_
(x, t)
_
R
3
f(x, t, )d, density,
v(x, t)
_
R
3
f(x, t, )d, momentum, v = (v
1
, v
2
, v
3
) uid velocity,
e(x, t)
_
R
3
v
2
2
f(x, t, )d, internal energy,
E(x, t)
_
R
3

2
2
f(x, t, )d = e +
1
2
[v[
2
, total energy,
p
ij
(x, t)
_
R
3
(
i
v
i
)(
j
v
j
)f(x, t, )d, P = (p
ij
)
1i,j3
, stress tensor,
q
i
(x, t)
_
R
3
(
i
v
i
)
1
2
[v [
2
f(x, t, )d, heat ux.
(1.1)
The most important equation for the kinetic theory for the gases is the
Boltzmann equation, [2],
t
f(x, t, ) +
x
f(x, t, ) =
1
k
Q(f, f)(x, t, ), (1.2)
The left hand side of the equation,
t
f +
x
f, is the transport term, measur
ing the time rate of change along the particle path. The Boltzmann equation
says that the change is due to the collision operator Q(f, f), which takes the
form of the binary collision:
Q(g, h)
1
2
_
R
3
S
2
()0
(g()h(
) h()g(
) + g(
)h(
)
+h(
)g(
))B(
, )d
d; (1.3)
_
= [(
) ] ,
+ [(
) ] .
(1.4)
The cross section B(
, ) = (
) . (1.5)
The Maxwellian, thermoequilibrium states have the dening property that
Q(M, M) = 0
2011] SOLVING BOLTZMANN EQUATION, PART I: GREENS FUNCTION 117
and are of the form
M = M
(,v,)
(x, t)
(2R(x, t))
3/2
e
v(x,t)
2
2R(x,t)
,
determined by the macroscopic variables. Here is the temperature dened
by p/ = R. For simplicity, in most part of our presentation, we will set
the mean free path k = 1.
The main purpose of this Part I and the forthcoming Part II of the
present paper is to present the Greens function approach for the study of
the solutions of the Boltzmann equation. The Greens function approach
aims at pointwise, more quantitative understanding of the solutions of the
Boltzmann equation. This is necessary for the qualitative understanding
of the rich phenomena that the kinetic theory can model, particularly the
nonlinear interior waves and the behavior of the solution near the boundary.
There are studies based on physical considerations and asymptotic theory
for the Boltzmann equation, see [42], [43], [18], [19], [20]. We start with
the construction and analysis of the Greens function in this Part I. The
construction involves explicit inversion of the Fourier transform for the uid
like waves and Picard iterations for particle, particlelike and other singular
waves. The Greens functions are for the linearized Boltzmann equation. The
simplest situation is to consider the Boltzmann equation linearized around
a global Maxwellian M:
f = M +
Mg
so that the linearization yields
_
_
g
t
+
x
g = Lg, linearized Boltzmann equation,
Lg =
2Q(
Mg, M)
M
, linearized collision operator.
(1.6)
The most basic Greens function is for the initial value problem for the above
linearized Boltzmann equation. Because we are considering the Boltzmann
equation linearized around a xed Maxwellian, the equation is of constant
coecients in (x, t). Thus the equation is both time and space translational
invariant and so the Greens function has the property:
G(x, x
0
, t, s, ;
0
) = G(x x
0
, t s, ;
0
)
118 TAIPING LIU AND SHIHHSIEN YU [June
and satises
_
G
t
+
x
G = LG,
G(x, 0, ;
0
) = (x)(
0
).
(1.7)
The Greens function G(xx
0
, t s, ;
0
) describes the propagation of the
perturbation over the Maxwellian M when at time s the perturbation consists
of particles concentrated at the location x = x
0
and with the microscopic
velocity concentrated at
0
. The basic reason for the study of the Greens
function is that it can be used to study the general solutions. The simplest
situation is to consider the initial value problem for the linear Boltzmann
equation with a given source h = h(x, t, ):
_
g
t
+
x
g = Lg + h,
g(x, 0, ) = g
0
(x, ).
(1.8)
Multiply the equation with the Greens function and integrate to yield the
solution representation:
g(x, t, ) =
_
R
3
_
R
3
G(x y, t, ;
0
)g
0
(y,
0
) d
0
dy
+
_
t
0
_
R
3
_
R
3
G(x y, t s, ;
0
)h(y, s,
0
) d
0
dyds. (1.9)
We will construct this Greens function rst for the 1dimensional case, x
R, and then for the 3dimensional case, x R
3
. In preparation for these
constructions, we make the following preliminaries. In Section 2, we will
rst review the basics for the Boltzmann equation and in Section 3 for the
linearized Boltzmann equation. For the study of the uid aspects of the
Boltzmann equation, we consider in Section 4 the Euler and NavierStokes
equations in gas dynamics. The construction of the Greens function is done
rst for the case of plane waves, i.e. the case when the space variable is
1dimensional, x R,
_
G
t
+
1
x
G = LG,
G(x, 0, ;
0
) =
1
(x)(
0
);
(1.10)
2011] SOLVING BOLTZMANN EQUATION, PART I: GREENS FUNCTION 119
and the solution formula for the general initial value problem becomes
_
_
g
t
+
1
x
g = Lg,
g(x, 0, ) = g
0
(x, ),
g(x, t, ) =
_
R
_
R
3
G(x y, t, ;
0
)g
0
(y,
0
) d
0
dy.
(1.11)
We will consider the uidlike waves in Section 5 and the particlelike and
other singular waves in Section 6. The uidlike waves are studied by the
Fourier transform and the spectral analysis near the origin. The particlelike
waves are constructed by Picard iterations and their regularity properties are
studied by the Mixture Lemma. To combine these two types of waves for the
complete picture of the Greens function requires intricate analysis.
In Section 7 we construct the Greens function for the initial value prob
lem for the general waves in 3dimensional space. There are Huygens waves
and other uidlike waves.
After the Greens function for a global Maxwellian is constructed for the
initial value problem, we will use it to construct the Greens function for the
initialboundary value problem in Section 8.
These construction are pointwisely explicit and will be used in the forth
coming Part II for the study of nonlinear waves and boundary layers for the
Boltzmann equation. Our presentation is essentially selfcontained. The
existing materials, [31], [32], for the initial value problem, [33], for initial
boundary value problem, are reorganized with new perspectives. Some issues
are claried and new materials are added. We raise open problems from time
to time. For the Greens function approach aimed at the understanding of
the dissipation eects of the boundary, see [34] and [47].
2. Boltzmann Equation
In this section we review two basic properties of the Boltzmann equation
(1.2): the conservation laws and the HTheorem.
2.1. Conservation laws
The collision operator Q(f, f), (1.3), redistributes the density function
120 TAIPING LIU AND SHIHHSIEN YU [June
f(x, t, ) as a function of the microscopic velocity . Nevertheless, on the
macroscopic level, the collision operator conserves the mass, momentum and
energy:
_
R
3
_
_
_
1
1
2
[[
2
_
_
_
Q(g, h)d = 0,
_
_
_
mass
momentum
energy
_
_
_
. (2.1)
The above is proved by simple change of variables, noting that the transfor
mation (1.4) has Jacobian one:
_
R
3
()Q(g, h)()d =
1
4
_
R
3
(() + (
) (
) (
))Q(g, h)()d.
(2.2)
The key property is thus that the functions
0
() 1,
i
()
i
, i = 1, 2, 3,
4
() [[
2
/2, =
j
, j = 0, . . . , 4,
(2.3)
are collision invariants in the following sense.
Denition 2.1. A function () of the microscopic velocity is a collision
invariant if
() + (
) = (
) + (
) (2.4)
under the transformation (1.4).
In fact, the relation (1.4) is equivalent to the fact that the functions
1, , [[
2
/2 are collision invariant. Thus any function in the span of these
functions are also collision invariant and has the property that the integration
of them times the collision operator vanishes. There is the basic theorem
of Boltzmann saying that these are all the collision invariants; its proof is
omitted, [8].
Theorem 2.2 (Boltzmann). Any collision invariant is a linear combination
of 1, , [[
2
/2.
Multiply 1, , [[
2
/2 with the Boltzmann equation (1.2) and integrate
2011] SOLVING BOLTZMANN EQUATION, PART I: GREENS FUNCTION 121
to yield the conservation laws for the macroscopic variables of (1.1):
_
t
+
x
(v) = 0, conservation of mass,
t
(v) +
x
(v v +P) = 0, conservation of momentum,
t
(E) +
x
(vE +Pv +q) = 0, conservation of energy.
(2.5)
Remark 2.3. There are 5 conservation laws: 1 for the mass, 3 for the
momentum, and 1 for the energy. On the other hand, there are 14 unknowns:
1 for the density , 3 for the uid velocity v, 6 for the symmetric stress
tensor P, 3 for the heat ux q, and 1 for the internal energy e. Thus
the conservation laws are not a system of selfcontained partial dierential
equations. The macroscopic quantities are computed as moments of the
density function f(x, t, ). The stress tensor consists of second moments and
the heat ux consists of third moments. To derive equations for these would
require the fourth moments and so on. For instance, one may multiply
the Boltzmann equation by (
i
v
i
)(
j
v
j
) and integrate to derive the
evolutionary equation for the stress tensor p
ij
, a second moment:
t
(p
ij
) +
x
_
R
3
(
i
v
i
)(
j
v
j
)fd =
_
R
3
(
i
v
i
)(
j
v
j
)Q(f, f)d.
This would involve the third moment, the integration of (
i
v
i
)(
j
v
j
)f.
The evolution of the third moments then involves the fourth moments and
so on. Also the right hand side is not zero anymore, as (
i
v
i
)(
j
v
j
)
is not a collision invariant. Thus the Boltzmann equation is equivalent to
a system of innite partial dierential equations. Another way to view the
kinetic theory is to consider the distribution function f(x, t, ), for each xed
space and time variables (x, t), as a function of the microscopic variables .
The space of functions in is innite dimensional.
For the uid dynamics, the conservation laws are closed by making
assumptions on the dependence of the stress tensor and heat ux on the
conserved quantities and their dierentials. For the Euler equations in gas
dynamics,
p
ij
E
= p
ij
, q
E
= 0, Euler stress and heat ux; (2.6)
122 TAIPING LIU AND SHIHHSIEN YU [June
and for the NavierStokes equations,
_
_
p
ij
NS
= p
ij
[
v
i
x
j
+
v
j
x
i
2
3
3
k=1
v
k
x
k
ij
]
B
3
k=1
v
k
x
k
ij
,
NavierStokes stress;
q
NS
=
x
, NavierStokes heat ux.
(2.7)
We will derive the viscosity and the heat conductivity later. They are
functions of the temperature
= (), = ().
For the indepth study of the uid dynamics from the point of view of the
kinetic theory, the readers are referred to [42], [43].
2.2. HTheorem and Maxwellians
The Boltzmann equation has the socalled molecular chaos hypothesis
built in the collision operator in that the loss term f()f(
)f(
[f
]Bdd
d 0. (2.8)
The above is proved by simple change of variables, again using fact that the
transformation (1.4) has Jacobian one, (2.2). From (2.8), the Boltzmann
Theorem yields that
_
R
3
log fQ(f, f)d = 0, if and only if log f is a collision invariant,
in the span of 1, , [[
2
/2. (2.9)
In other words, log f, as a function of the microscopic velocity , is quadratic,
or, equivalently, f is gaussian in . Direct calculations show that the dis
tribution function is explicitly given in terms of the macroscopic variables:
2011] SOLVING BOLTZMANN EQUATION, PART I: GREENS FUNCTION 123
f(x, t, ) =
(x, t)
(2R(x, t))
3/2
e
v(x,t)
2
2R(x,t)
M
[
(, v, )]. (2.10)
Here we have introduced the macroscopic quantity , the temperature, through
the ideal gas relation between the temperature , the pressure p and the den
sity :
p = R. (2.11)
The states M
(,v,)
are called the Maxwellian distributions. They are the
thermoequilibrium states in that
Q(M, M) = 0. (2.12)
The HTheorem is obtained by integrating the Boltzmann equation (1.2)
times log f
t
H +
x
H 0, H
_
R
3
f log fd, H
_
R
3
f log fd. (2.13)
Thus a Boltzmann solution has the tendency to reach the thermoequilibrium
and the Boltzmann process is irreversible.
Remark 2.4. From the kinetic theory point of view, the uid dynamics and
thermodynamics are the study around the Maxwellians. The fact that the
Maxwellian distribution (2.10) is determined by only two state variables
and , and is consistent with the basic axiom in thermodynamics that there
are exactly two independent state variables. One can choose any two state
variables and all the other state variables are functions, the constitutive
relations, of the two chosen state variables. With the given distribution
(2.10), this thermodynamics axiom clearly holds. For the nonequilibrium
situation one can dene, symbolically, the pressure
p
p
11
+ p
22
+ p
33
3
=
1
3
_
R
3
[v [
2
f(x, t, )d, pressure,
and we have from (1.1) that the internal energy and the pressure are related
as:
p =
2
3
e. (2.14)
124 TAIPING LIU AND SHIHHSIEN YU [June
This is the relation for the monatomic gases and, when combined with the
ideal gas relation (2.11), we have
e =
3
2
R. (2.15)
In fact, one often takes (2.15) as the dening property of the temperature
in terms of the internal energy e. The monatomic gases have the 3 degrees
of freedom of translational energy. For the nonmonatomic gases, there are
other degree of freedom of rotational and vibrational energy, for instance.
For a gas with , 3, degrees of freedom, we would have e =
2
R.
2.3. Onedimensional ows
For plane waves, the density function f(x, t, ) depends only on 1di
mensional space variable x
1
, x = (x
1
, x
2
, x
3
). We will write x
1
= x. The
Boltzmann equation becomes
t
f(x, t, ) +
1
x
f(x, t, ) = Q(f, f)(x, t, ), (2.16)
Note that the microscopic velocity = (
1
,
2
,
3
) must remain 3dimensional.
The Greens function G(x, t, ;
0
), x R, for the initial value problem, c.f.
(1.7), becomes
_
G
t
+
1
x
G = LG for x R,
G(x, 0, ;
0
) =
1
(x)
3
(
0
).
(2.17)
Here we have highlighted the dimension of the delta functions. For conve
nience, we assume that the density distribution
f(x, t, ) is even in
2
and
3
. (2.18)
It is easy to see that if this assumption is made for the initial data, it remains
so for later time. With this, the uid velocity v = (v
1
, v
2
, v
3
) becomes
v = (v
1
, 0, 0) (v, 0, 0), (2.19)
and we will write the Maxwellians as
M
[,v,]
M
[,v,]
. (2.20)
2011] SOLVING BOLTZMANN EQUATION, PART I: GREENS FUNCTION 125
The thermoequilibrium manifold, the collection of all Maxwellians, is then
3dimensional. There are 3 eective collision invariants:
1,
1
,
[[
2
2
, 1dimensional collision invariants. (2.21)
3. Linearized Boltzmann Equation
Consider the solution f of the Boltzmann equation (1.2) as a perturba
tion of the Maxwellian M = M
[,v,]
:
_
_
f = M +
Mg,
g
t
+
x
g = Lg + (g),
Lg =
2Q(
Mg, M)
M
, linearized collision operator,
(g) =
Q(
Mg,
Mg)
M
, nonlinear term.
(3.1)
The Boltzmann equation linearized around M with the weight function
M
is
g
t
+
x
g = Lg. (3.2)
3.1. Linear collision operator
Any Maxwellian
M = M
[,v,]
is a thermoequilibrium state, Q(
M,
M) =
0. The equilibrium manifold is a 5dimensional manifold of the Maxwellians,
parametrized by (, v, ). As a consequence, the kernel of linearized collision
operator Lg = 2Q(
Mg, M)/
M, (3.1), it is in the
126 TAIPING LIU AND SHIHHSIEN YU [June
span of
M,
M,
M, [[
2
M,
i
=
i
M, i = 1, 2, 3,
4

2
2

2
2
(3.4)
2011] SOLVING BOLTZMANN EQUATION, PART I: GREENS FUNCTION 127
the linearized collision operator has the explicit form, e.g. [9],
_
_
Lg() = ( + K)g() ()g() +
_
R
3
K(,
)g(
)d
,
()
1
2
_
2e

2
2
+ 2([[ +
1
[[
)
_

0
e
2
2
d
_
,
C
1
(1 +[[) () C
2
(1 +[[) for some C
1
, C
2
> 0,
K(,
)
2
2[
[
exp
_
([[
2
[
[
2
)
2
8[
[
2
[
[
2
8
_
[
2
exp
_
([[
2
+[
[
2
)
4
_
.
(3.5)
The linearized collision operator around a Maxwellian M
[,v,]
is a simple
translation and dilation of the above. Here, as in what follows, we have
taken, for simplicity, the gas constant R = 1.
The Hilbert space L
2
,
sup
R
3
[g()[(1 +[[)
, 0. (3.6)
The integral operator K has some boundedness and smoothing properties.
The following lemma follows from direct computations.
Lemma 3.2. The operator K is compact and the integral operators K
, K
,
dened by the kernel K
, K
,+1
C()j
L
,
,
Kj
L
,0
C
1
j
L
2
.
(3.7)
The integral operator K has limited smoothing property in , (3.7).
This is due to the singularity of the kernel K(,
) at =
, (3.5). We
128 TAIPING LIU AND SHIHHSIEN YU [June
decompose it into
_
_
K = K
0
+ K
1
= K
0,D
+ K
1,D
,
K
i
z()
_
R
3
K
i
(,
)z(
) d
for i = 0, 1,
K
0
(,
) =
0
_

D
0
_
K(,
),
K
1
(,
) =
_
1
0
_

D
0
__
K(,
),
0
(r) 1 for r [1, 1],
supp(
0
) [2, 2],
0
C
c
(R),
0
0.
(3.8)
Here the cuto parameter D will be chosen to be small. We may write the
linearized Boltzmann equation (3.2) as:
t
g +
1
x
g + ()g = (K
0
+ K
1
)g. (3.9)
The operator K
0
shares the same smoothing property as K and has strength
of the order of the cuto parameter D:
K
0
h
L
,+1
C
Dh
L
,
for 0. (3.10)
The operator K
1
is a smoothing operator because it does not inherit the
singular nature of the kernel K(,
) at =
.
Lemma 3.3. The operator K
1
given in (3.8) is a smoothing operator in :
for any h L
2
, i 0,
K
1
h
H
i
= O(1)h
L
2
.
Here H
i
.
Proof. From the denition of K
1
in (3.8),
K
1
h()
_
R
3
h(
__
1
0
_
[
[
D
0
__
K(,
)
_
d
.
The function K(,
) is smooth for [
[ > D
0
. Thus,
__
1
0
_

D
0
__
K(,
)
_
is a globally smooth function. It is easy to see that for any i
[[ 0 the function
((1
0
(

D
0
))K(,
)) L
1
to L
2
):
G
L
2
x
(L
2
)
1. (3.11)
Proof. The form of the linear collision operator with the weight of
M
serves the convenient purpose that the collision operator is selfadjoint in
the Hilbert space L
2
:
(Lg, h) = (g, Lh), (g, h) = (g, h)
L
2
_
R
3
g()h()d. (3.12)
This is shown by simple changes of variables in the integrations such as (2.2),
_
_
(Lg, h) = (g, Lh)
=
1
16
_
R
3
_
R
3
_
S
2
MM
[
g
+
g
M
g
]
[
h
+
h
M
h
]Bdd
d,
(3.13)
making use of the fact that the Maxwellians have the dening property that
log M is collision invariant, or, M
= MM
dx =
_
R
3
(Lg, g)
L
2
dx 0. (3.14)
Thus the Greens function as an operator in propagating the solution of the
initial value problem, (1.8), (1.9), is contractive in L
2
x
(L
2
) and so (3.11)
holds.
3.2. Macromicro projections
One forms the orthogonal basis for the kernel of the collision operator
130 TAIPING LIU AND SHIHHSIEN YU [June
(3.3):
_
0
M
1/2
,
i
(
i
v
i
)M
1/2
, i = 1, 2, 3,
4
1
6
([ v[
2
3)M
1/2
.
(3.15)
The macro projection P
0
and the micro projection P
1
are
P
0
g
4
j=0
(g,
j
)
j
, P
1
I P
0
. (3.16)
We will write
g
0
P
0
g, g
1
P
1
g, g = g
0
+ g
1
, (g
0
, g
1
) = 0. (3.17)
The macro projection P
0
consists of the isotropic part P
iso
0
and the momen
tum part P
m
0
:
P
0
= P
iso
0
+ P
m
0
, P
iso
0
g = (g,
0
)
0
+ (g,
4
)
4
, P
m
0
3
i=1
(g,
i
)
i
. (3.18)
Clearly, the linear collision operator L satises
P
0
L = LP
0
= 0, P
1
L = LP
1
= L; (3.19)
and, from the Boltzmann Theorem, the nonlinear collision term (g), (3.1),
also satises
P
0
= 0, P
1
= . (3.20)
Theorem 3.5. The spectrum (L) of the linearized collision operator L has
the following property:
(1) There is the eigenvalue 0 with corresponding eigenfunctions the 5dimensional
collision invariants, (3.3).
(2) Besides the 0 eigenvalues, the maximum value of the spectrum is a nega
tive number
0
. In other words, there is a spectrum gap between 0 and
the rest of the spectrum.
2011] SOLVING BOLTZMANN EQUATION, PART I: GREENS FUNCTION 131
Consequently,
(Lg, g)
1
(P
1
g, P
1
g), (3.21)
for any function g L
2
.
Proof. From (3.5), the collision operator can be viewed as the compact
perturbation K of the multiplicative operator (). From the Weyls The
orem, the essential spectrum
ess
= y : y
0
of L is the values taken
by (). As () is an increasing function of [[,
0
(0) = min
(), (3.22)
As L is selfadjoint and nonpositive, the discrete eigenvalues
discrete
(L) =
(L)
essential
(L) ((0), 0]. Therefore, there are nite eigenvalues in in
terval [
0
+, 0] for any small positive . In particular, the zero eigenvalue
is isolated. Thus the maximum of the rest of the spectrum
1
is negative,
0
>
1
> 0. Finally, (3.21) follows from the fact that the micro projection
P
1
projects to the orthogonal complement of the kernel of L. This establishes
(3.21) and the theorem is proved.
The above theorem is due to the eorts of Hilbert, Carleman, and Grad.
For constructive estimates of the spectrum gap, see [1], [37], [38].
Corollary 3.6. There exist positive constants C
1
, C
2
such that
(Lg, g) C
1
((1 +[[)P
1
g, P
1
g), (3.23)
[([[g, g)[ C
2
[(g, g) + (Lg, g)]. (3.24)
Proof. From the expression L = K () and the estimates, (3.5),
() D
1
(1 +[[), K
L
2
= D
2
< ,
for some positive constants D
1
, D
2
, and so we have
(Lg, g) = (LP
1
g, P
1
g) D
1
((1 +[[)P
1
g, P
1
g) + D
2
(P
1
g, P
1
g). (3.25)
The estimate (3.23) is shown with C
1
= D
1
1
/(D
2
+
1
) by adding (3.21)
times D
2
/
1
and (3.25). We have from the CauchySchwarz inequality that
[([[g, g)[ = [([[(g
0
+ g
1
), g
0
+ g
1
)[
132 TAIPING LIU AND SHIHHSIEN YU [June
= O(1)[((1 +[[)P
0
g, P
0
g) + ((1 +[[)P
1
g, P
1
g)].
Thus (3.24) follows from (3.23) if
((1 +[[)P
0
g, P
0
g) = O(1)(P
0
g, P
0
g),
which holds trivially because, after the macro projection P
0
, it is a nite
dimensional situation. This completes the proof of the corollary.
3.3. Macroscopic variables
The uid variables for the conservation laws
_
_
(g, M
1/2
) = (g
0
, M
1/2
),
m
i
(g,
i
M
1/2
) = (g
0
,
i
M
1/2
), m (m
1
, m
2
, m
3
),
E (g,
1
2
[[
2
M
1/2
) = (g
0
,
1
2
[[
2
M
1/2
).
(3.26)
There is some ambiguity with earlier notations, as, for instance, the here
is not the density, but the perturbation of the density. There is no confusion
here and we use these for the sake of notational simplicity. We will also use
the projection to the orthogonal basis
i
, i = 0, 1, 2, 3, 4:
_
_
(g,
0
) = (g
0
,
0
),
m
i
(g,
i
) = (g
0
,
i
),
E (g,
4
) = (g
0
,
4
),
(3.27)
g
0
(x, t, ) = (x, t)
0
() +
3
i=1
m
i
(x, t)
i
() +
E(x, t)
4
(). (3.28)
4. Euler and NavierStokes
For the discussion of this section, the mean free path k is retrieved to
highlight the dependence of the hydrodynamics dissipation parameters, the
viscosities and heat conductivity on k.
2011] SOLVING BOLTZMANN EQUATION, PART I: GREENS FUNCTION 133
Boltzmann equation has its uid dynamics aspects. The most basic equa
tions for gas dynamics are the Euler equations. The Euler equations in gas
dynamics can be directly derived from the Boltzmann equation and provide
the basic information on the propagation of uidlike waves in the kinetic
theory. The NavierStokes equations and the Boltzmann equation share
the basic property that they are both dissipative. Boltzmann equation is
dissipative partly as a consequence of the HTheorem. The heat conduc
tivity and viscosity coecients for the NavierStokes equations are also the
basic dissipation parameters for the Boltzmann solutions. We will derive
the compressible NavierStokes equations in the spirit of ChapmanEnskog
expansion.
4.1. Euler equations
Each Maxwellian is a constant solution of the Boltzmann equation,
(2.12). In general, local Maxwellians do not form a solution of the Boltz
mann equation. As the mean free path k tends to zero, for smooth solutions,
one expects Q(f, f) to go to zero:
Q(f, f)(x, t, ) = k[
t
f(x, t, ) +
x
f(x, t, )] 0.
And so, by the HTheorem, (2.12), f M
f
, where M
f
are the Maxwellians
determined by the macroscopic variables of the solution f. We call
(M
f
)
t
+
x
(M
f
) = 0 (4.1)
the Euler equations in the kinetic form. Assuming that the distribution
function is locally Maxwellian
f = M
f
,
then, by direct calculations, it is easy to see that
stress tensor P = pI, the pressure p; heat ux q = 0.
Here I is the 3 by 3 identity matrix. Thus in this zero mean free path limit,
k 0+, the conservation laws (2.5) are simplied to the Euler equations in
134 TAIPING LIU AND SHIHHSIEN YU [June
gas dynamics
_
t
+
x
(v) = 0,
t
(v) +
x
(v v + pI) = 0,
t
(E) +
x
(vE + pv) = 0.
(4.2)
The state variables are related by the constitutive relations for the monatomic
gases:
p = R =
2
3
e = p
0
e
As
5
3
. (4.3)
Here the rst relation is the ideal gas law, (2.11). We will take the gas
constant R = 1. The second equality comes from the monatomic gases
law (2.14). From these we have the last equality by the thermal dynamics
relation de = pd(1/) + ds, with s the entropy, for any choice of positive
constants A and p
0
.
The Euler equations are selfcontained, as the stress tensor is now a
given function of the conserved quantity and the heat ux is zero. So the
number of the dependent variables is now 14 6 3 = 5, the same as the
number of equations.
Similarly, the linear Euler equations are the projection of the linearized
Boltzmann equation (3.2) to the tangent space of the equilibrium Maxwellian
states manifold at the base Maxwellian M = M
0
= M
[
0
,v
0
,
0
]
. Thus the linear
Euler equations in the kinetic form is the macro projection of the linearized
Boltzmann equation, (3.19):
(P
0
g)
t
+
x
P
0
P
0
g = 0. (4.4)
The resulting conservation laws, in terms of the macroscopic variables are
the linear Euler equations in gas dynamics. There are two versions. The
rst is using the conservative macroscopic variables (3.26). This one is the
same as the linearized version of the full Euler equations (4.2):
_
t
+
x
m = 0,
m
t
+
2
3
x
E = 0,
E
t
+
5
2
x
m = 0.
(4.5)
2011] SOLVING BOLTZMANN EQUATION, PART I: GREENS FUNCTION 135
Here we have, for simplicity, assumed that the base state for the linearization
is taken to be
0
= 1, v
0
= 0,
0
= 1. The second version is using the
macroscopic variables through the orthogonal projections (3.27), (3.28):
_
_
t
+v
0
x
+
x
m = 0,
m
t
+v
0
x
m+
x
+
_
2
3
x
E = 0,
E
t
+v
0
x
E +
_
2
3
x
m = 0.
(4.6)
4.2. Euler, Euler ux projections
In the linear Euler equations (4.4), the ux operators P
0
i
P
0
, i = 1, 2, 3,
is on the 5dimensional space
j
, j = 0, . . . , 4. For deniteness, we con
sider P
0
1
P
0
. By straightforward calculations, we have the following Euler
characteristics
1
i
and Euler characteristic directions E
1
i
:
_
_
P
0
1
E
1
j
=
1
j
E
1
j
, j = 1, . . . , 5,
1
1
= v
1
c,
1
2
= v
1
,
1
3
= v
1
+c,
1
4
=
1
5
= v
1
,
E
1
1
=
_
3
10
_
1
2
1
+
_
1
5
4
,
E
1
2
=
_
2
5
0
+
_
3
5
4
,
E
1
3
=
_
3
10
0
+
_
1
2
1
+
_
1
5
4
,
E
1
4
=
2
,
E
1
5
=
3
.
(4.7)
Here we have taken the base state to be (, v, ) and so the sound speed c is
c =
_
5
3
. (4.8)
The Euler characteristic directions are orthogonal:
(E
1
j
, E
1
k
) = 0 for
1
j
,=
1
k
. This is seen easily as follows:
1
j
(E
1
j
, E
1
k
) = (P
0
1
E
1
j
, E
1
k
) = (
1
E
1
j
, E
1
k
) = (E
1
j
,
1
E
1
k
)
= (E
1
j
, P
0
1
E
1
k
) =
1
k
(E
1
j
, E
1
k
).
136 TAIPING LIU AND SHIHHSIEN YU [June
The above Euler characteristic directions have been normalized:
(E
1
i
, E
1
j
) =
ij
, i, j = 1, . . . , 5, (4.9)
so that the macro projection can also be written as
P
0
g =
4
j=0
(g, E
1
j
)E
1
j
.
We dene the Euler projections B
j
:
B
j
g = (g, E
1
j
)E
1
j
, j = 1, . . . , 5. (4.10)
For the 1dimensional Boltzmann equation, (2.16), the linearized Euler
equations are
P
0
g
t
+ (P
0
1
P
0
g)
x
= 0. (4.11)
The kernel of the linearized collision operator is the span of
L
j
= 0, j = 0, 1, 4,
0
M,
1
=
1
M,
4
[[
2
2
M,
1dimensional linear collision kernel, (4.12)
and the macro and micro projections become, with v = (v, 0, 0),
_
_
P
0
g (g,
0
)
0
+ (g,
1
)
1
+ (g,
4
)
4
; P
1
I P
0
,
0
M
1/2
,
1
(
1
v)M
1/2
,
4
1
6
([ v[
2
3)M
1/2
.
(4.13)
The Euler characteristics are
_
_
P
0
1
E
j
=
1
j
E
j
, j = 1, 2, 3,
1
= v c,
2
= v,
3
= v +c,
E
1
=
_
3
10
_
1
2
1
+
_
1
5
4
,
E
2
=
_
2
5
0
+
_
3
5
4
,
E
3
=
_
3
10
0
+
_
1
2
1
+
_
1
5
4
.
(4.14)
2011] SOLVING BOLTZMANN EQUATION, PART I: GREENS FUNCTION 137
t
x
dx
dt
=
1
dx
dt
=
2
dx
dt
=
3
Figure 1: Euler waves.
The Euler projections B
j
are dened as:
B
j
g = (g, E
j
)E
j
, j = 1, 2, 3. (4.15)
The initial value problem for the Euler equations
_
P
0
g
t
+ (P
0
1
P
0
g)
x
= 0,
P
0
g(x, 0, ) = g(x, ),
(4.16)
is decoupled by the Euler projections:
_
_
g
j
(g, E
j
), g
j
( g, E
j
),
(g
j
)
t
+
j
(g
j
)
x
= 0,
g
j
(x, 0) = g
j
(x).
(4.17)
This can be solve by the characteristic method to yield the Euler waves,
Figure 1:
_
g(x, t, ) =
3
j=1
g
j
(x, t)E
j
(),
g
j
(x, t) = g
j
(x
j
t).
(4.18)
The Euler ux projections
B
j
are dened for nonzero Euler characteristic
speeds
j
as follows:
B
j
g =
1
j
(
1
g, E
j
)E
j
, j = 1, 2, 3,
P
0
3
j=1
B
j
. (4.19)
138 TAIPING LIU AND SHIHHSIEN YU [June
It is easy to see that
P
0
=
3
j=1
B
j
, P
1
B
i
= B
i
P
1
= 0,
(4.20)
B
i
E
j
=
ij
E
i
, B
i
B
j
=
ij
B
i
, i, j = 1, 2, 3, P
1
B
i
= 0.
Although B
i
P
1
= 0, it is not true in general that
B
i
P
1
= 0.
For the study of initialboundary and the boundary value problems, it is
essential to dierentiate the direction of the Euler waves. For this, we dene
the upwind Euler projection B
+
and the downwind Euler projection B
:
B
+
i
>0
B
i
, B
i
<0
B
i
. (4.21)
Similarly, the upwinddownwind Euler ux projections
B
are dened as
B
+
i
>0
B
i
,
B
i
<0
B
i
. (4.22)
4.3. NavierStokes dissipation parameters
The full Euler equations (4.2) are derived by assuming that the distri
bution function is locally Maxwellian, f = M
f
. For the derivation of the
NavierStokes equations, the distribution function is assumed to deviates
slightly from the local Maxwellian. For our purpose of the construction of
the Greens function, we will consider the case of linear NavierStokes equa
tions only. In the linear case, the Euler equations (4.6) are derived from the
linear Boltzmann equation (3.2) by considering only the macro projection
of the macro part g
0
= P
0
g of the Boltzmann solution g, (4.6). The macro
projection is the same as considering the conservation laws (4.6) induced by
the macro part. We now derive the linear NavierStokes equations by the
same procedure. First we write the linear Boltzmann equation (3.2) as:
(g
0
+ g
1
)
t
+
x
(g
0
+ g
1
) =
1
k
Lg
1
, g
0
P
0
g, g
1
P
1
g.
2011] SOLVING BOLTZMANN EQUATION, PART I: GREENS FUNCTION 139
The macro and micro projections of this equation are
(g
0
)
t
+
x
P
0
(g
0
+ g
1
) = 0, macro part of linear Boltzmann equation;
(4.23)
(g
1
)
t
+
x
P
1
(g
0
+ g
1
) =
1
k
Lg
1
, micro part of linear Boltzmann equation.
(4.24)
In the spirit of ChapmanEnskog expansion, we assume that the macro part
g
0
dominates the micro part g
1
and that the dierential
x,t
h is dominated
by the quantity h itself. Another way of thinking of this is to consider
the timeasymptotic dissipation of the Boltzmann solution toward a given
Maxwellian. As we will see, the macro part decays slower than the micro
part. Thus we will make the simplication of the micro part (4.24) of the
linear Boltzmann equation into
x
P
1
(g
0
)
1
k
Lg
1
, or g
1
kL
1
[
x
P
1
(g
0
)] ChapmanEnskog relation.
(4.25)
Plug the ChapmanEnskog relation (4.25) into the macro part (4.23) of the
Boltzmann equation:
(g
0
)
t
+
x
P
0
(g
0
) = k
x
P
0
(L)
1
[
x
P
1
(g
0
)]; or,
(g
0
)
t
+
x
P
0
(g
0
) = k
3
i,j=1
x
i
x
j P
0
i
(L)
1
[P
1
(
j
g
0
)],
NavierStokes equations in kinetic form. (4.26)
The linear NavierStokes equations in gas dynamics form are the con
servation laws, obtained by integrating the kinetic equation (4.26) by the
linear collision invariants, the kernel of the linear collision operator, (3.15),
(3.26). The integration of (4.26) times
0
yields the usual conservation of
mass:
t
+
x
m+ v
0
= 0.
The integration of (4.26) times
4
yields the conservation of energy:
E
t
+ v
0
E +
_
2
3
x
m =
x
E,
140 TAIPING LIU AND SHIHHSIEN YU [June
with the heat conductivity coecient :
= k
_
P
1
4
, L
1
P
1
4
_
. (4.27)
We now compute the integration of (4.26) times
l
, l = 1, 2, 3, to yield the
conservation of momentum and the viscosity coecients:
m
l
t
+
3
i=1
v
i
0
m
l
x
i
+
x
l
+
_
2
3
E
x
l
= k
3
i,j,n=1
2
m
n
(x, t)
x
i
x
j
_
l
, P
0
_
i
L
1
_
P
1
(
j
n
)
__
. (4.28)
Observe that
3
i,j,n=1
2
m
n
(x, t)
x
i
x
j
_
l
, P
0
_
i
L
1
_
P
1
(
j
n
)
__
=
3
i,j,n=1
2
m
n
(x, t)
x
i
x
j
_
P
1
_
l
_
, L
1
_
P
1
(
j
n
)
_
=
3
i,j,n=1
2
m
n
(x, t)
x
i
x
j
_
P
1
_
i
_
l
v
l
0
_
M
_
, L
1
_
P
1
(
j
(
n
v
n
0
)
M)
__
=
3
i,j,n=1
2
m
n
(x, t)
x
i
x
j
_
P
1
_
M
_
, L
1
_
P
1
(
j
M)
__
=
3
j=1
2
m
j
(x, t)
x
l
x
j
_
P
1
_
(
l
)
2
M
_
, L
1
_
P
1
((
j
)
2
M)
__
(i =l, n=j)
+
n =l
2
m
n
(x, t)
x
n
x
l
_
P
1
_
M
_
, L
1
_
P
1
(
l
M)
__
(j =l, i =n)
+
j=l
2
m
l
(x, t)
x
j
x
j
_
P
1
_
M
_
, L
1
_
P
1
(
j
M)
__
(n=l, i =j) .
Due to the rotational invariance of the L
1
and integration, for i ,= j,
_
P
1
_
M
_
, L
1
_
P
1
_
M
___
=
_
P
1
_
(
i
)
2
M
_
, L
1
_
P
1
_
(
j
)
2
M
___
2011] SOLVING BOLTZMANN EQUATION, PART I: GREENS FUNCTION 141
and
_
P
1
_
(
i
)
2
M
_
, L
1
_
P
1
_
(
i
)
2
M
___
=
_
P
1
_
_
i
+
j
2
_
2
M
_
, L
1
_
P
1
_
_
i
+
j
2
_
2
M
___
=
_
P
1
__
(
i
)
2
+ (
j
)
2
2
+
i
j
_
M
_
,
L
1
_
P
1
__
(
i
)
2
+ (
j
)
2
2
+
i
j
_
M
__ _
=
1
2
_
P
1
_
(
i
)
2
M
_
, L
1
_
P
1
_
(
i
)
2
M
___
+
3
2
_
P
1
_
M
_
, L
1
_
P
1
_
M
___
,
which implies that
_
P
1
_
(
i
)
2
M
_
, L
1
_
P
1
_
(
i
)
2
M
___
= 3
_
P
1
_
M
_
, L
1
_
P
1
_
M
___
. (4.29)
With these, (4.28) becomes
m
l
t
+
3
i=1
v
i
0
m
l
x
i
+
x
l
+
_
2
3
E
x
l
= 3
2
m
l
(x, t)
x
l
x
l
+
_
_
_
j=l
2
m
j
(x, t)
x
l
x
j
+
n =l
2
m
n
(x, t)
x
n
x
l
+
j=l
2
m
l
(x, t)
x
j
x
j
_
_
_
=
3
j=1
_
2
m
l
(x, t)
(x
j
)
2
+ 2
2
m
j
(x, t)
x
l
x
j
_
=
3
j=1
x
j
_
_
m
l
x
j
+
m
j
x
l
2
3
jl
3
n=1
m
n
x
n
_
+
5
3
jl
3
n=1
m
n
x
n
_
Here the viscosity and the bulk viscosity
B
are given as
k
_
P
1
_
M
_
, L
1
_
P
1
_
M
___
,
B
5
3
. (4.30)
Using (4.27) and (4.30) the linear NavierStokes equations in gas dynamics
142 TAIPING LIU AND SHIHHSIEN YU [June
form are then:
_
t
+
x
m+ v
0
x
= 0,
m
l
t
+
3
i=1
v
i
0
m
l
x
i
+
x
l
+
_
2
3
E
x
l
=
3
j=1
x
j
_
_
m
l
x
j
+
m
j
x
l
2
3
jl
3
n=1
mn
x
n
_
5
3
jl
3
n=1
mn
x
n
_
, l =1, 2, 3,
E
t
+ v
0
x
E +
_
2
3
x
m =
x
E.
(4.31)
Another form of the NavierStokes equations that are convenient for the
study of uid waves are obtained by integrating the kinetic equation (4.26)
times the Euler characteristic directions. We do this for 1D case so that the
NavierStokes equation in the kinetic form is
g
0
t
+
x
P
0
_
1
g
0
_
= k
2
x
2
P
0
_
1
L
1
_
P
1
(
1
g
0
)
_
. (4.32)
Express
g
0
=
3
j=1
g
j
E
j
in terms of the Euler characteristic directions E
j
, j = 1, 2, 3, (4.14). This
yields the NavierStokes equations in gas dynamics form:
_
_
(g
i
)
t
+
i
(g
i
)
x
=
3
j=1
A
ij
(g
j
)
xx
, i = 1, 2, 3,
A
ij
= k
_
P
1
1
E
i
, L
1
_
P
1
(
1
E
j
)
_
.
(4.33)
The dissipation parameters matrix A
ij
are related to the viscosity and
heat conductivity . For instance, from (4.14),
1
k
A
11
=
_
P
1
1
E
1
, L
1
_
P
1
(
1
E
1
)
__
=
1
5
_
P
1
1
_
_
3
2
_
5
2
1
+
4
_
, L
1
_
P
1
1
_
_
3
2
_
5
2
1
+
4
___
Note that
1
0
is a linear collision invariant and so P
1
0
= 0, and that
4
is odd in
1
and
1
1
is even in
1
and so (P
1
1
, L
1
4
) = 0. Thus
2011] SOLVING BOLTZMANN EQUATION, PART I: GREENS FUNCTION 143
it follows from (4.27), (4.30), and (4.29) that
A
11
=k
1
5
_
5
2
_
P
1
1
, L
1
_
P
1
1
_
+
_
P
1
4
, L
1
P
1
4
_
_
=
3
2
+
1
5
.
Similarly, we have
A
12
= k(P
1
1
E
1
, L
1
[P
1
(
1
E
2
)])
=
3
5
_
P
1
1
_
_
3
2
_
5
2
1
+
4
_
, L
1
_
P
1
1
_
_
2
3
0
+
4
___
=
3
5
_
P
1
4
, L
1
P
1
4
_
=
3
5
,
A
13
= k
_
P
1
1
E
1
, L
1
_
P
1
(
1
E
3
)
_
=
1
5
_
P
1
1
_
_
3
2
_
5
2
1
+
4
_
, L
1
_
P
1
1
_
_
3
2
0
+
_
5
2
1
+
4
___
=
1
5
_
5
2
_
P
1
1
, L
1
_
P
1
1
_
+
_
P
1
4
, L
1
P
1
4
_
_
=
3
2
+
1
5
,
A
22
= k
_
P
1
1
E
2
, L
1
_
P
1
(
1
E
2
)
_
= k
3
5
_
P
1
1
_
_
2
3
0
+
4
_
, L
1
_
P
1
1
_
_
2
3
0
+
4
___
= k
3
5
_
P
1
4
, L
1
P
1
4
_
=
3
5
,
A
23
= k
_
P
1
1
E
2
, L
1
_
P
1
(
1
E
3
)
_
= k
3
5
_
P
1
1
_
_
2
3
0
+
4
_
, L
1
_
P
1
1
_
_
3
2
0
+
_
5
2
1
+
4
___
= k
3
5
_
P
1
4
, L
1
P
1
4
_
=
3
5
,
A
33
= k
_
P
1
1
E
3
, L
1
_
P
1
(
1
E
3
)
_
= k
1
5
_
P
1
1
_
_
3
2
0
+
_
5
2
1
+
4
_
, L
1
_
P
1
1
_
_
3
2
0
+
_
5
2
1
+
4
___
= k
1
5
_
5
2
_
P
1
1
, L
1
_
P
1
1
_
+
_
P
1
4
, L
1
P
1
4
_
_
=
3
2
+
1
5
.
144 TAIPING LIU AND SHIHHSIEN YU [June
Due to the symmetry of A
ij
we have
[A
ij
] =
_
_
3
2
+
1
5
3
5
3
2
+
1
5
3
5
3
5
3
5
3
2
+
1
5
3
5
3
2
+
1
5
_
(4.34)
where the viscosity and heat conductivity are given by (4.30) and (4.27).
The dissipation parameters matrix [A
ij
] is nonnegative, and but not
positive denite. This is typical for systems of viscous conservation laws
with physical viscosity matrix, [25]. It follows from the general theory for
the system of hyperbolicparabolic equations that the dominant dissipation
waves are given by the diagonal system, [35]:
( g
j
)
t
+
j
( g
j
)
x
= A
j
( g
j
)
xx
, j = 1, 2, 3, (4.35)
with the dissipation parameters
_
_
A
1
A
11
= k
_
P
1
1
E
1
, L
1
_
P
1
(
1
E
1
)
_
=
3
2
+
1
5
,
A
2
A
22
= k
_
P
1
1
E
2
, L
1
_
P
1
(
1
E
2
)
_
=
3
5
,
A
3
A
33
= k
_
P
1
1
E
3
, L
1
_
P
1
(
1
E
3
)
_
= A
1
=
3
2
+
1
5
.
(4.36)
The Greens function
_
(
G
j
)
t
+
j
(
G
j
)
x
= A
j
(
G
j
)
xx
, j = 1, 2, 3,
G(x, 0) = (x),
(4.37)
consists of dissipation waves for this simplied system are the collection of
heat kernels H(x, t; A), Figure 2, c,f, Figure 1:
_
_
_
G
j
(x, t) = H(x
j
t, t; A
j
),
H(x, t; A)
1
4At
e
x
2
4At
.
(4.38)
In the kinetic form, (4.32), (4.33), this is translated into
G(x, t) =
3
j=1
H(x
j
t, t; A
j
)E
j
E
j
[, (4.39)
2011] SOLVING BOLTZMANN EQUATION, PART I: GREENS FUNCTION 145
t
x
dx
dt
=
1
dx
dt
=
2
dx
dt
=
3
Figure 2: NavierStokes leading waves.
containing the operator in the function of the microscopic velocity:
E
j
E
j
[g (E
j
, g)E
j
. (4.40)
5. 1D Greens Function, FluidLike Waves
In this and the following two sections, we will construct the Greens
function for the initial value problem for the linear Boltzmann equation.
This and the next sections will concentrate on the 1dimensional linearized
Boltzmann equation, cf. (3.2),
g
t
+
1
x
g = Lg. (5.1)
For simplicity, we will take the base Maxwellian for linearization, (3.1) to be
M = M
0
M
[1,0,1]
so that the explicit expression of the linearized collision
operator is given in (3.5) and the Euler characteristic values are
1
=
_
5
3
,
2
= 0,
3
=
_
5
3
.
The 1D Greens function G(x, x
0
, t, s, ;
0
) = G(xx
0
, t s, ;
0
) satises
(2.17):
_
G
t
+
1
x
G = LG for < x < , t > 0,
G(x, 0, ;
0
) = (x)
3
(
0
).
(5.2)
The Greens function G(x y, t s, ;
0
) describes the propagation of the
perturbation over the Maxwellian M
0
when at time the perturbation con
146 TAIPING LIU AND SHIHHSIEN YU [June
sists of particles concentrated at space x = 0 and with microscopic velocity
0
.
Consider the initial value problem
_
g
t
+
1
x
g = Lg,
g(x, 0, ) = g
0
(x, ).
(5.3)
Multiply the equation with the Greens function and integrate to yield the
solution representation for (5.3):
g(x, t, ) =
_
R
_
R
3
G(x y, t, ;
0
)g
0
(y,
0
) d
0
dy (locally in (x, t, )),
(5.4)
g(x, t) G
t
g
0
(x) (locally in (x, t) and in a semigroup format). (5.5)
Here the formula (5.4) is for pointwise expression; (5.5) is to view the Greens
function as an operator. Consider the Fourier transform in the space variable
x:
g(, t)
1
2
_
R
e
ix
g(x, t)dx, g(x, t)
1
2
_
R
e
ix
g()d.
Take the Fourier transform of (2.17) to obtain
_
_
_
G
t
+ i
1
G = L
G(x, 0, ;
0
) =
1
3
(
0
).
(5.6)
With this, one can use the isometry property of the Fourier transform in the
L
2
x
to study the dissipation property of the Boltzmann solutions. Instead,
we will invert the Fourier transform for the uidlike part of
G, to gain quan
titative information on the Greens function in the physical space variable
x. The formal expression of the Greens function is
G(x, t, ;
0
) =
1
2
_
R
e
ix+(i
1
+L)t
3
(
0
)d. (5.7)
Plug this into the solution formula (5.4) to yield
g(x, t, ) =
_
R
_
1
2
_
R
e
i(xy)+(i
1
+L)t
dg
0
(y, )
_
()dy. (5.8)
2011] SOLVING BOLTZMANN EQUATION, PART I: GREENS FUNCTION 147
Thus in the operator expression, c.f. (5.5), we can write
G(x, t)
1
2
_
R
e
ix+(i
1
+L)t
d. (5.9)
Remark 5.1. The operator i
1
+ L on functions of the microscopic ve
locity is the focus of our study for this section. As the operator has the
onedimensional Fourier parameter , we do not have complete information
of its spectrum. Nevertheless, the point spectrum for near origin has
explicit expression and correspond to the uidlike waves, similar to that
for the compressible NavierStokes equations. These waves dominate the
Greens function timeasymptotically. The construction and description of
these waves is the main concern of this section.
To obtain the complete description of the Greens function, there are two
further steps to take. The rst is to construct the singular waves through
a series of Picard iterations, done in Section 6. These two constructions of
waves oers two distinct decompositions of the Greens function, and allows
us nally to apply the soft analysis of weighted energy method and Sobolev
calculus toward the end of Section 6 for the complete pointwise description
of the Greens function.
5.1. Spectral near origin
From the expression (5.9) for the Greens function, it is natural to con
sider the spectrum of the operator i
1
+ L. We consider rst the situa
tion near the origin, [[ small. Let (, e) = ((), e()) be the eigenvalue
eigenfunctions for i
1
+ L:
(i
1
+ L)e = e. (5.10)
We know from Theorem 3.5 that the zero eigenvalue of the linear collision
operator L is isolated and with multiplicity 3, (4.12). Thus for [[ 1 the
spectrum of i
1
+ L should be also be 3 eigenvalues near zero.
The following Lemma 5.4 provides the spectrum information we need.
Statement (I)(III) of Lemma 5.4 are true for hard potentials with Grad
cuto. However, the analytic property, (IV), is only true for the hard sphere
148 TAIPING LIU AND SHIHHSIEN YU [June
model, see Remark 5.5. Since analyticity is required for the following anal
ysis, we will assume the hard sphere model throughout. Still, for the sake of
completeness, we point out the fact that (I)(III) are generally true for hard
potentials with Grad cuto.
Denition 5.2. For any complex , L
2
1
() = i
1
A
1
2
+ O([[
3
),
2
() = i
2
A
2
2
+ O([[
3
),
3
() = i
3
A
3
2
+ O([[
3
),
(5.13)
2011] SOLVING BOLTZMANN EQUATION, PART I: GREENS FUNCTION 149
cf. Figure 3, where A
j
= (P
1
1
E
j
, L
1
P
1
1
E
j
) is the NavierStokes
dissipation coecient, cf. (4.36).
In view of (5.13), for 0 < [[ 1, the eigenvalues
1
(),
2
(),
3
()
are distinct. After being normalized according to [e
j
, e
k
] =
jk
, cf.
Remark 5.3, the eigenvectors e
1
(), e
2
(), e
3
() take the following form:
e
j
() =E
j
+ i
_
k=j
A
jk
j
E
k
+ e
j
_
+O([[
2
),
e
j
=L
1
P
1
1
E
j
,
(5.14)
where A
jk
= (P
1
1
E
j
, L
1
P
1
1
E
k
) is the NavierStokes dissipation
coecient, cf. (4.34).
(III) For all 0 < 1, the semigroup e
(i
1
+L)t
can be decomposed as
e
(i
1
+L)t
=
+
{<}
1
2i
_
e
zt
_
z (i
1
+ L)
_
1
dz, (5.15)
where 

L
2
= O(1)e
a()t
, a() > 0 depends on (and therefore on
),
{}
is the indicator function, and can be any close curve that
lies entirely on Re z > and that encloses the three eigenvalues
1
(),
2
(),
3
(), Figure 3.
(IV) Particularly, for the hard sphere model,
j
(), e
j
() are holomorphic
in for all [[ 1. Consequently, (5.13) and (5.14) hold even for
complex .
Proof. The spectrum of i
1
+ L has been studied and this lemma has
been proved, [41], [36], [14], [45], by making use of the spectral gap at origin
for L, Theorem 3.5. Here we emphasize the computation of eigenvalues and
eigenvectors, (5.13) and (5.14), and prove (II) only.
Apply MacroMicro projection to the equation (i
1
+ L)e
j
=
j
e
j
:
iP
0
1
_
(P
0
e
j
) + (P
1
e
j
)
_
=
j
(P
0
e
j
), (5.16a)
iP
1
1
(P
0
e
j
) iP
1
1
(P
1
e
j
) + L(P
1
e
j
) =
j
(P
1
e
j
). (5.16b)
In view of (5.16a), it is convenient to set
j
= i
j
. From (5.16b) we can
solve P
1
e
j
in terms of P
0
e
j
as P
1
e
j
= i[L iP
1
1
i
j
]
1
P
1
1
(P
0
e
j
).
150 TAIPING LIU AND SHIHHSIEN YU [June
3
()
2
()
1
()
Im
Re
Re (z) =
Figure 3: Spectrum near origin.
Substituting this back to (5.16a), we obtain the following equation for P
0
e
j
:
_
P
0
1
+ iP
0
1
_
L iP
1
1
i
j
_
1
P
1
1
_
(P
0
e
j
) =
j
(P
0
e
j
). (5.17)
Through MacroMicro decomposition, we have transformed the original in
nite dimensional eigenequation into the 3 dimensional equation (5.17). Our
next step is to solve (5.17) by the implicit function theorem. We will do this
only for e
1
, since e
2
, e
3
can be treated similarly.
Instead of normalizing e
1
according to [e
1
, e
1
] = 1, we temporarily adopt
the normalization e
1
to e
1
by (E
1
, e
1
) = 1. Set
P
0
e
1
= E
1
+
2
E
2
+
3
E
3
.
Under this setup, (5.17) becomes
G
1
(
1
,
2
,
3
; )
1
+ i
_
P
0
1
(L iP
1
1
i
j
)
1
P
1
1
(E
1
+
2
E
2
+
3
E
3
), E
1
_
+
1
= 0,
G
2
(
1
,
2
,
3
; )
2
2
+ i
_
P
0
1
(L iP
1
1
i
j
)
1
P
1
1
(E
1
+
2
E
2
+
3
E
3
), E
2
_
+
1
2
= 0,
2011] SOLVING BOLTZMANN EQUATION, PART I: GREENS FUNCTION 151
G
3
(
1
,
2
,
3
; )
3
3
+ i
_
P
0
1
(L iP
1
1
i
j
)
1
P
1
1
(E
1
+
2
E
2
+
3
E
3
), E
3
_
+
1
3
= 0. (5.18)
This equation has a solution (
1
,
2
,
3
; ) = (
1
, 0, 0; 0). Moreover,
(G
1
, G
2
, G
3
)
(
1
,
2
,
3
)
(
1
,0,0,0)
=
1 0 0
0 (
2
1
) 0
0 0 (
3
1
)
,= 0.
Consequently, by the implicit function theorem, for [[ 1, (5.18) has
the solution (
1
(),
2
(),
3
()), smooth in , with (
1
(0),
2
(0),
3
(0)) =
(
1
, 0, 0).
Dierentiating (5.18) with respect to yields
(0) =i
_
P
0
1
L
1
P
1
1
E
1
, E
1
_
= iA
1
,
(0) =i
_
P
0
1
L
1
P
1
1
E
1
, E
2
_
1
1
=
iA
12
1
,
(0) =i
_
P
0
1
L
1
P
1
1
E
1
, E
3
_
1
1
=
iA
13
1
.
Hence
1
() = i
1
A
1
2
+O(
3
). Moreover, since P
1
e
1
= i[LiP
1
i
1
]
1
P
1
1
(P
0
e
1
),
e
1
() = E
1
+ i
_
A
12
1
E
2
+
A
13
1
E
3
+ L
1
P
1
1
E
1
_
+ O(
2
).
We now normalize e
1
: put e
1
=
e
[e
1
,e
1
]
. Note that [e
1
, e
1
] is generally a
complex number. By some direct computation, [e
1
, e
1
] = 1 + O(
2
) 1.
We unambiguously refer to
_
[e
1
, e
1
] as the complex square root closer to
1. Under this choice,
_
[e
1
, e
1
] is smooth in , and, in the case of (IV),
holomorphic in . Also we have (5.14):
e
1
() =
_
1 +O(
2
)
_
_
E
1
+i
_
A
12
1
E
2
+
A
13
1
E
3
+ L
1
P
1
1
E
1
__
+ O(
2
)
152 TAIPING LIU AND SHIHHSIEN YU [June
=E
1
+ i
_
A
12
1
E
2
+
A
13
1
E
3
+ L
1
P
1
1
E
1
_
+ O(
2
).
Remark 5.5. The linear collision operator L = + K has the property
that
() 1 +[[
,
where = 1 for the hard sphere models and 0 < < 1 for Grad cuto hard
potentials. The analyticity of the eigenvalues
i
() holds only for the hard
sphere models because of this. To see that, note that
1
+
i
=
_
E
1
, P
0
1
[L iP
1
1
+ i(
1
+ )]
1
P
1
1
(E
1
+ b
0
)
_
The key is the behavior at [[ = :
[L iP
1
1
+ i(
1
+ )]
1
n!
[[
n
n
()
e

2
2
,
for hard sphere model. Hence,
1
+
i
n=1
n(n + 1)(n 1)!
(n + 1)!
n+1
i
1
+
n=1
n+1
,
which converges as [[ 1. On the other hand, for hard potentials
1
+
i
n=1
n!
_
[[
n(1)
e

2
2
d
n=1
(n(1 ))
n+1
,
which diverges for any 0 < < 1 and ,= 0. Thus the complex analytic
method that we will use to obtain the explicit expression of the inverse
Fourier transform for the hard sphere models cannot be applied to the hard
potential models. There is the study for the hard potential models, [28],
which yields a version of the pointwise estimates for the Greens function
weaker than that of Theorem 5.9 for hard sphere models. The real analytic
method used in [28] is motivated by [29] for viscous conservation laws and
[10] for the boundary layers. It would be interesting to obtain, for the hard
potential models, similar estimates as for the hard sphere models in Theorem
5.9.
We now proceed to compute explicitly the second term of (7.21).
2011] SOLVING BOLTZMANN EQUATION, PART I: GREENS FUNCTION 153
Denition 5.6. For L
2
,
_
e
j
[e
j
[
_
[, e
j
]e
j
.
Lemma 5.7. For [[ 1,
1
2i
_
e
zt
_
z (i
1
+ L)
_
1
dz =
3
j=1
e
j
t
e
j
[e
j
[. (5.19)
Proof. Dene a subspace H
of L
2
as
H
: [, e
j
()] = 0, for j = 1, 2, 3 .
Let ) be the linear span. Clearly, L
2
= e
1
, e
2
, e
3
) H
3
1
e
j
[e
j
[ is the
projection onto e
1
, e
2
, e
3
) along H
, and 1
3
1
e
j
[e
j
[ is the projection
onto H
along e
1
, e
2
, e
3
). Moreover, since e
j
is an eigenvector of (i
1
+L)
and (i
1
+ L) is symmetric with respect to [, ], (5.12), both e
1
, e
2
, e
3
)
and H
are (z (i
1
+ L))invariant:
_
z (i
1
+ L)
_
e
1
, e
2
, e
3
) e
1
, e
2
, e
3
) ,
_
z (i
1
+ L)
_
H
.
(5.20)
Set
S
=
_
i
1
+ L
_
3
j=1
e
j
[e
j
[, T
=
_
i
1
+ L
__
1
3
j=1
e
j
[e
j
[
_
.
As a consequent of (5.20), (z S
) : e
1
, e
2
, e
3
) e
1
, e
2
, e
3
), (z T
) :
H
, and (z (i
1
+L))
1
= (z S
)
1
+(z T
)
1
. S
becomes
a 3 by 3 diagonal matrix under the basis e
1
, e
2
, e
3
, so by some direct
computations:
1
2i
_
e
zt
(z S
)
1
dz =
3
j=1
e
j
t
e
j
[e
j
[.
It remains only to show
_
e
zt
(zT
)
1
dz = 0. First, since (i
1
+L)
154 TAIPING LIU AND SHIHHSIEN YU [June
[(1 +[[ + O())
0
+ K], for [[ 1,
(i
1
+ L)
1
= L
1
+O(), (5.21)
where O() symbolizes an operator with O()
L
2
,
[, e
j
()] = 0 = [, E
j
] + [, O()] = (, E
j
) + O()
= (, E
j
) =  O().
Therefore, P
0
= O() . Combine this fact with (5.21) and Lemma 3.5
to obtain
_
_
T
1
_
_
L
2

L
2
=
_
_
_
_
L
1
+O()
__
P
1
+ O()
__
_
_
L
2

L
2
1
P
1

L
2
+ O() 
L
2

L
2
=
1
+ O().
Consequently, T
1
1
,
2
,
3
and
j
= O(), in the region enclosed by we may assume
[z[ [[ 1. Under this assumption, (z T
)
1
= T
1
(1 + z(T
)
1
+
(zT
1
)
2
+ . . .) is holomorphic in z and therefore
_
e
zt
(zT
)
1
dz =0.
5.2. Fluidlike Waves
Plugging (7.21) and (5.19) to (5.7), we have
G(x, t, ;
0
) =
1
2
F
1
+
3
j=1
1
2
_
e
ix+
j
()t
e
j
() [e
j
()[ d,
where F
1
symbolizes the inverse Fourier transform. Since F
1
: L
2
L
2
x
is an isometry,
_
_
F
1
_
_
L
2
x
(L
2
)
= O(1)e
t
C()
, (5.22)
for some C() > 0 depending on .
2011] SOLVING BOLTZMANN EQUATION, PART I: GREENS FUNCTION 155
Denition 5.8. Dene the uidlike waves as
G
L
(x, t, ; )
3
j=1
1
2
_
e
ix+
j
()t
e
j
() [e
j
()[ d. (5.23)
To highlight the dependence on t, we will frequently abbreviate G
L
(x, t, ;
) as G
t
L
.
Our goal of this subsection is to the extract leading terms, the main
uidlike waves, from G
L
.
Theorem 5.9. For any xed C
j
> A
j
, j = 1, 2, 3, there exists C > 0 and
> 0 such that
_
_
_
_
_
_
_
G
L
(x, t, ; )
3
j=1
e
(x
j
t)
2
4A
j
t
_
4A
j
(t + 1)
E
j
E
j
[
_
_
_
_
_
_
_
L
2
= O(1)
_
_
_
3
j=1
e
(x
j
t)
2
4C
j
t
t + 1
+ e
t
C
_
_
_. (5.24)
Note that here our estimate is pointwise in x, unlike (), which is L
2
in
x.
Proof. Pick C
j
> C
j
> C
j
> A
j
. We then x some > 0, so small such
that whenever [[ < 2:
(1)
j
, e
j
are holomorphic in .
(2) (5.13), (5.14), and Lemma 5.7 hold. Moreover,
j
()
_
i
j
A
j
2
_
< min
__
1
A
j
C
j
_
,
1
3
_
A
j
[[
2
. (5.25)
Because of the analyticity property of the eigenvalues and eigenfunc
tions, we can apply the complex analysis method to compute explicitly the
inverse Fourier transform by the contour integral. This will yield the uid
like behavior of the Greens function. By Cauchy integral theorem, we can
substitute the path [, ] by any other path, lying entirely in the ana
lytic region [[ < 2, with the same endpoints. We choose the contour
156 TAIPING LIU AND SHIHHSIEN YU [June
3
Im
Re
Re () = Re () =
Im() = c
Figure 4: Path of integration.
1
+
2
+
3
, Figure 4,
1
=
1
(c) : Re () = , Im() lies between 0 and c ,
2
=
2
(c) : Re () , Im() = c ,
3
=
3
(c) : Re () = , Im() lies between 0 and c ,
(5.26)
where c is a constant, specied later. On
2
, put Re () = u so that = u+ic.
Since
e
ix+
j
()t
= exp
_
A
j
t
_
i
x
j
t
2A
j
t
_
2
(x
j
t)
2
4A
j
t
+ O(
3
)t
_
,
it is desirable to set c =
x
j
t
2A
j
t
. However, in order that
1
+
2
+
3
lies
entirely in [[ < 2, [c[ cannot be arbitrarily large. For this reason, we will
process the integral on
2
in two separated situations.
Case 1:
x
j
t
2A
j
t
<
2
In this case, set c =
x
j
t
2A
j
t
.
On
2
,
e
ix+
j
()t
= e
(x
j
t)
2
4A
j
t
e
A
j
tu
2
+e
(x
j
t)
2
4A
j
t
e
A
j
tu
2
_
e
O(
3
)t
1
_
.
By (5.25),
e
(x
j
t)
2
4A
j
t
e
A
j
tu
2
_
O(1)[[
3
t, for [[
3
t < 1
e
O(
3
)t
, for [[
3
t > 1
_
= O(1)e
(x
j
t)
2
4A
j
t
e
A
j
tu
2
e
O(
3
)t
[[
3
t
2011] SOLVING BOLTZMANN EQUATION, PART I: GREENS FUNCTION 157
= O(1) exp
_
(x
j
t)
2
4A
j
t
A
j
tu
2
+ min
__
1
A
j
C
j
_
,
1
3
_
tA
j
_
u
2
+
_
x
j
t
2A
j
t
_
2
__
[[
3
t
= O(1) exp
_
(x
j
t)
2
4A
j
t
A
j
tu
2
+
_
1
A
j
C
j
_
(x
j
t)
2
4A
j
t
+
A
j
tu
2
3
_
_
[u[
3
t +
[x
j
t[
3
t
2
_
= O(1)e
(x
j
t)
2
4C
j
t
e
A
j
tu
2
3
_
[u[ +
x
j
t
t
_
.
Therefore,
1
2
_
2
e
ix+
j
()
e
j
() [e
j
()[d
=
1
2
_
(x
j
t)
2
4A
j
t
e
u
2
tA
j
_
E
j
E
j
[ +[[O(1)
_
du
+
1
2
_
(x
j
t)
2
4C
j
t
e
u
2
t
A
j
3
_
[u[ +
x
j
t
t
_
e
j
() [e
j
()[du
=
e
(x
j
t)
2
4A
j
t
4
_
1
e
A
j
tu
2
du
_
E
j
E
j
[
+e
(x
j
t)
2
4C
j
t
_
e
u
2
t
A
j
4
1
t + 1
O(1)du
=
_
_
_
e
(x
j
t)
2
4A
j
t
_
4A
j
t
E
j
E
j
[ +
e
(x
j
t)
2
4C
j
t
_
(t + 1)t
O(1)
_
_
_
_
1
A
j
t
A
j
t
e
u
2
du
_
=
e
(x
j
t)
2
4A
j
t
_
4A
j
(t + 1)
E
j
E
j
[
+e
(x
j
t)
2
4A
j
t
_
1
t
_
1
A
j
t
A
j
t
e
u
2
du
_
t + 1
_
O(1)
+
e
(x
j
t)
2
4C
j
t
_
(t + 1)
_
1
t
_
A
j
t
A
j
t
e
u
2
du
_
O(1)
158 TAIPING LIU AND SHIHHSIEN YU [June
=
e
(x
j
t)
2
4A
j
t
_
4A
j
(t + 1)
E
j
E
j
[ +
e
(x
j
t)
2
4C
j
t
t + 1
O(1).
Case 2:
x
j
t
2A
j
t
>
2
In this case, set c =
2
sign (x
j
t).
Note that in this case we have [x
j
t[ > A
j
t. On
2
, by (5.25),
e
ix+
j
()t
exp
_
i(x
j
t) A
j
t
2
+
A
j
3
[[
2
t
_
=exp
_
[x
j
t[
2
+ A
j
t
_
2
4
u
2
_
+
A
j
t
3
_
2
4
+ u
2
__
exp
_
A
j
t
_
2
2
2
3
__
= O(1)e
t
C
.
Consequently,
1
2
_
2
e
ix+
j
()
e
j
() [e
j
()[d = O(e
t
C
).
For both cases, on
1
,
3
, Im() takes the same sign as x
j
t. There
fore, [e
i(x
j
t)
[ 1. Moreover, for both cases, [Im()[ [c[
2
. Combin
ing this fact with (5.25), we have
e
i(x
j
)t
2
A
j
t+O(
3
)t
exp
_
A
j
t
_
2
4
_
+
A
j
t
3
_
2
+
2
4
__
= O(1)e
t
C
.
This implies
1
2
_
1
or
3
e
ix+
j
()
e
j
() [e
j
()[d = O(e
t
C
).
Remark 5.10. The computation of the inverse Fourier transform is an es
sential part of the study of the Greens function here. The approach goes
back to Zeng 1994, [48], LiuZeng 1997, [35] for the viscous conservation
laws. This is generalized to the Boltzmann equation, LiuYu 2004 [31]. For
2011] SOLVING BOLTZMANN EQUATION, PART I: GREENS FUNCTION 159
the generalization to the 3D case, LiuYu 2006 [32], to be presented later,
the inversion of the Fourier transform using the complex analytic method re
quires additional thinking. There is an essential dierence between Greens
function for the viscous conservation laws and that for the Boltzmann equa
tion in that the former contains heat kernel with singularity, c.f. (4.39), while
the laters singularity resides not in the heat kernels, (5.24). Instead, the
singularity for the Boltzmann equation is contained in the essential kinetic
waves that will be constructed in the next section.
5.3. Scale separations
Boltzmann equation has much richer wave phenomena than the equa
tions for the uid dynamics. We illustrate here a basic separation of scales
property, that the micro part decays at a faster rate than the macro part.
Theorem 5.11. For any xed C
j
> A
j
, j = 1, 2, 3, there exists C and
such that
G
L
(x, t, ; )P
1

L
2
, P
1
G
L
(x, t, ; )
L
2
=O(1)
_
_
_
3
j=1
e
(x
j
t)
2
4C
j
t
t + 1
+ e
t
C
_
_
_,
(5.27a)
P
1
G
L
(x, t, ; )P
1

L
2
=O(1)
_
_
_
3
j=1
e
(x
j
t)
2
4C
j
t
(t + 1)
3
2
+ e
t
C
_
_
_.
(5.27b)
Proof. Since P
1
(E
j
E
j
[) = (E
j
E
j
[) P
1
= 0, Theorem 5.9 immediately
implies (5.27a).
For (5.27b), after referring to the proof of Theorem 5.9, one nds that:
In Case 1,
1
2
_
1
or
3
e
ix+
j
()
e
j
() [e
j
()[d = O(e
t
C
),
In Case 2,
1
2
_
1
+
2
+
3
e
ix+
j
()
e
j
() [e
j
()[d = O(e
t
C
).
160 TAIPING LIU AND SHIHHSIEN YU [June
Now that P
1
O(e
t/C
)P
1
= O(e
t/C
), it remains only to consider
Case 1:
1
2
_
2
e
ix+
j
()
P
1
_
e
j
() [e
j
()[
_
P
1
d.
P
1
e
j
= O() implies P
1
_
e
j
() [e
j
()[
_
P
1
= O(
2
). Therefore,
1
2
_
2
e
ix+
j
()
P
1
_
e
j
() [e
j
()[
_
P
1
d =
_
2
e
ix+
j
()
[[
2
dO(1)
=
_
(x
j
t)
2
4C
j
t
e
u
2
t
A
j
3
_
u
2
+
_
x
j
t
t
_
2
_
duO(1)
= e
(x
j
t)
2
4C
j
t
_
1
t + 1
e
u
2
t
A
j
4
duO(1)
=
e
(x
j
t)
2
4C
j
t
(t + 1)
3
2
O(1).
6. 1D Greens Function, ParticleLike Waves
The Boltzmann equation is the mesoscopic equation, being between the
microscopic interacting particle systems and the macroscopic uid dynamics
equations. In the last section we have shown that the uidlike waves for
the Boltzmann solution have as their leading terms closely related to the
Euler and NavierStokes waves. In this section we nish the construction of
the Greens function by considering the short waves in the Greens function.
However, because the spectrum the short waves represent is not explicit,
the construction is indirect and elaborate, starting with the construction of
the singular waves. These waves contains particle wave, particlelike waves,
essential kinetic waves and a series of increasingly smooth waves. We now
outline the denition of these waves. For this, we need the following solution
operators, c.f. (3.8):
Denition 6.1. Denote by S
t
and O
t
D
the solution operators of the equa
tions
_
h
t
+
1
x
h + ()h = 0,
h(x, 0) = h
0
(x), x R,
(6.1)
2011] SOLVING BOLTZMANN EQUATION, PART I: GREENS FUNCTION 161
h(x, t) S
t
h
0
(x),
and
_
j
t
+
1
x
j + ()j = K
0
j,
j(x, 0) = j
0
(x), x R,
(6.2)
j(x, t) O
t
D
j
0
(x).
These operators are studied in the rst subsection. For the complete
construction of the singular waves, the rst step is to use the damped trans
port operator (6.1) and the essential kinetic operator (6.2) to extract the
particlelike waves from the Greens function. Recall the Greens function,
(2.17),
G
t
+
1
x
G+ ()G = KG, x R, R
3
,
G(x, 0, ;
0
) = (x)
3
(
0
), x R.
The rst term is the particle waves dened by
_
h
0
t
+
1
x
h
0
+ ()h
0
= 0,
h
0
(x, 0) = (x)
3
(
0
).
(6.3)
We then dene the particlelike waves h
j
, j = 1, 2, . . . , as follows: We have
(Gh
0
)
t
+
1
x
(Gh
0
) + ()(G h
0
) = K(G h
0
) + Kh
0
,
(Gh
0
)(x, 0, ;
0
) = 0.
Thus we dene the second term h
1
by
_
h
1
t
+
1
x
h
1
+ ()h
1
= Kh
0
= KS
t
(x)
3
(
0
),
h
1
(x, 0) = 0.
(6.4)
In general, the particlelike waves are dened through the Picard iterations
as:
_
h
j
t
+
1
x
h
j
+ ()h
j
= Kh
j1
,
h
j
(x, 0) = 0, j = 1, 2, . . . .
(6.5)
162 TAIPING LIU AND SHIHHSIEN YU [June
The Greens function minus these particlelike waves
g
k
G
k
i=0
h
i
(6.6)
satises
_
g
kt
+
1
g
kx
= Lg
k
+ Kh
k
,
g
k
(x, 0, ) = 0.
(6.7)
The source Kh
k
is bounded for k 2, Lemma 6.9; and as it turns out, this
is so also for the 3 D case. We next use the essential kinetic operator O
t
D
,
(6.2), (3.8), to construct an essential kinetic wave
h:
_
h
t
+
1
h
x
+ ()
h = K
0
h + Kh
2
,
h(x, 0, ) = 0.
(6.8)
The function
h is named the essential kinetic wave as it is localized. More
over, the new source K
1
_
(Gh
0
h
1
h
2
h)
t
+
1
(Gh
0
h
1
h
2
h)
x
= L(Gh
0
h
1
h
2
h) + K
1
h,
(Gh
0
h
1
h
2
h)(x, 0, ) = 0.
(6.9)
With the source smooth in microscopic velocity, we use the Picard iter
ations, c.f. (6.5), again:
_
_
g
0
t
+
1
x
g
0
+ ()g
0
= K
1
h,
g
j+1
t
+
1
x
g
j+1
+ ()g
j+1
= Kg
j
,
g
j
(x, 0) = 0, j = 0, 1, 2, . . . .
(6.10)
This is the nal step in the construction of the singular waves. It is shown
by a Mixture Lemma that the smoothness of the source in induces an
increasingly more smooth functions g
j
, as j increases.
The construction in the last section of the uidlike waves and the con
struction of singular waves in this section oer two decompositions of the
Greens function. This allows us to apply the energy method and Sobolev
2011] SOLVING BOLTZMANN EQUATION, PART I: GREENS FUNCTION 163
calculus to gain global pointwise analysis of the Greens function.
6.1. Essential kinetic operators
The operator S
t
, (6.1), is a damped transport equation and is solved by
the characteristic method:
S
t
h
0
(x, ) = e
()t
h
0
(x
1
t, ). (6.11)
From this we have immediately the following lemma.
Lemma 6.2. For any 0, the operator S
t
satises
_
_
_
S
t

L
x
(L
,
)
e
0
t
,
S
t

L
2
x
(L
2
)
e
0
t
.
(6.12)
The operator O
t
D
, (6.2), captures the singular part of the linear Boltz
mann operator.
Lemma 6.3. There exist positive constants D
0
and C
1
such that for any
D (0, D
0
) the operator O
t
D
satises
O
t
D

L
2
x
(L
2
)
C
1
e
0
t/2
.
Proof. First, we regard O
t
D
as an operator on L
2
x
(L
2
j(, t) = e
(i
1
()+K
0
)t
g
0
().
Since the operator K
0
, (3.8), is symmetry and K
0

L
2
0
+ O(1)D <
0
/2. This implies that there exist positive constants
C
1
, D
0
such that, for D (0, D
0
)
e
(i
1
()+K
0
)t

L
2
C
1
e
0
t/2
for any R.
From this
O
t
D
g
0

2
L
2
x
(L
2
)
=F(O
t
D
g
0
)
2
L
2
(L
2
)
(C
1
)
2
e
0
t
 g
0

2
L
2
(L
2
)
=(C
1
)
2
e
0
t
g
0

2
L
2
x
(L
2
)
.
This lemma results in the existence of the operator O
t
D
in the functional
space L
2
x
(L
2
x
(L
,
) for any
0. There exist positive constants D
0
, C
1
such that, for any D (0, D
0
),
O
t
D
g
0

L
x
(L
,
)
C
1
e
0
t/2
g
0

L
x
(L
,
)
.
Proof. From Lemma 3.2
K
0
h
L
,+1
C
Dh
L
,
for 0.
This and Lemma 6.3 yield
_
_
_
_
_
t
0
_
s
k
0
S
ts
1
K
0
S
s
1
s
2
K
0
S
s
2
s
3
K
0
S
s
k
s
k+1
K
0
S
s
k+1
ds
k+1
ds
1
_
_
_
_
L
x
(L
,
)
(C
)
k+1
D
k+1
e
0
t/2
.
Thus, Picards iteration gives a convergent geometric sequence in L
x
(L
,
)
for suciently small D > 0:
O
t
D
= S
t
+
_
t
0
S
ts
1
K
0
S
s
1
ds
1
+
_
t
0
_
s
1
0
S
ts
1
K
0
S
s
1
s
2
K
0
S
s
2
ds
2
ds
1
+
+
_
t
0
_
s
k
0
S
ts
1
K
0
S
s
1
s
2
K
0
S
s
2
s
3
K
0
S
s
k
s
k+1
K
0
S
s
k+1
ds
k+1
ds
1
+ , (6.13)
2011] SOLVING BOLTZMANN EQUATION, PART I: GREENS FUNCTION 165
and the lemma follows.
The following lemma yields the signicant hyperbolic property of the
operator S
t
and O
t
D
. We will use the crucial property that, for the hard
sphere model, (3.5),
1
2
()
1
[[ as [[ and so, for some positive
constant ,
() = O(1)(1 +[[), [
1
t[ ()t for R
3
. (6.14)
Lemma 6.5. For any given 0, there exists suciently small D > 0
such that
S
t
g
0
(x)
L
,
O(1)e
2
0
t/3
_
max
yx<t
g
0
(y)
L
,
+ max
yx>t
e
yx/3
g
0
(y)
L
,
_
, (6.15)
O
t
D
g
0
(x)
L
,
O(1)e
0
t/2
_
max
yx<t
g
0
(y)
L
,
+ max
yx>t
e
yx/4
g
0
(y)
L
,
_
, (6.16)
where
0
is given in (3.22) and in (6.14).
Proof. We use the representation (6.11) for S
t
.
For [
1
[ 1, we have from (6.14),
[S
t
g
0
(x, )[ e
()t
(1 +[[)
g
0
(x
1
t, )
L
,
e
2
0
t/3
(1 +[[)
max
yx<t
g
0
(y)
L
,
. (6.17)
From (6.14), for [
1
[ > 1
[S
t
g
0
(x, )[ e
()t/32
0
t/3
(1 +[[)
g
0
(x
1
t, )
L
,
e

1
t/32
0
t/3
(1 +[[)
g
0
(x
1
t, )
L
,
max
yx>t
e
xy/32
0
t/3
(1 +[[)
g
0
(y)
L
,
. (6.18)
The estimate (6.15) follows from (6.17) and (6.18). From the construction
of O
t
D
in Lemma 6.4, one can view O
t
D
as a small perturbation of S
t
. From
166 TAIPING LIU AND SHIHHSIEN YU [June
(3.10) and (6.15), one has that
_
_
_
_
_
t
0
_
s
k
0
S
ts
1
K
0
S
s
1
s
2
K
0
S
s
2
s
3
K
0
S
s
k
s
k+1
K
0
S
s
k+1
g
0
ds
k+1
ds
1
_
_
_
_
L
,
(C
D)
k+1
e
0
t/2
_
max
yx<t
g
0
(y)
L
,
+ max
yx>t
e
1
yx/4
g
0
(y)
L
,
_
.
(6.19)
Thus, the Picards iteration in (6.13) converges for suciently small D > 0
and (6.16) follows.
6.2. Particlelike waves
In this section we compute and estimate the particlelike waves h
0
, h
1
, h
2
,
(6.3), (6.4). First, by (6.11), we have
h
0
(x, t, ;
0
) = e
()t
1
(x
1
t)
3
(
0
). (6.20)
From (6.20) we compute h
1
as the following:
Lemma 6.6. Set t
1
t
(x
1
0
t)
1
0
. We have
h
1
(x, t, ;
0
) =
_
_
_
e
()(tt
1
)(
0
)t
1
K(,
0
)
(
1
1
0
)
, for (
x
t
1
)(
x
t
1
0
) < 0,
0 , otherwise.
(6.21)
In particular,
[h
1
(x, t, ;
0
)[ = O(1)
_
_
_
e
0
t
2
0
x
2

1
1
0

K(,
0
), for [
1
1
0
[
x
1
0
t
t
,
0 , otherwise.
(6.22)
Proof. By the Duhamels principle, (6.4), and (6.11), we have
h
1
(x, t; ,
0
) =
_
t
0
e
(
0
)s
e
()(ts)
K(,
0
)(x
1
(ts)
1
0
s)ds. (6.23)
The above integral is zero unless the characteristic line through (x, t) and
2011] SOLVING BOLTZMANN EQUATION, PART I: GREENS FUNCTION 167
y
s
dy
ds
=
1
0
dy
dt
=
1
(x, t)
s = t
1
Figure 5: Characteristics.
with speed
1
intersects with the source (x
1
0
s) at some time s = t
1
during the time period (0, t), Figure 5,
x
1
(t t
1
) =
1
0
t
1
.
This yields the expression (6.21). Note that we have
[x[ = [
1
(t t
1
) +
1
0
t
1
[ [
1
[(t t
1
) +[
1
0
[t
1
,
and so the estimate (6.22) follows from the linear growth
0
(1 + [[) of the
function (), (3.5).
Lemma 6.7.
K(h
1
)(x, t, )
= O(1)e
0
2
(t+x)
e

0

2
32
_
1 +
log
1
0
2
+
log
x
1
0
t
t
x
1
0
t
t
<1
_
, (6.24)
where
{}
is the indicator function.
Proof. From the explicit expression (3.5), we obtain
K(,
) = O(1)
1
[
[
e

2
9
. (6.25)
168 TAIPING LIU AND SHIHHSIEN YU [June
Combining this with the estimate (6.22) for h
1
, we have
K(h
1
)(x, t, )
= O(1)e
0
2
(t+x)
_
[
1
1
0
[>
x
1
0
t
t
1
[
1
1
0
[
e

0

2
9
[
0
[
e

2
9
[
[
d
= O(1)e
0
2
(t+x)
e
2
2
+2
0

2
32
_
[
1
1
0
[>
x
1
0
t
t
1
[
1
1
0
[
e
(
1
9
1
16
)
0

2
[
0
[
e
(
1
9
1
16
)
2
[
[
d
= O(1)e
0
2
(t+x)
e

0

2
32
_

1
1
0
>
x
1
0
t
t
1
[
1
1
0
[
e
(
1
9
1
16
)
0

2
[
0
[
e
(
1
9
1
16
)
2
[
[
d
= O(1)e
0
2
(t+x)
e

0

2
32
__
D
1
+
_
D
2
_
(. . .)
O(1)e
0
2
(t+x)
e

0

2
32
(I
1
+ I
2
),
where
T
1
_
1 > [
1
1
0
[ >
[x
1
0
t[
t
_
, T
2
_
[
1
1
0
[ > 1
_
.
We further split I
1
into three parts:
I
1
=
__
A
1
+
_
A
2
+
_
A
3
_
1
[
1
1
0
[
e
(
1
9
1
16
)
0

2
[
0
[
e
(
1
9
1
16
)
2
[
[
d
I
11
+ I
12
+ I
13
,
where
/
1
_
T
1
: [(
2
2
0
)
2
+ (
3
3
0
)
2
[ < 1 &[
0
[ < [
[
_
,
/
2
_
T
1
: [(
2
2
)
2
+ (
3
3
)
2
[ < 1 &[
0
[ > [
[
_
,
/
3
T
1
_
/
1
/
2
_
.
2011] SOLVING BOLTZMANN EQUATION, PART I: GREENS FUNCTION 169
For I
11
,
[I
11
[
_
A
1
1
[
1
1
0
[
1
[
0
[
1
[
[
d
_
A
1
1
[
1
1
0
[
1
[
0
[
2
d
2
_
1
x
1
0
t
t
_
1
0
_
2
0
1
z
r
z
2
+ r
2
ddrdz =
_
1
x
1
0
t
t
4
z
_
1
0
r
z
2
+ r
2
drdz
_
1
x
1
0
t
t
4
[ log z[
z
dz = 4
_
log
x
1
0
t
t
_
2
,
where we use the cylindrical coordinates:
r =
_
(
2
2
0
)
2
+ (
3
3
0
)
2
, z = [
1
1
0
[, = tan
1
2
2
0
3
0
.
Next, we compute I
12
under the following three dierent conditions:
2 < [
1
1
0
[, 1 < [
1
1
0
[ < 2, and [
1
1
0
[ < 1. For [
1
1
0
[ > 2, since
/
2
= , I
12
= 0. For 1 < [
1
1
0
[ < 2,
[I
12
[
_
A
2
2
[
1
1
0
[[
[
d
. (6.26)
Without lost of generality, assume 1 <
1
1
0
< 2. We consider the following
cylindrical coordinates
r =
_
(
2
2
)
2
+ (
3
3
)
2
, z =
1
1
0
, = tan
1
2
3
.
Thus,
[I
12
[
_
1
x
1
0
t
t
_
4
0
_
1
0
2r
[z[
_
r
2
+ (z (
1
1
0
))
2
drddz
_
1
x
1
0
t
t
2
[z[
dz 4
log
x
1
0
t
t
4
_
1 +
log
x
1
0
t
t
2
_
.
(6.27)
170 TAIPING LIU AND SHIHHSIEN YU [June
For a = [
1
1
0
[ 1, we have
[I
12
[
_
A
2
1
[
1
1
0
[[
[[
0
[
d
_
A
2
1
[
1
1
0
[[
[
2
d
_
1
x
1
0
t
t
_
2
0
_
1
0
r
[z[(r
2
+ (z a)
2
)
drddz
_
1

x
1
0
t
t

log [z a[
z
dz. (6.28)
We shall estimate the integral on ([
x
1
0
t
t
[,
a
2
), and (
a
2
, 2a), (2a, 1) separately.
In the rst interval,
_ a
2

x
1
0
t
t

log [z a[
z
dz (2 +[ log a[)
_ a
2

x
1
0
t
t

1
z
dz
(2 +[ log a[)
log [
x
1
0
t
t
[
C(1 +[ log a[
2
+
log [
x
1
0
t
t
[
2
).
In the second interval, we let =
z
a
. Thus,
_
2a
a
2
log [z a[
z
dz =
_
2
1
2
log [a(1 )[
d
2 log 2[ log a[ +
_
2
1
2
log [1 [
d
C(1 +[ log a[
2
).
For the last interval,
_
1
2a
log [z a[
z
dz [ log a[
_
1
2a
1
z
dz C[ log a[
2
. (6.29)
Therefore, we can conclude
I
12
= O(1)
_
1 +
log
1
0
2
+
log
x
1
0
t
t
2
_
. (6.30)
2011] SOLVING BOLTZMANN EQUATION, PART I: GREENS FUNCTION 171
Similarly,
[I
13
[
_
A
3
1
[
1
1
0
[
e
5
144

0

2
e
5
144

2
d
= O(1)
_
1

x
1
0
t
t

1
z
dz = O(1)
_
log [
x
1
0
t
t
[
_
= O(1)
_
1 +
log [
x
1
0
t
t
[
2
_
. (6.31)
In T
2
, note that [
0
[ > [
1
1
0
[ > 1 . Thus,
[I
2
[
_
D
2
1
[
[
e
5
144

2
d
= O(1). (6.32)
Combine all the inequalities above we can conclude (6.24).
Apply the Duhamel principle to (6.5), we obtain a integral representa
tion of h
2
:
h
2
(x, t, ) =
_
t
0
e
()(ts)
K(h
1
)(x (t s)
1
, s, )ds. (6.33)
Lemma 6.8.
h
2
(x, t, ) = O(1)e
0
3
(t+x)
e

0

2
32
(1 +
log [
1
1
0
[
2
). (6.34)
Proof. Using the estimate (6.24), we obtain
[h
2
(x, t, )[
_
t
0
O(1)e
()(ts)
e
0
2
(s+x
1
(ts))
e

0

2
32
_
1 +
log
1
0
2
+
log
x
1
0
t + (
1
1
0
)s
s
x
1
0
t+(
1
1
0
)s
s
<1
_
ds
e
0
2
(t+x)
e

0

2
32
_
t
0
_
1 +
log
1
0
2
+
log
x
1
0
t + (
1
1
0
)s
s
2
172 TAIPING LIU AND SHIHHSIEN YU [June
x
1
0
t+(
1
1
0
)s
s
<1
_
ds. (6.35)
For simplicity, put = (
1
1
0
), = x
1
t. The problem is now
reduced to the estimate of the following integral
R =
_
t
0
log
x
1
0
t + (
1
1
0
)s
s
x
1
0
t+(
1
1
0
)s
s
<1
ds
=
_
t
0
log
s
s
[
s
s
[<1
ds. (6.36)
In order to prove the lemma, it is enough to show R = O(1)t(1 +
log [[
2
).
Since we are considering (6.36), in order that ,= 0, we require
s < s < s. (6.37)
Express (6.37) in seven dierent cases, classied by the signs of , , Figure
6:
_
_
1. 0 < s < , for = 0, 1 < < 1,
2.
+1
< s <
1
, for > 0, > 1,
3.
+1
< s < , for > 0, 0 < < 1,
4.
+1
< s < , for > 0, 1 < < 0,
5.
+1
< s <
1
, for < 0, < 1,
6.
+1
< s < , for < 0, 1 < < 0,
7.
+1
< s < , for < 0, 0 < < 1.
For case 1, we can evaluate the integral (6.36) directly and obtain
R =
log [[
2
t.
Case 5, 6, and 7 are similar to case 2, 3, and 4 respectively, so, without lost
of generality, we consider only case 2, 3, and 4.
2011] SOLVING BOLTZMANN EQUATION, PART I: GREENS FUNCTION 173
s
+1
s
s
s
Figure 6: range of s.
For case 2, since > 1,
2
<
+1
. Therefore,
R
_
t
log
s
s
2
ds
_
t
2
2
_
log
_
2
+ 2
log
s
1
2
ds
2t
_
log
_
2
+ 2
_
_ 2
n
t>
2
o
+
_
t
2
n
t>
2
o
_
log
s
1
2
ds
2t
_
log
_
2
+
n
t>
2
o
_
_
2
_
2
1
2
_
log
1
z
1
_
2
dz + 2
_
t
2
log
s
1
2
ds
_
_
=2t
_
log
_
2
+
n
t>
2
o
_
O(1)
+ 2t
_
log 2
_
2
_
= O(1)t(1 + (log )
2
).
For case 3, since 0 < < 1,
R
_
t
2
_
log
_
2
ds
=
_
_
min
n
t,
2
o
2
+
_ 2
n
t>
2
o
+
_
t
2
n
t>
2
o
_
_
log
_
2
ds
174 TAIPING LIU AND SHIHHSIEN YU [June
_
min
n
t,
2
o
2
_
log
_
2
ds + O(1)t
_
1 + (log )
2
_
.
For
2
< s <
2
, 0 < <
s
< 2. Therefore,
R
_
min
n
t,
2
o
2
_
log
_
2
ds + O(1)t
_
1 + (log )
2
_
t
_
2(log 2)
2
+ 2(log )
2
_
+ O(1)t
_
1 + (log )
2
_
= O(1)t
_
1 + (log )
2
_
.
For case 4,
1 [[
< s = [[ < [[ +
s
< 1.
Therefore,
R
_
t
1
log
_
[[ +
s
_
2
ds t
_
log [[
_
2
.
The next lemma shows that the source K(h
2
) for the next Picard iter
ation (6.5) is nite. That the source is gaussian in microscopic velocity is
an interesting fact, which follows from the detailed estimates in the above
lemma.
Lemma 6.9.
[K(h
2
)(x, t, )[ Ce
0
3
(t+x)
e

0

2
128
. (6.38)
Proof. Applying the estimate for h
2
, Lemma 6.8, and the estimate (6.25)
for K, we obtain
[K(h
2
)(x, t, )[ = O(1)
_
R
3
1
[
[
e

2
9
e
0
3
(t+x)
e

0

2
32
_
1 +
log [
1
1
0
[
2
_
d
= O(1)e

0

2
128
e
0
3
(t+x)
_
R
3
1
[
[
e
(
1
9
1
64
)
2
(1 +
log [
1
1
0
[
2
)e

0

2
64
d
.
2011] SOLVING BOLTZMANN EQUATION, PART I: GREENS FUNCTION 175
By CauchySchwarz inequality,
_
R
3
1
[
[
e
(
1
9
1
64
)
2
(1 +
log [
1
1
0
[
2
)e

0

2
64
d
__
R
3
1
[
[
2
e
2(
1
9
1
64
)
2
d
_1
2
__
R
3
(1 +
log [
1
1
0
[
2
)
2
e

0

2
32
d
_1
2
.
This completes the proof of the lemma.
6.3. Essential kinetic waves
With the estimate of the source K(h
2
) in Lemma 6.9, we now study the
essential kinetic wave
h, (6.8).
Lemma 6.10. For any 0, there exists D > 0 such that

h
L
,
= O(1)e
1
(x+t)
D
. (6.39)
Proof. With the operator O
t
D
of (6.2), we have by Duhamel principle that
h =
_
t
0
O
ts
D
Kh
2
(s)ds =
_
t
0
O
ts
D
O(1)e
0
3
(s+x)
e

0

2
128
ds,
where we have used Lemma 6.9 for Kh
2
. The estimate (6.39) now follows
from Lemma 6.5 on the exponential decay in time and the hyperbolicity
property of the operator O
D
. We omit the details.
With this we have
_
g
t
+
1
g
x
+ ()g = Kg + K
1
h,
g(x, 0, ) = 0;
(6.40)
g G[h
0
+ h
1
+ h
2
]
h. (6.41)
We have the following estimate on the smoothness in microscopic velocity
for the source K
1
h.
176 TAIPING LIU AND SHIHHSIEN YU [June
Lemma 6.11. For any xed natural number l 0,

l
(K
1
h)
L
2
x
(L
2
)
= O(1)e
1
t
3D
. (6.42)
Proof. From Lemma 3.3, we have K
1
h
H
i
= O(1)h
L
2
,
= O(1)e
1
(x+t)
D
j
. (6.43)
Proof. This is proved by induction on j and uses Lemma 6.5. The procedure
is similar to the proof of Lemma 6.10; details are omitted.
We have, c.f. (6.9),
_
_
(G
2
i=0
h
i
j
i=0
g
i
)
t
+
1
(G
2
i=0
h
i
j
i=0
g
i
)
x
= L(G
2
i=0
h
i
j
i=0
g
i
) + Kg
j
,
(G
2
i=0
h
i
j
i=0
g
i
)(x, 0, ) = 0.
(6.44)
Using the notion of damped transport operator S and the integral operator
K, the Picard iteration (6.10) has the representation:
g
j+1
=
_
t
0
S
ts
1
Kg
j
(s
1
)ds
1
=
_
t
0
_
s
1
0
S
ts
1
KS
s
1
s
2
Kg
j1
(s
2
)ds
2
ds
1
.
and so inductively we have
g
j
=
_
t
0
_
s
1
0
_
s
j
0
S
ts
1
KS
s
1
s
2
KS
s
2
s
3
K S
s
j
s
j+1
K
1
h(s
j+1
)
ds
j+1
ds
1
, j = 0, 1, 2, . . . . (6.45)
A major point is that the above repeated convolutions of the damped trans
port operator S and the integral operator K yield the smoothness in (x, t) as
2011] SOLVING BOLTZMANN EQUATION, PART I: GREENS FUNCTION 177
a consequence of the smoothness of the source K
1
_
t
0
_
s
1
0
_
s
2k1
0
S
ts
1
KS
s
1
s
2
KS
s
2
s
3
K S
s
2k1
s
2k
KS
s
2k
g
0
ds
2k
ds
1
.
(6.47)
The Picard iteration for solving the initial value problem
t
h +
1
x
h
Lh = 0, h(x, 0) h
0
(x), can be rewritten as
h(x, t) = S
t
h
0
(x) +
_
t
0
S
ts
KS
s
h
0
ds(x) +M
t
1
h
0
(x)
+
j=1
__
t
0
S
ts
KM
s
j
h
0
ds(x) +
_
t
0
_
t
1
0
S
tt
1
KS
t
1
t
2
KM
t
2
j
h
0
dt
2
dt
1
(x)
_
.
(6.48)
This series is an asymptotic expansion, useful for identifying the singular
parts of the solution.
178 TAIPING LIU AND SHIHHSIEN YU [June
Remark 6.14. The Mixture Lemma states that the mixture of the two op
erators S and K in M
t
k
transports the regularity in the microscopic velocity
to the regularity of the space time (x, t). The Mixture Lemma is similar in
spirit as the wellknown Averaging Lemma, [15], [16], [3], [27], (see also [40]
for the gliding regularity for the nondissipative Landau damping.) These
two Lemmas have been introduced independently and used for dierent pur
poses.
Lemma 6.15 (Mixture Lemma). There exist C
k
> 0, k = 1, 2, . . . , such
that
_
_
_
k
x
M
t
k
g
0
_
_
_
L
2
x
(L
2
)
C
k
e
0
t
2
k
l=0
_
_
_
l
1
g
0
_
_
_
L
2
x
(L
2
)
. (6.49)
Proof. In order to explain the main ideas, we will go to some details for the
cases of k = 1 and k = 2. The proof uses the characteristic method, [31],
[26]. For k = 1, we write down the explicit form of the Mixture operator by
spelling out the operator S
t
h(x, ) = e
()t
h(x
1
t, ), (6.11):
M
t
1
g
0
(x, ) =
_
t
0
_
s
1
0
_
R
3
_
R
3
e
()(ts
1
)(
1
)(s
1
s
2
)(
2
)s
2
K (,
1
) K (
1
,
2
)
g
0
(x
1
(t s
1
)
1
1
(s
1
s
2
)
1
2
s
2
,
2
)d
2
d
1
ds
2
ds
1
. (6.50)
Set
A(,
1
,
2
, t, s
1
, s
2
) e
()(ts
1
)(
1
)(s
1
s
2
)(
2
)s
2
K (,
1
) K (
1
,
2
) ,
(6.51)
z x
1
(t s
1
)
1
1
(s
1
s
2
)
1
2
s
2
,
and we have
x
M
t
1
g
0
(x, ) =
_
t
0
_
s
1
0
_
R
3
_
R
3
A(,
1
,
2
, t, s
1
, s
2
)
x
g
0
(z,
2
)d
2
d
1
ds
2
ds
1
.
(6.52)
Next, we use the change of variables and the chain rule:
V
1
1
, V
2
1
2
,
V
1
1
g
0
(z,
2
) = s
1
x
g
0
(z,
2
)
1
2
g
0
(z,
2
). (6.53)
2011] SOLVING BOLTZMANN EQUATION, PART I: GREENS FUNCTION 179
By the switching the order of dierentiations and applying the integration
by parts, we obtain
x
M
t
1
g
0
(x, )
=
_
t
0
_
s
1
0
_
R
3
_
R
3
1
s
1
_
A(, V
1
, V
1
V
2
, t, s
1
, s
2
)
1
2
g
0
(z,
2
)
g
0
(z,
2
)
V
1
1
A(, V
1
, V
1
V
2
, t, s
1
, s
2
)
_
dV
2
dV
1
ds
2
ds
1
. (6.54)
To estimate the L
2
norm of
x
M
t
1
g
0
(x, ), we only need the integrability
of
1
s
1
_
A+
V
1
1
A
_
. Note that, as the result of the above switching of the
order of dierentiation, it yields the integrability
_
t
0
_
s
1
0
1
s
1
ds
2
ds
1
= t. If we
were to switch the dierentiation with respect to x with the dierentiation
with respect to V
1
2
, then we would get the nonintegrable factor
1
s
2
, instead
of
1
s
1
. This would implies that the characteristic method only works for the
short time scale. On the other hand, we now have the partial dierentiation
of the function A. The second point is then to show the integrability of
V
1
1
A. We have from (6.51)
V
1
1
A e
3
4
0
t
K(
1
,
2
)
1
1
K(,
1
). (6.55)
From the Holder inequality and Fubini Theorem, we conclude
_
_
x
M
t
1
g
0
_
_
L
2
x
(L
2
)
C
1
e
0
t
2
_
g
0

L
2
x
(L
2
)
+
_
_
1g
0
_
_
L
2
x
(L
2
)
_
. (6.56)
For k = 2, there are more changes of variables that have to be done in
the right order. We have
2
x
M
t
2
g
0
(x, )
=
_
t
0
_
s
1
0
_
s
2
0
_
s
3
0
_
R
3
_
R
3
_
R
3
_
R
3
B(,
1
,
2
,
3
,
4
, t, s
1
, s
2
, s
3
, s
4
)
2
x
g
0
(w,
4
)d
4
d
3
d
2
d
1
ds
4
ds
3
ds
2
ds
1
(6.57)
where
B(,
1
,
2
,
3
,
4
, t, s
1
, s
2
, s
3
, s
4
)
K (,
1
) K (
1
,
2
) K (
2
,
3
) K (
3
,
4
)
180 TAIPING LIU AND SHIHHSIEN YU [June
e
()(ts
1
)(
1
)(s
1
s
2
)(
2
)(s
2
s
3
)(
3
)(s
3
s
4
)(
4
)s
4
(6.58)
w x
1
(ts
1
)
1
1
(s
1
s
2
)
1
2
(s
2
s
3
)
1
3
(s
3
s
4
)
1
4
s
4
. (6.59)
Again, we change the variables and make the switching of the dierentiations:
V
1
=
1
, V
2
=
1
2
, V
3
=
2
3
, V
4
=
3
4
,
V
1
1
g
0
(w,
4
) = s
1
x
g
0
(w,
4
)
1
2
g
0
(w,
4
),
V
1
3
g
0
(w,
4
) = s
3
x
g
0
(w,
4
)
1
2
g
0
(w,
4
),
2
x
g
0
=
1
s
1
s
3
_
V
1
1
V
1
3
g
0
+
V
1
1
1
2
g
0
+
V
1
3
1
2
g
0
+
2
1
2
g
0
_
. (6.60)
The key observation is that we switch the second derivative with respect
to x evenly to V
1
1
and V
1
3
. There are two reasons for doing this. The rst
is that the resulting factor
1
s
1
s
3
is integrable. Another is that B keeps the
integrability after integration by parts. We set s
0
t, s
5
0 so that
B =
4
i=0
_
e
(
P
i
j=1
V
j)(s
i
s
i+1
)
_
i=1
_
2
2[V
i
[
exp
_
([
i1
j=1
V
j
[
2
[
i
j=1
V
j
[
2
)
2
8[V
i
[
2
[V
i
[
2
8
_
[V
i
[
2
exp
_
i1
j=1
V
j
[
2
+[
i
j=1
V
j
[
2
4
__
. (6.61)
With this and the boundedness of [
()[, we have
V
1
1
B
= O(1)e
3
4
0
t
__
1
1
K(,
1
)
_
K (
1
,
2
) K (
2
,
3
) K (
3
,
4
)
+K(,
1
)K (
1
,
2
) K (
2
,
3
) K (
3
,
4
)
_
,
V
1
3
B
= O(1)e
3
4
0
t
_
K(,
1
)K (
1
,
2
)
_
1
3
K (
2
,
3
)
_
K (
3
,
4
)
+K(,
1
)K (
1
,
2
) K (
2
,
3
) K (
3
,
4
)
_
,
V
1
3
V
1
1
B
= O(1)e
3
4
0
t
__
1
1
K(,
1
)
_
K(
1
,
2
)
_
1
3
K (
2
,
3
)
_
K (
3
,
4
)
2011] SOLVING BOLTZMANN EQUATION, PART I: GREENS FUNCTION 181
+
_
1
1
K(,
1
)
_
K (
1
,
2
) K (
2
,
3
) K (
3
,
4
)
+K(,
1
)K (
1
,
2
)
_
1
3
K (
2
,
3
)
_
K (
3
,
4
)
+K(,
1
)K (
1
,
2
) K (
2
,
3
) K (
3
,
4
)
_
, (6.62)
for some 0 < < 1. From these we obtain
_ _ _ _
V
1
1
BdV
4
dV
3
dV
2
dV
1
= O(1)e
3
4
0
t
,
_ _ _ _
V
1
3
BdV
4
dV
3
dV
2
dV
1
= O(1)e
3
4
0
t
, (6.63)
_ _ _ _
V
1
3
V
1
1
BdV
4
dV
3
dV
2
dV
1
= O(1)e
3
4
0
t
.
We have
2
x
M
t
2
g
0
(x, )
=
_
t
0
_
s
1
0
_
s
2
0
_
s
3
0
_
R
3
_
R
3
_
R
3
_
R
3
1
s
1
s
3
_
g
0
V
1
3
V
1
1
B
1
2
g
0
__
V
1
1
B +
V
1
3
B
_
+ B
2
1
2
g
0
_
dV
4
dV
3
dV
2
dV
1
ds
4
ds
3
ds
2
ds
1
.
(6.64)
Apply (6.63) to the above, we conclude from the Holder inequality and Fubini
Theorem that
_
_
2
x
M
t
2
g
0
_
_
L
2
x
(L
2
)
C
2
e
0
t
2
_
g
0

L
2
x
(L
2
)
+
_
_
1g
0
_
_
L
2
x
(L
2
)
+
_
_
_
2
1
g
0
_
_
_
L
2
x
(L
2
)
_
.
(6.65)
This completes the proof of the Mixture Lemma.
Lemma 6.16. The wave g
j
of (6.45) satisfy, for some positive constant D
k
,

l
x
(g
2k
)
L
2
x
(L
2
)
= O(1)e
1
t
D
k
, k = 0, 1, 2 , 0 l k. (6.66)
Proof. This is direct consequence of the smoothness of K
1
h in microscopic
velocity, Lemma 6.10 and the Mixture Lemma, (6.49).
182 TAIPING LIU AND SHIHHSIEN YU [June
6.5. Global wave structure of the Greens function
We now put together the previous pieces of the waves in the Greens
function to gain global understanding of the Greens function. There is a
missing information on the pointwise description outside the nite Mach re
gion. This missing information is obtained through the following process.
First, the regularity resulting from Picard iterations through the Mixture
Lemma allows use to apply the Sobolev calculus if we have pointwise infor
mation on the Sobolev norms. The pointwise description of these Sobolev
norms is obtained by a weighted energy estimate in the proof of the follow
ing main theorem for the 1D Greens function. In this theorem, we describe
only the Greens function itself. We will be interested in the dierential
of the Greens function. For that we need to include other singular waves
besides the kineticlike waves.
Lemma 6.17. The Greens function minus the singular waves satises
_
_
_
_
_
l
x
_
G
2
j=0
h
j
h
2k
j=0
g
j
__
_
_
_
_
L
2
x
(L
2
)
= O(1), k = 0, 1, 2 , 0 l k.
(6.67)
Proof. The function G
2
j=0
h
j
h
2k
j=0
g
j
satises the linearized
Boltzmann equation with the source Kg
2k
, (6.44). From Lemma 6.16, we
know the smoothness and exponential decay in time of the source. Thus
the lemma follows from the boundedness of the solution operator for the
linearized Boltzmann equation, (3.14), and the Duhamels principle:
_
_
_
_
_
l
x
_
G
2
j=0
h
j
h
2k
j=0
g
j
__
_
_
_
_
L
2
x
(L
2
)
=
_
_
_
_
_
t
0
G
ts
K
l
x
(g
2j
)(s)ds
_
_
_
_
L
2
x
(L
2
_
t
0
_
_
G
ts
_
_
L
2
x
(L
2
)
K
L
2
_
_
_
l
x
(g
2j
)
_
_
_
L
2
x
(L
2
)
ds
= O(1)
_
t
0
e
(ts)
C
ds = O(1). (6.68)
j=1
1
_
4A
j
(t + 1)
e
(x
j
t)
2
4A
j
t
E
j
E
j
[ , (6.70)
where (
j
, E
j
), j = 1, 2, 3, are Euler characteristics, (4.14), and A
j
is the
NavierStokes dissipation parameters, (5.13); the particlelike part:
G
K
h
0
+ h
1
+ h
2
, (6.71)
(6.20), (6.21), (6.22), (6.33), (6.34); and the remainder G
R
, which is a
bounded function satisfying, for any xed constants C
j
> A
j
, j = 1, 2, 3,
and some positive constant C,
G
R

L
2
= O(1)
3
j=1
1
t + 1
e
(x
j
t)
2
4C
j
t
+ O(1)e
x+t
C
. (6.72)
Moreover, the Greens function as an operator has the following properties:
GP
1

L
2
, P
1
G
L
2
= O(1)
3
j=1
1
t + 1
e
(x
j
t)
2
4C
j
t
+ O(1)e
x+t
C
(6.73)
P
1
GP
1

L
2
= O(1)
3
j=1
1
(t + 1)
3
2
e
(x
j
t)
2
4C
j
t
+ O(1)e
x+t
C
. (6.74)
Proof. The study of the structure of the Greens function is done according
to the following two cases:
Case 1. Within nite Mach region, [x[ Mt, for some constant M >
[
j
[, j = 1, 2, 3, and suciently large.
By (), G G
L

L
2
x
(L
2
)
decays exponentially in time. As the essential
kinetic waves h
j
, j = 0, 1, 2, and
h, (6.20) Lemmas 6.6, Lemmas 6.8, Lemma
6.10, and g
j
, j = 0, 1, . . ., Lemma 6.12, also decay exponentially in time, we
have
_
_
_
_
_
GG
L
j=0
h
j
h
2
j=0
g
j
_
_
_
_
_
L
2
x
(L
2
)
= O(1)e
Ct
. (6.75)
184 TAIPING LIU AND SHIHHSIEN YU [June
The uidlike wave G
L
is smooth and bounded, Theorem 5.9. This fact and
Lemma 6.17, applied here with k = 1, yield
_
_
_
_
_
GG
L
j=0
h
j
h
2
j=0
g
j
_
_
_
_
_
H
1
x
(L
2
)
= O(1). (6.76)
By Sobolev inequality, we have from (6.75) and (6.76) that for some constant
C > 0,
_
_
_
_
_
GG
L
j=0
h
j
h
2
j=0
g
j
_
_
_
_
_
L
x
(L
2
)
= O(1)e
t
C
. (6.77)
Thus in the nite Mach region, for some constant C
1
> 0,
_
_
_
_
_
GG
L
j=0
h
j
h
2
j=0
g
j
_
_
_
_
_
L
2
= O(1)e
t+x
C
1
. (6.78)
The theorem then follows from Theorem 5.9 and Theorem 5.11.
Case 2. Outside nite Mach region, [x[ > Mt.
Outside of the nite Mach region, (6.75) still holds. Therefore, it re
mains to obtain the pointwise estimate in the space variable x. The Greens
function minus the essential kinetic waves satises, (6.44):
_
_
R
t
+
1
R
x
= LR + S,
R(x, 0, ) = 0,
R GG
K
2
j=0
g
j
,
S Kg
2
.
(6.79)
We apply the weighted energy method to this equation with a weight w(x, t)R:
w(x, t) e
xMt
N
, (6.80)
for some large constants M, N to be chosen later. Multiply (6.79) with
2011] SOLVING BOLTZMANN EQUATION, PART I: GREENS FUNCTION 185
w(x, t) and integrate to obtain
d
dt
_
R
1
2
w(R, R)dx+
_
R
1
2N
w(R, (M
1
x
[x[
)R)+w(R, LR)dx=
_
R
w(R, S)dx.
(6.81)
There are terms on the left hand side of (6.81) that are of good sign:
_
R
w[
1
2N
(R, MR) + (R, LR)]dx =
_
R
w[
M
2N
(R, R)dx + (R, LR)]dx.
From (3.23) and (3.24),
[
_
R
1
N
w(R,
1
x
[x[
R)dx[ C
2
_
R
1
N
w[(R, R) + (LR, R)]dx.
Thus we may choose
M
2N
=
C
1
N
= 2,
and yield
d
dt
_
R
1
2
w(R, R)dx +
_
R
w(R, R)dx
_
R
w(R, S)dx. (6.82)
We conclude from the CauchySchwarz inequality and the estimate of the
source S by (6.43) that
d
dt
_
R
w(R, R)dx+
_
R
w(R, R)dx=O(1)
_
R
w(S, S)dx=O(1)
_
R
we
1
(x+t)
D
2
dx.
(6.83)
By further restriction of the choice of the weight function w(x, t), (6.80),
1
D
2
2
N
,
we have
d
dt
_
R
w(R, R)dx+
_
R
w(R, R)dx = O(1)
_
R
e
xMt
N
e
1
(x+t)
D
2
dx=O(1)e
4t
,
(6.84)
and we obtain the desired energy estimate
_
R
w(R, R)(x, t)dx+
_
t
0
_
R
w(R, R)(x, s)dxds
_
R
w(R, R)(x, 0)dx+O(1)
186 TAIPING LIU AND SHIHHSIEN YU [June
=O(1). (6.85)
Similarly, since the function R is in H
1
x
(L
2
= O(1)e
xMt
N
= O(1)e
C(x+t)
, for [x[ > Mt. (6.87)
This completes the proof of the theorem.
7. 3D Greens Function
In this section we consider the 3D Greens function for the initialvalue
problem, (1.7):
_
G
t
+
x
G = LG for x R
3
,
G(x, 0, ;
0
) = (x)(
0
).
(7.1)
Here (
0
) was denoted by
3
(
0
) in the last two sections for 1D case
to emphasize that it is a 3dimensional delta function. Here both (x) and
(
0
) are 3dimensional delta functions. We will derive the explicit form
of the Greens function and give pointwise estimate of the remaining terms.
As in the 1D case, we construct the essential kinetic waves and uidlike
waves separately. However, there are essential dierences with the 1D case.
For instance, there is Huygens principle for 3D case. The complex analytic
method for inverting the Fourier transform is also more sophisticated than
the 1D case.
7.1. Kinetic waves
As with the 1dimensional case, we construct the kinetic waves using
2011] SOLVING BOLTZMANN EQUATION, PART I: GREENS FUNCTION 187
the Picards iteration,c.f. (6.3), (6.5):
_
h
0
t
+
x
h
0
+ ()h
0
= 0,
h
0
(x, 0) = (x)(
0
);
(7.2)
_
h
j
t
+
x
h
j
+ ()h
j
= Kh
j1
,
h
j
(x, 0) = 0, j = 1, 2, . . . .
(7.3)
The dierence here is that the delta function (x) =
3
(x) is now 3
dimensional. Meanwhile, there is a stronger dispersion for the 3D case. As
a consequence, the same number of iterations will result in a bounded source.
The rst term is the same as before, (6.20), an exponentially decaying delta
function along the characteristic direction:
h
0
(x, t, ,
0
) S
t
(x)(
0
) = e
()t
(x t)(
0
). (7.4)
The second term is:
h
1
(x, t, ,
0
)
_
t
0
S
ts
Kh
0
(x, s, ,
0
)ds
=
_
t
0
e
()(ts)
_
R
3
K(,
)h
0
(x (t s), s,
,
0
)d
ds
=
_
t
0
_
R
3
e
()(ts)()s
K(,
)(x (t s)
s)(
0
)d
ds
=
_
t
0
e
()(ts)(
0
)s
K(,
0
)(x (t s)
0
s)ds (7.5)
which is still a generalized function, though the generalized part is of lower
dimension. From direct calculations,
h
2
(x, t, ,
0
)
_
t
0
S
ts
Kh
1
(x, s, ,
0
)ds
=
_
t
0
_
R
3
_
s
0
e
()(ts)()(s)(
0
)
K(,
)K(
,
0
)
(x (t s)
(s )
0
)dd
ds
188 TAIPING LIU AND SHIHHSIEN YU [June
=
_
t
0
_
s
0
exp
_
()(t s) (
x (t s)
0
s
)(s ) (
0
)
_
K(,
x (t s)
0
s
)K(
x (t s)
0
s
,
0
)
1
(s )
3
dds.(7.6)
From the explicit expression in (3.5),
K(,
) = O(1)
1
[
[
exp
_
[
2
8
_
and hence
K(,
x (t s)
0
s
)
= O(1)
[s [
[(t ) +
0
x[
exp
_
[(t ) +
0
x[
2
8[s [
2
_
,
K(
x (t s)
0
s
,
0
)
= O(1)
[s [
[x (t s)
0
s[
exp
_
[x (t s)
0
s[
2
8[s [
2
_
.
For (x, t, ,
0
) R
3
R
+
R
3
R
3
satisfying
min
0st
[x (t s)
0
s[ > 0,
we have
K(,
x (t s)
0
s
)K(
x (t s)
0
s
,
0
)
1
(s )
3
= O(1),
0 s t.
We denote
o (x, t, ,
0
) : 0
x
1
1
t
1
0
1
=
x
2
2
t
2
0
2
=
x
3
3
t
3
0
3
t.
Then for (x, t, ,
0
)
_
R
3
R
+
R
3
R
3
_
o, we have [h
2
(x, t, ,
0
)[ <
.
Moreover, we have
()(t s) + (
x (t s)
0
s
)(s ) + (
0
)
2011] SOLVING BOLTZMANN EQUATION, PART I: GREENS FUNCTION 189
0
(1 +[[)(t s) +
0
(1 +[
x (t s)
0
s
[)(s ) +
0
(1 +[
0
[)
0
(t +[x[) for (x, t, ,
0
) / o.
Therefore, there exists a C > 0 such that
_
_
h
2
(x, t, ,
0
)
_
_
L
2
O(1)e
(t+x)
C
.
This implies
_
_
Kh
2
_
_
L
O(1)e
(t+x)
C
and so
_
_
h
3
(x, t, ,
0
)
_
_
L
O(1)e
(t+x)
C
.
As with the 1D case, we can stop at h
2
and construct
h as well as g
j
, j =
0, 1, . . .. The Mixture Lemma can also be generalized easily to the present
3D case. We omit the details.
7.2. Euler waves
Before we study the uidlike waves for the Boltzmann equation, we
consider the Euler waves. We use the conservative form of the linearized
Euler equations (4.5):
_
t
+
x
m = 0,
m
t
+
2
3
x
E = 0,
E
t
+
5
2
x
m = 0.
This is the linearization with base state
0
= 1, v
0
= 0,
0
= 1 and so the
sound speed c is given as
c
2
=
5
0
3
=
5
3
.
From this we derive the wave equation for
t
:
_
5
2
E
_
t
= 0,
190 TAIPING LIU AND SHIHHSIEN YU [June
(
t
)
tt
=
x
_
2
3
x
E
_
=
x
_
5
3
_
= c
2
x
(
t
). (7.7)
The wave equation can be solved explicitly by the Kirchhos formula and
exhibits the Huygens principle for the 3D situation we consider here. With
t
thus determined, the other variables can also be explicitly constructed:
_
_
(x, t) = (x, 0) +
_
t
0
t
(x, s)ds,
E(x, t) = E(x, 0) +
5
2
_
t
0
t
(x, s)ds,
m(x, t) = m(x, 0) +
2
3
_
t
0
x
E(x, s)ds.
(7.8)
Note that, although
t
satises the Huygens principle, the formula for the
other variables involves time integration and so in general are not zero in
side the acoustic cone and decay algebraically there. The viscous version is
studied in Lemma 7.3 later.
To draw the comparison with the Boltzmann equation, we take Fourier
transform of the Euler equations
_
_
_
m
E
_
_
_+i
_
_
_
0
t
0
0 0
2
3
0
5
2
t
0
_
_
_
_
_
_
m
E
_
_
_ = 0; (7.9)
_
_
_
(, t)
m(, t)
E(, t)
_
_
_=
_
_
_
_
1
i sin(ct)
c
t 2
3c
2
(1+cos(c[[t))
0 I +
cos(ct)1

2
t
2i sin(ct)
3c
0
5i sin(ct)
2c
t
cos(c[[t)
_
_
_
_
_
_
_
(, 0)
m(, 0)
E(, 0)
_
_
_.
(7.10)
We therefore will be interested in the wellknown inversion of the Fourier
transform of the following types:
Theorem 7.1 (Kirchho). The inversion of the Fourier transform of g w
and g w
t
, where
w =
sin(c[[t)
c[[
, w
t
= cos(c[[t),
are
w g(x) =
t
4
_ _
y=1
g(x +cty)dS(y), (7.11)
2011] SOLVING BOLTZMANN EQUATION, PART I: GREENS FUNCTION 191
w
t
g(x) =
1
4
_ _
y=1
g(x+cty)dS(y) +
ct
4
_ _
y=1
g(x+cty) ydS(y).
(7.12)
The Kirchhos formula for the Greens function of the Euler equations
involves the function g as functions. For study of the dissipative Navier
Stokes and the Boltzmann equation, we consider g as the heat kernel. The
following lemma gives estimates of the viscous version of Huygens principle.
Lemma 7.2. For any positive integer l,
w
e
x
2
C(t+1)
(t + 1)
l
2
= O(1)
e
(xct)
2
2C(t+1)
(t + 1)
l
2
, (7.13)
w
t
x
2
C(t+1)
(t + 1)
l
2
= O(1)
e
(xct)
2
2C(t+1)
(t + 1)
l+1
2
. (7.14)
Proof. By the Kirchho formula (7.11),
J
1
w
e
x
2
C(t+1)
(t + 1)
l
2
=
t
4
_ _
y=1
e
xcty
2
C(t+1)
(t + 1)
l
2
dS(y).
There are two cases.
Case 1. [[x[ ct[ O(1)
1 +t.
J
1
= O(1)(t + 1)
_ _
y=1
e
O(1)ty
2
(t + 1)
l
2
dS(y) = O(1)(t + 1)
1
(t + 1)
l
2
_
1
1 +t
_
2
= O(1)
1
(1 + t)
l
2
. (7.15)
Case 2. [[x[ ct[ O(1)
1 +t.
In this case,
min
y=1
[x cty[ = [[x[ ct[,
192 TAIPING LIU AND SHIHHSIEN YU [June
and so
e
xcty
2
C(t+1)
e
xcty
2
2C(t+1)
(xct)
2
2C(t+1)
; (7.16)
and
J
1
= (t + 1)
_ _
y=1
e
xcty
2
2C(t+1)
xcty
2
2C(t+1)
(t + 1)
l
2
dS(y)
(t + 1)e
(xct)
2
2C(t+1)
_ _
y=1
e
xcty
2
2C(t+1)
(t + 1)
l
2
dS(y)
e
(xct)
2
C(t+1)
(t + 1)
l
2
. (7.17)
This proves (7.13); (7.14) is shown similarly.
The following will be needed for the viscous version of the procedure
(7.8) in yielding the algebraic decaying rate inside the acoustic cone.
Lemma 7.3. For [x[ < ct
_
t
0
_
_ _
y=1
e
xcy
2
Ct
(1 + t)
5/2
dS(y)
_
d
C
(1 + t)([x[ +
t + 1)
; (7.18)
and for [x[ > ct
_
t
0
_
_ _
y=1
e
xcy
2
Ct
(1 + t)
5/2
dS(y)
_
d
Ce
(xct)
2
2Ct
(1 + t)
2
. (7.19)
Proof. From (7.16),
_
t
0
_
_ _
y=1
e
xcy
2
Ct
(1 + t)
5/2
dS(y)
_
d
_
t
0
_
_ _
y=1
e
xcy
2
2Ct
e
(xc)
2
2Ct
(1 + t)
5/2
dS(y)
_
d
(7.20)
There are three cases.
2011] SOLVING BOLTZMANN EQUATION, PART I: GREENS FUNCTION 193
Case 1. [x[ O(1)
1 + t.
_
t
0
_
_ _
y=1
e
xcy
2
Ct
(1 + t)
5/2
dS(y)
_
d
_
_
1+t
0
+
_
t
1+t
__
_ _
y=1
e
xcy
2
2Ct
e
(xc)
2
2Ct
(1 + t)
5/2
dS(y)
_
d
1+t
0
(1 +t)
5/2
d + O(1)
_
t
1+t
(t + 1)
5
2
te
(xc)
2
2Ct
2
d =
O(1)
(1 + t)
3
2
.
Case 2.
1 + t [x[ ct +
1 + t.
_
t
0
_
_ _
y=1
e
xcy
2
Ct
(1 + t)
5/2
dS(y)
_
d
_
_
1+t
0
+
_
t
1+t
__
_ _
y=1
e
xcy
2
2Ct
e
(xc)
2
2Ct
(1 + t)
5/2
dS(y)
_
d
1+t
0
e
(xc)
2
2Ct
(1 +t)
5/2
d + O(1)
_
t
1+t
(t + 1)
5
2
te
(xc)
2
2Ct
2
d =
O(1)
(1 + t)[x[
.
Case 3. [x[ ct +
1 + t.
_
t
0
_
_ _
y=1
e
xcy
2
Ct
(1 + t)
5/2
dS(y)
_
d
_
_ t
2
0
+
_
t
t
2
__
_ _
y=1
e
xcy
2
2Ct
e
(xc)
2
2Ct
(1 + t)
5/2
dS(y)
_
d
_ t
2
0
e
c
2
t
4C
e
(xc)
2
2Ct
(1 + t)
5/2
d + O(1)
_
t
t
2
(t + 1)
5
2
te
(xc)
2
2Ct
2
d
=
O(1)e
(xct)
2
2Ct
(1 + t)
2
.
This completes the proof of the lemma.
194 TAIPING LIU AND SHIHHSIEN YU [June
7.3. Spectrum near origin
Like the onedimensional case, we have
Lemma 7.4. Consider the spectrum Spec () of the operator i+L,
R
3
.
(I) For any 0 < 1, there corresponds = () > 0 such that
(i) For [[ > ,
Spec () z C : Re (z) < .
(ii) For [[ , the spectrum within the region z C : Re (z)
consisting of exactly ve eigenvalues
1
(),
2
(),
3
(),
4
(),
5
():
Spec () z C : Re (z) =
1
(),
2
(),
3
(),
4
(),
5
() .
(II) For all 0 < 1, the semigroup e
(i+L)t
can be decomposed as
e
(i+L)t
=
+
{<}
1
2i
_
e
zt
_
z (i + L)
_
1
dz, (7.21)
where 

L
2
= O(1)e
a()t
, a() > 0 depends on (and therefore on
), and can be any close curve that lies entirely on Re z > and
that encloses the ve eigenvalues
1
(),
2
(),
3
(),
4
(),
5
().
We now compute the eigenvalues and eigenfunctions:
(i + L)
j
() =
j
()
j
(), R
3
. (7.22)
To simplify 7.22, we invoke a symmetric property of L. Consider the action
of a three dimensional orthonormal transformation Q O(3) on L
2
:
O(3) L
2
L
2
R
3
_
S
2
_
f((Q)
)g(
) f(Q)g(
)
_
B(Q
,
)d
.
We change variables
= Q and
= Q
_
(Q)
= Q [(Q
)
]
= Q [(Q Q
) Q]Q
= Q [(
) ]Q = Q
+ [(Q
)
]
= Q
+ [(Q Q
) Q]Q
= Q
+ [(
) ]Q = Q
,
Therefore,
_
QQ(f, g)
_
() =
_
QR
3
_
QS
2
_
f(Q
)g(Q
) f(Q)g(Q
)
_
B(Q Q
, Q)dQdQ
=
_
R
3
_
S
2
_
f(Q
)g(Q
) f(Q)g(Q
)
_
B(
, )dd
= Q(Qf, Qg)().
The invariance of L under Q can be proven similarly. We omit the details.
Let g O(3) be an orthonormal transformation that sends

to (1, 0, 0).
Apply g
1
to (7.22), we have, by Lemma 7.5,
i
_
g
1
_
g
1
j
+g
1
L
j
= i
_
g
_
g
1
_
g + L
_
g
1
j
= i
_
g + L
_
g
1
j
= i
_
1
[[ + L
_
g
1
j
=
j
g
1
j
. (7.24)
Therefore, the original equation (7.22) is reduced to
_
i
1
[[ + L
_
e
j
([[) =
j
([[)e
j
([[), (7.25)
196 TAIPING LIU AND SHIHHSIEN YU [June
with
j
=
j
,
j
= ge
j
.
Apply MacroMicro projection to the equation (i
1
[[ + L)e
j
=
j
e
j
:
i[[P
0
1
_
(P
0
e
j
) + (P
1
e
j
)
_
=
j
(P
0
e
j
), (7.26a)
i[[P
1
1
(P
0
e
j
) i[[P
1
1
(P
1
e
j
) + L(P
1
e
j
) =
j
(P
1
e
j
). (7.26b)
Set i[[ = and
j
=
j
. From (7.26b) we can solve P
1
e
j
in terms of P
0
e
j
:
P
1
e
j
= [L P
1
j
]
1
P
1
1
(P
0
e
j
). (7.27)
Substituting this back to (7.26a), we obtain the following equation for P
0
e
j
:
_
P
0
1
+ P
0
1
_
L P
1
j
_
1
P
1
1
_
(P
0
e
j
) =
j
(P
0
e
j
). (7.28)
For notation simplicity, put
I(, ) =
_
P
0
1
+ P
0
1
_
L P
1
_
1
P
1
1
_
.
When = 0, (7.28) has degeneracy:
2
=
4
=
5
= 0, so implicit function
theorem does not apply directly. However, as it turns out,
Lemma 7.6. (I(, )E
1
j
, E
1
k
) = 0, if j ,= k and j, k = 4 or 5, for any , .
Consequently, under the basis E
1
1
, E
1
2
, E
1
3
, E
1
4
, E
1
5
, the matrix representation
of I(, ) becomes:
_
_
_
_
_
_
_
3 3 0
0
_
_
_
_
_
_
_
.
Proof. Consider the reection J
2
O(3) : (
1
,
2
,
3
) (
1
,
2
,
3
).
Clearly, P
0
, P
1
, L,
1
,
j
, , as linear operators, commute with J
2
. Therefore,
I commutes with J
2
. Consequently, since E
1
4
=
2
M is an odd function
of
2
, IE
1
4
is also an odd function of
2
. Now that E
1
1
, E
1
2
, E
1
3
, E
1
5
are even
2011] SOLVING BOLTZMANN EQUATION, PART I: GREENS FUNCTION 197
function of
2
,
(IE
1
j
, E
1
k
) =
_
R
3
IE
1
j
()E
1
k
()d = 0, for j = 4, k = 1, 2, 3, 5. (7.29)
The case j = 5 can be proven similarly. We omit the details.
With the aid of Lemma 7.6, we can compute the eigenvalues and eigen
functions. By (7.29), P
0
e
1
, P
0
e
2
, P
0
e
3
E
1
1
, E
1
2
, E
1
3
_
and P
0
e
4
, P
0
e
5
E
1
4
, E
1
5
_
.
Therefore, the problem of solving for (
j
, P
0
e
j
) for j = 1, 2, 3 is reduced to
the 1dimensional case.
For j = 4, 5,
4
(),
5
() can be solved from
(I(,
4
)E
1
4
, E
1
4
) =
4
,
4
[
=0
= 0,
(I(,
5
)E
1
5
, E
1
5
) =
5
,
5
[
=0
= 0.
(7.30)
Consider E O(3) : (
1
,
2
,
3
) (
1
,
3
,
2
).
(IE
4
, E
4
) = (EIE
4
, EE
4
) = (IEE
4
, EE
4
) = (IE
5
, E
5
).
This, together with (7.30), imply
4
=
5
. Since the eigenspace for
4
=
5
always degenerates with multiplicity 2, we gain extra freedom in specifying
e
4
, e
5
. However, the choice must obey [e
j
, e
k
] =
jk
. As it turns out (see the
proof of the lemma below), we can choose
P
0
e
4
E
1
4
_
, P
0
e
5
E
1
5
_
.
Lemma 7.7. View as a real variable. For [[ 1,
j
(), j = 1, . . . , 5, are
real and analytic functions with
j
() =
j
+ A
j
+ O(
2
), for j = 1, 2, 3, (7.31a)
4
() =
5
() = A
4
+ O(
2
), (7.31b)
where A
j
is the NavierStokes dissipation coecients. Note that A
4
= A
5
.
Moreover,
P
0
e
j
() =
j1
()E
1
1
+
j2
()E
1
2
+
j3
()E
1
3
, for j = 1, 2, 3,
P
0
e
4
() =
44
()E
1
4
,
P
0
e
5
() =
55
()E
1
5
,
198 TAIPING LIU AND SHIHHSIEN YU [June
where
jk
() is a real and analytic function with
jj
= 1 + O() for j = 1, 2, 3, 4, 5,
jk
= O(), for j ,= k.
44
=
55
.
Proof. Since (IE
j
, E
k
) is real ( is thought of as a real variable here),
j
(),
jk
() are real. (7.31b) follows at once after we dierentiate (7.30).
It remains only to compute P
0
e
4
, P
0
e
5
. Consider the ansatz:
P
0
e
4
=
44
E
1
4
, P
0
e
5
=
55
E
1
5
,
44
,
55
R. (7.32)
Recall that P
1
e
j
= [L P
1
1
j
]
1
P
1
1
(P
0
e
j
), (7.27). Therefore e
4
, e
5
are real and
[e
4
, e
5
] =(e
4
, e
5
)=
2
44
55
_
_
LP
1
4
_
1
E
1
4
,
_
LP
1
5
_
1
E
1
5
_
.
This is zero, since (LP
1
4
)
1
E
1
4
is odd in
2
while (LP
1
5
)
1
E
1
5
is even in
2
. The validity of the ansatz (7.32) is justied. Subsequently,
44
,
55
can be solved from the normalization condition [e
j
, e
j
] = (e
j
, e
j
) = 1:
44
=
1
1 +
_
_
_
_
_
LP
1
4
_
1
E
1
4
_
_
_
_
2
L
2
=
1
1 +
_
_
_
_
_
LP
1
5
_
1
E
1
5
_
_
_
_
2
L
2
=
55
= 1 + O(
2
).
We continue with more detailed analysis of (
j
, e
j
). Consider the reec
tion J
1
: (
1
,
2
,
2
) (
1
,
2
,
3
). Note that
(
1
+ L)J
1
e
j
() = J
1
(
1
+ L)e
j
() = J
1
(
1
() + L)e
j
()
=
j
()J
1
e
j
() =
_
j
()
_
J
1
e
j
().
Thus((
j
()), J
1
e
j
()) is an eigenvalueeigenfunction pair. Note also
that, for j = 1, 2, 3, P
0
J
1
e
j
= J
1
P
0
e
j
E
1
1
, E
1
2
, E
1
3
_
, and this implies
1
(),
2
(),
3
()
1
(),
2
(),
3
(). Since
1
(),
2
(),
3
()
are distinct (for small ), by the fact that
1
(0) = c =
3
(0),
2
(0) = 0
2011] SOLVING BOLTZMANN EQUATION, PART I: GREENS FUNCTION 199
and by continuity, we conclude
1
() =
3
(),
2
() =
2
(),
J
1
e
1
() = e
3
(), J
1
e
2
() = e
2
(),
for all small.
For
4
=
5
, a similar argument yields
4
() =
4
(). As for the
eigenvectors, P
0
J
1
e
4
(), P
0
J
1
e
5
()
E
1
4
, E
1
5
_
. Moreover, P
0
J
1
e
4
() is
odd in
2
. Therefore, P
0
J
1
e
4
()
E
1
4
_
and J
1
e
4
() = e
4
(). Similarly,
J
1
e
5
() = e
5
().
Summarizing, we have
1
() =
3
(), (7.33a)
j
() =
j
(), for j = 2, 4, 5, (7.33b)
J
1
e
1
() = e
3
(), (7.33c)
J
1
e
j
() = e
j
(), for j = 2, 4, 5. (7.33d)
Now we switch back to the variable i[[ = . Although so far we have
been working with real , since
j
, e
j
are analytic in , (7.33) holds for all
complex , as long as [[ 1. For simplicity, set
L
j
= 1 +
_
L iP
1
j
()
_
1_
iP
1
_
. (7.34)
Lemma 7.8. For R
3
with [[ 1,
1
() =i[[
_
c + ([[
2
)
_
A
1
[[
2
+ O([[
4
),
2
() =A
2
2
+ O([[
4
)
3
() =i[[
_
c + ([[
2
)
_
A
1
[[
2
+ O([[
4
),
4
() =A
4
[[
2
+ O([[
4
),
5
() =A
4
[[
2
+ O([[
4
),
A
j
=(P
1
1
E
1
j
, L
1
P
1
1
E
1
j
),
(7.35)
for some analytic function : R R. Furthermore, there exists analytic
200 TAIPING LIU AND SHIHHSIEN YU [June
functions a
j
k,l
: R R such that
1
() = L
1
__
a
0
1,1
([[
2
) + i[[a
0
1,2
([[
2
)
_
0
+
_
a
1
1,1
([[
2
) + i[[a
1
1,2
([[
2
)
_
3
k=1
k
[[
+
_
a
4
1,1
([[
2
) +i[[a
4
1,2
([[
2
)
_
4
_
, (7.36a)
2
() = L
2
_
a
0
2,1
([[
2
)
0
+ ia
1
2,2
([[
2
)
3
j=1
j
+ a
4
2,1
([[
2
)
4
_
, (7.36b)
3
() = L
3
__
a
0
1,1
([[
2
) i[[a
0
1,2
([[
2
)
_
_
a
1
1,1
([[
2
) i[[a
1
1,2
([[
2
)
_
3
k=1
k
[[
+
_
a
4
1,1
([[
2
) i[[a
4
1,2
([[
2
)
_
4
_
, (7.36c)
4
() = L
4
_
a
2
4,1
([[
2
)g
2
_
, (7.36d)
5
() = L
4
_
a
2
4,1
([[
2
)g
3
_
, (7.36e)
with
a
0
1,1
(0) =
_
3
10
, a
1
1,1
(0) =
_
1
2
, a
4
1,1
(0) =
_
1
5
, a
0
2,1
(0) =
_
2
5
,
a
4
2,1
(0) =
_
3
5
, a
2
4,1
(0) = 1.
Proof. From (7.31a), we have
1
(i[[) = i[[
_
c+A
1
(i[[)+
k=1
([[
2
)
k
1
(2k)!
d
2k
1
d
2k
(0)
+i[[
k=1
([[
2
)
k
1
(2k + 1)!
d
2k+1
1
d
2k+1
(0)
_
i[[
_
c + A
1
(i[[) ([[
2
)
+i[[
k=1
([[
2
)
k
1
(2k+1)!
d
2k+1
1
d
2k+1
(0)
_
=i[[
_
c + ([[
2
)
_
A
1
[[
2
+ O([[
4
).
2011] SOLVING BOLTZMANN EQUATION, PART I: GREENS FUNCTION 201
By (7.33),
3
(i[[) =
1
(i[[)
= i[[
_
cA
1
(i[[)([[
2
)i[[
k=1
([[
2
)
k
1
(2k + 1)!
d
2k+1
1
d
2k+1
(0)
_
= i[[
_
c + ([[
2
)
_
A
1
[[
2
+ O([[
4
).
Similarly, for j = 2, 4, 5, by (7.33),
j
(i[[) =i[[
_
A
j
(i[[) + i[[
k=1
([[
2
)
k
1
(2k + 1)!
d
2k+1
j
d
2k+1
(0)
_
=A
j
[[
2
+ O([[
4
).
This proves (7.35). (7.36) follows by similar computations. We omit the
details. Note that
g P
1
1
[[ = P
1
g =g
_
1 +
_
L iP
1
1
[[
j
()
_
1_
iP
1
1
[[
_
_
=
_
1 +
_
L iP
1
j
()
_
1_
iP
1
_
_
g,
g
_
M
_
= (g)
1
M =
3
k=1
k
[[
M.
7.4. Fluidlike waves, I
This and next subsections are for the study of the uidlike waves. For
the 3D case considered here, there is a geometrically richer class of uidlike
waves as hinted by the study of the Euler waves. In this subsection, we will
explicitly constructed the leading uidlike waves. The next subsection will
be concerned with the remaining uidlike waves, which decay at faster rate
than the leading uidlike waves considered here.
As the one dimensional case, we have
1
2i
_
e
zt
_
z (i + L)
_
1
dz =
5
j=1
e
j
t
j
[
j
[.
202 TAIPING LIU AND SHIHHSIEN YU [June
This implies
G =
1
2
F
1
+
5
j=1
1
(2)
3
___
<
e
ix+
j
()t
j
[
j
[d,
with
_
_
F
1
_
_
L
2
x
(L
2
)
= O(1)e
t
C()
.
We again dene the uidlike waves in the Greens function as:
G
L
(x, t)
j=1
1
(2)
3
___
<
e
ix+
j
()t
j
() [
j
()[d
=
___
<
e
ix
_
e
1
()t
1
() [
1
()[ + e
3
()t
3
() [
3
()[
_
d
+
___
<
e
ix
e
2
()t
2
() [
2
()[d
+
___
<
e
ix
e
4
()t
_
4
() [
4
()[ +
5
() [
5
()[
_
d. (7.37)
We arrange the pairing in (7.37) to dene the following pairings:
G
L
(, t) =
G
H
(, t) +
G
C
(, t) +
G
R
(, t) +
G
PR
1
(, t) +
G
PR
2
(, t), (7.38)
where
Huygens Pairing
G
H
(, t) =
j=1,3
e
j
()
j
() [
j
()[
j=1,3
e
j
()
L
j
P
m
0
j
() [L
j
P
m
0
j
()[,
Contact Pairing,
G
C
(, t) = e
2
()
2
() [
2
()[,
Rotational Pairing,
G
R
(, t) = e
4
()
_
j=4,5
j
() [
j
()[
2011] SOLVING BOLTZMANN EQUATION, PART I: GREENS FUNCTION 203
t
ct
t
R
3
t
ct
R
3
(a) Huygens wave. (b) Solid Huygens wave.
Figure 7: Huygens wave and solid Huygens wave.
+
j=1,3
L
j
P
m
0
j
() [L
j
P
m
0
j
()[
_
,
1st Riesz Pairing,
G
PR
1
(, t) =
j=1,3
e
j
()
L
j
P
m
0
j
() [L
j
P
m
0
j
()[
e
A
1
(
2
)
j=1,3
L
j
P
m
0
j
() [L
j
P
m
0
j
()[,
2nd Riesz Pairing,
G
PR
2
(, t) = (e
A
1
(
2
)
e
4
()
)
j=1,3
L
j
P
m
0
j
() [L
j
P
m
0
j
()[,
(7.39)
We dene the leading uidlike waves to be those induced by the quadratic
terms, in the Fourier variable , in the eigenvalues and the constant term in
the eigenfunctions. This is made denite in the proof of the following theo
rem, (7.44). There are richer wave patterns for the 3D case. The Huygens
wave for the Euler equations is now a dissipative version G
0
H
, with essential
support of width
t
R
3
Figure 8: Diuse wave.
Theorem 7.9. The leading uidlike waves are G
0
= G
0
H
+ G
0
C
+ G
0
R
+
G
0
PR
1
+G
0
PR
2
G
0
H
(x, t) =
ct
4
_ _
y=1
H
1
(x +cty)dS(y) +
1
4
_ _
y=1
H
2
(x +cty)dS(y)
+
ct
4
_ _
y=1
H
2
(x +cty) ydS(y),
H
1
=
1
15
3
j=1
((4A
1
(1 + t))
3/2
e
x
2
4A
1
t
)
x
j
_
M
_
[[
2
+[[
2
M
_
_
,
H
2
=
1
15
(4A
1
(1 + t))
3/2
e
x
2
4A
1
t
[[
2
M
_
[[
2
,
G
0
C
(x, t) =
1
10
(4A
2
(1 + t))
3/2
e
x
2
4A
2
t
([[
2
5)
M
_
([[
2
5)
,
G
0
PR
1
(x, t) =
_
t
0
c
2
4
_ _
y=1
H
3
(x +cy)dS(y)d
H
3
=
3
j,k=1
((4A
1
(1 + t))
3/2
e
x
2
4A
1
t
)
x
j
x
k
j
M
_
,
2011] SOLVING BOLTZMANN EQUATION, PART I: GREENS FUNCTION 205
G
0
R
(x, t) = (4A
4
(1 + t))
3/2
e
x
2
4A
4
t
3
j=1
M
_
,
G
0
PR
2
(x, t) = (A
1
A
4
)
3
j,k=1
_
_
t
0
exp
x
2
4(max{A
1
,A
4
}t+A
1
A
4
)
(4(maxA
1
, A
4
(1 +t) +[A
1
A
4
[))
3/2
d
_
x
j
x
k
, (7.40)
and
_
_
G
0
H
(x, t)
L
2
C(1 + t)
2
e
(xct)
2
Ct
,
G
0
C
(x, t)
L
2
, G
0
R
(x, t)
L
2
, G
0
PR
2
(x, t)
L
2
C(1 + t)
3
2
e
x
2
Ct
,
G
0
PR
1
(x, t)
L
2
C
_
(1 + t)
2
e
(xct)
2
Ct
+ (1 + t)
3
2
e
x
2
Ct
+
xct
(1 +t)
3
2
(1 +
x
2
t
)
3
2
_
.
(7.41)
Proof. From (7.35) and (7.36), we can see that the leading term of
G
L
(, t)
is
e
ictA
1

2
t
gE
1
1
[gE
1
1
[ + e
A
2

2
t
gE
1
2
[gE
1
2
[ +e
ictA
1

2
t
gE
1
3
[gE
1
3
[
+e
A
4

2
t
gE
1
4
[gE
1
4
[ + e
A
4

2
t
gE
1
5
[gE
1
5
[
(
1
+(
2
+(
3
+(
4
+(
5
.
Direct computations yield
(
2
e
A
2

2
t
gE
1
2
[gE
1
2
[ = e
A
2

2
t
1
10
([[
2
5)
M[([[
2
5)
M[
G
1
C
.
For (
1
+(
3
, we use symmetry of (
1
,
1
) and (
3
,
3
) to calculate
(
1
+(
3
e
ictA
1

2
t
gE
1
1
[gE
1
1
[ + e
ictA
1

2
t
gE
1
3
[gE
1
3
[
= e
ictA
1

2
t
(
1
30
[[
2
M
1
2
3
j=1
j
[[
M)
206 TAIPING LIU AND SHIHHSIEN YU [June
[(
1
30
[[
2
M
1
2
3
j=1
j
[[
M)[
+e
ictA
1

2
t
(
1
30
[[
2
M +
1
2
3
j=1
j
[[
M)
[(
1
30
[[
2
M +
1
2
3
j=1
j
[[
M)[
= e
A
1

2
t
(e
ict
+ e
ict
)
_
1
30
[[
2
M[[[
2
M[
+
1
2
3
j,k=1
k
[[
2
j
M[
k
M[
_
e
A
1

2
t
(e
ict
e
ict
)
1
2
15
3
j=1
j
[[
_
M[[[
2
M[
+[[
2
M[
j
M[
_
= e
A
1

2
t
cos(c[[t)
_
1
15
[[
2
M[[[
2
M[ +
3
j,k=1
k
[[
2
j
M[
k
M[
_
+e
A
1

2
t
sin(c[[t)
c[[
3
j=1
ic
j
15
_
M[[[
2
M[ +[[
2
M[
j
M[
_
G
1
H
+ e
A
1

2
t
cos(c[[t)
3
j,k=1
k
[[
2
j
M[
k
M[.
In order to estimate (
4
+(
5
, we recall that g/[[ = (1, 0, 0). Let /[[, ,
be a basis for R
3
satisfying
g
_
_
/[[,
,
,
_
(1, 0, 0),
(0, 1, 0),
(0, 0, 1),
(7.42)
Using the above relation, the operator gE
1
4
[gE
1
4
[ + gE
1
5
[gE
1
5
[ can be
rewritten as
_
gE
1
4
[gE
1
4
[ +gE
1
5
[gE
1
5
[
_
h
= g(
2
M)
_
g(
2
M)h(
)d
+g(
3
M)
_
g(
3
_
M
)h(
)d
M
_
_
M
h(
)d
M
_
_
M
h(
)d
M
_
_
[(
) + (
)]
_
M
h(
)d
_
=
M
_
_
[
[[
)
[[
]
_
M
h(
)d
_
. (7.43)
Thus we have
(
4
+(
5
= e
A
4

2
t
_
3
j=1
M[
j
M[
3
j,k=1
k
[[
2
j
M[
k
M[
_
G
1
R
e
A
4

2
t
3
j,k=1
k
[[
2
j
M[
k
M[.
The two terms not yet given names are
e
A
1

2
t
cos(c[[t)
3
j,k=1
k
[[
2
j
M[
k
M[
e
A
4

2
t
3
j,k=1
k
[[
2
j
M[
k
M[
= e
A
1

2
t
(cos(c[[t) 1)
3
j,k=1
k
[[
2
j
M[
k
M[
+(e
A
1

2
t
e
A
4

2
t
)
3
j,k=1
k
[[
2
j
M[
k
M[
G
1
PR
1
+
G
1
PR
2
.
Thus we have the following named terms:
_
G
1
H
= e
A
1

2
t
cos(c[[t)
1
15
[[
2
M[[[
2
M[,
+e
A
1

2
t
sin(ct)
c
3
j=1
ic
j
15
_
M[[[
2
M[+[[
2
M[
j
M[
_
,
G
1
C
= e
A
2

2
t 1
10
([[
2
5)
M[([[
2
5)
M[,
G
1
R
= e
A
4

2
t
3
j=1
M[
j
M[,
G
1
PR
2
= (e
A
1

2
t
e
A
4

2
t
)
3
j,k=1
k

2
j
M[
k
M[,
G
1
PR
1
= e
A
1

2
t
(cos(c[[t) 1)
3
j,k=1
k

2
j
M[
k
M[.
(7.44)
208 TAIPING LIU AND SHIHHSIEN YU [June
To extract the leading waves (7.40) it is necessary to compute the inverse
Fourier transform of (7.44) explicitly. We carry the calculation for G
0
C
.
F
1
x
(
2
1
(2)
3
_
<
e
ix
e
A
2

2
t
gE
1
2
[gE
1
2
[d
=
1
(2)
3
_
[
2
,
2
]
3
e
ix
e
A
2

2
t
E
1
2
[E
1
2
[d +O(e
t
C
)
=
3
j=1
_
1
2
_
2
2
e
ix
j
j
e
A
2

j

2
t
d
j
_
E
1
2
[E
1
2
[ +O(e
t
C
).
We now introduce contours
j
1
+
j
2
+
j
3
j
1
=
j
: Re(
j
) =
2
, Im(
j
) lies between 0 and
ix
j
2A
2
t
,
j
2
=
j
:
2
Re(
j
)
2
, Im(
j
) =
ix
j
2A
2
t
,
j
3
=
j
: Re(
j
) =
2
, Im(
j
) lies between 0 and
ix
j
2A
2
t
,
(7.45)
to move the path [
2
,
2
] to
j
1
+
j
2
+
j
3
, then use that Ree
ix
j
j
e
A
2

j

2
t
A
2
2
t on both
j
1
and
j
3
1
2
_
2
2
e
ix
j
j
e
A
2

j

2
t
d
j
=
1
2
_
j
2
e
ix
j
j
e
A
2

j

2
t
d
j
+ O(1)e
t
C
=
1
_
4A
2
(1 +t)
e
x
j

2
4A
2
t
+ O(1)
1
1 + t
e
x
j

2
4A
2
t
+ O(1)e
t
C
.
Thus we have
1
(2)
3
_
<
(
2
d = (4A
2
(1+t))
3
2
e
x
2
4A
2
t
1
10
([[
2
5)
M[([[
2
5)
M[
+O(1)
e
x
j

2
4A
2
t
(1 + t)
2
+O(e
t
C
).
The Fourier inversion of
G
1
H
and
G
1
PR
1
are done using the Kirchho formulas
in Theorem 7.1. Unlike the heat kernel, our domain of integration is [[ < ,
2011] SOLVING BOLTZMANN EQUATION, PART I: GREENS FUNCTION 209
which doesnt yield the singularity at t = 0. A slight generalization of these
formulas of the following form is used:
cos(c[[t) 1
[[
2
= c
2
_
t
0
sin(c[[)
c[[
d,
and so the inverse Fourier of
cos(c[[t) 1
[[
2
g
is
c
2
_
t
0
4
_ _
y=1
g(x +cy)dS(y).
Note also that the term
j
corresponds to /x
j
. We omit the details. This
establishes the explicit form of the leading uidlike waves (7.40).
It remains to study the estimates (7.41). The second one is easy to see.
We use Theorem 7.1 and Lemma 7.2 to obtain the rst one. We consider
the last Riesz leading wave for small time and large time scale, i.e. t 1
and t 1.
Case 1 t 1 We consider two subcase [x[ ct
1
2
and [x[ ct
1
2
For [x[ ct
1
2
,note that [x[
2
/t is bounded above and so 1 Ce
x
2
/t
,
and, for some positive constant C
G
0
PR
1
(x, t)
L
2
C
_
t
0
c
2
(1 + t)
3/2
t
_ _
y=1
e
x+cy
2
8A
1
t
dS(y)d
C(1 + t)
3/2
C(1 + t)
3/2
e
x
2
Ct
.
For [x[ ct
1
2
, we have
G
0
PR
1
(x, t)
L
2
C
_
t
0
c
2
(1 + t)
3/2
t
_ _
y=1
e
x+cy
2
8A
1
t
dS(y)d
Ce
(xct)
2
16A
1
t
_
t
0
c
2
(1 + t)
3/2
t
_ _
y=1
e
x+cy
2
16A
1
t
dS(y)d
210 TAIPING LIU AND SHIHHSIEN YU [June
C(1 + t)
3/2
e
(xct)
2
16A
1
t
C(1 + t)
2
e
(xct)
2
16A
1
t
.
Case 2 t 1 We have three cases [x[ ct
1
2
, ct
1
2
[x[ ct and [x[ ct.
The case [x[ ct
1
2
is dealt with the same way as above. For ct
1
2
[x[ ct,
in order to directly calculate the x
j
and x
k
derivatives in G
0
PR
1
(x, t) we rst
observe that
_
t
0
c
2
4(4A
1
(1 + t))
3/2
_ _
y=1
e
x+cy
2
4A
1
t
dS
y
d = Ct
1
2
_
y1
1
[y[
e
x+cty
2
4A
1
t
dy,
where C is agenetic constant independent of (x, , t). We now adopt the
spherical coordinate and let x to be the polar axis.
_
y1
1
[y[
e
x+cy
2
4A
1
t
dy = C
_
1
0
_
2
0
_
0
re
x
2
+2ctxr cos +c
2
t
2
r
2
4A
1
t
sin dddr
= C
1
[x[
_
1
0
_
e
(xctr)
2
4A
1
t
e
(x+ctr)
2
4A
1
t
_
dr. (7.46)
Here we change a variable again to obtain
_
1
0
_
e
(xctr)
2
4A
1
t
e
(x+ctr)
2
4A
1
t
_
dr
=
1
ct
_
_
x
xct
e
s
2
4A
1
t
ds
_
ct+x
x
e
s
2
4A
1
t
ds
_
=
1
ct
_
2
_
x
0
e
s
2
4A
1
t
ds
_
ct+x
ctx
e
s
2
4A
1
t
ds
_
. (7.47)
We now are ready to take derivatives with respect to spatial variables
2
x
j
x
k
_
1
[x[
_
2
_
x
0
e
s
2
4A
1
t
ds
_
ct+x
ctx
e
s
2
4A
1
t
ds
__
=
_
jk
[x[
3
+
3x
j
x
k
[x[
5
__
2
_
x
0
e
s
2
4A
1
t
ds
_
ct+x
ctx
e
s
2
4A
1
t
ds
_
+
_
jk
[x[
2
3x
j
x
k
[x[
4
__
2e
x
2
4A
1
t
e
(ct+x)
2
4A
1
t
_
_
jk
[x[
2
3x
j
x
k
[x[
4
_
e
(ctx)
2
4A
1
t
x
j
x
k
[x[
3
_
[x[
2A
1
t
e
x
2
4A
1
t
ct +[x[
2A
1
t
e
(ct+x)
2
4A
1
t
_
2011] SOLVING BOLTZMANN EQUATION, PART I: GREENS FUNCTION 211
+
x
j
x
k
[x[
3
[x[ ct
2A
1
t
e
(ctx)
2
4A
1
t
/
1
+/
2
+/
3
+/
4
+/
5
.
it is easy to see that [/
1
[, [/
2
[ and [/
4
[ Ct
1/2
[x[
3
and for [x[ ct
1
2
1
[x[
3
(1 +c
2
)
3
_
c
2
t +c
2
[x[
2
_
3
2
(
1 +c
2
c
2
)
3
t
3
2
(1 +
[x[
2
t
)
3
2
.
For [/
3
[ and [/
5
[ we can estimate
_
_
_
([x[
2
+ t
1
2
[x[
1
)e
(ctx)
2
4A
1
t
Ct
1
e
c
2
t
2
32A
1
t
e
(ctx)
2
8A
1
t
, [x[ ct/2;
([x[
2
+ t
1
2
[x[
1
)e
(ctx)
2
4A
1
(1+t)
Ct
3
2
e
(ctx)
2
4A
1
t
, [x[ ct/2.
We combine estimates of /
i
to conclude that
G
0
PR
1

L
2
C
_
t
3
2
(1 +
[x[
2
t
)
3
2
+ t
2
e
(ctx)
2
8A
1
t
_
.
For [x[ ct,
G
0
PR
1

L
2
C
_
t
0
c
2
(1 + t)
3/2
t
_ _
y=1
e
x+cy
2
8A
1
t
dS(y)d
Ce
(xct)
2
16A
1
t
_
t
0
c
2
(1 + t)
3/2
t
_ _
y=1
e
x+cy
2
16A
1
t
dS(y)d.
We break the time integral domain into two parts
G
0
PR
1
(x, t)
L
2
Ce
(xct)
2
16A
1
t
_
_
t
1/2
0
+
_
t
t
1/2
_
c
2
(1 + t)
3/2
t
_ _
y=1
e
x+cy
2
16A
1
t
dS(y)d
Ce
(xct)
2
16A
1
t
_
_
t
1/2
0
(1 + t)
3/2
t
d +
_
t
t
1/2
(1 + t)
3/2
t
t
[x[
e
(xc)
2
16A
1
t
d
_
,
where we use (7.46) and (7.47) and simple calculation to yield
G
0
PR
1
(x, t)
L
2
C(1 + t)
2
e
(xct)
2
16A
1
t
212 TAIPING LIU AND SHIHHSIEN YU [June
+C(1 + t)
3
2
[x[
1
e
(xct)
2
16A
1
t
_
t
t
1/2
e
(xc)
2
16A
1
t
d
G
0
PR
1

L
2
C(1 + t)
2
e
(xct)
2
16A
1
t
+ C(1 + t)
1
[x[
1
e
(xct)
2
16A
1
t
.
From [x[ ct equivalently [x[
1
(ct)
1
and so
G
0
PR
1

L
2
C(1 + t)
2
e
(xct)
2
16A
1
t
, for [x[ ct.
This completes the study of leading waves.
7.5. Fluidlike waves, II
To study the remaining uidlike waves, we rst study the analyticity
of the pairings (7.39).
Proposition 7.10. There exist operators H
j
, H
jk
, C
j
, R
jk
, R
0
jk
, P
jkl
,
P
1
jk
, P
2
jk
, which are bounded operators in L
2
G
H
(, t) =e
A
1
(
2
)t
_
cos(c[[t)
_
cos([[([[
2
)t)
_
1
15
[[
2
M[[[
2
M[ +
3
j=1
j
H
j
_
+
sin([[([[
2
)t)
c[[
_
c
15
3
j=1
i
j
([[
2
M[
j
M[
+
j
M[[[
2
M[) +
3
j,k=1
k
H
jk
__
+
sin(c[[t)
c[[
_
cos([[([[
2
)t)
_
c
15
3
j=1
i
j
([[
2
M[
j
M[
+
j
M[[[
2
M[) +
3
j,k=1
k
H
jk
_
+
sin([[([[
2
)t)
c[[
c
2
[[
2
_
1
15
[[
2
M[[[
2
M[ +
3
j=1
j
H
j
___
G
C
(, t) =e
A
2
(
2
)t
_
1
10
([[
2
5)
M[([[
2
5)
M[ +[[
2
C
0
2011] SOLVING BOLTZMANN EQUATION, PART I: GREENS FUNCTION 213
+
3
j=1
j
C
j
_
G
R
(, t) =e
A
4
(
2
)t
_
3
j=1
(
j
[
j
[ +
3
k=1
k
R
jk
) +
3
j,k=1
k
R
0
jk
_
G
PR
1
(, t) =e
A
1
(
2
)t
_
cos(c[[t)(cos([[([[
2
)t) 1)
j,k=1
k
[[
2
_
j
[
k
[ +
3
l=1
l
P
jkl
+[[
2
P
1
jk
_
+
sin(c[[t)
c[[
_
sin([[([[
2
)t)
[[
j,k=1
c
j
k
_
j
[
k
[ +
3
l=1
l
P
jkl
+[[
2
P
1
jk
_
+cos([[([[
2
)t)
3
j,k=1
k
P
2
jk
_
+cos(c[[t)
sin([[([[
2
)t)
c[[
3
j,k=1
k
P
2
jk
_
,
G
PR
2
(, t) = (e
A
1
(
2
)t
e
A
4
(
2
)t
)
j,k=1
k
[[
2
_
j
[
k
[ +
3
l=1
l
P
jkl
+[[
2
P
1
jk
_
(7.48)
Proof. Huygens Pairing
G
H
(, t)
First, we consider the macro part of Huygens pairing using (7.35), (7.36),
G
H
1
(, t)
= e
A
1
(
2
)t
2 cos([[A
1
1
([[
2
)t)
_
(a
0
1,1
([[
2
)
2
[[
2
a
0
1,2
([[
2
)
2
)
0
[
0
[
+(a
4
1,1
([[
2
)
2
[[
2
a
4
1,2
([[
2
)
2
)
4
[
4
[ + (a
0
1,1
([[
2
)a
4
1,1
([[
2
)
a
0
1,2
([[
2
)a
4
1,2
([[
2
))(
0
[
4
[ +
4
[
0
[) + (a
0
1,1
([[
2
)
2
a
1
1,2
([[
2
)
2
+a
1
1,1
([[
2
)
2
a
0
1,2
([[
2
)
2
)
3
j=1
j
(
0
[
j
[ +
j
[
0
[)
214 TAIPING LIU AND SHIHHSIEN YU [June
+(a
4
1,1
([[
2
)
2
a
1
1,2
([[
2
)
2
+ a
1
1,1
([[
2
)
2
a
4
1,2
([[
2
)
2
)
3
j=1
j
(
4
[
j
[+
j
[
4
[)
_
e
A
1
(
2
)t
2i sin([[A
1
1
([[
2
)t)
[[
_
2i[[
2
a
0
1,1
([[
2
)a
0
1,2
([[
2
)
0
[
0
[
2i[[
2
a
4
1,1
([[
2
)a
4
1,2
([[
2
)
4
[
4
[ [[
2
(a
0
1,1
([[
2
)a
4
1,2
([[
2
)
+a
4
1,1
([[
2
)a
0
1,2
([[
2
))(
0
[
4
[ +
4
[
0
[) + (a
0
1,1
([[
2
)a
1
1,1
([[
2
)
+[[
2
a
0
1,2
([[
2
)a
1
1,2
([[
2
))
3
j=1
j
(
0
[
j
[ +
j
[
0
[)
+(a
4
1,1
([[
2
)a
1
1,1
([[
2
)
+[[
2
a
4
1,2
([[
2
)a
1
1,2
([[
2
))
3
j=1
j
(
4
[
j
[ +
j
[
4
[)
_
Evaluation of a
k
1,j
([[
2
) at = 0 in (7.36) gives
G
H
1
() = e
A
1
(
2
)t
cos([[A
1
1
([[
2
)t)
_
1
15
[[
2
M[[[
2
M[
+[[
2
H
00
+
3
j=1
j
H
0j
_
+e
A
1
(
2
)t
sin([[A
1
1
([[
2
)t)
c[[
_
c
15
3
j=1
i
j
([[
2
M[
j
M[
+
j
M[[[
2
M[) +[[
2
H
04
_
,
where H
0j
are analytic in for j = 0, 1, 2, 3, 4. We now consider one of the
remaining micro parts
G
H
1
(, t) =e
1
()t
(P
0
1
() [P
1
1
()[ P
m
0
1
() [(L
1
1)P
m
0
1
()[)
+e
3
()t
(P
0
3
() [P
1
3
()[ P
m
0
3
() [(L
3
1)P
m
0
3
()[)
and note that P
1
1
and P
1
3
are related by
_
_
P
1
1
() = (L
1
1)P
0
1
()
= (L iP
1
+A
1
([[
2
) + i[[A
1
1
([[
2
))
1
(P
1
i )P
0
1
(),
P
1
3
() = (L
3
1)P
0
3
()
= (L iP
1
+A
1
([[
2
) i[[A
1
1
([[
2
))
1
(P
1
i )P
0
3
().
(7.49)
2011] SOLVING BOLTZMANN EQUATION, PART I: GREENS FUNCTION 215
Here we expand the operator (LiP
1
+A
1
([[
2
)+i[A
1
1
([[
2
))
1
, [[ 1,
as follows
1
L iP
1
+ A
1
([[
2
) i[[A
1
1
([[
2
)
=
1
(L iP
1
+ A
1
([[
2
))(1 +
iA
1
1
(
2
)
LiP
1
+A
1
(
2
)
)
=
1
(L iP
1
+ A
1
([[
2
))
j=0
_
i[[A
1
1
([[
2
)
L iP
1
+ A
1
([[
2
)
_
j
=
1
iA
1
1
(
2
)
LiP
1
+A
1
(
2
)
(L iP
1
+ A
1
([[
2
))
j=0
(1)
j
_
[[
2
A
1
1
([[
2
)
2
(L iP
1
+ A
1
([[
2
))
2
_
j
=
1
iA
1
1
(
2
)
LiP
1
+A
1
(
2
)
(L iP
1
+ A
1
([[
2
))(1 +

2
A
1
1
(
2
)
2
(LiP
1
+A
1
(
2
))
2
)
Thus (L iP
1
+ A
1
([[
2
) i[[A
1
1
([[
2
))
1
can be expressed by the
summation of two operators, both analytic in and P
0
1
and P
0
3
are
related
L
1
1 = M
1
() +[[M
2
(),
L
3
1 = M
1
() [[M
2
(),
P
0
1
() = (a
0
1,1
([[
2
) + i[[a
0
1,2
([[
2
))
0
+ (a
1
1,1
([[
2
) + i[[a
1
1,2
([[
2
))
k=1
k
[[
+ (a
4
1,1
([[
2
) + i[[a
4
1,2
([[
2
))
4
H
1
+[[(H
2
+
H
4
[[
2
),
P
0
3
() = (a
0
1,1
([[
2
) i[[a
0
1,2
([[
2
))
0
(a
1
1,1
([[
2
) i[[a
1
1,2
([[
2
))
k=1
k
[[
+ (a
4
1,1
([[
2
) i[[a
4
1,2
([[
2
))
4
H
1
[[(H
2
+
H
4
[[
2
),
P
m
0
1
() = (a
1
1,1
([[
2
) + i[[a
1
1,2
([[
2
))
3
k=1
k
[[
H
3
+
H
4
[[
,
216 TAIPING LIU AND SHIHHSIEN YU [June
P
m
0
3
() = (a
1
1,1
([[
2
) i[[a
1
1,2
([[
2
))
3
k=1
k
[[
H
3
H
4
[[
.
Thus we may rewrite
G
H
1
(, t)
= e
A
1
(
2
)t
2 cos([[A
1
1
([[
2
)t)
_
(H
1
H
3
) [M
1
H
1
+M
2
([[
2
H
2
+ H
4
)[ + H
2
[[[
2
(M
2
H
1
+M
1
H
2
) +M
1
H
4
[
+H
3
[M
1
(H
1
H
3
) +[[
2
M
2
H
2
[ + H
4
[M
1
H
2
M
2
(H
1
H
3
)[
_
e
A
1
(
2
)t
2i sin([[A
1
1
([[
2
)t)
[[
_
(H
1
H
3
) [[[
2
(M
2
H
1
+M
1
H
2
)
+H
4
[ +[[
2
H
2
[M
1
H
1
+M
2
([[
2
H
2
+ H
4
)[
+[[
2
H
3
[M
1
H
2
M
1
(H
1
H
3
)[+H
4
[M
1
(H
1
H
3
)+[[
2
M
2
H
2
[
_
= e
A
1
(
2
)t
2 cos([[A
1
1
([[
2
)t)
_
[[
2
H
1
10
+
3
j=1
j
H
1j
_
e
A
1
(
2
)t
sin([[A
1
1
([[
2
)t)
c[[
_
[[
2
H
2
10
+
3
j,k=1
k
H
1jk
_
,
where H
l
01
, H
1j
and H
1jk
is analytic in for l = 1, 2 j, k = 1, 2, 3. We now
sum them up to obtain
G
H
(, t) =e
A
1
(
2
)t
cos([[A
1
1
([[
2
)t)
_
1
15
[[
2
M[[[
2
M[+
3
j=1
j
H
j
_
+e
A
1
(
2
)t
sin([[A
1
1
([[
2
)t)
c[[
_
c
15
3
j=1
i
j
([[
2
M[
j
M[
+
j
M[[[
2
M[) +
3
j,k=1
k
H
jk
_
,
then trigonometric equality gives the result for the Huygens pairing.
2011] SOLVING BOLTZMANN EQUATION, PART I: GREENS FUNCTION 217
Contact pairing
G
C
(, t)
We recall that from (7.36b)
2
() = L
2
P
0
2
()
=
_
1 + [L iP
1
+ A
2
([[
2
)]
1
iP
1
_
(a
0
2,1
([[
2
)
0
+ia
1
2,2
([[
2
)
3
j=1
j
+ a
4
2,1
([[
2
)
4
),
and that the micro part is zero when = 0. Thus we can calculate contact
pairing as follows:
G
C
(x, t)
= e
A
2
(
2
)t
2
() [
2
()[
= e
A
2
(
2
)t
_
P
0
2
() [P
0
2
()[ + P
0
2
() [(L
2
1)P
0
2
()[
+(L
2
1)P
0
2
() [
2
()[
_
= e
A
2
(
2
)t
_
1
10
([[
2
5)
M[([[
2
5)
M[ +[[
2
C
0
+
3
j=1
j
C
j
_
,
for some C
j
analytic in .
1nd Riesz Pairing
G
PR
1
(, t)
From (7.35), (7.36a), (7.36c) and (7.39) we have
G
PR
1
(, t) = e
A
1
(
2
)tiA
1
1
(
2
)t
L
1
P
m
0
1
() [L
1
P
m
0
1
()[
e
A
1
(
2
)t
L
1
P
m
0
1
() [L
1
P
m
0
1
()[
+e
A
1
(
2
)t+iA
1
1
(
2
)t
L
3
P
m
0
3
() [L
3
P
m
0
3
()[
e
A
1
(
2
)t
L
3
P
m
0
3
() [L
3
P
m
0
3
()[
= e
A
1
(
2
)t
(cos([[A
1
1
([[
2
)t) 1)
_
L
1
P
m
0
1
() [L
1
P
m
0
1
()
+L
3
P
m
0
3
() [L
3
P
m
0
3
()
_
e
A
1
(
2
)t
sin([[A
1
1
([[
2
)t)
_
L
1
P
m
0
1
() [L
1
P
m
0
1
()
L
3
P
m
0
3
() [L
3
P
m
0
3
()
_
,
218 TAIPING LIU AND SHIHHSIEN YU [June
= e
A
1
(
2
)t
2(cos([[A
1
1
([[
2
)t) 1)
_
((1 +M
1
)i[[a
1
1,2
([[
2
) +[[M
2
a
1
1,1
([[
2
))
j=1
j
[[
j
[((1 +M
1
)i[[a
1
1,2
([[
2
) +[[M
2
a
1
1,1
([[
2
))
3
k=1
k
[[
k
[
+((1 +M
1
)a
1
1,1
([[
2
) + i[[
2
M
2
a
1
1,2
([[
2
))
j=1
j
[[
j
[((1 +M
1
)a
1
1,1
([[
2
) + i[[
2
M
2
a
1
1,2
([[
2
))
3
k=1
k
[[
k
[
_
+e
A
1
(
2
)t
2 sin([[A
1
1
([[
2
)t)
_
((1 +M
1
)i[[a
1
1,2
([[
2
) +[[M
2
a
1
1,1
([[
2
))
j=1
j
[[
j
[((1 +M
1
)a
1
1,1
([[
2
) + i[[
2
M
2
a
1
1,2
([[
2
))
3
k=1
k
[[
k
[
+((1 +M
1
)a
1
1,1
([[
2
) + i[[
2
M
2
a
1
1,2
([[
2
))
j=1
j
[[
j
[((1 +M
1
)i[[a
1
1,2
([[
2
)
+[[M
2
a
1
1,1
([[
2
))
3
k=1
k
[[
k
[
_
,
and that a
1
1,1
(0) =
_
1/2 and the microscopic parts containing one more .
Thus we obtain
G
PR
1
(, t) = e
A
1
(
2
)t
(cos([[A
1
1
([[
2
)t) 1)
j,k=1
k
[[
2
_
j
[
k
[ +
3
l=1
l
P
jkl
+[[
2
P
1
jk
_
+e
A
1
(
2
)t
sin([[A
1
1
([[
2
)t)
c[[
3
j,k=1
k
P
2
jk
,
for some P
jkl
P
1
jk
and P
2
jk
analytic in . Here we have used the trigono
metric equalities as in the Huygens pairing case.
2011] SOLVING BOLTZMANN EQUATION, PART I: GREENS FUNCTION 219
2nd Riesz Pairing
G
PR
2
(, t)
G
PR
2
(, t) = (e
A
1
(
2
)t
e
A
4
(
2
)t
)
_
((1 +M
1
)i[[a
1
1,2
([[
2
) +[[M
2
a
1
1,1
([[
2
))
j=1
j
[[
j
[((1 +M
1
)i[[a
1
1,2
([[
2
) +[[M
2
a
1
1,1
([[
2
))
k=1
k
[[
k
[ + ((1 +M
1
)a
1
1,1
([[
2
) + i[[
2
M
2
a
1
1,2
([[
2
))
j=1
j
[[
j
[((1 +M
1
)a
1
1,1
([[
2
) + i[[
2
M
2
a
1
1,2
([[
2
))
3
k=1
k
[[
k
[
_
= (e
A
1
(
2
)t
e
A
4
(
2
)t
)
j,k=1
k
[[
2
_
j
[
k
[ +
3
l=1
l
P
jkl
+[[
2
P
1
jk
_
.
Rotational pairing
G
R
(, t)
We have
G
R
(, t) = e
A
4
(
2
)t
_
4
() [
4
()[ +
5
() [
5
()[
+L
1
P
m
0
1
() [L
1
P
0
1
()[ +L
3
P
m
0
3
() [L
3
P
0
3
()[
_
.
First, we recall that from the above cases
L
1
P
m
0
1
() [L
1
P
0
1
()[ +L
3
P
m
0
3
() [L
3
P
0
3
()[
a
1
1,1
([[
2
)
2
3
j,k=1
k
[[
2
(1 +M
1
)
j
[(1 +M
1
)
k
[.
We now use (7.36d) and (7.36e)
_
P
0
4
() = a
2
4,1
([[
2
)g
2
,
P
0
5
() = a
2
4,1
([[
2
)g
3
.
to yield
4
() [
4
()[ +
5
() [
5
()[
220 TAIPING LIU AND SHIHHSIEN YU [June
= a
2
4,1
([[
2
)
2
_
L
4
g
2
[L
4
g
2
[ +L
4
g
3
[L
4
g
3
[)
_
a
2
4,1
([[
2
)
2
(
3
j=1
L
4
j
[L
4
j
[
3
j,k=1
k
[[
2
L
4
j
[L
4
k
[),
Here our calculations are similarly to (7.43) and we have used g(/[[, ,
) = ((1, 0, 0), (0, 1, 0), (0, 0, 1)). We now observe that, for small ,
L
4
(1 +M
1
)
=
__
L iP
1
+ A
4
([[
2
)
_
1
_
LiP
1
+A
1
([[
2
)
_
1
_
1+
[[
2
A
1
1
([[
2
)
2
(LiP
1
+A
1
([[
2
))
2
_
1
_
(iP
1
)
= [[
2
N (iP
1
),
for some N analytic in . Thus we have
3
j,k=1
k
[[
2
_
a
2
4,1
([[
2
)
2
L
4
j
[L
4
k
[a
1
1,1
([[
2
)
2
(1+M
1
)
j
[(1+M
1
)
k
[
_
=
3
j,k=1
k
R
0
jk
,
and conclude that
G
R
(, t) = e
A
4
(
2
)t
_
3
j=1
(
j
[
j
[ +
3
k=1
k
R
jk
) +
3
j,k=1
k
R
0
jk
_
.
This completes the proof of the proposition.
With the analyticity property, we adopt complex analysis technique to
estimate the pairings of (7.39). We state rst a general lemma.
Lemma 7.11. Suppose that f(, t) = O(1)e
O(
4
)t
is analytic in for [[ <
1. Then in the region of [x[ < (M+1)ct, M any given positive constant,
there exists a constant C such that the following inequality holds:
_
<
e
ix
e
A
2
t
f(, t)d C
_
(1 + t)
3+
2
e
x
2
Ct
+ e
t/C
_
.
2011] SOLVING BOLTZMANN EQUATION, PART I: GREENS FUNCTION 221
Proof. We choose an orthogonal transformation O
x
which maps
x
x
to
(1, 0, 0), i.e.,
O
x
x
[x[
= (1, 0, 0). (7.50)
We change variables = O
x
, then by virtue of the orthogonality of
the transformation,
x = O
x
x O
x
= [x[(1, 0, 0) = [x[
1
, [[ = [[, and
i
= (O
1
x
)
i
.
(7.51)
Thus we may rewrite
_
<
e
ix
1
e
A
2
t
f(, t)d
=
_
<
e
ix
e
A
2
t
(O
1
x
)
f(O
1
x
, t)d
We will prove this lemma only in the case of = 0. The extra factor
i
,
i = 1, 2, 3 result in the extra timedecaying factor t
1
2
; details are omitted.
We rst divide the integral into two region [[ < = B (B
c
[[ <
) D
1
D
2
where
B [
2
,
2
] [
2
,
2
] [
2
,
2
].
On D
2
, [[ >
2
and so
_
D
2
e
ix
1
e
A
2
t
f(O
1
x
, t)d Ce
t/C
. (7.52)
To estimate the integration on D
1
part, we use analyticity of f(O
1
x
, t)
to equate the integral with respect to
1
on [
2
,
2
] with the integral on
(
2
,
2
,
x
tM
),
(
2
,
2
,
[x[
tM
) =
1
2
3
=
1
: Re(
1
) =
2
, 0 < Im() <
[x[
tM
1
:
2
< Re(
1
) <
2
, Im(
1
) =
[x[
tM
.
222 TAIPING LIU AND SHIHHSIEN YU [June
1
: Re(
1
) =
2
, 0 < Im(
1
) <
[x[
tM
,
B
2
= (
2
,
3
) R
2
[[
2
[, [
3
[ <
2
,
here we take M large enough such that on the contour we have the growth
rate
[f(O
1
x
, t)[ O(1)e
o(1)
2
.
We now break the domain of integration into two parts
_
D
1
e
ix
1
e
A
2
t
f(O
1
x
, t)d
=
_
B
2
_
3
e
ix
1
e
A
2
t
f(, t)d
1
d
2
d
3
+
_
B
2
_
2
e
ix
1
e
A
2
t
f(, t)d
1
d
2
d
3
=
1
+
2
.
Note that [[ >
2
on
1
and
3
, and
1
=
_
B
2
_
3
e
ix
1
e
A
2
t
f(, t)d Ce
t/C
. (7.53)
We now estimate the integral over
2
2
_
B
2
_
2
e
ix
1
e
A
2
t+o(1)
2
t
d
1
d
2
d
3
= e
x
2
4At
_
B
2
_
2
e
At
1
ix
2At
At
2

2
At
3

2
+o(1)
2
t
d
1
d
2
d
3
= e
x
2
4At
_
B
2
_
2
2
e
Atu+
ix
Mt
ix
2At
At
2

2
At
3

2
+o(1)(u+
ix
Mt

2
+
2

2
+
3

2
)t
dud
2
d
3
= e
x
2
4At
(1(1
2A
M
)
2
)
_
B
2
_
2
2
e
Au
2
t2Aut(
ix
Mt
ix
2At
)At
2

2
At
3

2
+o(1)(u+
ix
Mt

2
+
2

2
+
3

2
)t
dud
2
d
3
,
2011] SOLVING BOLTZMANN EQUATION, PART I: GREENS FUNCTION 223
and so
[
2
[ Ce
x
2
Ct
_
B
2
_
2
2
e
A
2
u
2
t
At
2

2

2
At
2

3

2
dud
2
d
3
C
1
(1 + t)
3
2
e
x
2
Ct
, (7.54)
We combine (7.52), (7.53) and (7.54) to complete the proof.
Theorem 7.12. For any given Mach number M > 1, there exists C > 0
such that for [x[ (M + 1)ct,
G
H
(x, t) G
0
H
(x, t)
L
2
C
_
e
(xct)
2
Ct
(1 + t)
5/2
+ e
t/C
_
,
G
PR
1
(x, t) G
0
PR
1
(x, t)
L
2
C
_
e
(xct)
2
Ct
(1 + t)
5/2
+
e
x
2
Ct
(1 + t)
5/2
+
xct
(1 + t)
2
(1 +
[x[
2
t
)
3
2
+ e
t/C
_
,
G
C
(x, t)G
0
C
(x, t)
L
2
, G
PR
2
(x, t)G
0
PR
2
(x, t)
L
2
, G
R
(x, t)G
0
R
(x, t)
L
2
C
_
e
x
2
Ct
(1 + t)
2
+ e
t/C
_
. (7.55)
Proof. It is enough to provide the proof for G
P
(x, t) G
1
P
(x, t)
L
2
for any
pairing P since we have showed the remainder of G
1
P
decay faster when the
explicit leading uidlike waves G
0
P
(x, t) has been taken away in Theorem
7.9. The Huygens wave are calculated using Theorem 7.1 and Lemma 7.2:
G
H
(x, t) G
1
H
(x, t)
L
2
=
_
_
_
1
(2)
3
_
</2
e
ix
(
G
H
(, t)
G
1
H
(, t))d
_
_
_
L
2
.
From Proposition 7.10,
G
H
(, t)
G
1
H
(, t)
= e
A
1

2
t
(e
O(
4
)t
1)
_
cos(c[[t)
_
cos([[([[
2
)t)
_
1
15
[[
2
M[[[
2
M[ +
3
j=1
j
H
j
_
224 TAIPING LIU AND SHIHHSIEN YU [June
+
sin([[([[
2
)t)
c[[
_
c
15
3
j=1
i
j
([[
2
M[
j
M[ +
j
M[[[
2
M[)
+
3
j,k=1
k
H
jk
__
+
sin(c[[t)
c[[
_
cos([[([[
2
)t)
_
c
15
3
j=1
i
j
([[
2
M[
j
M[ +
j
M[[[
2
M[) +
3
j,k=1
k
H
jk
_
+
sin([[([[
2
)t)
c[[
c
2
[[
2
_
1
15
[[
2
M[[[
2
M[ +
3
j=1
j
H
j
___
+e
A
1

2
t
_
cos(c[[t)
_
(cos([[([[
2
)t) 1)
1
15
[[
2
M[[[
2
M[
+cos([[([[
2
)t)
3
j=1
j
H
j
+
sin([[([[
2
)t)
c[[
_
c
15
3
j=1
i
j
([[
2
M[
j
M[
+
j
M[[[
2
M[) +
3
j,k=1
k
H
jk
__
+
sin(c[[t)
c[[
_
(cos([[([[
2
)t) 1)
c
15
3
j=1
i
j
([[
2
M[
j
M[
+
j
M[[[
2
M[) + cos([[([[
2
)t)
3
j,k=1
k
H
jk
+
sin([[([[
2
)t)
c[[
c
2
[[
2
_
1
15
[[
2
M[[[
2
M[ +
3
j=1
j
H
j
___
cos(c[[t)O([[
4
)te
A
1

2
t
H
1
+
sin(c[[t)
c[[
O([[
4
)te
A
1

2
t
H
2
+cos(c[[t)e
A
1

2
t
H
3
+
sin(c[[t)
c[[
e
A
1

2
t
H
4
.
We have
_
_
_
1
(2)
3
_
<
e
ix
sin(c[[t)
c[[
e
A
1

2
t
H
j
d
_
_
_
L
2
,
_
_
_
1
(2)
3
_
<
e
ix
cos(c[[t)e
A
1

2
t
H
j
d
_
_
_
L
2
=
_
_
_w
t
1
(2)
3
_
<
e
ix
e
A
1

2
t
H
j
d
_
_
_
L
2
.
To apply Theorem 7.1 and Lemma 7.2 we only need to estimate
_
_
_
_
</2
e
ix
e
A
1

2
t
H
j
d
_
_
_
L
2
. Note that
[ cos([[([[
2
)t)[,
sin([[([[
2
)t)
[[
Ce
O(
3
)t
,
and that H
1
and H
2
have extra O([[
4
)t. These induce extra (1+t)
1
decay
by Lemma 7.11:
_
_
_
_
</2
e
ix
O([[
4
)te
A
1

2
t
H
1
d
_
_
_
L
2
C
_
e
x
2
Ct
(1 + t)
5/2
+ e
t/C
_
,
_
_
_
_
</2
e
ix
O([[
4
)te
A
1

2
t
H
2
d
_
_
_
L
2
C
_
e
x
2
Ct
(1 + t)
3
+ e
t/C
_
Similarly, for H
3
and H
4
we use the following equality
cos([[([[
2
)t) 1 = [[
2
([[
2
)
_
t
0
sin([[([[
2
))
[[
d,
to obtain
_
_
_
_
<
e
ix
O([[
4
)te
A
1

2
t
H
3
d
_
_
_
L
2
C
_
e
x
2
C(1+t)
t
2
+ e
t/C
_
,
_
_
_
_
<
e
ix
O([[
4
)te
A
1

2
t
H
4
d
_
_
_
L
2
C
_
e
x
2
C(1+t)
t
5/2
+ e
t/C
_
.
We now consider 1st Riesz wave.
G
PR
1
(, t)
G
1
PR
1
(, t)
= e
A
1

2
t
(e
O(
4
)t
1)
_
(cos(c[[t) cos([[([[
2
)t) 1)
226 TAIPING LIU AND SHIHHSIEN YU [June
j,k=1
k
[[
2
_
j
[
k
[ +
3
l=1
l
P
jkl
+[[
2
P
1
jk
_
+
sin(c[[t)
c[[
_
sin([[([[
2
)t)
[[
3
j,k=1
c
j
k
_
j
[
k
[ +
3
l=1
l
P
jkl
+[[
2
P
1
jk
_
+ cos([[([[
2
)t)
3
j,k=1
k
P
2
jk
_
+cos(c[[t)
sin([[([[
2
)t)
c[[
3
j,k=1
k
P
2
jk
_
+e
A
1

2
t
_
cos(c[[t)
_
(cos([[([[
2
)t) 1)
3
j,k=1
k
[[
2
_
j
[
k
[
+
3
l=1
l
P
jkl
+[[
2
P
1
jk
_
sin([[([[
2
)t)
c[[
3
j,k=1
k
P
2
jk
_
+(cos(c[[t) 1)
3
j,k=1
k
[[
2
_
3
l=1
l
P
jkl
+[[
2
P
1
jk
_
+
sin(c[[t)
c[[
_
sin([[([[
2
)t)
[[
3
j,k=1
c
j
k
_
j
[
k
[
+
3
l=1
l
P
jkl
+[[
2
P
1
jk
_
+ cos([[([[
2
)t)
3
j,k=1
k
P
2
jk
__
O([[
4
)te
A
1

2
t
_
cos(c[[t)T
1
+
sin(c[[t)
c[[
T
2
j,k=1
k
[[
2
_
j
[
k
[ +
3
l=1
l
P
jkl
+[[
2
P
1
jk
__
+e
A
1

2
t
_
cos(c[[t)T
3
+
sin(c[[t)
c[[
T
4
+ (cos(c[[t) 1)T
5
j,k=1
k
P
1
jk
_
.
We now use Theorem 7.1 and Lemma 7.2 and Lemma 7.11 to obtain
_
_
_
_
<
e
ix
O([[
4
)te
A
1

2
t
T
j
d
_
_
_
L
2
x
2
Ct
(1 + t)
5/2
+ e
t/C
_
, for j =1, 2,
_
_
_
_
<
e
ix
e
A
1

2
t
T
j
d
_
_
_
L
2
C
_
e
x
2
Ct
(1 + t)
5/2
+ e
t/C
_
, for j =3, 4,
_
_
_
_
<
e
ix
O([[
4
)te
A
1

2
t
3
j,k=1
k
[[
2
_
j
[
k
[
+
3
l=1
l
P
jkl
+[[
2
P
1
jk
_
d
_
_
_
L
2
C
_
e
x
2
Ct
(1 + t)
5/2
+ e
t/C
_
,
_
_
_
_
<
e
ix
e
A
1

2
t
3
j,k=1
k
P
1
jk
_
d
_
_
_
L
2
C
_
e
x
2
Ct
(1 + t)
5/2
+ e
t/C
_
.
To calculate cos(c[[t) 1)T
5
, we write
cos(c[[t) 1)T
5
= (cos(c[[t) 1)
3
j,k,l=1
l
[[
2
P
jkl
=
_
t
0
sin(c[[)
c[[
3
j,k,l=1
c
2
l
P
jkl
d,
and so the inverse Fourier transform of cos(c[[t) 1)T
5
satises
_
_
_
_
<
e
ix
e
A
1

2
t
cos(c[[t) 1)T
5
d
_
_
_
L
2
=
_
_
_
_
t
0
_
<
e
ix
sin(c[[)
c[[
3
j,k,l=1
c
2
l
P
jkl
dd
_
_
_
L
2
=
_
_
_
_
t
0
w(x, )
_
<
e
ix
3
j,k,l=1
c
2
l
P
jkl
dd
_
_
_
L
2
.
The above has one more factor comparing to
G
1
PR
1
(, t) which induces
extra (1 + t)
1/2
decay. We can apply similar argument to G
C
(x, t)
G
1
C
(x, t)
L
2
, G
R
(x, t) G
1
R
(x, t)
L
2
and G
PR
2
(x, t) G
1
PR
2
(x, t)
L
2
to
complete the proof.
228 TAIPING LIU AND SHIHHSIEN YU [June
7.6. Global wave pattern for Greens function
With the construction of essential kinetic waves and the uidlike waves
in the nite Mach region, the global structure of the Greens function can
now be studied following the basic procedure of the 1D case in the last
two sections. The Mixture Lemma for the 3D case is a straightforward
generalization of the 1D case; and similarly for the weighted energy method.
We will not elaborate this and only state the result.
Theorem 7.13. The Greens function, (7.1),
G = G
0
+G
K
+G
R
, (7.56)
consists of the leading uidlike waves G
0
of Theorem 7.9, the essential ki
netic waves G
K
= h
0
+h
1
+h
2
given by (7.4), (7.5)(7.6), and the remainder
G
R
satisfying
G
R

L
2
=O(1)
_
e
(xct)
2
Ct
(1 +t)
5/2
+
1
(1+t)
2
(1+
x
2
t
)
3
2
{x<ct}
+
e
x
2
C(1+t)
(1 + t)
2
+e
(x+t)/C
.
(7.57)
Moreover,
GP
1

L
2
+ P
1
G
L
2
= O(1)
_
e
(xct)
2
Ct
(1 + t)
5/2
+
1
(1 + t)
2
(1 +
x
2
t
)
3
2
{x<ct}
+
e
x
2
C(1+t)
(1 +t)
2
+ e
(x+t)/C
; (7.58)
P
1
GP
1

L
2
= O(1)(t + 1)
1/2
_
e
(xct)
2
Ct
(1 + t)
5/2
+
1
(1 + t)
2
(1 +
x
2
t
)
3
2
{x<ct}
+
e
x
2
C(1+t)
(1 + t)
2
+ e
(x+t)/C
. (7.59)
Remark 7.14. The Boltzmann equation and the NavierStokes equations
share some similarity, for instance, in the largetime behavior of uidlike
waves. On the other hand, the Boltzmann equation is semilinear hyper
bolic; while the NavierStokes equations are hyperbolicparabolic. Thus the
2011] SOLVING BOLTZMANN EQUATION, PART I: GREENS FUNCTION 229
small time behaviors are quite dierent. In particular, the Greens function
for the linear NavierStokes equations contains the singularity of the heat
kernel type; while that for the Boltzmann equation does not. Thus in the
above theorem we have expression like (t +1)
instead of t
. This is seen
analytically as follows: For the hyperbolicparabolic systems, the inversion
of the Fourier transform is done for the Fourier variables in the whole space,
as the spectrum is known there. For the Boltzmann equation, the spectrum
is known explicitly only near origin. Meanwhile, it is possible to identify
the singular waves, the essential kinetic waves. The Greens function minus
these singular waves can be as smooth as needed by the Mixture Lemma.
Thus the singularity near the initial time is not of the heat kernel type.
8. InitialBoundary Value Problem
In this section we use the 1dimensional Greens function G(x, t), (5.2),
Theorem 6.18, for the initial value problem to study the pointwise structure
of the solution for the initialboundary value problem:
_
_
g
t
+
1
g
x
= Lg, x > 0
g(x, 0) = I
0
(x),
g(0, t, )[
1
>0
= b
+
(t, ).
(8.1)
The Boltzmann equation is semilinear hyperbolic with characteristic lines
d
dt
x(t) =
1
.
The characteristic lines pointing toward the interior region x > 0 from the
boundary x = 0 correspond to
1
> 0. Thus the boundary data b
+
is dened
only for positive
1
, c.f. [11], [17], [33].
As the boundary x = 0 is stationary, we will see that it is important
to specify the sign of the Euler characteristics
j
, j = 1, 2, 3, of the base
Maxwellian M = M
[1,u,]
of the linearization. One of the main purposes is
to understand the boundary eect on the wave propagation. For this, we
230 TAIPING LIU AND SHIHHSIEN YU [June
choose a localized initial value:
_
_
_
supp(I
0
) [1, 1],
sup
x
I
0
(x, )
L
,3/2
1,
(8.2)
and the imposed boundary data tends to a constant state exponentially in
time:
sup
1
>0
(1 +[[)
4
[b
+
[ e
t/D
for some D > 3/ min
i
>0
i
. (8.3)
Here, the Mach number of the Maxwellian state M
[1,u,]
is assumed to satisfy
[u[/
_
5/3 ,= 0, 1; and b
+
is a given boundary data for
1
> 0, t > 0 with
sucient decay rate sup
1
>0
(1 + [[)
3
[b
+
(t, )[ < for all t > 0. When
all the Euler characteristics are of the same sign, one can use the energy
method; otherwise, there is no straightforward energy method for the study
of the initialboundary value problem (8.1). The energy method yields the
existence of the solutions. Our aim is to study the pointwise behavior of
solutions without the same sign property of the Euler characteristics. We
will design a series of approximations through two approximate solution
operators.
8.1. Two approximate solution operators
If the full boundary values g(0, t) = b(t, ), R
3
is known, then there
is the solution formula
g(x, t) =
_
0
G(x y, t)I
0
(y)dy +
_
t
0
G(x, t )
1
b(, )d. (8.4)
We introduce the rst approximate solution operator using (8.4):
_
_
H[I
0
, b](x, t) : a solution of linearized Boltzmann equation in x > 0,
H[I
0
, b](x, t)
_
0
G(x y, t)I
0
(y)dy +
_
t
0
G(x, t )
1
b(, )d.
(8.5)
Remark 8.1. As the boundary value should only be posted for
1
> 0,
the function g(x, t) = H[I
0
, b](x, t) satises only the rst two equations of
(8.1) and not the third equation for the boundary values. In other words,
2011] SOLVING BOLTZMANN EQUATION, PART I: GREENS FUNCTION 231
in general, g(0, t, ) ,= b(t, ), even for
1
> 0. Nevertheless, with the ex
plicit expression of the Greens function, Theorem 6.18, the formula (8.4)
oers pointwise expression and will be useful for our analysis of the initial
boundary value problem (8.1).
The second approximation solution operator is by the damped equation
with physical boundary data b
+
_
_
I
[I
0
, b
+
](x, t) : a solution of damped Boltzmann equation in x 0,
h(x, t) I
[I
0
, b
+
](x, t),
_
_
h
t
+
1
h
x
= Lh B
+
h,
h[
x=0,
+
>0
= b
+
,
h(x, 0) = I
0
(x).
(8.6)
Here, the Euler waves propagating toward the gas region x > 0 is damped
through the upwind Euler projection operator B
+
, (4.21). We will also write
B
[I
0
, b
+
](t) I
[I
0
, b
+
](0, t). (8.7)
Remark 8.2. The operator I
[I
0
, b
+
] does satisfy the physical boundary
value
B
[I
0
, b
+
](t, )[
1
>0
= b
+
(t, ).
On the other hand, I
[I
0
, b
+
] is not an exact solution operator for the origi
nal Boltzmann equation. Nevertheless, as an approximate operator with the
damped coecient taken to be small, the operator is accurate in restoring
the full boundary values as the waves being damped are leaving the bound
ary. In fact, in the iterations below, this operator is used mainly for the
purpose of restoring the full boundary values. Moreover, as the Euler waves
with positive speed are damped, we can use the weighted energy method to
estimate the solutions of (8.6).
We now design an iterated scheme based on the above two operators
for the construction of the original initialboundary value problem (8.1).
232 TAIPING LIU AND SHIHHSIEN YU [June
Consider the iterations:
_
_
q
1
(x, t) = I
[I
0
, b
+
](x, t),
b
1
(t) = q
1
(0, t),
g
1
(x, t) = H[I
0
, b
1
](x, t),
d
1,+
(t) (g
1
(0, t) b
+
(t))[
1
>0
;
and, for n 2,
q
n
(x, t) = I
[0, d
n1,+
](x, t),
b
n
(t) = q
n
(0, t),
g
n
(x, t) = H[0, b
n
](x, t),
d
n,+
(t) = (g
n
(0, t) d
n1,+
)[
1
>0
.
(8.8)
As noted above, the functions g
n
(x, t) are solutions of the Boltzmann equa
tion, but the boundary condition is not satised in general. However, the
discrepancy of the boundary values will shown to decrease and, if the se
ries
n=1
g
n
(x, t) converges, then it is easy to see that it converges to the
solution of the initialboundary value problem (8.1).
8.2. Weighted energy method
The initialboundary value problem for the damped Boltzmann equation
with 0 < 1,
_
_
q
t
+
1
q
x
= Lq B
+
1
q, x > 0,
q[
x=0,
1
>0
= q
+
(t)[
1
>0
,
q[
t=0
= u
0
,
(8.9)
is studied by the weighted energy method. Integrate the rst equation in
(8.9) taken inner product with 2q and multiplied with the weight w(x) = e
x
:
d
dt
_
0
e
x
(q, q)dx +
_
0
e
x
(q, 2B
+
1
q
1
q)dx (q(0, t),
1
q(0, t))
=
_
0
e
x
(2q, Lq). (8.10)
2011] SOLVING BOLTZMANN EQUATION, PART I: GREENS FUNCTION 233
We now study the second integral. Decompose the solution into
q = P
0
q + P
1
q = q
0
+ q
1
=
3
j=1
(q, E
j
)E
j
+ q
1
=
3
j=1
q
0j
E
j
+ q
1
.
The macro part of the integrant for the second integral above is
(q
0
, 2B
+
1
q
0
1
q
0
) = 2
j
>0
j
(q
0j
)
2
j=1
j
(q
0j
)
2
j=1
[
j
[(q
0j
)
2
(q
0
, q
0
), (8.11)
where
min
j
>0
j
.
The estimate for the micro part is provided by the term (2q, Lq). Thus we
have control of the energy of both parts by using (3.23) and (3.24) and
conclude that, for suciently small c.f. (6.81),
d
dt
_
0
e
x
(q, q)dx +
2
3
_
0
e
x
(q, q)dx 2(q(0, t),
1
q(0, t))
2(q
+
,
1
q
+
)
+
+ 2
_
0
e
x
(u
0
, u
0
)dx, (8.12)
Here,
(f, h)[
_
R
3
,
1
>0
f()h()d.
This leads to the estimate
_
0
e
x
(q, q)dx[
_
0
e
2
(t)
2(q(0, t),
1
q(0, t))
dt
_
0
e
2
(t)
2(q(0, t),
1
q(0, t))
+
dt + e
2
_
0
e
x
(u
0
, u
0
)dx. (8.13)
By the above estimates, one has
_
0
e
2
t
2(b
1
(0, t),
1
b
1
(0, t))
dt O(1)
1
, (8.14)
234 TAIPING LIU AND SHIHHSIEN YU [June
_
0
e
x
(q, q)dx[
O(1)
e
/D
, (8.15)
where D is given in (8.3).
Remark 8.3. The operator I
[I
0
, b
+
] is used in the iterations (8.8) to restore
the full boundary values b(t) from the given values b
+
(t). From (8.13) we
have
_
t
0
e
2
(ts)
(b(s), [
1
[b(s))ds
2
_
t
0
e
2
(ts)
(b
+
(s), [
1
[b
+
(s))ds + 4e
_
0
e
x
(I
0
, I
0
)dx, (8.16)
b(t) I
[I
0
, b
+
](0, t).
Thus the weighted energy method yields the estimate for the full boundary
values b(t) in terms of the boundary data b
+
and the initial data I
0
. This
turns out to be sucient for further analysis.
8.3. Pointwise estimates
For the operator I
[0, d
n1,+
](x, t) in (8.8), we need an estimate for the
function of the form H[I
0
, b][
x=0
for
1
> 0. In carrying out the estimate for
H[0, b
n
](x, t) in the iteration (8.8), what we have from the previous iteration
is the boundary estimate of the form (8.16). We carry out these estimates
in this subsection. For this, we need to make essential use of the decompo
sition of the Greens function into uidlike, essential kinetic parts, and the
remaining parts, G = G
F
+ G
K
+G
R
, (7.56), (6.71), Theorem 5.11, (8.14),
and (8.15).
The rst term b
1
(0, t) = I
[I
0
, b
+
](0, t) in the iteration is estimated by
the weighted energy method and shown to satises estimate of the form
(8.14). We then study the next term in the iteration. The component h
0
in (6.71), G
K
= h
0
+ h
1
+ h
2
, contributes
_
t
0
h
0
1
b
1
d to H[I
0
, b
1
] which
has pointwise estimates when b
1
has pointwise structure for
1
> 0, with
the same sign as x. This is due to the deltafunction in h
0
. The compo
nent
_
t
0
(h
1
+ h
2
)
1
b
1
d gives pointwise structure for all x because of the
regularizing eect from K in constructing h
1
and h
2
. Thus we conclude that
2011] SOLVING BOLTZMANN EQUATION, PART I: GREENS FUNCTION 235
_
t
0
G
K
(x, t)
1
b
1
d has pointwise estimate given that the boundary values
b
1
are given with estimate of the form (8.14):

_
t
0
G
K
(x, t )
1
b
+
(t )d
L
2
O(1)
1
e
(x+t)/C
0
for x > 0 and some C
0
> 0. (8.17)
The detailed calculations are straightforward and omitted.
By the pointwise structure of the uidlike wave component in (6.70)
and the remainder component in (6.72) together with (8.14), one has that
_
_
_
_
_
_
t
0
G
F
(x, t )
1
b
1
()d
_
_
_
_
_
L
2
=
_
_
_
_
_
_
t
0
G
F
(x, t )e
/4
e
/4
1
b
1
()d
_
_
_
_
_
L
2
O(1)
__
t
0
G
F
(x, t )
2
L
2
e
/2
d
_
1
2
__
t
0
e
/2
(b
1
(), [
1
[b
1
())d
_
1
2
O(1)
1
_
_
_
j
>0
e
(x
j
t)
2
C
0
t
1 + t
+O(1)e
x+t
C
0
_
_
_ (8.18)
for some C
0
> 0. The estimates in (8.17) and (8.18) give the pointwise
estimate of the operator H[I
0
, b
1
].
For the convergence of the iterations, we will need more detailed struc
ture of the Greens function stated in Theorem 5.11. We can also construct
the Greens function, G
D
with similar property for the damped equation
h
t
+
1
h
x
Lh = B
+
1
h in a whole space problem.
The estimates of b
1
1
>0
b
0+
[
1
>0
.
We know the full values of the function b
1
(t) = q(0, t) of the solution of
236 TAIPING LIU AND SHIHHSIEN YU [June
an initialboundary problem of the damped Boltzmann equation:
_
_
q
t
+
1
q
x
= Lq B
+
1
q, x > 0,
q[
x=0,
1
>0
= b
+
(t),
q[
t=0
= I
0
.
(8.19)
By the Greens identity we have in turn the following representation of the
boundary data b
+
[
1
>0
:
b
+
[
1
>0
=
__
0
G(y, t)I
0
(y)dy +
_
t
0
G(0, t )
1
q(0, )d
_
1
>0
_
t
0
_
0
G(y, t )
1
B
+
1
qdyd
1
>0
. (8.20)
By the denitions of H[I
0
, b
1
] and that b
1
(t) = q(0, t), one has that
d
1,+
= b
+
[
1
>0
H[I
0
, b
1
](0, t)[
1
>0
=
__
t
0
_
0
G(y, t )B
+
1
q(y, s)dyd
_

1
>0
. (8.21)
The combination of the energy estimate and separation scale of the Greens
function is used in the next step:
_
_
_
_
_
t
0
_
0
G(y, t )B
+
1
q(y, s)dyd
_
_
_
_
L
2
=
_
_
_
_
_
t
0
_
0
G(y, t )e
x/2
B
+
1
q(y, s)e
x/2
dyd
_
_
_
_
L
2
O(1)
_
t
0
_
_
0
_
j
<0
e
(y
j
(t))
2
C
0
(t)
(t + 1)
2
+
j
>0
e
(y
j
(t))
2
C
0
(t)
(t + 1)
+ e
xy+t
C
0
_
e
y
dy
_
1/2
_
_
0
e
y
(q, q)dy
_
1/2
d
2011] SOLVING BOLTZMANN EQUATION, PART I: GREENS FUNCTION 237
O(1)
_
t
0
_
e
K
0
(t)
(t + 1)
3/2
+e
(t)
+ e
(t)
C
0
_
1/2
e
/D
d
= O(1)
1/4
e
t/D
. (8.22)
Here, we have used the fact that G(x, t)B
+
gains an extra decaying factor
0f 1/
t for
j
< 0. (8.22) gives the estimate of d
0+
in  
L
2
,+
. We need
to improve it into a pointwise estimate in
1
> 0. For this, one uses the
Greens function G
D
(x, t) for the damped Boltzmann equation (
t
+
1
L+B
+
1
)h = 0 for a whole space problem and the estimate (8.14), and the
representation
q(x, t) =
_
0
G
D
(x y, t)u
0
(y)dy +
_
t
0
G
D
(x, t )
1
b
1
()d
to yield a pointwise q(x, t)
L
2
1
q(y, )dyd
as the solution of the following initial value problem with a given inhomo
geneous term Heaviside (x)B
+
1
q with pointwise structure in (x, t, ):
_
(
t
+
1
x
L)D = Heaviside (x)B
+
1
q,
D(x, 0) 0.
where Heaviside (x) is the Heaviside function. With the estimate on 
_
t
0
_
0
G(x y, t )B
+
1
q(y, )dyd
L
2
1
>0
(1 +[[)
4
[d
1,+
(t, )[ O(1)
1/4
e
t/D
. (8.23)
The estimate (8.23) is compared to that for the given boundary data b
+
in
(8.3). Then, one can repeat the procedure from (8.9) by replacing b
+
with
238 TAIPING LIU AND SHIHHSIEN YU [June
d
1,+
and I
0
with 0. We therefor show that the iteration scheme satises
_
_
sup
1
>0
(1 +[[)
4
d
n,+
(t) C
1
n/5
e
t/D
,
g
n
(x, t)
L
2
C
1
1+
n1
5
_
_
_
j
>0
e
(x
j
t)
2
C
0
t
1 + t
+ O(1)e
x+t
C
0
_
_
_
(8.24)
for some C
1
> 0. This yields the convergence of the iteration scheme and
the series g(x, t) =
j=1
g
j
solves the initial boundary value problem (8.1)
with
g(x, t)
L
2
O(1)C
1
1
_
_
_
j
>0
e
(x
j
t)
2
C
0
t
1 +t
+ O(1)e
x+t
C
0
_
_
_. (8.25)
This structure of g(x, t) and that of the G(x, t) are used to generate the
solution of an given initial boundary value problem with any y > 0:
_
_
(
t
+
1
x
L)h = 0,
h(0, t)[
1
>0
= 0,
h(x, 0) = 0 for [x y[ 1,
h(, 0)
L
x
(L
,4
)
1.
(8.26)
The solution satises, for x, t > 0,
h(x, t) O(1)
_
_
_
3
j=1
e
(xy
j
t)
2
C
0
(1+t)
1 + t
+ e
xy+t
C
0
_
_
_
+O(1)
1
_
t
0
3
j,i=1
e
(x
j
(t))
2
C
0
(t+1)
e
(y+
i
))
2
C
0
(+1)
_
(t + 1)( + 1)
d. (8.27)
For the case
1
<
2
< 0 <
3
, the solution has the essential support as
depicted in Figure 9. The Greens function G
b
for the initialboundary value
2011] SOLVING BOLTZMANN EQUATION, PART I: GREENS FUNCTION 239
problem
_
_
(
t
+
1
x
L)G
b
= 0,
G
b
(0, t)[
1
>0
= 0,
G
b
(x, 0) = (x x
0
)
3
(
0
), x
0
> 0,
(8.28)
can also be constructed. Because of the boundary, the Greens function is
not translational invariant in x and is of the general form G
b
(x, x
0
, t, ;
0
).
The procedure in the preceding sections of constructing the essential kinetic
and uidlike waves are needed. The construction of the uidlike waves
have been outlined above. The construction of essential kinetic waves are
similar. We conclude with the form of the Greens function, again for the
case of
1
<
2
< 0 <
3
:
_
_
G
b
= G
F
b
+G
K
b
+G
R
b
,
G
K
b
= h
0
+ h
1
+ h
2
,
h
0
= e
()t
(x x
0
1
t)
3
(
0
),
h
1
=
_
_
0, for (x x
0
1
0
t)(
1
t x + x
0
) < 0,
0, for (x x
0
1
0
t)(
1
t x + x
0
) > 0,
x
0
/
1
0
+ x/
1
< t,
1
> 0,
1
0
< 0,
e
nu()(tt
1
)nu(
0
)t
1K(
0
)
1
0
)
, otherwise,
t
1
t (x
x
0
1
0
t)/(
1
1
0
),
[h
2
[ = O(1)e
0
(t+xx
0
)/3
1+
0


0

(1 +[[)
1
,
G
F
b

L
2
= O(1)[
3
j=1
(1 + t)
1/2
e
(x
j
tx
0
)
2
Ct
+
2
j=1
[(t + 1)(1 + x
0
)]
1/2
e
(x
3
t
3
x
0
/
j
)
2
Ct
],
G
R
b

L
2
= O(1)e
(x+x
0
+t)
C
.
(8.29)
Moreover, the algebraic decay rates are higher when the Greens function
acts on the micro part of the initial data.
240 TAIPING LIU AND SHIHHSIEN YU [June
t
x
x
0
dx
dt
=
3
dx
dt
=
3
dx
dt
=
3
dx
dt
=
1
dx
dt
=
2
Figure 9: Solution for the case
1
<
2
< 0 <
3
.
Acknowledgment
The authors would like to thank the referee for his critical comments.
The rst author appreciates the company and helps of IKun Chen, Hung
Wen Kuo, Se Eun Noh and LiCheng Tsai at Academia Sinica during the
writing of this paper.
References
1. Baranger, C. and Mouhot, C., Explicit spectral gap estimates for the linearized Boltz
mann and Landau operators with hard potentials. Rev. Mat. Iberoamericana, 21
(2005), no. 3, 81941.
2. Boltzmann, L., Vorlesungen uber Gastheorie, 2 Vols., Barth, Leipzig (189698). English
translation by S.G. Bush: Lectures on Gas Theory, University of California Press,
California (1964).
3. Bouchut, F., Hypoelliptic regularity in kinetic equations. J. Math. Pures Appl. (9) 81
(2002), no. 11, 1135159.
4. Carleman, T., Sur La Theorie de l