[petsc-users] GAMG and linear elasticity

[Mark F. Adams]

> 
> You probably forgot to set the near null space.  You can use
> MatSetNearNullSpace (and maybe MatNullSpaceCreateRigidBody) or the more
> hacky (IMO) PCSetCoordinates.  It's important to have translational

No debate this is a hack and but people use it and I don't want to force them to commute the RBM (error prone and tedious).  I've never heard of MatNullSpaceCreateRigidBody, and its easy so I'd recommend using it.  I'll fix ex55 & ex56 to use this one day.

> *and* rotational modes in the near null space that GAMG uses to build a
> coarse space.

This can be tricky in testing because if you just crush a cube (not a bad test problem) then there are not rotational modes in the solution and so rotational bodes don't help.  Which, as Jed points out correctly is not a good general conclusion.