Summer Course at PKU (July 2020)

Introduction to Kinetic Theory – Lecture Notes∗

Jingwei Hu

July 5, 2020

Contents

1 Introduction 2

2 The Boltzmann equation for hard spheres 2

2.1 Heuristic derivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
2.2 Formal derivation from the Liouville equation (BBGKY hierarchy) . . . 5

3 The Boltzmann equation for general (repulsive) intermolecular poten-

tials 9

4 Basic properties of the Boltzmann equation 12

4.1 Collision invariants and local conservation laws . . . . . . . . . . . . . . 13
4.2 Boltzmann’s H-theorem and Maxwellian . . . . . . . . . . . . . . . . . . 15
4.3 Boundary condition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16

∗
These lecture notes are for course use only and will be kept updating. Please do not copy or
distribute for any other purposes. Please send corrections or comments to jingweihu@purdue.edu.

1
1 Introduction
These lecture notes are a collection of materials related to various aspects of modern
kinetic theory, including physical derivation, mathematical theory, and numerical meth-
ods. The main focus is on the Boltzmann-like collisional kinetic equations and their
numerical approximations. To begin with, let us take a look at Figure 1 to understand
the role of kinetic theory in multiscale modeling hierarchy.

Figure 1: Role of kinetic theory in multiscale modeling hierarchy.

2 The Boltzmann equation for hard spheres

Proposed by Ludwig Boltzmann in 1872, the Boltzmann equation is one of the fun-
damental equations in kinetic theory. It describes the non-equilibrium dynamics of a gas
or system comprised of a large number of particles. In this very first part of the course,
we derive the Boltzmann equation for hard sphere molecules. For better understanding,
we start with a heuristic derivation and then discuss a more formal derivation from the
Liouville equation.

2.1 Heuristic derivation

This part of the presentation mainly follows [2, Chapter 1.2].
Let us start with the function P (1) (t, x1 , v1 ), which is the one-particle probability
density function (PDF). P (1) dx1 dv1 gives the probability of finding one fixed particle
(say, the one labeled by 1) in an infinitesimal volume dx1 dv1 centered at the point
(x1 , v1 ) of the phase space, where x1 ∈ R3 is the position and v1 ∈ R3 is the particle
velocity.

2
 When two particles (say, particles 1 and 2) collide, momentum and energy must be
conserved (mass is always conserved). Let v1 , v2 be the velocities before a collision and
(v10 , v20 ) the velocities after a collision. From

v1 + v2 = v10 + v20 , |v1 |2 + |v2 |2 = |v10 |2 + |v20 |2 , (2.1)

one can derive that

v10 = v1 − [(v1 − v2 ) · ω]ω, v20 = v2 + [(v1 − v2 ) · ω]ω, (2.2)

where ω is the impact direction (the unit vector connecting the centers of particles 1
and 2). Note from (2.2) that

v20 − v10 = (v2 − v1 ) − 2[(v2 − v1 ) · ω]ω, (2.3)

i.e., the relative velocity undergoes a specular reflection at the impact (see Figure 2).

v20 v10

x2 v2 v1
radius 2r
!
x1

Figure 2: Illustration of particle collisions.

In the absence of collisions and external forces, P (1) would remain unchanged along
the trajectory of particle 1. That is, P (1) satisfies

∂P (1)
+ v1 · ∇x1 P (1) = 0. (2.4)
∂t
Now with collisions, one would expect

∂P (1)
+ v1 · ∇x1 P (1) = G − L, (2.5)
∂t
where L dx1 dv1 dt gives the probability of finding particles with position between x1
and x1 + dx1 and velocity between v1 and v1 + dv1 that disappear from these ranges of

3
values because of a collision in the time interval between t and t + dt (L is often called
the loss term of the collision operator), and G dx1 dv1 dt gives the analogous probability
of finding particles entering the same range in the same time interval (G is often called
the gain term of the collision operator). To count these probabilities, imaging particle
1 as a sphere at rest and endowed with twice the actual radius r and the other particles
being the point masses with velocity v2 − v1 (see Figure 2). Fixing particle 1, there are
N − 1 particles (assume there are a total of N particles) that will collide with it, and
they are to be found in the cylinder of height |(v2 − v1 ) · ω| dt and base area (2r)2 dω.
Then
Z Z
L dx1 dv1 dt = (N − 1) P (2) (t, x1 , v1 , x1 + 2rω,v2 )|(v2 − v1 ) · ω| dt
3 2
R S− (2.6)
× (2r)2 dω dv2 dx1 dv1 ,

where P (2) is the two-particle PDF, and S− 2 is the hemisphere corresponding to (v −

2
v1 ) · ω < 0. Therefore,
Z Z
2
L = (N − 1)(2r) P (2) (t, x1 , v1 , x1 + 2rω, v2 )|(v2 − v1 ) · ω| dω dv2 . (2.7)
R3 2
S−

Similarly,
Z Z
2
G = (N − 1)(2r) P (2) (t, x1 , v1 , x1 + 2rω, v2 )|(v2 − v1 ) · ω| dω dv2 , (2.8)
R3 2
S+

2 is the hemisphere corresponding to (v − v ) · ω > 0.

where S+ 2 1
Now we make two crucial assumptions:

• Assume N → ∞, r → 0, but N r2 is finite. This is the so-called Boltzmann-Grad

limit.

• Assume P (2) (t, x1 , v1 , x2 , v2 ) = P (1) (t, x1 , v1 )P (1) (t, x2 , v2 ) for two particles that
are about to collide. This is the molecular chaos assumption.

Then L becomes
Z Z
2
L = N (2r) P (2) (t, x1 , v1 , x1 , v2 )|(v2 − v1 ) · ω| dω dv2
R3 2
S−
Z Z (2.9)
2 (1) (1)
= N (2r) P (t, x1 , v1 )P (t, x1 , v2 )|(v2 − v1 ) · ω| dω dv2 ,
R3 2
S−

where we used the assumption 1 in the first equality and assumption 2 in the second

4
equality. For G, we have
Z Z
G = (N − 1)(2r)2 P (2) (t, x1 , v10 , x1 + 2rω, v20 )|(v2 − v1 ) · ω| dω dv2
R3 2
S+
Z Z
= N (2r)2 P (1) (t, x1 , v10 )P (1) (t, x1 , v20 )|(v2 − v1 ) · ω| dω dv2 (2.10)
R3 2
S+
Z Z
= N (2r) 2
P (1) (t, x1 , v10 )P (1) (t, x1 , v20 )|(v2 − v1 ) · ω| dω dv2 ,
R3 2
S−

where the first equality is because P (2) is continuous at a collision, the second equality is
obtained for the same reason as above for L (since (v2 −v1 )·ω > 0 implies (v20 −v10 )·ω < 0),
and the third one is a simple change of variable ω → −ω.
Putting together G and L, we have

∂P (1)
Z Z
+v1 · ∇x1 P (1) = N (2r)2 |(v2 − v1 ) · ω|
∂t R3 2
S− (2.11)
× [P (1)
(t, x1 , v10 )P (1) (t, x1 , v20 ) −P (1)
(t, x1 , v1 )P (1)
(t, x1 , v2 )] dω dv2 .

In this course we will often consider the one-particle number distribution function f
(i.e., f = N P (1) ), then f satisfies (changing x1 → x, v1 → v, v2 → v∗ , ω → −ω)

∂f
Z Z
+ v · ∇x f = (2r) 2
|(v − v∗ ) · ω|[f 0 f∗0 − f f∗ ] dω dv∗ , (2.12)
∂t R3 (v−v∗ )·ω<0

where f , f∗ , f 0 , f∗0 are short hand notations for f (t, x, v), f (t, x, v∗ ), f (t, x, v 0 ), f (t, x, v∗0 ),
and
v 0 = v − [(v − v∗ ) · ω]ω, v∗0 = v∗ + [(v − v∗ ) · ω]ω. (2.13)

Equation (2.12) is the Boltzmann equation for hard spheres.

It is often convenient to integrate ω over the whole sphere S 2 rather than hemisphere,
which yields
∂f
Z Z
+ v · ∇x f = 2r2 |(v − v∗ ) · ω|[f 0 f∗0 − f f∗ ] dω dv∗ . (2.14)
∂t R 3 S 2

2.2 Formal derivation from the Liouville equation (BBGKY hierarchy)

In this section, we give a formal derivation of the Boltzmann equation starting from
the Liouville equation. The rigorous derivation was an open and challenging problem
for a long time. In 1973, Lanford showed that, although for a very short time, the
Boltzmann equation can be derived from the mechanical systems.
This part of the presentation mainly follows [1, Chapter 3.2], where one can also find
the rigorous treatise.

5
 Consider N hard spheres of radius r. Let xi , vi denote the position and velocity of
particle i, then the state of the system is given by

(x1 , v1 , . . . , xN , vN ) ∈ ΩN × R3N = Λ,

where
ΩN = {(x1 , . . . , xN ) | |xi − xj | > 2r, i 6= j},

∂Λ = {(x1 , v1 , . . . , xN , vN ) | |xi − xj | = 2r, i 6= j},

since the particles cannot overlap.

Let P (N ) (t, x1 , v1 , . . . , xN , vN ) be the N -particle PDF, then P (N ) satisfies the Liou-
ville equation
N
∂P (N ) X
+ vi · ∇xi P (N ) = 0. (2.15)
∂t
i=1

Define the s-particle PDF as

Z
(s)
P (t, x1 , v1 , . . . , xs , vs ) = P (N ) dxs+1 dvs+1 . . . dxN dvN , (2.16)

then integrating (2.15) one obtains

∂P (s)
+ I1 + I2 = 0, (2.17)
∂t
with
s Z
X
I1 = vi · ∇xi P (N ) dxs+1 dvs+1 . . . dxN dvN ,
i=1
N Z (2.18)
X
(N )
I2 = vi · ∇xi P dxs+1 dvs+1 . . . dxN dvN .
i=s+1

For I2 , applying the divergence theorem (one can refer to Figure 2 again but with
(x1 , v1 ) replaced by (xi , vi ), (x2 , v2 ) by (xj , vj ), and ω by ωij ), one has
s
X N
X Z
2
I2 = (2r) vi · ωij P (N ) (t, x1 , v1 , . . . , xi−1 , vi−1 , xj − 2rωij , vi , . . . , xN , vN )
j=1 i=s+1

× dωij dxs+1 . . . dxi−1 dxi+1 . . . dxN dvs+1 . . . dvN

N
X N
X Z
+ (2r)2 vi · ωij P (N ) (t, x1 , v1 , . . . , xi−1 , vi−1 , xj − 2rωij , vi , . . . , xN , vN )
j=s+1, i=s+1
j6=i

× dωij dxs+1 . . . dxi−1 dxi+1 . . . dxN dvs+1 . . . dvN .

(2.19)

6
The second sum in the above equation is completely zero by the Liouville theorem (it is
the integral of N (N ) relative to the dynamics of the last N − s particles).
P
i=s+1 vi · ∇xi P
Using the symmetry of P (N ) , the first term can be further reduced to
s Z
X
I2 = (N − s)(2r)2 vs+1 · ωs+1,j P (N ) (t, x1 , v1 , . . . , xs , vs , xj − 2rωs+1,j , vs+1 , . . . , xN , vN )
j=1

× dωs+1,j dxs+2 . . . dxN dvs+1 . . . dvN

s
X Z
= (N − s)(2r)2 vs+1 · ωs+1,j P (s+1) (t, x1 , v1 , . . . , xs , vs , xj − 2rωs+1,j , vs+1 ) dωs+1,j dvs+1 .
j=1
(2.20)

For I1 , it can be shown that (see below)

s
X s Z
X
I1 = vi · ∇xi P (s) − (N − s)(2r)2 vj · ωs+1,j
i=1 j=1
(s+1)
×P (t, x1 , v1 , . . . , xs , vs , xj − 2rωs+1,j , vs+1 ) dωs+1,j dvs+1 ,
(2.21)

where the second term is due to the integration domain depends on xi .

Putting together I1 and I2 , (2.17) becomes
s s
∂P (s) X X Z
+ vi · ∇xi P (s) = (N − s)(2r)2 (vj − vs+1 ) · ωs+1,j
∂t (2.22)
i=1 j=1
(s+1)
×P (t, x1 , v1 , . . . , xs , vs , xj − 2rωs+1,j , vs+1 ) dωs+1,j dvs+1 .

This is the so-called BBGKY hierarchy for hard spheres (the equation of P s depends
on P s+1 ), named after Bogoliubov, Born, Green, Kirkwood, and Yvon. In particular,
taking s = 1 in (2.22) gives

∂P (1)
Z
(1) 2
+ v1 · ∇x1 P = (N − 1)(2r) (v1 − v2 ) · ω21 P (2) (t, x1 , v1 , x1 − 2rω21 , v2 ) dω21 dv2
∂t
Z
2
= (N − 1)(2r) (v2 − v1 ) · ω12 P (2) (t, x1 , v1 , x1 + 2rω12 , v2 ) dω12 dv2
Z
2
= (N − 1)(2r) |(v2 − v1 ) · ω12 |P (2) (t, x1 , v1 , x1 + 2rω12 , v2 ) dω12 dv2
(v2 −v1 )·ω12 >0
Z
2
− (N − 1)(2r) |(v2 − v1 ) · ω12 |P (2) (t, x1 , v1 , x1 + 2rω12 , v2 ) dω12 dv2 .
(v2 −v1 )·ω12 <0
(2.23)

This is the same as equation (2.5) with (2.8) and (2.7) derived in the previous section.
The rest of the derivation is the same. That is, the first BBGKY hierarchy yields the
Boltzmann equation.

7
 It remains to prove (2.21). Note that in the two-particle case,
Z
v 1 · ∇ x1 P (2) (t, x1 , v1 , x2 , v2 ) dx2 dv2
|x −x |>2r
"Z 1 2 #
1
Z
(2) (2)
= lim P (t, x1 + tv1 , v1 , x2 , v2 ) dx2 dv2 − P (t, x1 , v1 , x2 , v2 ) dx2 dv2
t→0 t |x1 +tv1 −x2 |>2r |x1 −x2 |>2r
"Z #
1
Z
(2) (2)
= lim P (t, x1 + tv1 , v1 , x2 + tv1 , v2 ) dx2 dv2 − P (t, x1 , v1 , x2 , v2 ) dx2 dv2
t→0 t |x1 −x2 |>2r |x1 −x2 |>2r
"Z #
1
Z
= lim P (2) (t, x1 + tv1 , v1 , x2 + tv1 , v2 ) dx2 dv2 − P (2) (t, x1 + tv1 , v1 , x2 , v2 ) dx2 dv2
t→0 t |x1 −x2 |>2r |x1 −x2 |>2r
"Z #
1
Z
+ P (2) (t, x1 + tv1 , v1 , x2 , v2 ) dx2 dv2 − P (2) (t, x1 , v1 , x2 , v2 ) dx2 dv2
t |x1 −x2 |>2r |x1 −x2 |>2r
Z
= (v1 · ∇x2 + v1 · ∇x1 )P (2) (t, x1 , v1 , x2 , v2 ) dx2 dv2
|x1 −x2 |>2r
Z Z
2
=(2r) v1 · ω21 P (t, x1 , v1 , x1 − 2rω21 , v2 ) dω21 dv2 + v1 · ∇x1 P (2) (t, x1 , v1 , x2 , v2 ) dx2 dv2 .
(2)

(2.24)

Analogously,
s
X Z
vi · ∇xi P (N ) dxs+1 dvs+1 . . . dxN dvN
i=1
s Z N
!
X X
= v i · ∇ xk + v i · ∇ xi P (N ) dxs+1 dvs+1 . . . dxN dvN
i=1 k=s+1
Xs  Z
= (N − s)(2r)2 vi · ωs+1,i P (N ) (t, x1 , v1 , . . . , xs , vs , xi − 2rωs+1,i , vs+1 , . . . , xN , vN )
i=1
Z 
(N )
× dωs+1,i dvs+1 dxs+2 . . . dxN dvN + v i · ∇ xi P dxs+1 dvs+1 . . . dxN dvN .

(2.25)

This implies
s
X s
X Z
(s) 2
v i · ∇ xi P − (N − s)(2r) vi · ωs+1,i P (s+1) (t, x1 , v1 , . . . , xs , vs , xi − 2rωs+1,i , vs+1 ) dωs+1,i dvs+1
i=1 i=1
s Z
X
= vi · ∇xi P (N ) dxs+1 dvs+1 . . . dxN dvN ,
i=1
(2.26)

which is (2.21).

8
3 The Boltzmann equation for general (repulsive) inter-
molecular potentials
We have seen that the Boltzmann equation for hard spheres is given by (2.12) (2.13).
Another way of viewing (2.13) is to note that (v 0 , v∗0 ) must lie on the same sphere
determined by (v, v∗ ) (see Figure 3).

v⇤ v0
!

✓
x

v⇤0 y v

Figure 3: Velocities during a classical elastic collision. ω is the unit vector along the
direction of v 0 − v.

To include general interactions other than hard spheres, one can generalize the Boltz-
mann equation (2.12) as
∂f
Z Z
+ v · ∇x f = B(v − v∗ , ω)[f 0 f∗0 − f f∗ ] dω dv∗ . (3.1)
∂t R3 (v−v∗ )·ω<0

Here the function B is called the collision kernel and is defined as

b db
B(v − v∗ , ω) = |V | , (3.2)
sin θ dθ
where V = v − v∗ is the relative velocity, 0 < θ < π/2 is the angle between −V and
ω, b = b(|V |, θ) is the impact parameter (see Figure 4). From the classical scattering
theory, we know for hard spheres with radius r,

b = 2r sin θ. (3.3)

Therefore,
B = 4r2 cos θ|V | = 4r2 |V · ω|, (3.4)
i.e., it reduces to the hard sphere case derived before.
Corresponding to the integration over the whole sphere (2.14), we define
∂f
Z Z
+ v · ∇x f = Bω (|V |, | cos θ|)[f 0 f∗0 − f f∗ ] dω dv∗ , (3.5)
∂t R3 S 2

9
 V0
!

V
✓
b
scattering center

Figure 4: Illustration of particle scattering in a repulsive potential field (notation con-

sistent with Figure 3). b is the impact parameter, χ is the scattering angle, V = v − v∗
and V 0 = v 0 − v∗0 .

where
1 b db
Bω (|V |, | cos θ|) = |V | , (3.6)
2 sin θ dθ
and
v 0 = v − [(v − v∗ ) · ω]ω, v∗0 = v∗ + [(v − v∗ ) · ω]ω. (3.7)
This is what we are going to refer to as the ω-representation.
Another parametrization of the Boltzmann equation that uses the unit vector σ along
0
V reads
∂f
Z Z
+ v · ∇x f = Bσ (|V |, cos χ)[f 0 f∗0 − f f∗ ] dσ dv∗ , (3.8)
∂t R3 S 2

where
b db
Bσ (|V |, cos χ) = |V |Σ(|V |, χ), Σ(|V |, χ) = , (3.9)
sin χ dχ
Σ(|V |, cos χ) is the differential cross section with 0 < χ < π, and
v + v∗ |v − v∗ | v + v∗ |v − v∗ |
v0 = + σ, v∗0 = − σ. (3.10)
2 2 2 2
This is what we are going to refer to as the σ-representation. In particular, the hard
sphere collision kernel under this representation reads Bσ (|V |, cos χ) = r2 |V |.
Now for a general (repulsive) intermolecular potential φ(r) (r is the distance between
two particles), b is related to χ implicitly as follows
Z r0
dr
χ=π−2 h i1/2 , (3.11)
0 2 4φ(br−1 )
1 − r − m|V |2

where m is the single particle mass, and r0 is the positive root to the equation
4φ(br−1 )
1 − r2 − = 0. (3.12)
m|V |2

10
 Let’s take a close look of the inverse power law potential
K
φ(r) = , 2 < s ≤ ∞, K is some positive constant. (3.13)
rs−1
Then (3.11) becomes
Z r0 Z r0
dr dr
χ=π−2 h i1/2 = π − 2   s−1 1/2 , (3.14)
0 4Krs−1 0
1 − r2 − m|V |2 bs−1 1 − r − βr
2

  1
m|V |2 s−1
with β := 4K b. Thus the collision kernel Bσ is

  2   2
b db 4K s−1 β dβ
β dβ 4K s−1 s−5
Bσ = |V | = |V | = . |V | s−1
sin χ dχ m|V |2 sin χ dχ
sin χ dχ m
(3.15)
Since β can be solved implicitly from (3.14) to yield β = β(χ), (3.15) implies that
  2
s−5 4K s−1 β dβ
Bσ = bs (cos χ)|V | s−1 , bs (cos χ) = . (3.16)
m sin χ dχ

When s = 5, Bσ is a function of χ only which will lead to many simplifications

(usually referred to as Maxwell molecules). The hard sphere kernel can be considered
as a special case of (3.16) when s = ∞. Furthermore, (3.16) shows that the velocity
s−5
dependence in the collision kernel behaves like |V | s−1 +2 . When |V | is small, this is
integrable if s−5
s−1 + 2 > −1, i.e. s > 2. Note that s = 2 corresponds to the Coulomb po-
tential. Therefore, the Boltzmann equation should not be used to describe the Coulomb
interaction1 .
Based on (3.16), it is common in the kinetic literature to distinguish the kernel by
its velocity dependence:

Bσ = bλ (cos χ)|V |λ , −3 < λ ≤ 1, (3.17)

where λ > 0 is called the hard potential, λ < 0 is the soft potential, and λ = 0 is the
Maxwell kernel.
Let’s analyze a bit the asymptotic behavior of χ w.r.t. β. When β  1, (3.14) can
be approximated as
β 1
dr du
Z Z
χ≈π−2  s−1 1/2 = π − 2β = π − 2βA(s), (3.18)
(1 − us−1 )1/2

0 r 0
1− β

1
In this limit, one should consider the so-called Landau operator which is a diffusive type operator.
We will come back to this later in the course.

11
 dβ
so when β → 0, χ → π, and β dχ is well behaved. When β  1, (3.14) can be
approximated as
 s−1
Z 1 1 + 12 (1 − r2 )−1 r
β A(s)
χ≈π−2 dr = , (3.19)
0 (1 − r2 )1/2 β s−1

so when β → ∞, χ → 0, and
dβ 2
β ∼ χ− s−1 −1 , (3.20)
dχ
i.e., the collision kernel contains a nonintegrable singularity at χ = 0 for all s > 2 (except
s = ∞). This can be avoided either by cutting off β, so that the potential φ is zero for
large β, or by the less physical, but mathematically more tractable, method of directly
cutting off χ near 0, that is, eliminating grazing collisions from the collision term. This
is the so-called Grad’s angular cut-off assumption.

4 Basic properties of the Boltzmann equation

In this section, we derive some basic properties of the Boltzmann equation, which
we rewrite here for clarity2
∂f
+ v · ∇x f = Q(f, f ), x ∈ Ω ⊂ Rd , v ∈ Rd , d ≥ 2. (4.1)
∂t
Q(f, f ) is the so-called collision operator, which is a quadratic integral operator acting
only in the velocity space. In fact, it is convenient to introduce a bilinear form of Q as
(in both σ- and ω- representations):
Z Z
Q(g, f )(v) = Bσ (|v − v∗ |, cos χ)[g∗0 f 0 − g∗ f ] dσ dv∗
R d S d−1
Z Z (4.2)
0 0
= Bω (|v − v∗ |, | cos θ|)[g∗ f − g∗ f ] dω dv∗ ,
Rd S d−1

where
v + v∗ |v − v∗ | v + v∗ |v − v∗ |
v0 = + σ, v∗0 = − σ, cos χ = σ · (v\
− v∗ ), (4.3)
2 2 2 2

v 0 = v − [(v − v∗ ) · ω]ω, v∗0 = v∗ + [(v − v∗ ) · ω]ω, | cos θ| = |ω · (v\

− v∗ )|. (4.4)

Note that the physically relevant case is the dimension d = 3 as we considered in previous
sections. Here we assume d ≥ 2 for mathematical generality.
We first derive a very important formula of the collision operator using the ω-
representation.
2
We have deliberately ignored the forcing term like F (x) · ∇v f in the discussion so far. With this
term, the equation is the so-called Vlasov equation. We will come back to this later in the course.

12
Proposition 4.1. (Boltzmann’s lemma) For any functions ϕ(v), f (v) such that the
integrals make sense, one has
ϕ + ϕ∗ − ϕ0 − ϕ0∗
Z Z Z Z
Q(f, f )ϕ dv = Bω (|v − v∗ |, | cos θ|)[f 0 f∗0 − f f∗ ] dω dv dv∗
Rd Rd Rd S d−1 4
Z Z Z
= Bω (|v − v∗ |, | cos θ|)f f∗ [ϕ0 − ϕ] dω dv dv∗ .
Rd Rd S d−1
(4.5)

Proof.
Z Z Z Z
Q(f, f )ϕ dv = Bω (|v − v∗ |, | cos θ|)[f 0 f∗0 − f f∗ ]ϕ dω dv dv∗
Rd Rd Rd S d−1
Z Z Z
= Bω (|v − v∗ |, | cos θ|)[f 0 f∗0 − f f∗ ]ϕ∗ dω dv dv∗
d d d−1
ZR ZR ZS
ϕ + ϕ∗
= Bω (|v − v∗ |, | cos θ|)[f 0 f∗0 − f f∗ ] dω dv dv∗
Rd Rd S d−1 2
ϕ0 + ϕ0∗
Z Z Z
= Bω (|v − v∗ |, | cos θ|)[f f∗ − f 0 f∗0 ] dω dv dv∗
d d d−1 2
ZR ZR ZS
ϕ + ϕ∗ − ϕ0 − ϕ0∗
= Bω (|v − v∗ |, | cos θ|)[f 0 f∗0 − f f∗ ] dω dv dv∗ ,
Rd Rd S d−1 4
(4.6)

where in the second line we swapped v and v∗ (hence v 0 and v∗0 ); in the fourth line we
changed (v, v∗ ) to (v 0 , v∗0 ) (hence (v 0 , v∗0 ) becomes (v, v∗ )) for a fixed ω and used the fact
that dv dv∗ = dv 0 dv∗0 (the transform has the unit Jacobian).
The second equality in (4.5) is obtained by changing (v, v∗ ) to (v 0 , v∗0 ) only to the
gain term.

4.1 Collision invariants and local conservation laws

Definition 4.2. A collision invariant is a continuous function ϕ = ϕ(v) such that for
each v, v∗ ∈ Rd and ω ∈ S d−1 , one has

ϕ + ϕ∗ = ϕ0 + ϕ0∗ . (4.7)

Since during collisions, mass, momentum and energy are conserved, it is obvious that
functions 1, v, and |v|2 , and any linear combination of them are the collision invariants.
In fact, it can be shown that these are the only collision invariants (this is a non-trivial
result, for proof one may refer to [3, p. 36-42]).
Using the Boltzmann’s lemma, it is clear that

Corollary 4.3.
Z Z Z
Q(f, f ) dv = Q(f, f )v dv = Q(f, f )|v|2 dv = 0. (4.8)
Rd Rd Rd

13
 Using the Corollary 4.3, if we multiply the Boltzmann equation (4.1) by m, mv,
m|v|2 /2, and integrate w.r.t. v, we obtain
 Z Z


 ∂t mf dv + ∇x · mvf dv = 0,
Rd Rd



 Z Z
∂t mvf dv + ∇x · mv ⊗ vf dv = 0, (4.9)

 Rd Rd Z
1 1

 Z
 ∂t m|v|2 f dv + ∇x · mv|v|2 f dv = 0.


Rd 2 Rd 2

These are the local conservation laws (conservation of mass, momentum, and energy).
To better view the connection of f (number distribution function) and macroscopic
quantities such as density, temperature, etc., let us define
1
Z Z
n= f dv, ρ = mn, u = vf dv, (4.10)
Rd n Rd

where n is the number density, ρ is the mass density, and u is the bulk velocity. Further,
with the peculiar velocity
c = v − u, (4.11)

we define
1 1
Z Z Z
T = |c|2 f dv, P= mc ⊗ cf dv, q= mc|c|2 f dv, (4.12)
dRn Rd Rd Rd 2

where T is the temperature, P is the stress tensor, and q is the heat flux vector. R = kB /m
is the gas constant (kB is the Boltzmann’s constant).
Finally, the pressure p is defined as
1
p= tr(P) = ρRT. (4.13)
d
With the above definitions, we can recast the local conservation laws (4.9) using
macroscopic quantities:



 ∂t ρ + ∇x · (ρu) = 0,


∂t (ρu) + ∇x · (ρu ⊗ u + P) = 0, (4.14)



 ∂ E + ∇ · (Eu + Pu + q) = 0,

t x

where E = d2 ρRT + 12 ρu2 is the total energy. The system (4.14) is completely equivalent
to (4.9), hence to the Boltzmann equation. Note that this system is not closed because
P and q, generally speaking, cannot be represented in terms of ρ, u, and T .

14
4.2 Boltzmann’s H-theorem and Maxwellian
Proposition 4.4. (Boltzmann’s H-theorem)
Z
Q(f, f ) ln f dv ≤ 0, (4.15)
Rd

and the equality holds if and only if f = exp(a + b · v + c|v|2 ).

Proof. Taking ϕ = ln f in the Boltzmann’s lemma yields

1
Z Z Z Z
Q(f, f ) ln f dv = − Bω [f 0 f∗0 − f f∗ ][ln(f 0 f∗0 ) − ln(f f∗ )] dω dv dv∗ ≤ 0,
Rd 4 Rd Rd S d−1
(4.16)
where the inequality is due to ln x is a monotonically increasing function, so (x−y)(ln x−
ln y) ≥ 0 for any x, y > 0. The equality holds iff ln f is a collision invariant, i.e.,
f = exp(a + b · v + c|v|2 ), with a, b, c being some constants.

If a function f is of the form exp(a + b · v + c|v|2 ), it can be rewritten as

!
b 2 |b|2
f = exp c v + − +a . (4.17)
2c 4c

2
For f to be integrable, c must be negative. Choosing c0 = −c, b0 = − 2c
b
, a0 = exp(− |b|
4c +
a) gives
f = a0 exp(−c0 |v − b0 |2 ). (4.18)

Using the definition of n, u and T given in the previous section, we can see that3

|v − u|2
 
n
f= exp − := M. (4.19)
(2πRT )d/2 2RT

(4.19) is called the Maxwellian.

Corollary 4.5. The following statements are equivalent.

Z
Q(f, f ) ln f dv = 0 ⇐⇒ f = M ⇐⇒ Q(f, f ) = 0. (4.20)
Rd

Corollary 4.6.
Z Z
∂t f ln f dv + ∇x · vf ln f dv = −D(f ) ≤ 0, (4.21)
Rd Rd

where
1
Z Z Z
D(f ) = Bω [f 0 f∗0 − f f∗ ][ln(f 0 f∗0 ) − ln(f f∗ )] dω dv dv∗ . (4.22)
4 Rd Rd S d−1
R∞ 2
1 R ∞ 2 −αv2
1
3
Note the Gaussian integrals −∞
e−αv dv = π
α
2
, −∞ v e dv = 1
2α
π
α
2
.

15
 If we assume f decays fast enough as x → ∞, or is periodic in x, then (4.21) upon
further integration in x yields
d
H(t) ≤ 0, (4.23)
dt
R R
where H(t) := Rd Rd f ln f dv dx is the H function. (4.23) shows that H is always
non-increasing and reaches its minimum value iff f reaches the Maxwellian (local equi-
librium). This is consistent to the second law of thermodynamics.

4.3 Boundary condition

The commonly used boundary condition for the Boltzmann equation consists of the
following: for a boundary point x ∈ ∂Ω and outward pointing normal n(x),

• Inflow boundary:

|v − u0 |2
 
n0
f (t, x, v) = exp − , v · n < 0, (4.24)
(2πRT0 )d/2 2RT0

where n0 (t, x), u0 (t, x), and T0 (t, x) are the prescribed density, velocity and tem-
perature.

• Maxwell diffusive boundary:

f (t, x, v) = ρw (t, x)fw (t, x, v), (v − uw ) · n < 0, (4.25)

with
|v − uw |2
 
fw (t, x, v) = exp − , (4.26)
2RTw
where uw (t, x) and Tw (t, x) are the wall velocity and temperature. ρw (t, x) is
determined by R
(v−uw )·n≥0 (v − uw ) · nf dv
ρw (t, x) = − R . (4.27)
(v−uw )·n<0 (v − uw ) · nfw dv

• Reflective boundary:

f (t, x, v) = f (t, x, v − 2[(v − uw ) · n]n), (v − uw ) · n < 0. (4.28)

16
References
[1] F. Bouchut, F. Golse, and M. Pulvirenti. Kinetic Equations and Asymptotic Theory.
Series in Applied Mathematics. Gauthier-Villars, 2000.

[2] C. Cercignani. Rarefied Gas Dynamics: From Basic Concepts to Actual Calculations.
Cambridge University Press, Cambridge, 2000.

[3] C. Cercignani, R. Illner, and M. Pulvirenti. The Mathematical Theory of Dilute

Gases. Springer-Verlag, 1994.

17