Regularized symmetric indefinite systems in interior point methods for linear and quadratic optimization

摘要: This paper presents linear algebra techniques used in the implementation of an interior point method for solving programs and convex quadratic with constraints. New regularization Newton systems applicable to both symmetric positive definite indefinite are described. They transform latter quasidef-inite known be strongly factorizable a form Cholesky-like factorization.Two different techniques,primal; dual, very well suited (infeasible) primal-dual algorithm. particular algorithm, extension multiple centrality correctors, is implemented our solver HOPDM. Computational results given illustrate potential advantages approach when applied solution large programs.