MueLu: filtered matrix lumping may be buggy

[aprokop]

@trilinos/muelu

Summary

A potential bug in MueLu's FilteredAFactory lumping procedure affecting both non-kokkos and kokkos branches.

Description

In MueLu_FilteredAFactory_def.hpp the lumping procedure seem to only use local indices for determining which element is diagonal. This is a problem for situations when row map and column map do not "match". In that situation, the dropped entries may be added to non-diagonal element thus exacerbating situation.

Who is affected

Potentially anybody using non-standard column map in conjuction with filtering. Panzer?

How to address

As a first step, we need to identify if anybody is getting hit by this. I propose adding a check isSameMap similarly to TentativePFactory. If we do get somebody affected, we need to decide what to do. Going to global indices is going to be very expensive.

[jhux2]

As a first step, we need to identify if anybody is getting hit by this. I propose adding a check isSameMap similarly to TentativePFactory.

Seems like a reasonable first step.

[aprokop]

Wait until #1695 is fixed so this does not cost more. Once that's done, this check would not cost us anything for trivial maps.

[github-actions]

This issue has had no activity for 365 days and is marked for closure. It will be closed after an additional 30 days of inactivity.
If you would like to keep this issue open please add a comment and remove the MARKED_FOR_CLOSURE label.
If this issue should be kept open even with no activity beyond the time limits you can add the label DO_NOT_AUTOCLOSE.

[github-actions]

This issue was closed due to inactivity for 395 days.

[jhux2]

Not sure that this was ever addressed.