Block immobility Convergence of the diffusion rate
where a
i
ξ :
=
k
i
−1 k
=ξ
1 N
k i
c N
i k
−1
ξ ∈
i
. 4.4.39
Lemma 4.4.3 below gives us an estimate of how close the expectation of functions f
i
of large k
i
−1-blocks is to their equilibrium value with respect to the dynamics described by the operator A
i x
ki
.
Lemma 4.4.3 For i
∈
N
let X
N
i
be a solution of 4.1.5 with initial condition 4.4.22 and let τ
i
be as in Section 4.4.2. For i ∈
N
, let f
i
∈
C
2
K
i
, let ∇
ξ
f
i
: =
∂ ∂
x
α ξ
f
i α
=1,...,d
, 4.4.40
and let ∇
ξ
f
i ∞
be the supremum norm of |∇
ξ
f
i
| the Euclidean norm of ∇
ξ
f
i
. Then there exist constants M
1
, M
2
such that for all i ∈
N
, x ∈ K
Ni
and ω ∈
C
ξ
i
, K
[0, ∞
E
ω
A
i x
ki
f
i
X
N
i
τ
i
≤ 2λ
−1 i
f
i ∞
+ M
1
c
k
i
−1
N
1 −
3 2
k
i
i
λ
1 2
i
+ M
2
c
k
i
N
−k
i
i ξ
∈
i
∇
ξ
f
i ∞
. 4.4.41
Here, in the left-hand side of 4.4.41, we lift the function A
i x
k
f
i
to the space K
Ni
in the obvious way. Proof of Lemma 4.4.3: Use the fact that the process M in 4.4.25 is a martingale
and apply optional stopping to see that
E
ω
[ f
i
X
N
i
τ
i
] − E
ω
[ f
i
X
N
i
0] = E
ω τ
i
A f
i
X
N
i
t dt ,
4.4.42 where for x
∈ K
Ni
A f
i
x =
ξ ∈
i
,α ∞
k =1
c N
k −1
[x
k,α ξ
− x
α ξ
]
∂ ∂
x
α ξ
f
i
x +
ξ ∈
i
,α
gx
ξ ∂
∂ x
α ξ
2
f
i
x, 4.4.43
and we have lifted the function f
i
to K
Ni
in the obvious way. Note that ξ
i
∈
i
, so that in our summations we do not have to write ξ
= ξ
i
.
Formula 4.4.42 can be rewritten as
∞
dt E
ω
f
i
X
N
i
t λ
−1 i
e
−tλ
i
− E
ω
f
i
X
N
i
β
i
t =
∞
dt E
ω t
du A f
i
X
N
i
u λ
−1 i
e
−tλ
i
=
∞
dt
t
du E
ω
A f
i
X
N
i
u λ
−1 i
e
−tλ
i
= −e
−tλ
i
t
du E
ω
A f
i
X
N
i
u
∞ t
=0
−
∞
dt −e
−tλ
i
E
ω
A f
i
X
N
i
t =
∞
dt E
ω
A f
i
X
N
i
t e
−tλ
i
, 4.4.44
leading to the equation E
ω
A f
i
X
N
i
τ
i
= λ
−1 i
E
ω
f
i
X
N
i
τ
i
− E
ω
f
i
X
N
i
. 4.4.45
Here A f
i
X
N
i
τ
i
=
ξ ∈
i
∞ k
=1
c N
i k
−1 α
[X
N
i
, k,α
ξ
τ
i
− X
N
i
,α ξ
τ
i
]
∂ ∂
x
α ξ
f
i
X
N
i
τ
i
+
ξ ∈
i
gX
N
i
ξ
τ
i α
∂ ∂
x
α ξ
2
f
i
X
N
i
τ
i
. 4.4.46
In view of our heuristic reasoning in Section 4.4.1, we write this as A f
i
X
N
i
τ
i
=
ξ ∈
i
k
i
−1 k
=1
c N
i k
−1 α
[X
N
i
, k,α
ξ
τ
i
− X
N
i
,α ξ
τ
i
]
∂ ∂
x
α ξ
f
i
X
N
i
τ
i
+
ξ ∈
i
gX
N
i
ξ
τ
i α
∂ ∂
x
α ξ
2
f
i
X
N
i
τ
i
+
ξ ∈
i
c N
i k
i
−1 α
[x
k
i
,α
− X
N
i
,α ξ
τ
i
]
∂ ∂
x
α ξ
f
i
X
N
i
τ
i
+
ξ ∈
i
c N
i k
i
−1 α
[X
N
i
, k
i
,α
τ
i
− x
k
i
,α
]
∂ ∂
x
α ξ
f
i
X
N
i
τ
i
+
ξ ∈
i
∞ k
=k
i
+1
c N
i k
−1 α
[X
N
i
, k,α
τ
i
− X
N
i
,α ξ
τ
i
]
∂ ∂
x
α ξ
f
i
X
N
i
τ
i
. 4.4.47
Here the first two terms represent the ‘internal’ evolution of the k
i
− 1-block around the origin. The third term comes from their interaction with the k
i
-block average, which for an appropriate choice of λ
i
is essentially fixed to its value at time 0. The fourth term is an error term compensating for the fact that the k
i
-block is not completely fixed at its value at time 0. The fifth term describes the small
interaction with the k-blocks for k k
i
. Combining 4.4.45 and 4.4.47, we arrive at
E
ω
A
i x
ki
f
i
X
N
i
τ
i
= λ
−1 i
E
ω
f
i
X
N
i
τ
i
− E
ω
f
i
X
N
i
β
i
t −
c N
i k
i
−1 ξ
∈
i
E
ω α
[X
N
i
, k
i
,α
τ
i
− x
k
i
,α
]
∂ ∂
x
α ξ
f
i
X
N
i
τ
i
−
∞ k
=k
i
+1
c N
i k
−1 ξ
∈
i
E
ω α
[X
N
i
, k,α
τ
i
− X
N
i
,α ξ
τ
i
]
∂ ∂
x
α ξ
f
i
X
N
i
τ
i
. 4.4.48
From this it follows that E
ω
A
i x
ki
f
i
X
N
i
τ
i
≤ 2λ
−1 i
f
i ∞
+ c
N
i k
i
−1
E
ω
X
N
i
, k
i
τ
i
− x
k
i
ξ ∈
i
∇
ξ
f
i ∞
+2R c
N
i k
i
ξ ∈
i
∇
ξ
f
i ∞
, 4.4.49
where we use that N
i
≥ 2 and c 1. We now note that by Corollary 4.4.2 E
ω
X
N
i
, k
i
τ
i
− x
k
i
2
≤ E
ω
X
N,k
τ
i
− x
k
i
2
≤ Mλ
i
N
−k
i
i
. 4.4.50
Inserting this into 4.4.49, we arrive at 4.4.41.