Two efficient nonlinear conjugate gradient methods for Riemannian manifolds

In this paper, we address some of the computational challenges associated with the RMIL+ conjugate gradient parameter by proposing an efficient conjugate gradient (CG) parameter along with its generalization to the Riemannian manifold. This parameter ensures the good convergence properties of the CG method in Riemannian optimization and it is formed by combining the structures of two classical CG methods. The extension utilizes the concepts of retraction and vector transport to establish sufficient descent property for the method via strong Wolfe line search conditions. Additionally, the scheme achieves global convergence using the scaled version of the Ring-Wirth nonexpansive condition. Finally, numerical experiments are conducted to validate the scheme’s effectiveness. We consider both unconstrained Euclidean optimization test problems and Riemannian optimization problems. The results reveal that the performance of the proposed method is significantly influenced by the choice of line search in both Euclidean and Riemannian optimizations.