A nonmonotone trust-region algorithm with nonmonotone penalty parameters for constrained optimization (Q1883470)

The authors consider general nonlinear programming problems in the form \[ \min f(x) \quad\text{s.t. }c(x)= 0\quad\text{and }1\leq x\leq u. \] For these problems, a nonmonotone trust-region algorithm with nonmonotone penalty parameters is presented. The given algorithm combines an successive quadratic programming approach with a trust-region strategy to globalize the process. The global convergence theory for the given algorithm is developed without regularity assumptions. Numerical experiments are presented.