Consider the context of constrained function minimization. The well-known KKT conditions do not require the objective or constraint functions to be convex, but they do require them to be differentiable. On the other hand, if the problem is convex, then the KKT conditions generalize using the notion of subgradient.
Question: Does there exist a generalization of the KKT conditions for non-convex problems, where the objective function is differentiable everywhere except from the origin?
