O n the B o l t z m a n n Equation

Part I: Existence
LEW ARKERVD
Communicated by J.L. LIONS
O. Introduction
This is the first of two papers considering the initial-value problem for the
spatially homogeneous, non-linear Boltzmann equation in the absence of exterior
forces:

df
dt =Qf

(o.l)

(t>0), f ( 0 ) = f o > 0 ,

with collision operator

(0.2)
Here (f|

Qf(vt)=

R3 x B2(r)

(J* (f Of) (vl, vz)- (f | f) (v,, vz))k(v. v2)d vz d u.

(vl, v2)=f(vl) g(v2), and u is the impact parameter:

u~B 2(r)= {u sR2; l u I--<r}.

A physical interpretation of the solutionfis the following. If E is any measurable subset of R 3, then ~f(v)dv is the " n u m b e r " of particles contained in E.

J~*(usB 2(r)) is induced by a C Ldiffeomorphism J of R 3 x R 3 x B 2 (r) having the

following physical meaning. If we write
'
1)2)=Ju(U1,132),
Pl,
'

then v~ and v~ signify the velocities after the collision of two colliding particles,
which initially had the velocities v1 and v2.
On physical grounds we impose certain restrictions on J. Let us introduce
the mappings
1: Ra

p"

R3

(Vl, o2)--'1,

R3..-.,R 3: (Vl, v2)--,Vl + v2,

T: R 3 x Ra.--,R: (01, o2)--*l v 112+1v212,

,S,: R 3 R3"-~'R 3 R 3 : (1)1, 02)--I.(02, o l ) .

Then the restrictions on J can be written concisely as follows:

(0.3)

1o J,=l,

1 Arch. Rational Mech. Anal., Vol. 45

poJ==p,

To J== T.

L. ARKERYD:

These express the conservation of number, momentum, and energy. In addition,

we impose requirements of symmetry and involution, respectively
~;o J . = J ~ o ~;

(0.4)
and

J o J = identity.

(0.5)

Remark 0.1. Some comments on the conditions (0.3)-(0.5) are in order. The
first identity in (0.3) is of course trivial. The two other ones can also be written
in the forms
v~+v;=vl+v2,
Iv~12+lv~12=lvll2+lv212.
A s a consequence
(0.6)

I v~ - v[ I -- Iv1 - v2 l,

which follows from the parallelogram identity. The Jacobian determinant of J

is 5:1 by (0.5), i.e., J is volume-preserving.
Finally we assume that the kernel k(v 1, v2) is measurable, non-negative and
invariant under 27% J*(Z* k=J* k=k) with
(0.7)

k(vl, v2) < C~(1 + Iv~ I*+lv2

I*).

For simplicity we normalize B2(r) by taking r = I/V~, so that

S du= 1, and

B2(r)

we write B 2 = B 2 (I/1/T). The letters C and K will stand for various constants.
Using (0.3)-(0.5), we can easily show that

S q~(v) Q f (v) d v = 0

(0.8)

R3

for any function ~p satisfying the functional equation of Cauchy type:

(0.9)

J : ( ~ r ~o)=~ r ~o,

provided all integrals involved are absolutely convergent. Here

<9 q~(v~, v2)=9(vO + q~(v2) .

On the other hand (0.8) implies that
(0.10)

[. ~p(v)f(v, Odv= [. cp(v)fo(v)dv

R3

R3

if f is a solution of (0.1). This includes the five basic conservation laws for the gas,
since (0.9) holds in particular f o r q, = 1, v, I v 12, by (0.3).
In regard to the Cauchy problem (0.1), the questions of existence, uniqueness,
positivity, and asymptotic behavior including equilibrium solutions, are of
major interest. They have been studied by T. C~LRLEM~ [3] for elastic collisions
and continuous initial data fo with (1 +lvlOfo (v)eL ~176
for some x > 6, by D. MORGENSTERN [7], [8] for Maxwellian molecules with a cut-off and initial d a t a f o e L 1,
and by A. JA. POVZNER [9] (existence and uniqueness) for collision operators of
our type, continuous kernels k([ vI - v2 ]) =<C(1 + [vt - v2 l) and initial data fo with

The Boltzmann Equation. I

fo(v) (I +Iv [4, eL 1, and by others. We mention also the results of H. GRAD [5] for
a special class of "hard" intermolecular forces and exponentially decreasing data.
In this paper we study the existence of a non-negative LX-solution of (0.1),
leaving the other problems mentioned for Part II, following immediately hereafter.
For J and k as described above and for 2 of (0.7) not greater than 2, this paper
proves the existence of a solution f(v, t ) > 0 (t > 0), such that (1 +Iv [2)f(v, t)~L 1
(t>0) if
(1 + lvlZ) fo(v),fologfo~L a,
i.e., under physically natural initial conditions. We discuss also the moments of fi
Our results include the existence results mentioned above. In fact, MORGENSTERN'S results are included in our studies of bounded kernels in Section 1, and
CARLEMAN'Sresults are included in Section 3. Moreover the condition f0 logfoeL 1
in Section 3 can be substituted by fo(v)(l+lvl4)eL 1, as well as by fo(v) (1+
[v 12 a) EL 1 for a continuous kernel k (vl, v2), as shown in Part II, respectively [2].
Thus the existence results of POVZNERare included.
The plan of the paper is as follows. We prove (Section I) an existence and
uniqueness theorem for the case when k is bounded. This depends on successive
approximations by a monotone approximation scheme, which seems to have been
introduced first by E. WILD [10] in a special case. In Section 2 we demonstrate the
o~"-theorem for these solutions. This enables us to get existence for unbounded
k by a weak Ll-compactness argument for approximating equations. An alternative is to use a constructive proof along the lines of the bounded case (cf. Part II).
The compactness follows from the conservation laws and the ~-theorem. Similar
compactness ideas have previously been employed in connection with the asymptotic behavior of the Boltzmann equation by T. CARLEMAN[3] and D. MORG~STERN [8] for elastic collisions and for Maxwellian molecules with a cut-off,
respectively.
1. Existence in the Bounded Case
We start by recalling some well-known general ideas.
Let X be a partially ordered Banach space (continuous ordering relation < )
having the Levi property, i.e., every sequence (f,)T with
sup IILII < ~ ,

0<f,<f,+x,

n=l,2,...,

(bounded and positive, monotone) has a limit f > 0 . A mapping

M: X ~ X
is said to be positive and monotone if

O<Mf<MgLwhen

0<f<g.

Suppose we have given a semi-group of positive and monotone contraction

operators {Z(t); t > 0 } on X with infinitesimal generator - d , and a positive,
monotone (non-linear) mapping ~ from the domain of//into X. We consider the
1"

L. ARKERYD :

initial value problem

df t_df=c~f (t>0),

(1.1)

f(0)=fo>0

dt

Let us define a sequence of successive approximations inductively by setting

(1.2)

fl=0 '

t
+ IZ(t-s)C~f"(s)
'
f ,+1 = Z ()fo
as

(n>l).

Then (f"(to))~~ is positive and monotone for each to, and so the sequence converges
to a limitf(to) by the Levi property, if it is bounded. Moreoverf(t) is a solution of
(1.1) under suitable conditions on a~. If g > 0 satisfies
t

g(t) = Z(t)fo + I Z ( t - s) C~g(s)ds,

0

then g > f by (1.2).

Suppose another initial-value problem is given, which is of the same type

af' q_d,f,=c~,f,
dt

(1.1')

(t>0),

f'(0)=f~,

and such that

O<'~(g)-C~'(g),

(1.3)

O<=Z(t)g - Z ' ( t ) g ,

if g is in the domain of d and 0____f The pair (1.1) and (1.1') is called a monotone
pair if moreover O<f~ ~-fo.
By induction the successive approximations f " and f ' " of (1.1) and (1.1')
given by (1.2) satisfy
(1.4)

O<<_f'"(t)<f"(t)

(t>0),

0 ~ f ' (t) ~f(t)

(t > 0).

and so in the limit

(1.5)

We now apply these considerations to (0.1) in the simplest case, namely when k
is bounded (2 = 0).
Proposition

1.1.

Assume that
O< k(Vl, v2)< K k .

(1.6)

Then there exists a unique, non-negative solution f(v, t)eLl(t>O) of (0.1) for
everyfo > 0 in L 1.
Proof. As the Banaeh space X we take L t, which evidently has the Levi property
under the usual ordering

f>g

if f(v)>g(v)

fora.e, veR 3.

We set
,r

Ilfll =

Slf(v)ldv,

R3

IIfL = S ( l + l v l 2 ) ~ I f ( O l d v .
R3

The Boltzmann Equation. I

From (1.6) it follows that the polarization Q (f, g) of Qfsatisfies

IIQ(f, g)ll <=2KkIlfll Ilgli 9

(1.7)

Hence Q is locally Lipschitz continuous in L ~. It follows that there exists a unique

solutionfof (0.1) for 0 < t < t o, where to only depends on K k and Ilfoll.
This solution satisfies

if(v, t) dv = ~fo (v) d r ,

(1.8)

by (0.10) with ~o= 1. Suppose the solution is non-negative for e v e r y f o > 0 . Then
we can taker(v, to) as initial value obtaining a solution for to < t < 2 to. By induction
this will give a solution for t > 0 . We end the proof by showing that the solution is
non-negative, using the argument from the beginning of this section.

By (1.8) f satisfies (0.1) rewritten in the form

df + h f = Q ' f
dt

(1.9)
with

h = C jfo(v2)dv2,
R3

(1.10)

Q'f(vl)= Qf(vO + Cf(vl) I f(v2) dv2.

R3

It is easy to see that Q' is positive and monotone if C is chosen sufficiently large,
since
~f (vl) f (v2) k(Vl, v2) d v2 < K k f (vl) j'f(v2) dr2.
Hence (1.9) is an equation of the type in (1.1) with
~=Q',

,4=h,

and

Z(t)=e -~'.

The corresponding successive approximations fn of (1.2) are non-negative

and satisfy

d .~ + hf ~= Q,f,-1
-d-[.l

(n>2).

Since the iterates are non-negative, an integration gives

t

Ilf"(t)ll = Ilfoll + .IS QI

n-1

/)

( , Qdv dz

0
t

+ II'C[f"-'(v,, T)f'-~ (v2, z)-fo(v2)fn(v~, z)] dv~ dv 2 dz.

0

If moreover IIf~-1 (t)ll < Ilfoll (t_~0), then J"Q f , - i = 0 by (0.8), and
t

Ilfn(t)ll<llfoll+O+ S[. C'fn - l v( 2, x)[fn-~(v~,x)-f"(vl,~)]dv~dv2dz<llfol].

0

L. ARKERYD:

The last inequality holds because f " - x < f L But llf 1(t)ll ~ lifo II sincef ~= 0, and so
by induction
llfnil < IIf011 ( n = l , 2 .... ).
It follows that the monotone sequencef x,f2, ... has a non-negative limit g = limf"
in L 1, and that
.++o
t

(1.11)

g(t)=

e-htfo + S e-h(t-s)Q' g(s) ds.

0

A differentiation in (1.11) implies that g satisfies (1.9). The operator Q' is locally
Lipschitz continuous. Hence the solution of (1.9) is unique, and f = g > 0. This ends
the proof of Proposition 1.1.
We next consider the moments of the solutions just obtained.

Proposition 1.2. Assume that

0 < f o (v) (1 + I vl2)mL~.
Then

f(v,t)(l+lvlZ)eL 1
and (0.10) holds for r

(t>0),

v, Ivl 2.

Proof. Consider the successive approximations

(1.2). Assume that

(l+lvl2)f"-l(v,t)eL 1

f~,f2 .... of f corresponding to

(t>0)

and

Ilf"-x(t)llz=Sf"-l(v, O(l +lvlZ)dv< Sfo(v)(l +lvl2)dv=llfoll2.

The assumption holds in particular f o r f I a s f I = 0. The iterates satisfy
IIf"(011 < Ilfoll,

ff-ff--[fn+hf"=Q'f "-~

(cf the previous proof). We multiply the last equation by (1 +lvl~) with IVlr.---min(I v I, m) and integrate, which gives

~f"(v, t) (1 +lVl2m)do
t

< Sfo (v) (1 +lvl~)dv+

S[. Qf"-' (v, T) (1 +lvl2m)dvd~

0

+ SS [Of "-1 (vl, z) f"-~ (v2, z)-- hf"(vl, z)] (1 + I vt I~) d vx a v2 dz.
0

If the last two terms have a non-positive limit as m tends to infinity, then

(l+lvl2)f"(v,t)EL 1,

IIf,(t)ll<llfoll

But we assumed (1 + Iv 12)f n- x(v, t) ~L 1 (t > 0), and so

lim ~Qf"-l(v,
m--+ oo

~) (1 +lvl~.)dv

(t=>0).

The Boltzmann Equation. I

exists and equals zero by (0.8). Finally

[Cf ~-~ (v~, z)f"-~ (v2, ~) - hf'(v~, ~)] (1 +[v~ 12) dvt dv 2
<=SCff-t(v2, ~) [ f " - t (Vl, z)-f"(vt, z)] (1 +l v, I~)dv, dv2<O.
Here we have used IIf"- 1(t)ll < Ilfoll. We conclude by induction that

(l+lvl2)f~(vt, t)eL1,

Ilff(t)ll2<=llfoll2 ( n = l , 2 .... ).

In the limit this gives

(l+lvl2)f(v, t)mL1

(t=0).

Hence (0.8) holds for q~= 1, v, I vl 2 and (0.10) follows for the same qes.
Another moment property is
Proposition 1.3. Assume that

O< fo(v)(l +lvl')eL 1,

and that (0.7) holds with 2 = min (x/2, 2) for some r > 2. If in addition k is bounded
as in (1.6), then the solution f o r Proposition 1.1 satisfies

f(v, t) (1 +Iv I")eL1

and

IIf(t)ll~<C,, Ilfoll,, (O<t<tD.

Here Ct, only depends on Ilfol12, tl, Ck, and r.
Remark 1.4. It is the "mild" influence of k on the moment estimate (Ck instead
of Kk) for bounded time intervals, guaranteed by Proposition 1.3, that implies the
energy conservation in the unbounded case below.

Remark 1.5. We show in Part II that the higher moments are uniformly bounded
in time under certain additional restrictions.
Proof of Proposition 1.3. The proposition follows from an inequafity ((1.13)
below) due to POVZN~ [9]. Together with (0.3)-(0.5), the elementary inequality

aS+bS<(a+b)S<aS+bS+fls(a~Sbtl-6)S+b~atl-~)~),

a, b>=O, 6s<-l<s,

gives
K

Ir

Ir

[(1 +[v~ 12)1- +(1 +lv~ [ 2)'2"] -- [(1 +101 I')1- +(1 +10212)i- 3
( r - a)

(1.12)

(K-a)

__<fl~ [ ( l + l v l l 2) 2 (l+lv212)-~ +(l+lvll2)-~ (1+1V212) 2 ]

for 0_< ~ __2. But after a change of variables

Ig

(1 + I V ] 2 ) 2

Qf(v)dv

R3
Ir

1r

=89 $ {['(1+ Iv~ 12)~+(1 +lv~12) T ]

R3
Ir

Ir

-- [(1 + I D1 12) 2 "4-(1 + I V212) 2 ]} k (v 1, v2) f (vl) f (v2) d v 1 d 02 d u,

L. ARKERVD:

and so applying (0.7) and (1.12) yields

/r

(1.13)

j" ( l + l v l 2) 2 Qf(v)dv< 89

~ [llfli~+~-~ Ilflla+ Ilfll~-~

R3

Ilfll~+~]

for 0 </7 < 2, which is the desired inequality.

Induction on the successive approximations of f corresponding to (1.2),
together with (1.13) for 2 < x < 4

and ~7= 2 ' gives the lemma for 2 < x < 4 , simi-

lady to Proposition 1.2. Using the result for x = 4 , we can repeat the proof for
4 < x < 6, taking 9 = 2, and by induction get the desired result for all x > 2.

2. The ,~t%Theorem in the Bounded Case

This section considers the ~f'-theorem, stating that

~ ( t ) =J'f(v, t)logf(v, t) dv
is non-increasing as a function of t for a solutionfof the Boltzmann equation (0.1).
Let us first look at some simple cases. Any function

f(vx, t)=exp(aiva 12-1-(b 9vl)-t-c),

a<0

satisfies Q ( f , f ) = 0 for all Q, and hence defines an equilibrium solution of the

Boltzmann equation, i.e. a time-independent solution. Evidently ~ ( t ) is constant
in time for an equilibrium solution. The function,C: (t) is constant in time for allfo
if J is the identity mapping, since in that case Q = 0 a n d f ( t ) = f o .
In the general case of (0.1), we have formally
d

d t "r176 =--d--[Sf(vl' t)logf(v 1, t) do 1

(2,1)

: J' (-di- f(vl' O) l~

0 d v' -I-S-dt -f(vl' Odvl

=S Qf(vl, t) logf(vl, t) dv 1 + ~ Qf(vl, t) dv 1

= 88~ [J* ( f Q f ) - f @f]k (vl, v2)
9l o g [ f @ flJ*(f@f)] du dvl dr2.
Here in the last step we have used (0.8) with 40= 1 for ~ Q f a n d a change of variables
in .[ Qf l o g f, employing (0.3)-(0.5). The last member of (2.1) appears repeatedly,
so for convenience we introduce the notation

N f = 88S [J* ( f @f ) - f @.f] k(v~, v2) logp f | f lJ* ( f @f)] du dv~ dv z.

Since the integrand of Nfis non-positive, it follows that
d

at : r 1 7 6

The Boltzmann Equation. I

This formal proof holds strictly on 0__ t<_ to, if

(2.2)

e,oexp(-Cro I vl2)<f(v, t)<K, o

(O<t<to),

for some eto, Cto, Kzo> 0. To get an Yg'-theorem for the more general solutions of
Proposition 1.1, we approximate (0.1) by equations with solutions bounded as in
(2.2), taking care in the limit of the extra terms which are introduced in (2.1)
(in integrated form).
We first describe the approximating equations. By use of (1.9) the Boltzmann
equation is written in the form (1.1). For convenience we take h > 1. With

Q~f (vl)=min(Q' f (vO, i),

fo,(vO=min(fo(vl), i),
an approximating equation with a bounded solution is

df

dt t-hf=Q~f (t>0),

f(O)=fo i.

To get the estimate from below by e exp(-Clv[2), the term m-1 exp(-Iv 12) is
added to fo giving the approximating initial value

f~(v~)=min(fo(Vl)+ m -1 e x p ( - I v112), i).

However, instead also of changing h accordingly by (1.10), for technical reasons
we

use

hj=C ~ [fo(v2)+ j - l exp(-lv212)]dv2

(j<m).

R 3

Thus the desired approximation of the Boltzmann equation with a solution

satisfying (2.2) for j < m is given by
(2.3)~i

d gJi+h j gmi=Q~
j
, .i (t>0),
dt
gmi

j
m
gmi(O)=f~i.

We notice the double notation

h|

Q~--Q',

ff~=fo~=fo.

Let us briefly discuss the properties of g ~ relevant for the o~F-theorem. Tile
solution of the Boltzmann equation (0.1) with initial value

fo(v)+j -1 exp(-I vl~)

for m = j and i= oo is one solution g}oo of (2.3)~-oo,hence also of the corresponding
integral equation, i.e.
t

(2.4)j~o

gjoo(t)J =e-h~'J~|

Ie-h't'-')Q'gj.(s)ds
0

It follows that the corresponding successive iterates (1.2) converge to a solution

bounded by ~ ~ and moreover to g~ ~ - by the Lipschitz continuity of Q', which
implies uniqueness in (2.4)~ ~.

10

L. ARKERYD:

The pair (2.3)~ and (2.3)~'v with

oo>_i>i'>>_O,

oo>j>j'>O,

oo>_m'>m>O,

is always monotone (cf. Section 1). It follows (cf. (1.4)) that the successive iterates
(1.2) of (2.3)~,r are bounded by those of (2.3)~i. We conclude that they are
bounded by gjoo for O<j<m, i>O, and so that they converge, moreover that they
converge to a solution gSn,i of (2.3)~i with
(2.5)

J
'
gmi>g~,i,

"> "',m'>m).
(i>l "',J=J

Finally for finite to,j, m, i, the term Z(t)fo in (1.2) provides the first inequality in
(2.2) with eito, Cito independent of i_>__1. In the limit (1.2) provides the right-hand
side of (2.2) with Kto < i, as h > 1.
Theorem 2.1. Let f(v, t) be the solution of the Boltzmann equation (0.1) with k
satisfying (1.6) and with initial value f o such that
fo logfo, fo (v) (1 -t-I vl2)~LI(Ra).
Then

f(t) logf(t)~L 1(R 3)

(t >_-0),

and
(t) = Sf (v, t) logf(o, t) d v

is non-increasing as a function of t.
Proof. We notice that

(l+lvlZ)g~i(v, t)~Ll (R s)

(re>j, t>=O).

This holds for the solution g~ oo of (0.1) by Proposition 1.2 and thus in general by
(2.5). By (2.2)

log~-Clvl2 <loggJm~<logK,
and so for m, i finite
g~i IoggJm~~LI"
From (2.2) and (2.3)~i it easily follows that
g~, log g~,l,,, dr1 - ~ g~, loggY,Iv dVl
d j
'"d
= i'~ (-d-fgr~,)logg~idvldt+!--d-~g~,dvldt
t"

(2.6)

t"

= j ~(Q'g~,-hj g~)loggJmidvt dt+ SS(Q; gJmi-hjg~)dvl dt

o

t"

it"

< ~ ~(Q' g ~ , - hj g~,)log g~,, dv I d t+ ~ ~ (Q' g~,- hj gJm,)do I dt

0

0
t'"

+~

( i - Q ' g~,)log-g'm,dVldt

(j,m,i<oo).

o O'g~l>t

This is our substitute for the formal computation in (2.1). The proof now
consists in letting i, m, and j tend to infinity in that order, eliminating the un-

The Boltzmann Equation. I

11

desired terms on the way. We consider separately each term in the last member of
(2.6), starting with the middle term:

~(Q' gS~ni-hjgJ~ni)do 1
=~Q gmidvx+SCgmi(VO[gmi(V2)-fo(v2)-d
J
J
J
.-1 e x p ( - I v2 [2)]dvtdv2du
Here the first term is zero by (0.8) with tp= 1, and the other term is non-positive
J
gm~(V2)
do2 <~ gJ oo(v2) do2 --~ [fo (v2) + j-x e x p ( - I v 212)] do2.

(2.7)
Hence

t"
j
~j9 (Q t gmi-hjg~,)dv,
dt<O.

(2.8)

We next consider the last term in (2.6)

t"

o~a , J > ( i - Q ' gimi)log-g~idv , dt

(2.9)

t"
<~

( i - Q ' g~o~)(logeto-C,olV, 12)dv, dt=o(1),

0 Q'g~oo>i

when i--*oo. Here we have used (2.2) with

to>t",

eto< 1,

Cto=>0.

Finally, the first term on the right side of (2.6) is rewritten by (2.7) (el. also (2.1)):
t t,

j
~ ( Q t gmi-hj
J
Y
g~i)loggmidvx
dt
0
t"

j v2, t ) - f o ( v 2 ) - J - l
= S S C g~i(v, t) [-gmi(
0

9exp(-Iv212] log g~(vl, t) dr1 dr2 du d t

(2.10)

t"

t'"

+ S~QgJ2,1ogg~idv, dt< ~ NgJ~,,dt

0

ttt

+ ~ c~'i.,(v2, t)(gL(o2, t)-g~ ~ (o. t))log- gL(o. t)do, do2 du.

0

By (2.8)-(2.10), the inequality (2.6) gives

t'"

-- ~ Ng~,dt+Sg~ilog + g~ile, dv,

0

(2.11)

<~gJilogg~,lodv,-$ g.,i
j log - gm~lt-dv,
j
,,,
+ C S I g~n,,(o,, t)(g]n,(v2, t ) - g ~ ( v 2 , t))
0

9log- gJ.,(v,, t) dv, do 2 du dr+o(1),

when i ~ oo. But

O > = g J i - - g ~ g ~ i o - gjJ ~o

(i>=io),

12

L. ARKERYD ."

and by (2.2)
0 __>log- gJmi> 1ogeto -- Cto [ v 12.
Hence all terms on the right in (2.11) can be estimated from above independently
of i > io. Then with gJm= tlimgJmi, Fatou's lemma implies

~'mlOg+ g ~ L 1

(t_-->0).

Moreover by Fatou's lemma and Lebesgue's theorem on dominated convergence,

applied to (2.11) when i-,oo, we have

S g~/.logg~l,,,dvl
t'l

(2.12)

< gr,
J loggmlodv,+C~.
J
J
'
=[.
[.gm(Vl,
t) (g~(v 2, t)--gJ~(v2,
t))
0

9log- gJ,,(vl, t)dvldv2dudt.

Here we have neglected the term on the left,
Ire

SNg

m,

with non-negative integrand.

By (2.5) the sequence (gJ~)m=j
9 co is non-increasing and non-negative, and so the
limit function g J= lim g~mis in L 1 together with gJ log + gJ.
m-*~

From
xlogx> -y+xlogy

(2.13)

(x>0, y > 0 )

follows
0> flog-

(2.14)

gJ~ --e-Ivl2--Iv 12gJ~L1.

Hence gJ log gJ belongs to L 1. From (2.12) we also get

S gJloggJlt,,dvl
t"

<=~ gJ log gJ I0 dr, + C ~ ~ gJ(vl, t) [gJ (v2, t ) - g~ ~o(v2, t)]

(2.15)

9log- gJ(vl, t) d v 1 d v 2 d u d t.
Using (2.14) and
$ gJ(v, t) lvlZdv<~gjo~(v, t)IvlZdv=Sgj~o(v, O) lvl2dv
(2.16)

<S gI~ (v, 0) lvl s dr,

we can estimate the right-hand side in (2.15) independently ofj. For the limit g of
the non-decreasing sequence (ga)~o=1 follows
g(v, t) (1 +lvlZ)mL1
by (2.16), and so
g log- g eL 1
by (2.14). Then
glog + g~L I,

The Boltzmann Equation. I

13

by Fatou's lemma applied to (2.15). Moreover g ~ satisfies

t

j
m
gmi(t)
=e - - h j t f~i+
Ie-hJO-S)Q~gYmi(s)ds.
0

We take the successive limits in this equation and get

t

g(t)=e-htfo+ Se-h(t-~)Q ' g(s)ds.

0

As in (2.4), from here it follows that g is the solutionfof the Boltzmann equation
(0.1). Then
lim [IgJ(t)-f(t)[[ = 0 ,
lim~ gJ(v, t ) a v = I f o ( v ) d v ,
and so
lim IIgj o~(t) - - f (t)II ~ lim [I (g~ ~ (v, t)-- gJ (v, t)) d v + 11gJ (t) -f (t)II]

=Ifo(v) av-lfo(V)aV+o=o.
We have thus proved
f ( v , t)= lim g~o (v, t)= lira gJ(v, t),
j--, oo
j ~ oo

which together with (2.15) implies

t'"

Sflog/I,,, dr1 < j'flogf Io dvl + C I If(v. t) If(v2, t) -f(v~, t)]

0

9l o g - f ( v 1, t)dv I dv 2 du d t = S f l o g f [ o d v 1,
i.e.

(2.17)

~(t")__<~e(0).

Now, knowing that f(v, t ) l o g f ( v , t ) e L 1 (t>0), we can start from any t ' > 0,
not necessarily from t'= 0 as above. Then (2.17) gives
<
M'(t " )=~r

(2.18)

'

(t">t').

This completes the proof.

3. Existence in the Unbounded Case

In this section we consider the Boltzmann equation (0.1) for the general type
of J and k defined in the Introduction. Our results are based on a compactness
argument.
Lemma3.1
such that

(cf. [8]). Let (f~)~ be a sequence of non-negative Ll(Ra)-functions

IIf, ll~<C~

( n = 1 , 2 .... )

for some x > 0 and such that

S f. (v) logf~ (v) d v < C.

113

14

L. ARKERYD:

Then (f~)T contains a sub-sequence (f~j)~~ 1 converging weakly to a function f ~ L 1,

and
lim ~f.~ (v) ~o(v) d v = Sf (v) r (v) d v,
j--* oo

if
O_K
<_ , <K.

(l +lvl)-~'q~(v)EL |
Proof. Assume that

K'

I L(v)(l +lvl2)~ dv=~

r--.oo Ivl~_r
lim

(3.1)
uniformly in n, and that

(3.2)

If.(v)log+f.(v)dv<=C '

( n = l , 2 .... ).

Set

og.N= {veR3 ; f.(v) > e~} .

From (3.2) follows

j IL(v)ldv= I f~(v) d v ~ N - 1 I f~(v)l~

~n N

~gn N

dv<=N-lc',

O)n N

and so

~lf,(v)ldv<e N ~ d v + N - l C '
03

~O\~OnN

for any measurable set o ) c R 3. This implies the uniform absolute continuity

of If..
The above argument is a special case of Nagumo's criterion (of. [6]). Now the
lemma follows from the uniform absolute continuity of ~f~ and (3.1) by compactness, the Dunford-Pettis theorem and ~mulian's theorem (cf. [4]).
But (3.1) and (3.2) are easily proved. First (3.1) follows from
Kt

I f~(v)(1-t-lvl2)--2-dv<=(l+r2) ~'-~

Ilf~ll~<=(1+r2)~'-~c~=~

Io1_~,
when r~oo. To prove (3.2), we notice that
0_->f, l o g - f . > -e-lvl~-Iul~f,
by (2.13) with y = exp ( - Iv IX). Then
S f. (v) log + f. (v) d v = I f. (v) logf. (v) d v - I f. (v) log- f. (v) d v
R3

R3

R 3

<=C+ S exp(-lvl~)dv+C~ =C''

R3

This completes the proof of the lemma.

Theorem 3.2. Assume that k satisfies (0.7) with 0 < 2 < 2 . There exists a solution

f > O of (O.1), such that

(l+lvl2)f(v, 0~LI(R a)

(t>0),

and such that

I ~o(v)f(v, t)dv= I ~P(V)fo(v) do
R 3

R 3

(t>0)

The BoltzmarmEquation. I

15

for cp= l, v, if
fologfo~L1, (l+[vlZ)fo(v)~L ~.

fo>0,

Proof. The idea is to approximate k with a sequence of bounded functions (k,)~~

kn(vi, v2)=min(k(va, v2), n).

The corresponding collision operators are denoted by Q,, and the solutions of the
equations

df
d--i=Q.f (t>0),

(0.1).

f(0)=fo,

are written

(L(t))?.

(3.3)

They exist by Proposition 1.1, satisfy

Sfn(v, t) qJ(v)dv=S fo(v)tp(v) dv

(3.4)

for tp(v)= 1, v, Iv[ z by Proposition 1.2, and

(3.5)

~f,(v, t)logf,(v, t)dv<=~f,(v, O)logfn(V, O)dv=~fo(v)logfo(v)dv

by Theorem 2.1. Using Lemma 3.1, we can choose a subsequence (f,j) of (3.3)
which converges in weak Ll-sense to a non-negative Ll-function for all rational
t>0. The sequence (f,) is equicontinuous in t. It follows that the subsequence
converges in weak Ll-sense to a non-negative Ll-function for all t >0. We write
the weak limit
lim f,j(t) =f(t).
j~oo

Again by Lemma 3.1, the relation (3.4) implies

~f(v, 0 r

dv=Sfo(v ) r

dv

f o r r = 1, v. M o r e o v e r

Qfnj~Qf
in weak LX-sense as 2 < 2. It follows thatfis a solution of (0.1).
In the same way, using also Proposition 1.3, we get
Theorem 3.3.

Assume k satisfies (0.7) with 0_<2<2. There exists a solution f>O

of(0.1) such that

(l+lvl~)f(v, t)~LI(R a)

and

(t>O),

q~(v)f(v, t)dv= ~ r
R3

(t>0)

R3

for tp=l, v, Ivl 2, ~f~>2 and

A_>0,
Moreover

fologfoeL~, (l+lvIDfo(v)~LL

IIf(0ll~_-<Ct, lifoII~

(O<t<q),

with Ctl only depending on

Ilfoll~, q, Ck, and x.

16

L. ARKERYD"The Boltzmann Equation. I

References
1. ARKERYD,L., On the Boltzmarm equation, Part II. The full initial-value problem, following
in this issue.
2. ARKERYD,L., An existence theorem for the Boltzmann equation. (1971), to appear.
3. CARLEMAN,T., Th6orie cin6tique des gaz. Uppsala 1957.
4. EDWARDS,R. E., Functional Analysis. N.Y. 1965 (see p. 274 and p. 549).
5. GRAD,H., Asymptotic equivalence of the Navier-Stekes and nonlinear Boltzmann equation.
Prec. of Syrup. in Appl. Math. 17, 154-183 (1965).
6. Mr
E. J., Integration. Princeton 1947 (see p. 176).
7. MORGENSTERN,D., General existence and uniqueness proof for spatially homogeneous
solutions of the Maxwell-Boltzmann equation in the case of Maxwellian molecules.
Prec. Nat. Acad. Sci. U.S.A. 40, 719-721 (1954).
8. ]V[ORGENSTERN,D., Analyticalstudies related to the Maxwell-Boltzmann equation. J. Rational
Mech. Anal. 4, 533-555 (1955).
9. POVZNER,A. JA., About the Boltzmann equation in kinetic gas theory. Mat. Shorn. 58 (100),
65-86 (1962).
10. WinD,E., On Boltzmann's equation in the kinetic theory of gases. Prec. Cambr. Phil. Sec.
47, 602-609 (1951).
Department of Mathematics
Chalmers University of Technology
and
University of G6teborg

(Received October 5, 1971)