Fast solver for multigrid coarse level

[yy.wayne]

Hello everyone!

In tutorials the coarest level multigrid is solved either by householder direct sovler for FullMatrix or by a iterative solver based on MGCoarseGridHouseholder and MGCoarseGridIterativeSolver, respectively. However I want to solve the coarest level with a sparse direct solver. Both 2 MGCoarseGrid class access MGCoarseGridBase, so SparseDirectUMFPACK cannot be used.

Take step-16 for example. The only way possible I my mind is using PETScMUMPS and put it in MGCoarseGridIterativeSolver, and matrix & vector type need to be changed to PETSc accordingly. Is this gonna work?

[Timo Heister]

We have a number of tests that do something more sophisticated like an
AMG (which is probably what you want to do if you are running in
parallel with many coarse cells). See for example
https://github.com/dealii/dealii/blob/915ca70fa7c3c2cfa68b632aa162226145b1907f/tests/multigrid/mg_coarse_01.cc#L434

You should be able to get SparseDirectUMFPACK to work in the same way.

> --
> The deal.II project is located at http://www.dealii.org/
> For mailing list/forum options, see https://groups.google.com/d/forum/dealii?hl=en
> ---
> You received this message because you are subscribed to the Google Groups "deal.II User Group" group.
> To unsubscribe from this group and stop receiving emails from it, send an email to dealii+un...@googlegroups.com.
> To view this discussion on the web visit https://groups.google.com/d/msgid/dealii/103ebcb1-ecfc-4bc8-bc06-a0ed07968834n%40googlegroups.com.



--
Timo Heister
http://www.math.clemson.edu/~heister/

[yy.wayne]

Thank you Heister,

AMG will not fit my problem (not converge), but I think SparseDirectUMFPACK can be applied this way.