A study of the complexity of an infeasible predictor-corrector variant of Mehrotra algorithm

Abstract

An infeasible version of the algorithm developed by Bastos and Paixão is analyzed and its complexity is discussed. We prove the efficiency of this infeasible algorithm by showing its complexity and Q-linear convergence. We start by demonstrating that, at each iteration, the step size computed by this infeasible predictor-corrector variant algorithm is bounded below 1/250 n 4 by and has O(n 4 | log(ε)|) iteration complexity; thus, proving that, similarly to what happens with the feasible version, this infeasible version of Bastos and Paixão algorithm has polynomial iteration complexity and is Q-linearly convergent. © 2014 Springer International Publishing.