Definitions Introduction and main result
By g
∗
we denote the unique solution continuous on K and twice continuously differentiable on K
◦
of the equation −
1 2
g
∗
= 1 on K
◦
g
∗
= 0 on ∂ K ,
4.1.17 with
=
α ∂
∂ x
α
2
the Laplacian. By
D
K
[0, ∞ we denote the space of cadlag functions from [0, ∞ to K ,
equipped with the Skorohod topology. We use the symbol ⇒ to denote weak
convergence of processes in path space, i.e., the weak convergence of their laws as probability measures on
D
K
[0, ∞.
d = 1: For one-dimensional K , exact results have been derived about the asymp-
totic behavior of ˆ X
N,k
. Let K = [0, 1]. Let
H
be the class of Lipschitz continuous functions g : [0, 1]
→ [0, ∞ satisfying gx = 0 ⇔ x ∈ {0, 1}. Then 4.1.16 has a unique strong solution for every g
∈
H
, x ∈ K and c ∈ [0, ∞. We cite the
following result from Dawson and Greven [10].
Proposition 4.1.1 If g
∈
H
, then for every k ≥ 0
ˆX
N,k
⇒ Z
g
k
, c
k
θ
as N → ∞,
4.1.18 where c
k
: = σ
k
c
k
, and g
k
∈
H
is the function g
k
: = σ
k
F
c
k −1
◦ · · · F
c
1
◦ F
c
g. 4.1.19
Here F
c
:
H
→
H
is a renormalization transformation given by F
c
gx : =
[0,1]
gyν
g,c x
d y x
∈ [0, 1], 4.1.20
where ν
g,c x
is the unique equilibrium distribution of 4.1.16. In Baillon et al. [1] the behavior of the function g
k
was studied in the limit as k
→ ∞. The main result of that paper is the following:
Proposition 4.1.2 For any g
∈
H
lim
k →∞
sup
x ∈[0,1]
|g
k
x − g
∗
x | = 0,
4.1.21 where the function g
∗
is given by g
∗
x : = x1 − x
x ∈ [0, 1].
4.1.22
Since c
k
→ c
∗
= c
1 − c
as k → ∞,
4.1.23 we may combine Propositions 4.1.1 and 4.1.2 to obtain see [21]
ˆX
N,k
t ⇒ Z
g
∗
, c
∗
θ
N → ∞, k → ∞,
4.1.24 where the limits need to be taken in the order indicated. The fact that large blocks
are always governed by the same diffusion function g
∗
, regardless of the diffusion function g for single components, is described by saying that systems of the form
4.1.5 exhibit universal behavior on large space-time scales. d
≥ 2: In den Hollander and Swart [21] we conjectured that the result in 4.1.24 generalizes to higher-dimensional K , where the function g
∗
is given by 4.1.17. However, we were only able to prove some partial results in this direction. The
main difficulty we encountered was that it is very hard to prove for a function g that the transformations F
c
k
are well-defined on g and on all its iterates F g,
F
c
◦ F g and so on. This requires that uniqueness holds for solutions of 4.1.16
for all x ∈ K , and not only for g itself, but also for its iterates. This is often already
a problem for g = g
∗
In the present paper, we do not attempt to prove seperate limit theorems for N
→ ∞ and k → ∞ such as Proposition 4.1.1 and Proposition 4.1.2, but instead let N and k tend to infinity together. In the case of one-dimensional K , the Propo-
sitions 4.1.1 and 4.1.2 already imply, through formula 4.1.24, that it is possible to choose N
i
, k
i
, tending to infinity as i → ∞, such that
ˆX
N
i
, k
i
t ⇒ Z
g
∗
, c
∗
θ
as i → ∞.
4.1.25 In this paper, we try to generalize this result to higher-dimensional K . Moreover,
we investigate how N
i
and k
i
must be chosen for the convergence in 4.1.24 to hold.
In what follows, we fix integers N
i
≥ 2, k
i
≥ 1 i ∈
N
tending to infinity as i
→ ∞. We write ˆX
i
: = ˆX
N
i
, k
i
β
i
: = β
N
i
, k
i
ˆ
F
i t
: =
F
i β
i
t
, 4.1.26
where
F
i t
t ≥0
is the filtration generated by the process X
N
i
, k
i
. Unfortunately, we do not know yet how to prove 4.1.25 in the case of higher-dimensional K . How-
ever, we can show that the drift and the diffusion rate of the process ˆ X
N
i
, k
i
converge to the functions x
→ c
∗
θ −x and x → g
∗
x, respectively. The following scaling theorem is our main result.
Theorem 4.1.3 Let N
i
≥ 2, k
i
≥ 1 i ∈
N
be integers tending to infinity as i
→ ∞ such that lim
i →∞
k
i
log N
i
= 0, 4.1.27
and assume that g
∗
∈
C
1
K . Then there exist ˆ
F
i t
t ≥0
-adapted processes ˆ B
i
= ˆ
B
i
t
t ≥0
and ˆ G
i
= ˆ G
i
t
t ≥0
with sample paths in
D R
d
[0, ∞,
D R
[0, ∞, re-
spectively, such that for each f ∈
C
2
K the process M
i
t
t ≥0
, given by
M
i
t : = f ˆX
i
t −
t α
ˆB
i,α
s
∂ ∂
x
α
f ˆX
i
s + ˆ
G
i
s
α ∂
∂ x
α
2
f ˆX
i
s ds
4.1.28 is an ˆ
F
i t
t ≥0
-martingale. Moreover, for each T 0 lim
i →∞
E
T
ˆB
i
t − c
∗
θ − ˆX
i
t
2
dt = 0.
4.1.29 and
lim
i →∞
E
T
ˆG
i
t − g
∗
ˆ X
i
t dt
2
= 0. 4.1.30
Formula 4.1.28 identifies ˆ B
i
and ˆ G
i
as the drift and the diffusion rate of the process ˆ
X
i
. Thus, formulas 4.1.29 and 4.1.30 show that these local character- istics of the process ˆ
X
i
converge, as i → ∞, and that their limits are universal in
the diffusion function g for single components. The fact that this happens for all N
i
, k
i
satisfying condition 4.1.27 is new even in the case of one-dimensional K . Theorem 4.1.3 is a universality result of the type we were originally after in den
Hollander and Swart [21]. The author believes that also the convergence in 4.1.25 holds under condition
4.1.27, but there are at present two technical obstacles to proving a result of that form.
The first difficulty comes from the fact that uniqueness of solutions to 4.1.16 for arbitrary g
= g
∗
, c = c
∗
and x = θ remains an open problem. The following
partial results are known. 1. Strong uniqueness is known to hold for θ
∈ K
◦
, c
∗
sufficiently large, and K satisfying mild regularity conditions Theorem 1.9 in Den Hollander
Swart [21].