On convergence of augmented Lagrangian method for inverse semi-definite quadratic programming problems (Q2379780)

The paper deals with an inverse problem associated with a semi-definite quadratic programming (SDQP) problem. The goal is to find a matrix and a vector of coefficients for the objective function of the SDQP problem which best approximate known estimates of these with respect to a certain norm and which simultaneously make a given feasible point of the SDQP problem optimal for it. For this inverse problem having about \(n^2/2\) variables, firstly the dual problem is set up which turns out to be a linearly positive semi-definite cone constrained semismoothly differentiable convex programming problem with only \(n\) variables. Then, for the dual problem, the global convergence of the augmented Lagrangian method is proven and the convergence rate of this method is established in dependency on the penalty parameter in the augmented Lagrangian. Numerical experiments with up to \(n=2000\) variables are presented.