Global Convergence of the Restarting Conjugate Gradient Algorithm with a Generalized Curry Linesearch

Abstract:

      This paper presents a new restarting conjugate gradient algorithm for unconstrained optimization problem: min f(x),x∈Rⁿ. This algorithm used a generalized Curry linesearch, parameter β_ｋ can be selected in a finite closed interval. Especially,it is allowed that β_ｋ is negative. The global convergence of this algorithms is proved under weaker conditions.