- Inexact Sequential Quadratic Optimization with Penalty Parameter Updates Inside the QP Resolve: Prolonged Model
Authors: James V. Burke, Frank E. Curtis, Hao Wang, Jiashan Wang
Summary: This paper focuses on the design of sequential quadratic optimization (generally often known as SQP) strategies for fixing large-scale nonlinear optimization issues. Essentially the most computationally demanding side of such an method is the computation of the search path throughout every iteration, for which we take into account using matrix-free strategies. Particularly, we develop a technique that requires an inexact clear up of a single QP subproblem to determine the convergence of the general SQP methodology. It’s recognized that SQP strategies could be affected by poor conduct of the worldwide convergence mechanism. To confront this subject, we suggest using an actual penalty perform with a dynamic penalty parameter updating technique to be employed inside the subproblem solver in such a manner that the ensuing search path predicts progress towards each feasibility and optimality. We current our parameter updating technique and show that, beneath affordable assumptions, the technique doesn’t modify the penalty parameter unnecessarily. We additionally talk about a matrix-free subproblem solver during which our updating technique could be integrated. We shut the paper with a dialogue of the outcomes of numerical experiments that illustrate the advantages of our proposed methods.
