After looking at the original paper on which the chow_patel_icc
preconditioner is based, it seems that convergence for the preconditioner
is guaranteed theoretically if a fixed point exists for the matrix.
However, overflow can occur in practice which will lead to lack of
convergence. In lieu of
Thank you for the quick reply Karl,
The symmetric matrices are almost certainly positive definite because the
system can be solved using the CG solver without preconditioning as well as
using the CG solver with the row_scaling preconditioner. It is only the
jacobi and chow_patel_icc
Hi Rick,
have you verified that your matrices are positive definite? The problems
with 'nan' usually stem from the lack of positive definiteness or zeros
on the diagonal (as they often show up in saddle point problems).
The code snippet you provide looks fine. 200 are a lot of
Hello all,
I am solving large sparse symmetric systems. As a result I have been using
the CG solver. However the only preconditioner that seems to work robustly
across different matrices is the row_scaling preconditioner. I have tried
jacobi_precond and chow_patel_icc_precond with the CG solver