with T
l
t :=
X
i∈I cardKi=l
Y
k∈Ki
s
k
X
n ,i
, 32
here by cardK we denote the cardinality of the set K. The expression 32 can be further recast as T
l
t =
X
K⊂{1,...,d} cardK=l
X
i∈I Ki=K
Y
k∈K
s
k
X
n ,i
= X
K⊂{1,...,d} cardK=l
Y
k∈K
s
k
X
i∈I Ki=K
X
n ,i
.
It should be clear that X
i∈I Ki=K
X
n ,i
= Y
k∈K
∆
k u
k
t
k
+1
S
n
Ut ,
where the symbol Π is intended as the composition product of differences operators. Recalling that s
k
= t
k
− b
k
u
k
t
k
∆b
k
u
k
t
k
+ 1, this leads to T
l
t =
X
K⊂{1,...,d} cardK=l
Y
k∈K
t
k
− b
k
u
k
∆b
k
u
k
t
k
+ 1 Y
k∈K
∆
k u
k
t
k
+1
S
n
Ut .
33
To complete the proof report this expression to the equation 31.
3.3 Rosenthal and Doob inequalities
When applied to our triangular array, Rosenthal inequality for independent non-identically dis- tributed random variables reads
E ¯
¯ ¯
¯ X
1≤ j ≤n
X
n , j
¯ ¯
¯ ¯
q
≤ c X
1≤ j ≤n
σ
2 n
, j
q2
+ X
1≤ j ≤n
E |X
n , j
|
q
, 34
for every q ≥ 2, with a constant c depending on q only. As in [11] we can also extend Doob inequality for independent non-identicaly distributed variables
E max
1≤k≤k
n
|S
n
k|
q
≤ p
p − 1
dq
E |S
n
k
n
|
q
, 35
for q 1.
4 Finite-dimensional distributions
4.1 Proof of the proposition 6
We have gt, s =
lim
mn→∞
µ
n
t ∧ s .
2270
Take p ∈ N, v
1
, . . . , v
p
∈ R and t
1
, . . . , t
p
∈ [0, 1]
d
. Note that for any t , s , r ∈ [0, 1]
d
we have
1{r ∈ [0, t ∧ s ]} = 1{r ∈ [0, t ] ∩ [0, s ]} = 1{r ∈ [0, t ]}1{r ∈ [0, s ]}. 36
Then
p
X
i=1 p
X
j=1
v
i
µ
n
t
i
∧ t
j
v
j
=
p
X
i=1 p
X
j=1
v
i
v
j
X
k≤k
n
1{B
n
k ∈ [0, t
i
∧ t
j
]}σ
2 n
,k
= X
k≤k
n
σ
2 n
,k
p
X
i=1
v
i
1{B
n
k ∈ [0, t
i
]}
2
≥ 0.
Since this holds for each n, taking the limit as mn → ∞ gives the positive definiteness of gt , s .
4.2 Proof of theorem 8
Consider the jump process defined as ζ
n
t =
X
1≤k≤k
n
1{Bk ∈ [0, t ]}X
n ,k
.
Now for each t
|Ξ
n
t − ζ
n
t | =
X
1≤k≤k
n
α
n ,k
X
n ,k
, where
α
n ,k
= |R
n ,k
|
−1
|R
n ,k
| − 1{Bk ∈ [0, t ]}.
Now |α
n ,k
| 1, and vanishes if R
n ,k
⊂ [0, t ], or R
n ,k
∩ [0, t ] = ∅. Actually α
n ,k
6= 0 if and only if
k ∈ I , where I is defined by 24. Thus E |Ξ
n
t − ζ
n
t |
2
= X
k∈I
α
n ,k
σ
2 n
,k
≤ X
k∈I
σ
2 n
,k
≤
d
X
l=1
∆b
l
u
l
t
l
+ 1. Using 9 we get
|Ξ
n
t − ζ
n
t |
P
− → 0, as mn → ∞.
We will concentrate now on finite-dimensional distributions of ζ
n
.
Fix t
1
, . . . , t
r
∈ [0, 1]
d
and v
1
, . . . , v
r
real, set V
n
=
r
X
p=1
v
j
ζ
n
t
p
= X
1≤k≤k
n
α
n ,k
X
n ,k
, where
α
n ,k
=
r
X
p=1
v
p
1{Bk ∈ [0, t
p
]}.
2271
Now using 36 we get b
n
:= E V
2
n
= X
k≤k
n
α
2 n
,k
σ
2 n
,k
= X
k≤k
n
X
p
X
q
v
p
v
q
1{Bk ∈ [0, t
p
]}1{Bk ∈ [0, t
q
]}σ
2 n
,k
= X
p
X
q
v
p
v
q
µ
n
t
p
∧ t
q
.
Letting mn to to infinity and using assumption 5, we obtain
b
n
−−−−−→
mn→∞
X
p
X
q
v
p
v
q
µt
p
∧ t
q
= E X
v
p
Gt
p 2
=: b. If b = 0, then V
n
converges to zero in distribution since E V
2 n
tends to zero. In this special case we also have
P
p
v
p
Gt
p
= 0 almost surely, thus the convergence of finite dimensional distributions holds.
Assume now, that b 0. For convenience put Y
n ,k
= α
n ,k
X
n ,k
and v = P
p
P
q
v
p
v
q
. Since Y
2
n ,k
≤ vX
2
n ,k
, Y
n ,k
satisfies the condition of infinitesimal negligibility. For mn large enough to have b
n
b2, we get
1
E V
2
n
X
1≤k≤k
n
E Y
2
n ,k
1{|Y
n ,k
|
2
ǫ
2
E V
2
n
} ≤
2v b
X
1≤k≤k
n
E X
2
n ,k
1
|X
n ,k
|
2
bǫ
2
2v .
Thus Lindeberg condition for V
n
is satisfied and that gives us the convergence of finite dimensional distributions.
5 Tightness results
5.1 Proof of theorem 3
We will use theorem 14. Using Doob inequality we have P sup
t ∈[ 0,1]
d
|Ξ
n
t | a = Pmax
k≤k
n
|S
n
k| a ≤ a
−2
E S
n
k
n
2
= a
−2
→ 0, as a → ∞,
thus condition i is satisfied. For proving ii note that due to the definitions of v , v
+
and v
−
we can write
Ξ
n
v − Ξ
n
v
+
=
l
X
i=1
Ξ
n
v + w
i−1
− Ξ
n
v + w
i
Ξ
n
v − Ξ
n
v
−
=
l
X
i=1
Ξ
n
v − w
i−1
− Ξ
n
v − w
i
2272
where l is the number of odd coordinates in 2
j
v , w
= 0, w
i
has 2
− j
in the first i odd coordinates of 2
j
v , and zero in other coordinates. So the condition ii holds provided one proves for every ǫ 0
lim
J →∞
lim sup
n→∞
Π
−
J , n; ǫ = 0, 37
lim
J →∞
lim sup
n→∞
Π
+
J , n; ǫ = 0,
38 where
Π
−
J , n; ǫ := P sup
j≥J
2
α j
max
r∈D
j
0≤ℓ≤2
j
|Ξ
n
r, s
ℓ
− Ξ
n
r
−
, s
ℓ
| ǫ ,
39 Π
+
J , n; ǫ := P sup
j≥J
2
α j
max
r∈D
j
0≤ℓ≤2
j
|Ξ
n
r
+
, s
ℓ
− Ξ
n
r, s
ℓ
| ǫ ,
40 with D
j
= {2l − 12
− j
; 1 ≤ l ≤ 2
j−1
}, r
−
= r − 2
− j
, ℓ = l
2
, . . . , l
d
, 2
j
= 2
j
, . . . , 2
j
vector of
dimension d − 1 and s
ℓ
= ℓ2
− j
. We prove only the 37, since the proof of 38 is the same. Denote by
v = r, s
ℓ
, and v
−
= r
−
, s
ℓ
. From 32 we have
Ξ
n
r, s
ℓ
− Ξ
n
r
−
, s
ℓ
= S
n
Uv − S
n
Uv
−
+
d
X
l=1
T
l
v − T
l
v
−
. To estimate this increment we discuss according to following configurations
Case 1. u
1
r = u
1
r
−
. Consider first the increment T
1
v − T
1
v
−
and note that by 33 with l = 1,
T
1
v =
X
1≤k≤d
v
k
− b
k
u
k
v
k
∆b
k
u
k
v
k
+ 1 ∆
k u
k
v
k
+1
S
n
Uv.
Recall the notation 5 and note that by definition v
2:d
= v
− 2:d
with U v = Uv
−
. Thus all terms indexed by k ≥ 2 disappear in difference T
1
v − T
1
v
−
. This leads to the factorisation T
1
v − T
1
v
−
= r − r
−
∆b
1
u
1
r + 1 ∆
1 u
1
r+1
S
n
Uv.
41 For l ≥ 2, T
l
v is expressed by 33 as
T
l
v =
X
1≤i
1
···i
l
≤d
v
i
1
− b
i
1
u
i
1
v
i
1
∆b
i
1
u
i
1
v
i
1
+ 1 . . .
v
i
l
− b
i
l
u
i
l
v
i
l
∆b
i
l
u
i
l
v
i
l
+ 1 ∆
i
1
u
i1
v
i1
+1
. . . ∆
i
l
u
il
v
il
+1
S
n
Uv.
In the difference T
1
v − T
1
v
−
all the terms for which i
1
≥ 2 again disappear and we obtain T
l
v − T
l
v
−
= r − r
−
∆b
1
u
1
r + 1 X
1i
2
···i
l
≤d
v
i
2
− b
i
2
u
i
2
v
i
2
∆b
i
1
u
i
1
v
i
1
+ 1 . . .
v
i
l
− b
i
l
u
i
l
v
i
l
∆b
i
l
u
i
l
v
i
l
+ 1 ∆
1 u
1
r+1
∆
i
2
u
i2
v
i2
+1
. . . ∆
i
l
u
il
v
il
+1
S
n
Uv. 42
2273
Since u
1
r = u
1
r
−
, we have b
1
u
1
r ≤ r r
−
b
1
u
1
r + 1, thus r − r