A variant of trust-region methods for unconstrained optimization (Q2518723)

The authors study alternatives to the traditional trust-region methods used to solve unconstrained optimization problems. They begin by introducing the general unconstrained optimization problem and presenting on overview of the relevant literature. In the second section they focus on the traditional trust-region methods using the quadratic model for solving this optimization problem and outline the characteristics, properties and shortfalls of these approaches. A further extension of the trust-region method is described in the third section which uses the conic model. In the fourth section, the authors present the main contribution of this article, namely a new trust-region method based on a linear model for unconstrained optimization. Several theorems are presented, with proof, including the convergence properties of the proposed algorithm. The last section of this paper includes the detailed results of the performed computational experimentation and a section of concluding remarks.