Module 4 : Solving Linear Algebraic Equations
Section 7 : Matrix Conditioning and Behavior of Solutions
 

Example 10 [8]

TaylorPhilllips
Consider the Hilbert matrix discussed in the module Problem Discretization using Approximation Theory. These matrices, which arise in simple polynomial approximation are notoriously ill conditioned and MATH as MATH For example, consider MATHThus, condition number can be computed as MATH MATH For n = 6, MATH MATH which is extremely bad.

Even for n = 3, the effects of rounding off can be quite serious. For, example, the solution of MATHis MATH If we round off the elements of $\QTR{bf}{H}_{3}$ to three significant decimal digits, we obtain MATHthen the solution changes to MATH The relative perturbation in elements of matrix $\QTR{bf}{H}_{3}$ does not exceed 0.3%. However, the solution changes by 50%! The main indicator of ill-conditioning is that the magnitudes of the pivots become very small when Gaussian elimination is used to solve the problem.