A Very Simple Polynomial-Time Algorithm for Linear Programming
Abstract
This document proposes a polynomial-time algorithm for linear programming. This algorithm augments the objective by a logarithmic penalty function and then solves a sequence of quadratic approximations of this program. This algorithms has a complexity of iterations and arithmetic operations, where m is a number of variables and L is the size of the problem encoding in binary. This algorithm does not require knowledge of the optimal value and generates a sequence of primal (Dual) feasible solutions that converges to an optimal primal (dual) solution (the latter property provides a particularly simple stopping criterion). Moreover, this algorithm is simple and intuitive, both in the description and in the analysis - in contrast to existing polynomial-time algorithms. It works exclusively in the original space. Keywords: Karmarkar methods; Quadratic approximation; Logarithmic penalty functions.
Document Details
- Document Type
- Technical Report
- Publication Date
- Sep 01, 1988
- Accession Number
- ADA202502
Entities
People
- Paul Tseng
Organizations
- Massachusetts Institute of Technology