A new modified RMIL CG method with global convergence properties

Abstract

The Conjugate Gradient (CG) method is an approach commonly used to solve large-scale optimization issues. This method is considered efficient for its properties of global convergence and low requirements for memory. In this study, we proposed a novel CG coefficient γk using the method developed by Rivaie-Mustafa-Ismail-Leong (RMIL). The suggested technique is shown to have global convergence under exact line search. This is reinforced by the numerical test results, which concurrently indicate that the new CG method is more efficient in comparison with the current CG methods.