The Lagrangian duality in nonconvex optimization and modifications of the Toeplitz-Hausdorff theorem (Q1386559)

Let \(Z\) be an affine subspace of the real linear space \(H\) and \(Y=\{ y \}\) a finite-dimensional ordered real linear space. The paper is concerned with the optimization problem \(F(z) \to \inf, z \in Z, G(z)\leq 0\), where \(F(z)={\mathbb B}_{F}(z)+\Phi_{F}(z),\quad G(z)={\mathbb B}_{G}(z)+ \Phi_{G}(z),\quad {\mathbb B}_{F}(z)=B_{F}(z,z),\quad {\mathbb B}_{G}(z)=B_{G}(z,z); \quad B_{F}:H\times H \to R\) and \(B_{G}:H\times H \to Y\) are bilinear symmetric mappings, \(\Phi_{F}:H \to R\) and \(\Phi_{G}: H \to Y\) are convex functions. The author studies the Lagrangian duality method and gives abstract conditions, under which the method is valid for the initial problem. With the use of the developed technique he also proves very interesting modifications of the following classical Toeplitz-Hausdorff theorem: the image of the unit sphere \(\{ z \in H: | z | =1 \}\) under a quadratic mapping \((y_{1},y_{2}): [{\mathbb B}_{1}(z),{\mathbb B}_{2}(z) ]\) into \(R^2\) is a convex set provided that \({\mathbb B}_{1}\) and \({\mathbb B}_{2}\) are continuous Hermitian forms on a complex Hilbert space \(H\). The author obtains similar results on the convexity of the image of a quadratic mapping \([{\mathbb B}_{1}(z),{\mathbb B}_{2}(z), \cdots {\mathbb B}_{k}(z) ]\), where \(k\geq 2\).