Hi all,
These days I am dealing with a very bad conditioned problem. ksp solver can not 
convergence.
Although direct linear solver such as mumps works, memory is a limitation to 
me. 
So I am still seeking a method to make iterative method work.

I know the bad condition number comes from a lumped boundary.
For this boundary, the sum of flux comes from all the boundary elements shoule 
equals to a fixed given flux. 
If I set the value of each boundary elements equals to a const value (first 
type boundary condition) 
the condition number of final matrix is in moderate range.

I image if I can do Gaussian elimination first to the boundary elements,
(create schur complement), is it possible to reduce condition number of origin 
matrix?
Then I can solve the problem by ksp method.

  

Reply via email to