Escolar Documentos
Profissional Documentos
Cultura Documentos
df
d--T=Qf
(0.1)
(t>0),
f(0)=fo=>0,
Qf(vl)=
[J*(f|
(01, v 2 ) - ( f |
R 3 B 2 (r)
The notations are those of Part I, just preceding. The impact parameter u is restricted
to a disc
B2(r)={u~R2; [ul<r},
and the kernel k(vt, v2) is a measurable, non-negative function with
(0.2)
(0<,l__<2)
and invariant under ~,*, J*, Z * k = J * k = k . Here J* is induced by a Cl-diffeomorphism J on R a x R 3 x B z (r), restricted in a certain way. With
1 : R 3 x Ra--*R: (vl, v l ) ~ l
p: R 3 x R a - - * R 3 : ( v l , v2)--~v 1 + v 2
lo./.=1,
poJ,=p,
(0.4)
~oJ,=J, oZ,
(0.5)
J o J = identity.
ToJ.=T,
18
L. ARKERYD:
For a discussion of these conditions, the Boltzmann equation and previous results
on the full initial-value problem, see Part I. There we have proved an existence
theorem for (0.1) under physically natural conditions on fo(fo logfo,fo(v).
(1 +Iv 12)~L1 (RS)).
The present paper considers the full initial-value problem (0.1), i.e. the questions of existence, uniqueness, positivity, and asymptotic behavior including
equilibrium solutions, under the additional hypothesis that the fourth moment of
fo exists (fo(v)Ivl%LX).
In Section 1 we give a constructive Ll-proof of existence and uniqueness for
(0.1), when fo(v)(l+lvl*)zL t (no entropy requirement), based on the monotonicity argument of the bounded case in Part I. For a continuous k(vt, v2) =
k(Ivl-v21) and 0___2<1 in (0.2), this corresponds to the results obtained by
A. JA. POVZNER[6] through a measure-compactness argument. We also prove the
~-theorem employing the method introduced in Part I for the bounded case.
The rest of the paper is devoted to the questions of positivity and asymptotic
behavior under some additional restrictions on J and k (Condition 2.2 and
Condition 3.1 below). Section 2 contains a strict positivity result for the solution
of the Boltzmann equation and demonstrates that the only equilibrium solutions
are the expected Maxwell distributions. The main step thereof has independent
interest, namely the proof that the only non-negative, measurable solutions of the
Cauchy equation
f(vl+ql)f(vl+q2)=f(vl)f(v1+qt+q2)
(a.e. q l l q 2 )
Section 3 proves that moments of high order are bounded uniformly in time
provided certain still higher moments exist initially. On the basis of the ~r
this section also demonstrates a weak Ll-convergence of the solution to the
corresponding Maxwell distribution when t approaches infinity.
1. Existence, Uniqueness and the .Yt~-Theorem
The existence proof is based on the same monotonicity argument as in Part I,
0__<~f_<__~3g if o<=f<=g.
Take h (v) such that
0~h(v)~C(l+lvl2).
The initial-value problem
- dd~ f +
(1.1)
hf=C~f (t>0),
f(0) =fo ,
(1.2)
fl=0,
f"+l=e-h'fo+ Se-h(tl)cSff(s)ds
0
(n>l).
19
h<h',
fd<fo,
and
~'g<~g
(g>0).
O<f'<f.
For details, see Part I.
As in Part I, we use C and K to denote various constants and set
/r
Ilfll= Sf(v)dv
and
R3
R3
(v~, v~)=Ju(v,,v2)
and
B 2=B2(l/V-~)
(giving
S du = 1).
(1.3)
B2
0<(1 + Iv ID fo (v)~L1(R 3)
for some x> 4. Then there exists a unique solution f > O of(0.1), such that
(l+lvl~)f(v,t)ELl(R3),
IIf(t)ll2=llfoll2
(t>0),
and
Ilf(t)ll~<Cto I]foll~ (t<to).
(1.4)
(l.5)m
f(0) =fo
with C,o only depending on Ilfol12, to>0, ~c and Ck of (0.2). Here we have used
Proposition 1.1-1.3 of Part I.
After introducing
20
L. ARr.ERYD:
we rewrite (1.5),~ as
(1.5)"
a/
d-----i-+hf=Q,f (t>0),
f(0) =f0,
jf(vl) g(v2) k(vl, v2) dr2 du < Ck(1 +l vl IX)f(vl) ~ g(v2) (1 +1021 a) dr2,
and thus Q" is positive, monotone for K sufficiently large, as 0 < ~ < 2. The operator
Q; defined by
Qmf=I 3* (f |
d v2 d u
o<o" f-Q~,f
(0<f),
O<Q'~f-Q~'f
(O<f,O<m-j).
and
It follows that the equations (1.5)" and
df
d----t4-hf = Q "" f (t>0),
(1.5)~
f(0)=fo
f " <fro.
Analogously two limit functionsf" and f ] ' with the same initial valuefo satisfy
fj'<f"
(1.9)
(j<m).
It follows that
(j<m).
(1.~o)
with
(1.11)
IIf(t)ll2 ~ Ilfol12,
and
t
(1.12)
21
Moreover by (1.7)
(t < to),
df ~_hf=Qf+KS(l+lvll2)(l+lv2]2 )
dt
9f(vl, t)f(v2, t)dudo2 ( t > 0 ) ,
(1.13)
f(0)=fo,
d---{s,,+hs,~=Q"fm-Q'~f~,',
with sm=f~-f". By (1.8) sm is non-negative. Now
k(vl, v2)-k,(vl, v2)=0,
say for
1
max (I vl I, I v2 I) < (m Ck 1 _ 1) a2
Then for t< to(to >0) using (0.3)--(0.5), we easily get
9(1+10112 ) d r ldv2dudz
<=O+tKCtollfoll+ sup
Ilsm(x)ll2+o(1)
O__.+~to
a s m - + o0.
It follows that
(1.14)
Ilsm(t)ll2~0
(m--+ oo)
!11,-+oo
22
L. ARKERVD:
(1.5)
Corollary 1.2. The integral equation (1.12) has a unique solution for which (1.6) and
(1.7) hold. This solution is also the solution of(0.1).
Proof. A solution of (1.12) satisfying (1.7) is also a solution of (1.13). The corollary
now follows by the uniqueness proof of Theorem 1.1 applied first to the iterated
solution (1.2) and the solution lim f " , then to the iterated solution (1.2) and an
arbitrary solution,
m-.|
Theorem 1.3. Suppose f(vl, t) is the solution of(0.1) with
(t >0),
and
~,o(t) = Sf (v, t) logf(v, t) d v
is non-increasing as a function of t.
Proof. This follows in the same way as Theorem 2.1 of Part I. The starting point
in Part I was the Boltzmann equation written as (1.I) (obtained by (1.9) of Part I).
Here (1.13) is of the desired type and a suitable starting point. We apply Corollary 1.2 above for the uniqueness argument in the last step of the proof.
Corollary 1.4. The sets
(0<x<2)
k(vl, v2)>0
(2.1)
a.e.,
as well as on the possibility of using the scattering angle instead of the impact
parameter as a variable of integration in Q.
Let us introduce some restrictions on J to handle the scattering angle. Set
ql = ql(vl,
23
We note that
(2.2)
q i + q2 = v2 Vl
-
(ql" q2) = 0 .
R3
and ~ in
S={~ee3; I~1=1}
are given. The set
(2.4)
is mapped into the ball
(2.5)
by the mapping
(2.6)
from R 3 x B 2 into
M = {(q,, ~)eR 3 x S; (q,, 0 > 0 } u {(0, 4); ~ S } .
Condition 2.2. Let k satisfy (2.1), and for a.e. Vl~R 3 let the mapping (2.7) be
bijective from R a x B 2 onto M after deleting a set of measure zero from each
space, (2.7) also allowing a formula for "changing variables":
R3 x B2
~k(r~,u)= M~ #/x(ql,r
u(qx, 0 ) .
Theorem 2.3. Let f(vl, t) be the solution of the Boltzmann equation (0.1) under
Condition 2.2 and with
O< f o ( v ) ( l +lvl*)~L t ,
Ilfoll > 0 .
Then
f(v, t ) > 0
24
L. ARKERYD
Remark 2.4. This weak positivity property is sufficient for the asymptotic
results of the present paper. However, it can easily be strengthened into an estimate
of f from below by a positive function, if Condition 2.2 is somewhat restricted.
Moreover the theorem itself is true with the same proof for the solutions of
Theorems 3.2-3.3 in Part I, as well as for the solution in [2].
Proof of Theorem 2.3. The positivity trivially follows from (1.12) without any
extra restrictions on J or k, w h e n f o > 0 a.e. I f f o > 0 only on some set of positive
measure, then the positivity is a consequence of the next two lemmas.
Lemma 2.5. Under the conditions of Theorem 2.3 there are r x > 0 and ~ ~ R a depending
on fo, such that
f(o, t)>O
(t>O, v~B3(rl, ~:)).
Lemma 2.6. Assume that the conditions of Theorem 2.3 hoM and that
f(v, t)>O
(t>O, w B S ( r , , O)
f>=fo e-h' ,
and that
t
f > Se-h('-OQ'~foe-h~dz.
(2.8)
We estimate the right member through an estimate of the first term in Q~. For
that purpose we are interested in
fo (o~)fo (v[) =fo (01 + ql) fo (or + q2).
The difference v2 -Vx = re is a function of Vl, ql and ~ by the bijectivity of (2.7).
Then q2 = v2 - Vl - qt andf(vx + q l ) f ( v l + q2) are functions of ql for vI and ~ given.
By hypothesis fo > 0 on a set O of positive measure in R 3. Starting from a Vitali
system of, say, cubes, O can be covered a.e. by a sequence of disjoint cubes (An)~'
with 13An approximating f2 arbitrarily well in measure. As a consequence, there
1
is a 6 > 0 and ~ R 3, such thatfo > 0 on a set of measure larger than, say, 2" "~n~53
contained in B 3 (6, vo), for every
B3(6, Vo)=B3( 4C5,O.
Let vleB3(26, ~), and let ~ e S be given. From the property that q l ~ q 2 is
measure preserving, which is an easy consequence of (2.2) and (2.3), we conclude
that
fo(vl + q l ) f o ( v l + q2) > 0
25
f(v 1, t)>0
(2.9)
q~N~,,
G~N~o.
(2.10)
t'
Here
du dv~ dv 2 .
26
L. ARKERYD"
Take Vl such that (2.11) holds for a.e. (v2, u)eR 3 x B 2. With (ql, 0 as new
variables (2.11) holds by Condition 2.2 for a.e. (qx, O e M . Taking (ql, q2) as new
variables, (2.11) is satisfied for a.e. (ql, q2)e M ' with
M ' = {(ql, q2)eR a x R3; (qx" q2)----0},
i.e.
(2.12)
C2/e>f(v)>eZ/c
a.e.
in some smaller ball. This implies, similarly to the proof of Lemma 2.6, that f is
uniformly bounded away from zero and infinity in the a.e. sense on all balls with
the same centre, hence also locally on R 3.
With vt such that (2.12) holds for a.e. (q~, qz)eM' and with
~b(v) = l o g f ( v + v t ) - l o g f ( v 0 ,
the equation (2.12) can be written
(2.13)
~/(ql)+~k(q2)=~/(ql +q2)
27
X2
Xa
Xl
=x~x~x3
X2
~ ~(x2e2)dx~/X~+ ~(x~e~)dx~/X3
1
ivl
X(v)= S~k(tv)dt= ~ ~,(tv/Ivl)dt/lvl
0
and so X is continuous in R 3 and (2.15) holds for all (ql, q2) eM'. By [3, p. 21-23]
this implies
X(v)=a' Ivl2 +(b '. v),
from which the lemma follows by taking the derivative in (2.14).
Theorem 2.9. Let f be the solution of the Boltzmann equation (0.1), with initial value
fo, satisfying
fo(v)(l+lvl4), fologfo~L 1.
Proof. The " i f " part is evident since Qf=O for the functions of (2.16). On the
other hand, if f is an equilibrium solution, then ~ ' ( t ) is constant, and so Nf=O
by (2.10). This implies (2.12), and then the "only if" part follows from Theorem 2.7.
Corollary 2.10. 3~'(t) is strictly decreasing until f becomes an equilibrium solution,
under the conditions of the previous theorem.
Proof. We have just seen that Nf(to)= 0 implies
f(v, to) = C exp(a [ v 12+ (b- v)).
28
L. ARrd/RYD:
Thisf(v, to), however, is also a solution for t ~ to, and by the uniqueness the only
solution. Hence Nf(t)<0 until f becomes an equilibrium solution, and ~ ( t ) is
strictly decreasing by (2.10).
3. Asymptotic Behavior of the Solution and the Higher Moments
To study the boundedness of the higher moments, we need a simple consequence of (2.2) and (2.3). Write as above
v~=vl+qj
( j = l , 2),
Vj=xV~+2vj
( j = l , 2)
and decompose
with tvj parallel to q~ (i= 1, 2). Then
Iv'llZ"+lv'212n-lvl12"-Iv212"
(3.1)
n-1
2 ,-1
j=l
v=l
~lvl2"f(v,t)dv
(n>l)
(3.2)
([vt [ > r ) ,
f2
n-1
~ ~, [xvllZVlzvllZ("-')du>>_Clvl[ z"
(3.3)
(vz~f2, lVxl>r),
B2 v = I
for some bounded set ~ c R 3 of positive measure and some C, r, and 5 > 0 depending on ilfo]l and Ilfoll2.
Remark 3.2. The inequality (3.2) holds e.g. in the physical case, i.e. for
k(Vl, v2)=C I v t - v 2 l .
Then
Ivl~N
I~I
and so
llf(t>l12jSCllf~l12j ( t 2 O )
fo(4
(1 + l v I K ) ~ ~ l
d t Jfrn(v17t ) I 01 12"dvl
30
L. ARKERYD"
do I d v2
t!
~ J,n[Vl, t)fm(V2,
t) km(vl, vz) dvldv2
ff-ff-[jfr.(Va, t) lv~lZ"dvx+C $
vzefl
n-2
'r |
Ilfm(t)ll2n-2
we next get
v2EIl
=Rm(t).
We conclude that
~C
(3.5)
~C
i vl ~")
by (1.4) and (1.15) as r > 2 n. Taking the limit in (3.5) and applying (3.2) we thus get
n-2
d z.
31
Sf(v,, t) I v, ]2nd v, < Sfo (v,) I v, I ~" e -~ c t/~ d v 1 + ~ e-" cr 0/2 C"
0
_<max
v 2ndv,
S~(v)dv=A,
~g/(v)vdv=B,
~(v)lvl2dv<=C.
Let
r
I v l ~ + ( b 9 v) + c)
be the unique exponentialfunction satisfying (3.6) with equality also in the third
relation. Then every ~ e c~ satisfies
(3.7)
and there is equality in (3.7) only for @=@oWe can now, at last, study the asymptotic behavior o f f by an Ll-analogue of a
method developed by T. CARLEMAN[3, p. 59--62] for the maximum-norm.
Theorem 3.6. Assume that Condition 2.2 and Condition 3.1 hold, 0__<2<2 in (0.2)
and that
for some K>4. Then the solutkm f of (O.1) converges in the weak Ll-sense to the
unique equilibrium function
~/o(v)=exp(a [vl2+(b 9v)+c),
which satisfies
tp(v) ~o (v) dv = ~ q~(v) f o (v) dv
for tp= l, v, lvl z.
Proof. g ( t ) is bounded from below by Gibbs' Lemma 3.5. Hence by (2.10) there
lim N f ( t . ) = O.
n.--~ oo
for a.e. (v,, v2, u)eR3 x R3 x B2. If not, then there is a bounded measurable set w
of measure m >0 in R3 x R~ x B2, and a constant Co,0< Co< 1, with
JAdvldv2du=Co.
0
This leads to a contradiction of (3.8), and thus (3.10) holds. We demonstrate the
contradiction.
Set
A ( t ) = f ( v ; ,t ) f ( v ; , t ) - f ( v l , t)f(v,, t ) ,
A ( t )= max ( A (t),0 ) .
+
7 hen
( ~ 2 0y>O),
,
33
has measure smaller than 1/N as 11 f (t) 11 = 11 fo 11. We conclude that there is a set
w, c w of measure m - 1/N, such that
on wv and
~ A ( t n V ) d v l d v 2 d u > C o / 2 (v>v,)
(3.12)
for some v,.
Take e > 0 such that
5 - J (A+(t,Y))2/2~l
du dv, dv2 -log(l + E ) J A (tnV)du dvl dv2
+
%1
Ov2
ov2
But by (3.12)
-2
as O<Co<l. Hence
-C
Nf( t n v ) s
16
c2
O < O
independent of v for
C2= min [tm e l , log(1 E ) ] . This, however, is impossible by (3.8), and so
(3.10) holds. Using Theorem 2.7, we conclude that $ (v) = $, (v) .
If i= lim t, < co,then $, (v) =f(v, i) by the continuity of f, and the theorem
n+ w
,,
34
lim S Nf(t)dt=O,
as N f < O and ~r (t) is bounded from below by Gibbs' Lemma 3.5. Hence
limNf(t+nO=O
(3.14)
for v>vl,
and
lim f (t + n~)=~,o
j~o3
GraBs, J. W., Collected Works, Vol. II. New Haven 1906 (see p. 130).