Lecture 12 of 'Scientific Computing' (wi4201)
previous lecture,
next lecture
Collegerama
,
slides
- The following subjects are discussed:
- Recap properties preconditioner
- BIM as preconditioner, be careful that the matrix M should be SPD
- Use Symmetric SOR (SSOR) instead of SOR
- The band of the Choleski factor is full, due to fill in
- Incomplete Choleski removes all or part of the fill in
- Historical remarks about ICCG (Incomplete Choleski CG)
- How to compute the Incomplete Choleski factor in an efficient way
- ICCG(0) and ICCG(3) compared
- Superlinear convergence and eigenvalues of the preconditioned
matrix
- Modified IC preconditioner
- Krylov methods for a general non-singular matrix
- No method exists with all nice properties of CG for a general non-singular matrix
- Three classes of methods: CGNR, Bi-CGSTAB, and GMRES
- Details of CGNR, CG applied to the normal equations
- Properties of CGNR
- Material is described in pages 108-118 of the lecture notes.
Recommended exercises: 7.2.1 - 7.2.4
Back to
Lectures of Scientific Computing page.