Block preconditioners for multiphysics (Teko?)

[jtostie]

Specifically, for coupled problems with potentially disparate units, figure out how to scale appropriately.

[bartgol]

I was thinking about this for a coupled 2D-3D problem in FELIX. I have never used teko (should look into that), so my tentative solution would have been to build a block triangular preconditioner. I would have done that simply dropping the undesided entries from the jacobian, quering the problem for the block columns (equations) to keep for each equation. I was also thinking of pre-building the prec graph inside the discretization, to avoid wasting space.

I am not yet solving the coupled problem, so that's why I haven't done this yet. But I'd be interested in developments in this direction.

[lxmota]

Before our LCM budget was reduced, we were thinking about doing some work along this lines this year. Now we will have to wait. There is significant interest in having this capability.