Modified Cholesky Factorizations for Sparse Preconditioners

From MaRDI portal





In large scale unconstrained optimization problems, preconditioned conjugate gradient techniques are often used to solve the Newton equations, however in certain applications, ``natural sparse preconditioners can be derived from the structure of the problem and can accelerate convergence significantly. This paper describes such an adaptation of two MC techniques --- GMW (Gill-Murray-Wright) and SE (Schnabel-Eskow) --- in a truncated Newton minimization method. The analysis suggests that, for the minimization method examined, SE may be better for highly nonlinear functions, while GMW may be more effective for functions that are well approximated by locally convex quadratic models.





Describes a project that uses

Uses Software






This page was built for publication: Modified Cholesky Factorizations for Sparse Preconditioners

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q5286358)