scientific article; zbMATH DE number 3919685
[For the entire collection see Zbl 0552.00011.] The intention of the paper is to show by means of specific examples, that the development of numerical software influences theoretical research in mathematical programming. The authors consider the implementation of quadratic programming methods and its relationship to the implementation of sequential quadratic (SQP)-programming methods for nonlinear optimization, which are based on formulating and solving a sequence of quadratic programming problems. They explain the inefficiencies of a black box implementation of an active set method and some possibilities to improve the initial phase I for finding a feasible point. Particular attention is given to improve the factorization of the working set also with the aim to handle linear and nonlinear constraints separately. Alternative possibilities for computing an initial projected Hessian and for representing quasi-Newton matrices are discussed. Further investigation of active sets separately for linear and nonlinear constraints, lead to a new subclass of SQP-methods. Also the representation of Hessian matrices for sparse quadratic programs is considered.
This page was built for publication:
Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q3694573)