Lecture 11 of 'Scientific Computing' (wi4201)
previous lecture,
next lecture
Collegerama
,
slides
,
Preconditioning 2019
Conference,
and
Preconditioning in the New Decade
- The following subjects are discussed:
- CG as a polynomial method
- CG started as a direction solution method in 1950
- due to rounding errors CG was not used until 1970, then it is used
as an iterative method
- use eigenvalues and eigenvectors, together with the polynomial
approach to analyse the convergence in more detail
- CG converges in m iterations if there are m different eigenvalues
- effective condition number
- Ritz matrix, Ritz values and Ritz vectors
- properties of Ritz values
- superlinear convergence
- if smallest Ritz value is close to the smallest eigenvalue this
eigenvalue is no longer determining the convergence rate
- sketch of the proof for this property
- preconditioner to decrease the condition number
- required properties of the preconditioner M
- choices of a preconditioner
- derivation of the Preconditioned CG (PCG) method
- work per iteration and memory
- properties of PCG
- take M as a diagonal matrix
- Material is described in pages 101-108 of the lecture notes.
Recommended exercises: 7.1.1 - 7.1.6
Back to
Lectures of Scientific Computing page.