A unified analysis for a class of long-step primal-dual path-following interior-point algor.pdf

A Unied Analysis for a Class of Long-StepPrimal-Dual Path-Following Interior-PointAlgorithms for Semidenite ProgrammingRenato D. C. Monteiro and Yin ZhangyJune 1996Revised May 1997AbstractWe present a unied analysis for a class of long-step primal-dual path-following algorithmsfor semidenite programming whose search directions are obtained through linearization ofthe symmetrized equation of the central path HP (XS)  [PXSP1 + (PXSP1)T ]=2 = I,introduced by Zhang. At an iterate (X;S), we choose a scaling matrix P from the class ofnonsingular matrices P such that PXSP1 is symmetric. This class of matrices includes thethree well-known choices, namely: P = S1=2 and P = X1=2 proposed by Monteiro, and thematrix P corresponding to the Nesterov-Todd direction. We show that within the class ofalgorithms studied in this paper, the one based on the Nesterov-Todd direction has the lowestpossible iteration-complexity bound that can provably be derived from our analysis. Morespecically, its iteration-complexity bound is of the same order as that of the correspondinglong-step primal-dual path-following algorithm for linear programming introduced by Kojima,Mizuno and Yoshise.Keywords: Semidenite programming, primal-dual, path-following, interior-point methods.Abbreviated Title: A class of primal-dual algorithms for SDPSchool of Industrial and Systems Engineering, Georgia Institute of Technology, Atlanta, GA 30332. This authorsresearch is supported in part by the National Science Foundation under grants INT-9600343 and CCR-970048 andthe Oce of Naval Research under grant N00014-94-1-0340.yDepartment of Computational and Applied Mathematics, Rice University, Houston, Texas 77005-1892. Thisauthors research is supported in part by DOE DE-FG02-93ER25171-A001.1 1 IntroductionSeveral authors have discussed generalizations of interior-point algorithms for linear programming(LP) to the context of semidenite programming (SDP). The landmark work in this direction is dueto Nester