GitLab

Created by: jmgate

@trilinos/epetra, @trilinos/thyra

Scenario

I have a matrix, A, and vector, x. A is a 2x2 Thyra::BlockedLinearOpBase, where A00 is a Thyra::LinearOpBase and the other blocks are Thyra::MultiVectorBases. Likewise, x is a Thyra::ProductVectorBase with two sub-vectors. I'm in a situation where I need to make A an Epetra_CrsMatrix and x an Epetra_Vector, but I'm at a bit of a loss to do just that. I see Thyra has a get_Epetra_Vector() routine, but that requires passing in an Epetra_Map. I see there's also a get_Epetra_Map() routine, which takes a Thyra::VectorSpaceBase, but if you give it a DefaultProductVectorSpace instead, it doesn't seem to work (I'm seeing MPI_Abort being called).

Questions

• Is there a way to convert the objects that I have to the ones that I need directly?
• Is there a way to cram a BlockedLinearOpBase into a single MultiVectorBase and a ProductVectorBase into a single VectorBase, such that these Epetra/Thyra wrappers work?
• If I'm able to convert the various pieces of my blocked objects into their Epetra counterparts, is there some way to cram four Epetra_CrsMatrixes into one, and two Epetra_Vectors into one?

Motivation and Context

In Charon I've got a ModelEvaluator decorator that takes the original physics ModelEvaluator and augments it with blocks corresponding to constraints/parameters. Solving the blocked system via a Teko blocked preconditioner is proving to be more finicky than we'd like in certain situations, so we're investigating collapsing the blocked system and throwing a standard solver at it to see what happens.

Additional Information

@mhoemmen, this seems like the kind of thing you'd chime in on.