On Mehrotra-Type Predictor-Corrector Algorithms

نویسندگان

  • Maziar Salahi
  • Jiming Peng
  • Tamás Terlaky
چکیده

In this paper we discuss the polynomiality of Mehrotra-type predictor-corrector algorithms. We consider a variant of the original prototype of the algorithm that has been widely used in several IPM based optimization packages, for which no complexity result is known to date. By an example we show that in this variant the usual Mehrotra-type adaptive choice of the parameter μ might force the algorithm to take many small steps to keep the iterates in a certain neighborhood of the central path, which is essential to prove the polynomiality of the algorithm. This example has motivated us to incorporate a safeguard in the algorithm that keeps the iterates in the prescribed neighborhood and allows us to get a reasonably large step size. This safeguard strategy is used also when the affine scaling step performs poorly that forces the algorithm to take pure centering steps. We proved that the modified algorithm, in the worst case, will terminate after at most O(n2 log (x0)T s0 2 ) iterations. By slightly modifying the Newton system in the corrector step, we reduce the iteration complexity to O(n log (x0)T s0 n ). To ensure the asymptotic convergence of the algorithm, we changed Mehrotra’s heuristic slightly. The new algorithms have the same iteration complexity as the previous one, but enjoy superlinear convergence rate as well.

برای دانلود رایگان متن کامل این مقاله و بیش از 32 میلیون مقاله دیگر ابتدا ثبت نام کنید

ثبت نام

اگر عضو سایت هستید لطفا وارد حساب کاربری خود شوید

منابع مشابه

Numerical Evaluation of SDPA (

SDPA (SemiDe nite Programming Algorithm) is a C++ implementation of a Mehrotra-type primal-dual predictor-corrector interior-point method for solving the standard form semide nite program and its dual. We report numerical results of large scale problems to evaluate its performance, and investigate how major time-consuming parts of SDPA vary with the problem size, the number of constraints and t...

متن کامل

Primal-Dual Path-Following Algorithms for Determinant Maximization Problems With Linear Matrix Inequalities

Primal-dual path-following algorithms are considered for determinant maximization problem (maxdet-problem). These algorithms apply Newton's method to a primal-dual central path equation similar to that in semideenite programming (SDP) to obtain a Newton system which is then symmetrized to avoid nonsymmetric search direction. Computational aspects of the algorithms are discussed, including Mehro...

متن کامل

A Mehrotra-Type Predictor-Corrector Algorithm for P∗(κ) Linear Complementarity Problems

Mehrotra-type predictor-corrector algorithm, as one of most efficient interior point methods, has become the backbones of most optimization packages. Salahi et al. proposed a cut strategy based algorithm for linear optimization that enjoyed polynomial complexity and maintained its efficiency in practice. We extend their algorithm to P∗(κ) linear complementarity problems. The way of choosing cor...

متن کامل

Polynomial Convergence of a Predictor-Corrector Interior-Point Algorithm for LCP

We establishe the polynomial convergence of a new class of pathfollowing methods for linear complementarity problems (LCP). Namely, we show that the predictor-corrector methods based on the L2 norm neighborhood. Mathematics Subject Classification: 90C33, 65G20, 65G50

متن کامل

ذخیره در منابع من


  با ذخیره ی این منبع در منابع من، دسترسی به آن را برای استفاده های بعدی آسان تر کنید

عنوان ژورنال:
  • SIAM Journal on Optimization

دوره 18  شماره 

صفحات  -

تاریخ انتشار 2007