A subspace implementation of quasi-Newton trust region methods for unconstrained optimization (Q2503148)

The authors study the subspace properties of trust region methods for unconstrained optimization. They begin with the formulation of the general unconstrained optimization problem of a continuously differentiable objective function. At each iteration, a trust region is defined for each iteration, which is a region near the current iterate point. The authors then continue by describing an algorithm based on trust methods, on which the subsequent analysis of subspace properties is based. The main part of the paper focuses on a series of theorems and lemmas which describe the properties of the subspace of the trial step, which are later used to produce am improved optimization algorithm. The article concludes with a series of numerical results.