Abstract:
The convergence rate as well as the robustness of Krylov
sub-space methods are improved by multiplying the coefficient
matrix of the linear system with a preconditioner matrix M. One
common approach for the construction of M bases on incomplete
factorisation of the coefficient matrix.
The presents presents the idea of recursive incomplete block
factorisation which is more suitable for parallel computer
architectures. Especially the problem of identifying sub-matrices,
which are suitable for elimination, is addressed. Such a
sub-matrix has to be easily invertible and has to ensure a stable
factorisation. Some examples illustrate the performance of the
method in practice.
![[Back]](/images/prevpage.gif)
![[Index]](/images/index.gif)
![[Help]](/images/help.gif)
![[MSI]](/images/msi.gif)
![[ANU Online]](/images/online.gif)
