Equivalence of saddle-points and optima, and duality for a class of non- smooth non-convex problems (Q1101679)

The author considers the problem of minimizing a locally Lipschitz function \(f: {\mathbb{R}}^ n\to {\mathbb{R}}\) subject to the constraints \(g_ i(x)\leq 0\), \(i=1,...,m\), where each \(g_ i:{\mathbb{R}}^ n\to {\mathbb{R}}\) is locally Lipschitz. After forming the Lagrangian \[ L(x;\lambda):=f(x)+\sum^{m}_{i=1}\lambda_ ig_ i(x), \] he uses invexity hypotheses on f, \(g_ i\), and a constraint qualification to prove that a point \(\hat x\) provides a global minimum for this problem iff there is a corresponding \({\hat \lambda}\in {\mathbb{R}}^ m_+\) satisfying the saddle-point condition \(L(\hat x;\lambda)\leq L(\hat x;{\hat \lambda})\leq L(x;{\hat \lambda})\), \(x\in {\mathbb{R}}^ n\), \(\lambda \in {\mathbb{R}}^ m_+\). The relation of this condition to the generalized Kuhn-Tucker and Fritz John optimality conditions is discussed, as are certain forms of duality. (A locally Lipschitz function \(f: {\mathbb{R}}^ n\to {\mathbb{R}}\) is ``invex'' if there exists a function \(\eta: {\mathbb{R}}^ n \times {\mathbb{R}}^ n \to {\mathbb{R}}^ n\) for which \(f(y)-f(x) \geq \max \{<\xi,\eta (x,y)>:\zeta\in \partial f(x)\}\), where \(\partial f(x)\) is Clarke's generalized gradient of f at x.)