STRICT MONOTONICITY AND IMPROVED COMPLEXITY IN THE STANDARD FORM PROJECTIVE ALGORITHM FOR LINEAR-PROGRAMMING

In a recent paper, Shaw and Goldfarb show that a version of the standard form projective algorithm can achieve O(square-root n L) step complexity, opposed to the O(nL) step complexity originally demonstrated for the algorithm. The analysis of Shaw and Goldfarb shows that the algorithm, using a constant, fixed steplength, approximately follows the central trajectory. In this paper we show that simple modifications of the projective algorithm obtain the same complexity improvement, while permitting a linesearch of the potential function on each step. An essential component is the addition of a single constraint, motivated by Shaw and Goldfarb's analysis, which makes the standard form algorithm strictly monotone in the true objective.