A proximal-point SQP trust region method for solving some special class of nonlinear semi-definite programming problems (Q2489329)

An important nonlinear and nonconvex problem in optimal control is one of designing static or reduced order output feedback compensators that meet desired performance criteria. According to the design goals (i.e., \(H_2\) or mixed \(H_2/H_\infty\) design, respectively), one can derive a nonlinear semi-definite programming (NSDP) formulation of specific control synthesis problem. The authors attempt to compute numerically an optimal output feedback reduced order controller and/or to establish an optimal static output feedback control law. The main goal of this paper is to propose a proximal-point sequential quadratic programming (SQP) algorithm that makes use of the trust region for finding an approximate solution to the NSDP problem. Due to the lack of positive definiteness of the Hessian of the Lagrangian or the non-convexity structure of these problems the SQP methods might sometimes not converge. By combining the proximal-point approach with an SQP trust region method for solving NSDP problem they attempt to enlarge the robustness of the regular SQP trust region method. Thereafter the constrained trust region method for a nonlinear semi-definite program representing the \(H_2\) synthesis problem is extended to solve a more general nonlinear semi-definite program representing the fixed-order \(H_2/H_\infty\) synthesis problem. The numerical results show that by combining the proximal-point approach with the SQP trust region method one might gain efficiency and also robustness of the method.