You are on page 1of 4

58 matrices and system of linear equations

3.3 LU factorisation

We can not in general expect a simple structured matrix


discussed in the Section 3.2. A better idea to write A in the
following form A = LU where L is lower triangular and U
is upper triangular. The advantage of this factorisation is
the following:

- Write Ax = b as LUx = b.

- Assume Ux = z and solve for z from Lz = b using


the forward substitution.

- Finally, we can solve for x from Ux = z using back


substitution, giving us the solution to the original
system.

Definition 3.3.1. We say A has LU factorization (or decom-


position) if A can be expressed as A = LU, where L is a
lower triangular matrix and U is an upper triangular matrix.

Note that for any invertible diagonal matrix D,


1
A = LU = LDD U = L̃Ũ.

Thus, LU factorisation is not unique.

Definition 3.3.2. A lower (upper) triangular matrix is said


to be unit lower (upper) triangular if the diagonal elements
lii = 1 (uii = 1).

3.3.1 Doolittle’s Factorisation

Definition 3.3.3. A matrix A is said to have a Doolittle


factorization if there exists a unit lower triangular matrix L
and an upper triangular matrix U such that A = LU.
3.3 LU factorisation 59

Theorem 3.3.1. Let n 2, and suppose that A 2 R n⇥n is such


that every leading principal submatrix A(k) 2 R k⇥k of A of order
k, with 1  k < n, is nonsingular. (Note that A itself is not
required to be nonsingular.) Then, A can be factorized in the
form A = LU, where L 2 R n⇥n is unit lower triangular and
U 2 R n⇥n is upper triangular.

The elements of L and U are given by:

lk,k = 1
k 1
uk,j = ak,j  lk,p u p,j for j = k, k + 1, . . . , n
p =1
!
k 1
1
li,k =
uk,k
ai,k  li,p u p,k for i = k + 1, k + 2, . . . , n
p =1

Pseudocode is given in Algorithm 9.

Theorem 3.3.2. The Doolittle’s factorisation is unique.

Proof. Let us assume A = L1 U1 = L2 U2 . After left multiply-


ing L2 1 and right multiplying U1 1 we get

L2 1 L1 = U2 U1 1

We recall the following results:

- inverse of a unit lower triangular matrix is unit lower


triangular.

- multiplication of two unit lower triangular matrices is


unit lower triangular.

- multiplication of two upper triangular matrices is up-


per triangular.
60 matrices and system of linear equations

Algorithm 9 Doolittle’s Algorithm


1: procedure Doolittle(A)
2: n size of A
3: initialize L and U to be n ⇥ n matrices with all ele-
ments equal to zero
4: for k 1 to n do
5: Lk,k 1
6: for j k to n do
7: Uk,j Ak,j Âkp=11 Lk,p U p,j
8: end for
9: for i k + 1 to⇣ n do ⌘
1 k 1
10: Li,k Uk,k A i,k  p =1 L i,p U p,k
11: end for
12: end for
13: return L, U
14: end procedure
3.3 LU factorisation 61

Thus L2 1 L1 is unit lower triangular and U2 U1 1 is upper


triangular. An upper triangular matrix that is also lower
triangular must be diagonal, that is,

L2 1 L1 = U2 U1 1 = D

where D is diagonal. Also since L2 1 L1 is unit lower trian-


gular we should have L2 1 L1 = I = U2 U1 1 . Hence, L1 = L2
and U1 = U2 .

Example 3.4. Find the Doolittle’s factorisation of the matrix


2 3
1 1 1
44 3 15 .
3 5 3

Example 3.5. Check whether the following matrix have the Doolit-
tle’s factorisation

0 1
.
2 1

Example 3.6. Find the Doolittle’s factorisation of the matrix


2 3
1 1 1
41 1 25 .
1 1 3

Example 3.7. Find the Doolittle’s factorisation (if exist) of the


matrix
2 3
1 2 3
42 4 55 .
1 3 4

You might also like