A
=
LDM
t
If
A
is nonsingular and
A
=
LU
, then we can set
D
= diag(
U
) and (since
D
is
nonsinglar)
M
t
≡
D

1
U
is a unit upper triangular matrix and
A
=
LDM
t
.
There is no inherent beneﬁt to this factorization over
LU
, but it can give us a
perspective from which to develop other algorithms. The idea is not to compute
LDM
t
from
LU
, but to derive a method to compute
L
,
D
and
M
directly. To that
end, consider the
k
th
column of
A
=
LDM
t
:
a
≡
Ae
k
=
LDM
t
e
k
≡
Ly.
(1)
Suppose we have already know the ﬁrst
k

1 columns of
L
and consider the blocked
treatment of the unit lower triangular system
a
=
Ly
:
±
L
11
0
L
21
L
22
² ±
y
1
y
2
²
=
±
a
1
a
2
²
,
(2)
where
L
11
is
k
×
k
and
known
, but the last column of
L
21
is something we would like
to compute. Forward substitution gives
y
1
as the solution to
L
11
y
1
=
a
1
. Now we
know the ﬁrst
k
elements of
y
, and (from (1))
y
=
DM
t
e
k
,
(3)
giving
e
t
k
y
=
e
t
k
DM
t
e
k
=
d
kk
e
t
k
M
t
e
k
=
d
kk
(right?).
D

1
y
=
M
t
e
k
with
M
t
unit
This is the end of the preview.
Sign up
to
access the rest of the document.
 Fall '10
 MARK
 Determinant, Triangular matrix, th column, L21 y1

Click to edit the document details