An efficient descent algorithm for a class of unconstrained optimization problems of nonlinear large mesh-interconnected systems

Presents an efficient descent algorithm for a class of unconstrained optimization problems of nonlinear large mesh-interconnected systems. This algorithm combines an approximate scaled gradient method with a textured decomposition-based block Gauss-Seidel method. The authors prove that their algorithm is globally convergent and that it is numerically stable. The authors also demonstrate the computational efficiency of the method compared with the Newton-like method associated with the sparse matrix technique through several numerical experiments