不假定凸性和精确线搜索时DFP算法的收敛性

Convergence of DFP Algorithms without Convexity and Exact Line Search Assumptions

摘要:

对非凸目标函数，Broyden变尺度算法的收敛性是一个没有完全解决的问题．本文主要针对DFP修正公式证明在不假定精确线搜索条件下,对光滑的目标函数，当DFP算法得到的点列收敛时，该点列一定趋向于稳定点．并指出对于其他Broyden算法结论都是成立的.

Abstract:

In this paper, the convergence of the Broyden algorithms without convexity and exact line search assumptions is discussed. It is proved that if the objective function is suitably smooth and the DFP algorithm produces a convergence point sequence, then the limit point of the sequence is a critical point of the objective function. We give mainly a proof for the DFP update, then point out that all the results are true for Broyden algorithms by a remark.