Nonlinear optimization: Characterization of structural stability (Q1177914)

Considered is the optimization problem \({\mathcal P}(f,H,G)\): \(\text{min} f(x)\) on \(M(H,G)=\{x\in R^ n\mid\;H(x)=0\), \(G(x)\geq0\}\), where \(H: R^ n\to R^ n\), \(G: R^ n\to R^ s\) and \(f: R^ n\to R\) are such that \(f\) and the components of \(H\) and \(G\) are in \(C^ 2(R^ n,R)\). The lower level sets for \({\mathcal P}(f,H,G)\) are defined by \({\mathcal L}^ t(f,H,G)=\{x\in M(H,G)\mid\;f(x)\leq t\}\). Two such problems are equivalent if each lower set of the one can be mapped homeomorphically onto a corresponding lower set of the other. In the case that \({\mathcal P}(\tilde f,\tilde H,\tilde G)\) is equivalent to \({\mathcal P}(f,H,G)\) for all \((\tilde f,\tilde H,\tilde G)\) in some neighborhood of \((f,H,G)\) in the \(C^ 2_ s\)-topology one says that \({\mathcal P}(f,H,G)\) is structurally stable. The main result of the paper is: The optimization problem \({\mathcal P}(f,H,G)\) with compact feasible set \(M(H,G)\) is structurally stable if and only if the following three conditions are satisfied: (C1) The Mangasarian-Fromowitz constraint-qualification is satisfied at every point of \(M(H,G)\); (C2) Every Kuhn-Tucker point of \({\mathcal P}(f,H,G)\) is strongly stable in the sense of Kojima; (C3) Different Kuhn-Tucker points have different (\(f-\)) values.