Duality for nonconvex optimization and its applications (Q1329205)

The main result of the paper is the following interesting theorem: Let \(Y\) be a (locally convex) normed space and \(g: Y\to \overline{\mathbb{R}}\) be (weakly) l.s.c. at \(y_ 0\in Y\). Suppose that there exists \(y^*_ 0\in Y^*\) such that \(g^*\) is (Gateaux) Fréchet differentiable at \(y^*_ 0\) and \(\nabla g^*(y^*_ 0)= y_ 0\). Then \(g^{**}(y_ 0)= g(y_ 0)\). The author applies the above result to duality schemes by using perturbation functions and Lagrangeans. Then he calculates the support function for the intersection of ellipsoids, which yields a class of dual norms, and constructs dual problems to the problem of minimax estimation.