Preconditioned conjugate gradient algorithms for nonconvex problems with box constraints (Q985699)

The authors consider the problem of minimizing a continuously differentiable function of \(n\) variables under box constraints. They propose a conjugate gradient and quasi-Newton algorithms for solving the problem. Unlike to algorithm by \textit{D. P. Bertsekas} [SIAM J. Control Optimization 20, 221--246 (1982; Zbl 0507.49018)], the authors use the quasi-Newton approximation instead of Hessian and different search rules. Unlike to \textit{R. Pytlak's} algorithm [SIAM J. Optim. 8, No. 2, 532--560 (1998; Zbl 0915.65060)] a preconditioner in defining the direction of descent is used. The algorithms, which are presented as two versions of one unifying general algorithm are globally convergent. Numerical results with test examples presented in the concluding part of the paper support the claim that the proposed algorithms are competitive with the ones known from the literature.