Introduction to duality in optimization theory

Abstract

A systematic exposition of duality theory is given on what appears to be the optimal level of generality. A condition is offered which implies that the ideal of duality theory is achieved. For the case of linear programming, our approach leads to two novel features. In the first place, primal and dual LP-problems and complementarity conditions are defined canonically, without choosing a matrix form. In the second place, without deriving the explicit form of the dual problem, we show that the following well-known fact implies that the condition mentioned above holds: the polyhedral set property is invariant under linear maps. We give a new quick algorithmic proof of this fact.