\(\mathcal{UV}\)-theory of a class of semidefinite programming and its applications (Q2240656)

Convex semidefinite optimization problems are converted into eigenvalue problems by means of exact penalty functions. First-and second-order derivatives of the \(\mathcal{U}\)-Lagrangian of the function of the largest eigenvalues are derived under a transversality condition holds. A conceptual algorithm with superlinear convergence for \(\mathcal{UV}\)-decomposition is proposed, and an application involving an opimization problem containing a bilinear matrix inequality in the constraints is studied.