Nonlinear semidefinite programming: sensitivity, convergence, and an application in passive reduced-order modeling (Q868464)

The paper deals with a numerical method for solving nonlinear semidefinite programs. The NLSDP model appears when the non polyhedral cone of semidefinite matrices is used to define the inequality constraints. The authors propose an algorithm that extends to the semidefinite setting the well-known sequential quadratic programming (SQP) method for standard nonlinear programming,. The so called sequential semidefinite programming (SSDP) model solves at each step a local approximation of the problem in form of an SDP, that do not change the semidefinite cone. The main result is a self-contained proof of the local quadratic convergence for the SSDP method. The proof base on a sensitivity for local solutions of quadratic semidefinite programs and is done using three main assumptioms: the local uniqueness of the solution; the strict complementarity and a strong second-order sufficient optimality condition. The description of the application of the NLSDP model to the passive reduced-order modeling is given and numerical results from a Matlab implementation SSDP for this problem are discussed.