to (3.2).

27/43
2.4 Pointwise bounds

28/43
Theorem 4.1 (Main theorem)
Assume that the initial data (≥ 0) is bounded and has finite mass, energy
and entropy, so that
Z Z
∥f0 ∥L∞ < ∞, (1 + |ξ|2 )f0 (ξ) dξ < ∞, H0 = f0 (ξ) ln f0 (ξ) dξ < ∞.

Moreover, assume that

Z
sup f (ξ) dξ < ∞,
E E

the supremum being taken over all two-dimensional planes E ⊂ R3 . Then

the solution remains uniformly bounded, i.e.,

∥f (·, t)∥L∞ ≤ C

for all t ≥ 0, where C depends only on physical quantities of initial data.

29/43
For gain term only, formally: (1st preparation, Carleman’s
representation)
 
ξ − ξ∗
Z Z
Q+ (f , g )(ξ) = dξ∗ dωB |ξ − ξ∗ |, · ω f (ξ ′ )g (ξ∗′ )
R3 S2 |ξ − ξ∗ |
We use the notations

ξ∗ − ξ = z, ξ ′ = ξ − (ξ − ξ∗ ) · ω ω = ξ + z · ω ω = ξ + z∥ ,
ξ∗′ = ξ∗ + (ξ − ξ∗ ) · ω ω = ξ + z − z · ω ω = ξ + z⊥

where

z∥ = z · ω ω, z⊥ = z − z∥ = z − z · ω ω (z⊥ · ω = z⊥ · z∥ = 0).

Then
Z Z  
z
Q+ (f , g )(ξ) = B |z|, · ω f (ξ + z∥ )g (ξ + z⊥ ) dω dz
R3 S2 |z|
where
2
dz dω = (2 d|z∥ | dz⊥ ) dω = (2 d|z∥ | dω) dz⊥ = dz∥ dz⊥ .
|z∥ |2

30/43
Make a change of variable ξ + z∥ = η, z = z∥ + z⊥ = η − ξ + z⊥ :
|z∥ |
 
dz∥
Z Z
Q+ (f , g )(ξ) = 2 dz⊥ B |z|, f (ξ + z∥ )g (ξ + z⊥ )
|z∥ |2 |z|
R3 z⊥ ·z∥ =0
Z Z  
dη |η − ξ|
=2 dz⊥ B |η − ξ + z⊥ |, f (η)g (ξ + z⊥ )
|η − ξ|2 |η − ξ + z⊥ |
R3η z⊥ ·(η−ξ)=0
Z Z
dη ′ ′
=2 dz⊥ B (| · · · |, | · · · |) f (η)g (z⊥ )
|η − ξ|2
R3η ′ −ξ)·(η−ξ)=0
(z⊥

η−ξ ′
z⊥
η

ξ ′
z⊥ −ξ
O

Ref.: Glassey’s book, The Cauchy Problem in Kinetic Theory, page 41-44.

31/43
For instance, hard-sphere model:
 
|ξ − ξ∗ | |ξ − ξ∗ |
B |ξ − ξ∗ |, · ω = |(ξ − ξ∗ ) · ω| = |ξ − ξ∗ | · ·ω
|ξ − ξ∗ | |ξ − ξ∗ |
= |z∥ |
Z Z
f (η)
∴ Q+ (f , g )(ξ) = 2 dη g (z) dz .
|η − ξ|2
R3 (z−ξ)·(η−ξ)=0
| {z }
dim=2 integration over the plane

η−ξ

32/43
2nd preparation:

Lemma 4.2
Suppose that f0 ≥ 0 with finite mass, energy and entropy. Then

R(f (t)) ≥ C (f0 )(1 + |ξ|),

where C (f0 ) depends only on ∥f0 ∥1,2 and H(f0 ).

Recall that:
• f (t, ξ) ≥ 0
Z Z
• f (t, ξ) dξ = f0 (ξ) dξ
Z Z
• |ξ|2 f (t, ξ) dξ = |ξ|2 f0 (ξ) dξ

33/43
Proof:
Z
R(f )(ξ) ≥ |ξ − ξ∗ |f (ξ∗ ) dξ∗
Z
≥ |ξ − ξ∗ |f (ξ∗ ) dξ∗
|ξ−ξ∗ |>A
Z Z !
≥A f (ξ∗ ) dξ∗ − f (ξ∗ ) dξ∗
|ξ−ξ∗ |≤A
Z Z !
1
≥ A ∥f ∥1 − f ln f χf >B − f χf ≤B
ln B |ξ−ξ∗ |≤A |ξ−ξ∗ |≤A
 
H(f ) 3
≥ A ∥f ∥1 − − BπA .
ln B

Therefore, R(f )(ξ) ≥ C .

34/43
 |ξ| ∥f ∥1,2
For > , we have
2 ∥f ∥1
Z
1
R(f )(ξ) ≥ |ξ − ξ∗ |f (ξ∗ ) dξ∗ ≥ |ξ|∥f ∥1 − ∥f ∥1,2 ≥ |ξ|∥f ∥1 .
2
Consequently,

R(f )(ξ) ≥ C (f0 )(1 + |ξ|), ∀ξ ∈ R3 .

35/43
 ∂t f = Q(f , f ) = Q+ (f , f ) − Q− (f , f )
Z Z
f (η)
= dη f (z) dz
|η − ξ|2
R3 (z−ξ)·(η−ξ)=0
ZZ
− f (t, ξ) |(ξ − ξ∗ ) · ω|f (ξ∗ ) dξ∗ dω
R3 ×S 2
| {z }
=R(f )(ξ)
Z
where R(f )(ξ) = 2π |ξ − ξ∗ |f (ξ∗ ) dξ∗ for hard spheres.
R3
Denote
Z
f (η) dη
J(ξ) =
3 |η − ξ|
ZR
I (E ) = f (z) dz, E : a 2D plane of R3
E

Then ∂t f ≤ J(ξ) sup I (E ) − C (f0 )f (t, ξ).

E

36/43
Claim: Jb = sup J(ξ) < +∞, Ib = sup I (E ) < +∞.
ξ E

Therefore:

∂t f (t, ξ) ≤ Jb· Ib − C (f0 )f (t, ξ), ∀t ≥ 0

( )
Jb· Ib
=⇒ f (t, ξ) ≤ max , ∥f0 ∥L∞
C (f0 )

37/43
Proof of claim:
Z Z
d d
I (E ) = f (z) dz = ∂t f (t, z) dz
dt dt
Z E E
Z
= Q+ (f , f )(t, z) dz − Q− (f , f )(t, z) dz
ZE Z E
Z
f (η)
= dz dη f (y ) dy (∗)
E R 3 |η − z|
(y −z)·(η−z)=0
Z
− f (t, z) · 2π R(f )(t, z) dz.
E | {z }
≥C (f0 )
Z Z
Here (∗) ≤ π f (η) dη f (y ) dy .
R3 R3

38/43
Informal view:
ZZ Z
f (η)f (y )
(∗) = 1(y −z)·(η−z)=0 dz dη dy
R3 ×R3 |η − y | E
ZZ
f (η)f (y ) 1
= lim |Sηy ∩ Eδ | dη dy .
R3 ×R3 |η − y | δ→0 2δ
Refer to the figure and
1 1
|Sηy ∩ Eδ | ≤ · π|η − y | · 2δ = π|η − y |.
2δ 2δ
ZZ
∴ (∗) ≤ π f (η)f (y ) dη dy .
R3 ×R3

2δ

Eδ
η y

z
E

39/43
Exercise 4.3
Give a rigorous proof based on the dual argugment.

d
∴ I (E ) ≤ π − C (f0 )I (E ), ∀ a plane E ⊂ R3 .
dt
 
π
∴ sup I (E )(t) ≤ max , sup I (E )(0) .
E C (f0 ) E

40/43
For J:
b
Z Z
d ∂t f (t, η) Q+ (f , f )(η) − Q− (f , f )(η)
J(ξ) = dη = dη
dt |η − ξ| |η − ξ|
ZZ Z !
1 1
≤ dσ(η) f (ξ1 )f (ξ2 ) dξ1 dξ2
|ξ1 − ξ2 | Sξ1 ξ2 |η − ξ|
Z Z
f (η)
− 2π dη · |ξ∗ − ξ|f (ξ∗ ) dξ∗
|η − ξ|
ZZ Z
1 dσ(η)
≤ f (ξ1 )f (ξ2 ) dξ1 dξ2 − J(ξ)C (f0 )
|ξ1 − ξ2 | Sξ ξ |η − ξ0 |
1 2

= 2π − C (f0 )J(ξ).
 
2π ξ1 + ξ2
∴ J(ξ)(t) ≤ max , J(ξ)(0) . Here ξ0 = .
C (f0 ) 2

ξ′
Sξ1 ξ2
ξ0
ξ1 ξ2

41/43
Exercise 4.4
Show that Z
1
sup dη = 2π|ξ1 − ξ2 |,
ξ Sξ1 ξ2 |η − ξ|
ξ1 + ξ2
where Sξ1 ξ2 = {η ∈ R3 : |η − ξ0 | = 21 |ξ1 − ξ2 |} with ξ0 = .
2

42/43
End of Lecture 2

43/43