Ordinary convex programs without a duality gap

In the Kuhn-Tucker theory of nonlinear programming, there is a close relationship between the optimal solutions to a given minimization problem and the saddlepoints of the corresponding Lagrangian function. It is shown here that, if the constraint functions and objective function arefaithfully convex in a certain broad sense and the problem has feasible solutions, then theinf sup andsup inf of the Lagrangian are necessarily equal.