Convergence properties of nonmonotone spectral projected gradient methods (Q557737): Difference between revisions

In a recent paper, a nonmonotone spectral projected gradient (SPG) method was introduced by \textit{E. G. Birgin} et al. [SIAM J. Optim. 10, No. 4, 1196--1211 (2000; Zbl 1047.90077)] for the minimization of differentiable functions on closed convex sets and extensive presented results showed that this method was very efficient. In this paper, the authors study the convergence properties of SPG methods. They remove various boundedness conditions that are assumed in existing results, such as boundedness from below of the objective function \(f\), boundedness of the sequence of iterates generated by this method or existence of accumulation point of that sequence. If the gradient of \(f\) is uniformly continuous, the convergence theory of this method is established and it is proved that the SPG method forces the sequence of projected gradients to zero. Moreover, under appropriate conditions it is shown that the SPG method has some encouraging convergence properties, such as the global convergence of the sequence of iterates generated by this method and the finite termination, which improve and generalize the corresponding results in previous papers.