Proximal interior point method for convex semi-infinite programming (Q2778680)

The authors propose an interior point type method for the solution of ill-posed convex semi-infinite programming problems. In fact, in each step of the algorithm a logarithmic barrier approach is used to solve a finite optimization problem which stems from a discretization of the original, infinite index set. Due to the ill-posedness of the original problem, the logarithmic barrier functions do not have to be strictly convex. Hence, a multi-step prox term is added to the barrier function. When the regularized barrier problem is solved with sufficient accuracy, the discretization of the index set, the barrier parameter, the prox parameter, and the accuracy parameter are updated for the next step. A special deleting rule allows one to drop a large part of the constraints in the discretized problem. Under mild assumptions the authors show that for an appropriate adjustment of the four mentioned parameters, the proposed method is well defined, convergent to a solution of the original problem and stable with respect to data perturbations. A number of numerical examples illustrate the performance of the algorithm.