[Trilinos-Users] Questions about the null space used in ML's MatrixFreePreconditioner

[Ramsey, James J CIV (US)]

I'm not entirely sure what to put as the argument to the NullSpace argument in ML's MatrixFreePreconditioner constructor. If I consider my *actual* global stiffness "matrix" of an elastic problem, where Dirichlet boundary conditions are applied to pin the structure and remove zero-body modes, the null space should be just a zero vector if it's nonsingular. However, from what I've seen in example usages of the MatrixFreePreconditioner class, that's not the null space used. For an elastic case, I've seen the null space take to be 6 vectors corresponding to the rigid body modes of an unpinned elastic body, e.g. [1 0 0 1 0 0 ...] for the first null space vector, [0 1 0 0 1 0 ...] for the second, etc. (See <http://link.springer.com/chapter/10.1007/3-540-26662-3_8>.)

Should I just use the null space of the global stiffness matrix of an unpinned elastic body, or are there additional wrinkles to consider?