On the use of ε-most-active constraints in an exact penalty function method for nonlinear optimization

This note presents an algorithm for nonlinear programming problems, which utilizes the ε-most-active constraint strategy in an exact penalty function method with trust region. The algorithm is particularly suitable for problems containing a large number of constraints. The global convergence of the proposed algorithm is proved. The results of limited computational experiments for discretized semi-infinite programming problems are also reported to demonstrate the effectiveness of the present approach.