A geometric framework for nonconvex optimization duality using augmented Lagrangian functions (Q2481369)

The authors present a geometric framework to analyze duality for nonconvex optimization problems using augmented Lagrangian functions. By developing a separation theorem for nonconvex sets via general concave surfaces, they derive necessary and sufficient conditions for establishing strong duality between geometric primal and dual problems. The implications of these conditions for constrained nonconvex optimization problems are discussed. Necessary and sufficient conditions for the convergence of penalty methods are also given together with implications of these conditions for the penalty methods for constrained nonconvex optimization problems.