Custom Solver Could avoid partial pivoting

[nkoskelo]

We currently solve, J^TJ = -J^T f in parallel with partial pivoting of Gaussian elimination. Since we know that the left hand side is positive semi definite, one does not need to do Gaussian elimination with partial pivoting. We could instead use Cholesky decomposition.

[coreyostrove]

Interesting, tell me more. Would there be any direct performance gains from doing so? And how would this interact with being run in a distributed setting (my recollection is the current gaussian elimination solver is written so as to take advantage of MPI).