The marginal value formula in input optimization (Q3974420)

After recalling some preliminary formulas about the theory of saddle points, an extension of the usual marginal formula to perturbations is given which is claimed to be the most complete formula for bi-convex mathematical programming models. This formula is then used to describe an optimal input on stable paths for bi-convex models. Two methods to solve the input optimization problem are presented, based on improvable stable paths and on stable perturbations. Both methods are of the descent type, for which convergence properties are established. The theoretical developments are complemented with many examples and pertinent applications including a problem of production optimization in a textile mill, an example of convexifiable constraints and an analysis of stable planning. Finally the role of this approach in a duality theory is emphasized. This article presents an extensive study of the capabilities of the marginal value formula in solving input optimization problems. The author demonstrates his mastership over the subject and his large knowledge of the related literature. It can however be deplored that a reader with less extensive knowledge of the subject will experience difficulties in following the expositions which are not always very clear nor straightforward.