Can we implement a Schur complement solver, as in http://arxiv.org/abs/1803.05478 ? We have encountered some numerical instability when Ubeta becomes too large---presumably the Schur solver does not suffer?
(Unfortunately it is only for exp discretization.)
Can we implement a Schur complement solver, as in http://arxiv.org/abs/1803.05478 ? We have encountered some numerical instability when Ubeta becomes too large---presumably the Schur solver does not suffer?
(Unfortunately it is only for exp discretization.)