Combining trust-region and line-search algorithms for minimization subject to bounds (Q1424055)

The trust region algorithm for minimizing a smooth function subject to box constraints is combined with a line search procedure. The authors show convergence of the procedure to a stationary point. The feature of the algorithm is it that the computation of only one descent direction per iteration is necessary. The line search is used to guarantee that this direction can be used to get the new iteration point. Usefulness of the algorithm is shown by presenting computational results for one example.