Two effective hybrid conjugate gradient algorithms based on modified BFGS updates (Q645035)

In this paper, following the successful approach proposed by \textit{N. Andrei} [Numer. Algorithms 47, No. 2, 143--156 (2008; Zbl 1141.65041)], [ibid., 54, No. 1, 23--46 (2010; Zbl 1192.65074)], the authors introduced the hybrid conjugate gradient methods based on two modified BFGS updates proposed by \textit{Y.-X. Yuan} [IMA J. Numer. Anal. 11, No. 3, 325--332 (1991; Zbl 0733.65039)] and \textit{D. Li} and \textit{M. Fukushima} [J. Comput. Appl. Math. 129, No. 1-2, 15--35 (2001; Zbl 0984.65055)]. The proposed methods are hybridizations of Hestenes-Stiefel and Dai-Yuan CG methods. Under proper conditions, the authors proved the global convergence of one of the proposed algorithms for uniformly convex functions and the other for general functions. To enhance the performance of the line search procedure for establishing the strong Wolfe conditions , a new procedure for computing the initial value of the steplength is proposed. Comparative numerical results obtained by implementation of the presented algorithms and two efficiently representative hybrid CG methods proposed by Andrei [loc. cit.], on a set of unconstrained optimization test problems from the CUTEr collection [\textit{N. I. M. Gould, D. Orban} and \textit{P. L. Toint}, CUTEr and SifDec: a constrained and unconstrained testing environment, revisited, ACM Trans. Math. Softw. 29, No. 4, 373--394 (2003; Zbl 1068.90526)] and in the sense of the performance profile introduced by \textit{E. D. Dolan} and \textit{J. J. Moré} [Math. Program. 91, No. 2 (A), 201--213 (2002; Zbl 1049.90004)], showed the efficiency of the proposed hybrid algorithms.