# 38 imagined structure of a top left and structure of

This preview shows page 1. Sign up to view the full content.

This is the end of the preview. Sign up to access the rest of the document.

Unformatted text preview: oting is performed and sparsity is not included. 52 Solution Techniques for Elliptic Problems for i = 2 to N do for k = 1 to i ; 1 do aik = aik =akk for j = k + 1 to N do aij = aij ; aik akj end for end for end for Figure 9.3.9: Gaussian elimination loop. The Gaussian elimination update in Figure 9.3.9 is aij := aij ; aik akj : Let the current level of ll of aij be denoted as levij . Then the size of updated element aij should be levij ; levik levkj = levij ; (levik +levkj ) : Roughly speaking, the size of the updated element aij will be the maximum of the two sizes levij and (levik +levkj ) . Thus, it makes sense to de ne the new level of ll as levij := min(levij levik + levkj ): It is common (cf. 4], Section 10.3) to shift all levels by -1 from this de nition. Thus, the initial level of ll of aij is set as aij 6= 0 levij := 0 ioftherwise or i = j : (9.3.33a) 1 The updated level of ll is levij := min(levij levik + levkj + 1): (9.3.33b) We see that levij cannot increase during the factorization. Thus, if aij 6= 0 in the original matrix A, levij never becomes nonzero. Equations (9.3.33) suggest a natural way of discarding elements during the incomplete factorization. With an ILU(p) algorithm, we'll keep all elements whose level of ll does not exceed p. The zero pattern of A for the ILU(p) procedure may be de ned as the set Zp(A) := f(i j )jlevij > pg: (9.3.33c) 9.3. Conjugate Gradient Methods 53 This de nition is consistent with the ILU(0) preconditioning described in Example 9.3.5. De ne levij according to (9.3.33a) for i = 2 to N do for k = 1 to i ; 1 do if levik p then aik = aik =akk for j = k + 1 to N do if levij p then aij = aij ; aik akj Update levij according to (9.3.33b) end if end for for j = k + 1 to N do if levij > p then aij = 0 levij = 1 end if end for end if end for end for Figure 9.3.10: Incomplete ILU(p) factorization. A pseudocode ILU(p) algorithm appears in Figure 9.3.10. It is intended to give the general idea of the procedure and, as stated, is not ve...
View Full Document

## This document was uploaded on 03/16/2014 for the course CSCI 6840 at Rensselaer Polytechnic Institute.

Ask a homework question - tutors are online