31

1.6 DOOLITTLES LU DECOMPOSITION

We shall now consider the LU decomposition of a general matrix. The
method we describe is due to Doolittle.
Let A = (aB
ij
B). We seek as in the case of a tridiagonal matrix, an LU decomposition
in which the diagonal entries lB
ii
B of L are all 1. Let L = (lB
ii
B) ; U = (uB
ij
B). Since L is a
lower triangular matrix, we have
lB
ij
B = 0 if j > i ; and by our choice, 1
ii
l = .
Similarly, since U is an upper triangular matrix, we have
uB
ij
B = 0 if i > j.
We determine L and U as follows : The 1P
st
P row of U and 1P
st
P column of L are
determined as follows :

=
=
n
k
k k
u l a
1
1 1 11

= lB
11
BuB
11
B since lB
1k
B = 0 for k>1
= uB
11
B since lB
11
B = 1.

Therefore
11 11
a u =
In general,

=
=
n
k
kj k j
u l a
1
1 1

=
11 1
.
j
l u since lB
1k
B = 0 for k>1
= uB
1j
B since lB
11
B = 1.
uB
1j
B = aB
1j
B . . . . . . . . . (I)

Thus the first row of U is the same as the first row of A. The first column of L is
determined as follows:

=
=
n
k
k jk j
u l a
1
1 1
= lB
j1
BuB
11
B since uB
k1
B = 0 if k>1
lB
j1
B = aB
j1
B/uB
11
B . . . . . . . . . (II)
Note : uB
11
B is already obtained from (I).

32
Thus (I) and (II) determine respectively the first row of U and first column of L.
The other rows of U and columns of L are determined recursively as given below:
Suppose we have determined the first i-1 rows of U and the first i-1 columns of L.
Now we proceed to describe how one then determines the iP
th
P row of U and iP
th
P
column of L. Since first i-1 rows of U have been determined, this means, uB
kj
B ; are
all known for 1 k i-1 ; 1 j n. Similarly, since first i-1 columns are known
for L, this means, lB
ik
B are all known for 1 i n ; 1 k i-1.
Now

=
=
n
k
kj ik ij
u l a
1

=
=
i
k
kj ik
u l
1
since lB
ik
B = 0 for k>i

ij ii
i
k
kj ik
u l u l + =

=
1
1


ij
i
k
kj ik
u u l + =

=
1
1
since lB
ii
B = 1.

=
=
1
1
i
k
kj ik ij ij
u l a u
. . . . . ... . . . .(III)
Note that on the RHS we have aB
ij
B which is known from the given matrix. Also the
sum on the RHS involves lB
ik
B for 1 k i-1 which are all known because they
involve entries in the first i-1 columns of L ; and they also involve uB
kj
B ; 1 k i-1
which are also known since they involve only the entries in the first i-1 rows of U.
Thus (III) determines the iP
th
P row of U in terms of the known given matrix and
quantities determined upto the previous stage. Now we describe how to get the
iP
th
P column of L :

=
=
n
k
ki jk ji
u l a
1

=
=
i
k
ki jk
u l
1
Since uB
ki
B = 0 if k>i
ii ji
i
k
ki jk
u l u l + =

=
1
1

(

=
1
1
1
i
k
ki jk ji
ii
ji
u l a
u
l ..(IV)

33
Once again we note the RHS involves uB
ii
B, which has been determined using (III);
aB
ji
B which is from the given matrix; lB
jk
B; 1 k i-1 and hence only entries in the first
i-1 columns of L; and uB
ki
B, 1 k i-1 and hence only entries in the first i-1 rows of
U. Thus RHS in (IV) is completely known and hence lB
ji
B, the entries in the iP
th
P
column of L are completely determined by (IV).
Summarizing, Doolittles procedure is as follows:
lB
ii
B = 1; 1P
st
P row U = 1P
st
P row of A ; Step 1 determining 1P
st
P row of U and
lB
j1
B= aB
j1
B/uB
11
B1P
st
P column of L.

For i 2; we determine, (by (III) and (IV)),

=
=
1
1
i
k
kj ik ij ij
u l a u
; j = i, i+1, i+2, .,n
(Note for j<i we have uB
ij
B = 0)
(

=
1
1
1
i
k
ki jk ji
ii
ji
u l a
u
l
; j = i, i+1, i+2,..,n
(Note for j<i we have lB
ji
B = 0)
We observe that the method fails if uB
ii
B = 0 for some i.
UExample:
Let
|
|
|
|
|
.
|

\
|

=
12 6 0 6
4 19 14 4
4 6 2 2
3 1 1 2
A
Let us determine the Doolittle decomposition for this matrix.

UFirst step:
1P
st
P row of U : same as 1P
st
P row of A.
Therefore uB
11
B = 2 ; uB
12
B = 1 ; uB
13
B = -1 ; uB
14
B = 3

U1UPU
st
UPU column of L:U
lB
11
B = 1;
lB
21
B = aB
21
B/uB
11
B = -2/2 = -1.
lB
31
B = aB
31
B/uB
11
B = 4/2 = 2.

34
lB
41
B = aB
41
B/uB
11
B = 6/2 = 3.

USecond step:U
2P
nd
P row of U : uB
12
B = 0 (Because upper triangular)
uB
22
B = aB
22
B lB
21
B uB
12
B = 2 (-1) (1) = 3.
uB
23
B = aB
23
B lB
21
B uB
13
B = 6 (-1) (-1) = 5.
uB
24
B = aB
24
B lB
21
B uB
14
B = - 4 (-1) (3) = -1.

2P
nd
P column of L : lB
12
B = 0 (Because lower triangular)
lB
22
B = 1.
lB
32
B = (aB
32
B lB
31
B uB
12
B) /uB
22
B

B = [14 (2)(1)]/3 = 4.
lB
42
B = (aB
42
B lB
41
B uB
12
B) /uB
22
B

B = [0 (3)(1)]/3 = -1.

UThird Step:U
3P
rd
P row of U: uB
31
B = 0 (because U is upper triangular )
uB
32
B = 0
uB
33
B = aB
33
B lB
31
BuB
13
B lB
32
B uB
23
B
= 19 (2) (-1) (4)(5) = 1.
uB
34
B = aB
34
B lB
31
BuB
14
B lB
32
B uB
24
B
= 4 (2) (3) (4)(-1) = 2.



3P
rd
P column of L : lB
13
B = 0 (because L is lower triangular)
lB
23
B B

B= 0
lB
33
B =1
lB
43
B = (aB
43
B lB
41
BuB
13
B lB
42
BuB
23
B)/ uB
33
B
= [-6 (3) (-1) (-1) (5)]/1
= 2.



35

UFourth Step: U
4P
th
P row of U: uB
41
B = 0
uB
42
B = 0 (because upper triangular)
uB
43
B = 0
uB
44
B = aB
44
B lB
41
B uB
14
B lB
42
B uB
24
B lB
43
B uB
34
B
= 12 (3) (3) (-1) (-1) (2) (2)
= -2.

4P
th
P column of L : lB
14
B = 0 = lB
24
B = lB
34
B Because lower triangular
lB
44
B = 1.
Thus
|
|
|
|
|
.
|

\
|

=
1 2 1 3
0 1 4 2
0 0 1 1
0 0 0 1
L ;
|
|
|
|
|
.
|

\
|

=
2 0 0 0
2 1 0 0
1 5 3 0
3 1 1 2
U . . . . . . . . . . .(V)
and
A = LU.
This gives us the LU decomposition by Doolittles method for the given A.
As we observed in the case of the LU decomposition of a tridiagonal matrix;
it is not advisable to choose the lB
ii
B as 1; but to choose in such a way that the
diagonal entries of L and the corresponding diagonal entries of U are of the same
magnitude. We describe this procedure as follows:

Once again 1P
st
P row and 1P
st
P column of U & L respectively is our first concern:
UStep 1:U aB
11
B = lB
11
B uB
11
Choose ( )
11 11 11 11 11
. sgn ; a a u a l = =
Next 1 11 1
1
n
ij k kj j
k
a l u l u
=
= =
as
1
0
k
l = for 1 k >
1
1 1
j
i j
a
u
l
=
Thus note that uB
1j
B have been scaled now as compared to what we did earlier.
Similarly,

36
1
1
1 1
j
j
a
l
u
=

These determine the first row of U and first column of L. Suppose we have
determined the first i-1 rows of U and first i-1 columns of L. We determine now
the iP
th
P row of U and iP
th
P column of L as follows:

=
=
n
k
ki ik ii
u l a
1


1
i
i k k i
k
l u
=
=
for lB
ik
B = 0 if k>i
ii
i
k
ii ki ik
u l u l

=
+ =
1
1

Therefore
1
1
,
i
ii ii ii ik ki i
k
l u a l u p

=
= =

say
Choose
ki
i
k
ik ii i ii
u l a p l

=
= =
1
1


i i ii
p p u sgn =

1 1
0
n i
ij ik kj ik kj ik
k k
a l u l u l
=
= = =

Q
for k i >
ij ii kj
i
k
ik
u l u l + =

=
1
1

|
.
|

\
|
=

=
kj
i
k
ik ij ij
u l a u
1
1
lB
ii
B

determining the iP
th
P row of U.

ki
n
k
jk ji
u l a

=
=
1



37

1
0
i
jk ki ki
k
l u u
=
= =

Q
if k i >

ii ji ki
i
k
jk
u l u l + =

=
1
1


|
.
|

\
|
=

=
1
1
i
k
ki jk ji ji
u l a l uB
ii
B ,

thus determining the iP
th
P column of L.
Let us now apply this to matrix A in the example in page 32.
UFirst Step:U
2 ; 2 2
11 11 11 11 11
= = = = u l a u l
2
3
;
2
1
2
1
11
14
14
11
13
13
11
12
12
= = = = = =
l
a
u
l
a
u
l
a
u
2
3
;
2
1
;
2
1
; 2
14 13 12 11
= = = = u u u u
21
21
11
2
2
2
a
l
u
= = =
2 2
2
4
11
31
31
= = =
u
a
l
2 3
2
6
11
41
41
= = =
u
a
l
Therefore
2 3
2 2
2
2
41
31
21
11
=
=
=
=
l
l
l
l


USecond Step:U
12 21 22 22 22
u l a u l =

38
( ) 3
2
1
2 2 = |
.
|

\
|
=
3 ; 3
22 22
= = u l
( )
23 21 13
23
22
a l u
u
l

=


( )
1
6 2
2
5
3 3
( | |

| (
\ .
= =

( )
24 21 14
24
22
a l u
u
l

=


( )
( )
3
4 2
2
1
3 3
( | |

| (
\ .
= =
Therefore
21 22 23 24
5 1
0; 3; ;
3 3
u u u u = = = =
( )
32 31 12
32
22
a l u
l
u

=


( )
1
14 2 2
2
3
( | |

| (
\ .
=

3 4 =

( )
42 41 12
42
22
a l u
l
u

=

( )
1
0 3 2
2
3
| | | |

| |
\ . \ .
=

3 =



39
Therefore
3
3 4
3
0
42
32
22
12
=
=
=
=
l
l
l
l

UThird Step:U
23 32 13 31 33 33 33
u l u l a u l =
( ) ( )
|
|
.
|

\
|
|
.
|

\
|
=
3
5
3 4
2
1
2 2 19
= 1
1 ; 1
33 33
= = u l
( )
34 31 14 32 24
34
33
a l u l u
u
l

=


( ) ( )
3 1
4 2 2 4 3
2 3
1
( | | | | | |

( | | |
\ . \ . \ .
=
= 2

2 , 1 ; 0 ; 0
34 33 32 31
= = = = u u u u

[ ]
43 41 13 42 23
43
33
a l u l u
l
u

=


( ) ( )
1 5
6 3 2 3
2 3
1
| |
| | | |

| | |
\ . \ .
\ .
=
= 2
Therefore
(
(
(
(

=
=
=
=
2
1
0
0
43
33
23
13
l
l
l
l


40

UFourth Step:U
34 43 24 42 14 41 44 44 44
u l u l u l a u l =
( ) ( ) ( )( ) 2 2
3
1
3
2
3
2 3 12
|
|
.
|

\
|
|
.
|

\
|
=
= -2

2 ; 2
44 44
= = u l
2 ; 0 ; 0 ; 0
44 43 42 41
= = = = u u u u

(
(
(
(
(

=
=
=
=
2
0
0
0
44
34
24
14
l
l
l
l


Thus we get the LU decompositions,
|
|
|
|
|
.
|

\
|

=
2 2 3 2 3
0 1 3 4 2 2
0 0 3 2
0 0 0 2
L ;
|
|
|
|
|
|
|
.
|

\
|

=
2 0 0 0
2 1 0 0
3
1
3
5
3 0
2
3
2
1
2
1
2
U

in which
ii ii
u l = , i.e. the corresponding diagonal entries of L and U have the
same magnitude.

UNote:U Compare this with the L and U of page 35. What is the difference.

The U we have obtained above can be obtained from the U of page 34 by
(1) replacing the numbers in the diagonal of the U of Page 34 by their square
roots of their magnitude and keeping the same sign. Thus the first
diagonal 2 is replaced by 2 ; 2P
nd
P diagonal 3 is replaced by 3 , third
diagonal 1 by 1 and 4P
th
P diagonal 2 by - 2 . These then give the
diagonals of the U we have obtained above.

41
(2) Divide each entry to the right of a diagonal in the U of page 35 by these
replaced diagonals.

Thus 1P
st
P row of the U of Page 34 changes to 1P
st
P row of U in page 40
2P
nd
P row of the U of Page 34 changes to 2P
nd
P row of U in page 40
3P
rd
P row of the U of Page 34 changes to 3P
rd
P row of U in page 40
4P
th
P row of the U of Page 34 changes to 4P
th
P row of U in page 40

This gives the U of page 40 from that of page 35.
The L in page 40 can be obtained from the L of page 35 as follows:
(1) Replace the diagonals in L by magnitude of the diagonals in U of page 40.
(2) Multiply each entry below the diagonal of L by this new diagonal entry.
We get the L of page 35 changing to the L of page 40.