This is actually the point, and is analogous to convex optimization, in that systems always evolve along the boundary of constraint, and in the linear case, it

This is actually the point, and is analogous to convex optimization, in that systems always evolve along the boundary of constraint, and in the linear case, it will always be an exchange of active parts of the problem.