You are on page 1of 12

APPLIED NUMERICAL METHODS

Dr. Khaled Ahmida Ashouri


Numerical Analysis Course/Academia
Dr. Khaled M. Ahmida

GAUSS ELIMINATION: LDU DECOMPOSITION


If matrix A is a non-singular, then A can be decomposed into LDU
decomposition (factorization):
A = LDU
Once we have LDU, Ax=b becomes LDUx=b, which must be easier to solve.
Ax  LDUx  LD(Ux)
 LDz  L(Dz)
 Ly=b
Ax=b  Ly=b
𝑙 ⋯ 0 𝑦 𝑏
⋮ ⋱ ⋮ ⋮ ⋮
𝑙 ⋯ 𝑙 𝑦 𝑏
Numerical Analysis Course/Academia
Dr. Khaled M. Ahmida

LDU: HOW TO!


Consider the matrix:

We want to make this = zero

To introduce zero in (3,1), we subtract 3*(line1) from (line3). This is equivalent


to multiplying by the following lower triangular matrix
Numerical Analysis Course/Academia
Dr. Khaled M. Ahmida

LDU: HOW TO!


Thus:

Since

Thus 
Numerical Analysis Course/Academia
Dr. Khaled M. Ahmida

LDU: HOW TO!


 (1)

Now, L2-L1 in last matrix results in, analogously: to make equal to zero

 1 0 0  1 1 1  1 1 1 
 1 1 0   1 2 4    0 1 3 

 0 0 1   0 6 24   0 6 24 

Again multiply by inverse 


Numerical Analysis Course/Academia
Dr. Khaled M. Ahmida

LDU: HOW TO!


Substituting in (1):

Simplifying:
Numerical Analysis Course/Academia
Dr. Khaled M. Ahmida

LDU: HOW TO!


 (2)

Now, L3-6L2, which is equivalent to:


Make = 0
1 0 0  1 1 1  1 1 1
0 1 0   0 1 3    0 1 3 

 0 6 1   0 6 24   0 0 6 

Again multiply by inverse 


 Substitute in (2)
Numerical Analysis Course/Academia
Dr. Khaled M. Ahmida

LDU: HOW TO!


Thus, finally,
Numerical Analysis Course/Academia
Dr. Khaled M. Ahmida

LDU: HOW TO!


Why we need LDU and not only LU?
Because LDU usually is guaranteed to be well-conditioned
Numerical Analysis Course/Academia

Pseudo algorithm Dr. Khaled M. Ahmida

LDU: ALGORITHM
 This algorithm creates L, D, U
Matrix D can be modified to look
like that of last slide
Numerical Analysis Course/Academia
Dr. Khaled M. Ahmida

LU: ALGORITHM
If [D] is the Identity matrix, then we can consider the system as LU
decomposition.

It also facilitates its computation in special cases.

For Ax=b  LUx=b  Ly=b

Then 1- solve the system Ly=b


2- solve the system Ux=y
It facilitates dense matrix computations.
Numerical Analysis Course/Academia
Dr. Khaled M. Ahmida

NUMERICAL STABILITY
How to Ensure Numerical Stability
 System of equations must be well conditioned
 Investigate condition number – if matrix is diagonally dominant or not
 Consistent with physics
 e.g. do not couple domains that are physically uncoupled or unrelated
 Consistent units
 e.g. do not mix known units with unknown units
 If Dimensions are unknown
 Normalize all unknowns consistently

You might also like