Further insight into the convergence of the Fletcher-Reeves method (Q1974184)

The paper deals with the Fletcher-Reeves (FR) conjugate gradient method for unconstrained optimization. It is shown that the FR method has a certain self-adjusting property, which is independent of the line search. Based on this, general conditions are given which ensure the global convergence of the FR method using generalized Wolfe line searches or Armijo line searches. A counterexample is constructed to show that the conditions cannot be relaxed in certain sense. Finally, some remarks are given. Since the generalized line search is a natural extension of the standard line search, it may be helpful for analyzing theoretical properties of other conjugate gradient methods.