A NEW CORRECTOR-PREDICTOR ALGORITHM FOR CONVEX QUADRATIC SEMIDEFINITE OPTIMIZATION

Abstract

In this paper, we propose a new corrector-predictor algorithm for convex quadratic semidefinite optimization problem based on a new proximity measure.  The search direction is obtained by an equivalent algebraic transformation of the centering equation.  At each iteration, the algorithm is composed of a corrector step and a predictor step.  The predictor step uses line search schemes requiring the reduction of the duality gap, while the corrector step is used to restore the iterates to the neighborhood of the central path.  Finally, the algorithm has the currently best-known iteration complexity.