Sup erpro cesses and McKean-Vlasov
equations with creation of mass
L. Overb eck
Department of Statistics,
University of California, Berkeley,
367, Evans Hall
Berkeley, CA 94720,
y
U.S.A.
Abstract
Weak solutions of McKean-Vlasov equations with creation of mass
are given in terms of sup erpro cesses. The solutions can b e approxi-
mated by a sequence of non-interacting sup erpro cesses or by the mean-
eld of multityp e sup erpro cesses with mean- eld interaction. The lat-
ter approximation is asso ciated with a propagation of chaos statement
for weakly interacting multityp e sup erpro cesses.
Running title: Sup erpro cesses and McKean-Vlasov equations .
1 Intro duction
Sup erpro cesses are useful in solving nonlinear partial di erential equation of
1+
the typ e f = f , 2 0; 1], cf. [Dy]. Wenowchange the p oint of view and
showhowtheyprovide sto chastic solutions of nonlinear partial di erential
Supp orted byanFellowship of the Deutsche Forschungsgemeinschaft.
y
On leave from the Universitat Bonn, Institut fur Angewandte Mathematik, Wegelerstr.
6, 53115 Bonn, Germany. 1
equation of McKean-Vlasovtyp e, i.e. wewant to nd weak solutions of
d d
2
X X
@ @ @
+ d x; + bx; : 1.1 = a x;
t i t t t t t ij t
@t @x @x @x
i j i
i=1 i;j =1
d
Aweak solution = 2 C [0;T];MIR satis es
s
Z
2
t
X X
@ @
a f = f + f + d f + b f ds:
s ij s t 0 i s s
@x @x @x
0
i j i
Equation 1.1 generalizes McKean-Vlasov equations of twodi erenttyp es.
Firstly,ifb = 0 it is the classical McKean-Vlasov equations asso ciated with
d
nonlinear IR -valued di usions and studied in many pap ers, e.g. [L,Oel,S1,S2].
If only b dep ends on then this equation is studied in [CR], [COR]. It is
used in applications for some biological problems, like conduction of nerve
impulse, cf. [GS,SP] and the references therein. Equations like 1.1 with
non-linear drift in the di usion and nonlinear reaction term are sometimes
called reaction-convection-di usion equation [Lo, p. 287] and can b e found
in many biological applications, cf. [M, Section 11.4].
Our main result concerning equation 1.1 is Theorem 3.1 in Section 3 and
claims that under Lipschitz conditions on the co ecients of 1.1 a unique
solution of 1.1 can b e found as the limit of the interacting one-typ e su-
P
1 N
i;N 1;N N;N
p erpro cess X as N !1where X ;:::;X is a N-typ e
i=1
N
sup erpro cess with mean- eld interaction. Obviously, this result should b e
emb edded in a Propagation of Chaos statementforweakly interacting N-
P
1
typ e sup erpro cesses including in particular tightness of i;N . This is
X
N
the topic of Section 2. However, this result do es not necessarily implies that
the limit pro cess gives a solution of 1.1, b ecause the application I from
d d
M M IR to M IR de ned by
1
Z
I mf := f md 1.2
fails to b e continuous. Finally, in Prop osition 3.3 we construct a unique
solution of 1.1 by a Picard-Lindelof approximation under stronger condition
as in Theorem 3.1. This approximation gives a solution of 1.1 as a limit of
intensity measures of sup erpro cesses with non-interactive immigration. 2
2 Weakly interacting and non-linear sup erpro cesses
In this section we construct a non-linear sup erpro cess as an accumulation
p oint of a sequence of weakly interacting multityp e sup erpro cesses.
2.1 Weakly interacting superprocesses
First wewant to describ e an N-typ e sup erpro cess with an interaction de-
p ending only on the empirical pro cess. We start with real-valued resp.
p ositive functions a ;d ; 1 i; j; k d; and b resp. c de ned on
ij k
d d
[0; 1 IR M M IR , where M E denotes the space of probability
1
1
measures on the Polish space E equipp ed with the top ology of weak conver-
d
gence. De ne for every s 0, m 2 M M IR the time-inhomogeneous
1
2 d
op erator Ls; mon C IR by
0
d d
2
X X
@ f @f
Ls; mf x:= a s; x; m x+ d s; x; m x: 2.1
ij i
@x @x @x
i j i
i;j =1 i=1
The op erator serves as a description of the one-particle motion. Let us de-
note the family Ls; m d by L.For the weakly interacting
s2[0;1;m2M M IR
1
sup erpro cess the function bs; x; m describ es the mean branching rate and
cs; x; m the variance in the branching rate while the empirical distribution
of the pro cess equals m.
In order to state now the basic de nition we need some notation. Let N 2 IN
+
d N d N
~
b e xed. If ~ = ;:::; 2 M IR ; f =f ;:::;f 2 B IR ;
1 N 1 N
b
+
where B are b ounded measurable non-negative functions, then
b
N
X
1
d
2 M M IR R~ :=
1
i
N
i=1
R
P
N
~
fd: The exp onential function e f ; where f = and ~ f :=
~ i i
i=1
f
~
is de ned by e ~=exp ~ f : Finally,we denote for a Polish space E
~
f
the space of E -valued continuous paths by C .
E
N
De nition 2.1 We cal l a measure P 2 M C d N an N-typeweakly
1
M IR 3
interacting measure-valued di usion starting at ~ if
0
Z
N
t
X
j
~ ~ ~
~ + X e e X Ls; RX f + 2.2 X e
0 t ~ ~ s j s ~
s
f f f
0
j =1
2
~ ~
bs; RX f cs; RX f ds
s j s
j
N 2
~ ~
is a P martingale for al l f =f ;:::;f with non-negative f 2 C . X
1 N i
0
denotes the coordinate process on C d N .
M IR
2.2 Propagation of Chaos
One main issue in this section is the question under which conditions the
N 1 1
sequence fP g is P -chaotic with a measure P 2 M C d .
N 2IN 1
M IR
1 N
De nition 2.2 Let P 2 M C d . We say that the sequence fP g
1 N 2IN
M IR
1
is P -chaotic if for every k 2 IN
N 1 k 1
P X ;:::;X = P as N k tends to in nity ; 2.3
1 k
i=1
cf. [S1,S2].
Intuitively, this means that the interaction disapp ears if the number of typ es
1
N tends to in nity.Typically the measure P is a solution of a non-linear
martingale problem. Roughly sp eaking, b ecause the law of large numb ers
N 1 1 1
~
implies RX ! P X , the measure P solves the martingale
s s
problem characterized by the martingale prop erty of the pro cess
Z
t
1 1 1 1
f e X e X + f + bs; P X e X X Ls; P X
f t f 0 f s s
s s
0
1 1 2