Thanks for the demo Lawrence, very useful! -- Marie On Tuesday, 8 December 2015, Lawrence Mitchell < lawrence.mitchell@imperial.ac.uk> wrote:
On 08/12/15 18:28, Justin Chang wrote:
Thanks Lawrence,
Just to clarify, does aP has to be provide on top of specifying these options?
Yes, that's right.
This demo, which I should finish, describes some of this stuff in the context of an H(div)-L2 poisson problem.
https://github.com/firedrakeproject/firedrake/blob/3a445bb321c27feb4999a14ba...
-pc_type fieldsplit -pc_fieldsplit_type additive
-fieldsplit_0_pc_type something_that_works_for_grad_div_helmholtz -fieldsplit_1_pc_type something_that_works_for_mass
What about the ksp's? Just use CG for outer loop and preonly for the two fields?
Hmph, your full system is not symmetric, right? As to the inner bits, solve them exactly to start, and then back things off (as is normal with block preconditioning).
Lawrence
-- -- Marie