Maintaining the positive definiteness of the matrices in reduced secant methods for equality constrained optimization (Q2641222)

An algorithm is proposed for minimizing a function on a manifold in \(R^ n\) which is locally a reduced secant method. The local method is extended to \(R^ n\) by using a nondifferentiable augmented Lagrangian whose decrease is obtained by both longitudinal search and transversal search. The main objective is to show that the longitudinal path can be designed to maintain the positive definiteness of the reduced matrices.