Closed rveltz closed 1 year ago
I don't know of a Krylov scheme to handle the generalized problem directly. So I'd advise a user-implemented x -> Bₚ \ x
at least, to alternate with the Aₚ
operators. (It may be more accurate to separate the A
and B
operations.) Can you afford a sparse direct solver? Otherwise it may be tricky to pick an adequate (but not too expensive) tolerance for an iterative method.
I thought you could do a gmres given that you have the Krylov infrastructure but you are right, better leave this on the user side
Hi,
Thanks a lot for your work!
Would it make sense to have a Matrix-free generalized periodic Schur decomposition? How would one provide Bₚ⁻¹? Just Bₚ?
Or maybe it is better to use
partial_pschur
where the user implementx -> B \ A*x
?Thanks for your input