Information Technology Reference
In-Depth Information
{
ψ
,k
:
≤
≤
∈∇
}
of
V
L
, we need to compute the stiff-
ness matrix entries
A
(
,k
),(,k)
=
a(ψ
,k
,ψ
,k
)
. Since
A
is, in general, densely
populated, we use wavelet compression to reduce the number of non-zero entries to
O
Using the basis
0
L, k
(N
L
)
. We need the following smoothness assumption on the Lévy density
k
:
∂
n
k(z)
C
0
C
n
n
|
−
α
−
1
−
n
,z
|
|≤
!|
z
=
0
,
∀
n
∈ N
0
,
(12.7)
for
C
0
,C >
0. These kind of estimates are called Caldéron-Zygmund estimates.
12.2.2 Matrix Compression
The compression scheme is based on the fact that the matrix entries
A
(
,k
),(,k)
can
be estimated a priori and therefore neglected if these are smaller than some cut-off
parameter. There are two reasons for an entry to be omitted. Either the distance of
the supports supp
ψ
,k
and supp
ψ
,k
or the distance of the singular supports (the
singular support of a wavelet is that subset of
G
where the wavelet is not smooth) is
large enough. The distance of support is denoted by
δ
:=
dist
{
supp
ψ
,k
,
supp
ψ
,k
}
,
and the distance of singular support
dist
if
≤
,
{
singsupp
ψ
,k
,
supp
ψ
,k
}
δ
sing
:=
dist
{
supp
ψ
,k
,
singsupp
ψ
,k
}
else.
Theorem 12.2.2
Let X be a Lévy process with Lévy density k satisfying
(
12.7
).
Define the compression scheme by
⎧
⎨
⎩
0
if δ>
B
,
,
A
(
,k
),(,k)
=
if δ<c
2
−
min
{
,
}
and δ
sing
>
B
,
,
0
A
(
,k
),(,k)
,
else,
with cut-of parameter
a
max
2
−
min
{
,
}
,
2
,a>
1
,
2
L(p
−
α/
2
)
−
(
+
)(p
+
p)
2
B
,
=
p
+
α
a
max
2
−
max
{
,
}
,
2
,
2
L(p
−
α/
2
)
−
(
+
)p
−
max
{
,
}
p
B
,
=
a>
1
.
p
+
α
The number of non-zero entries for the compressed matrix
A
is
O
(N
L
)
.
A proof can be found in [51, Theorem 11.1]. We call a compression scheme
first
compression
if
δ>
B
,
, i.e. if the distance of the wavelets is large enough, and
second compression
if
δ
sing
>
B
,
(and
δ<c
2
−
min
{
,
}
), i.e. if the distance of the
singular support is large enough (compare with Fig.
12.2
).
We give an example for the matrix compression.
Search WWH ::
Custom Search