Computation of condition numbers for linear programming problems using Peña’s method

Publisher Website
Google Scholar
Abstract
We present the computation of the condition numbers for linear programming (LP) problems from the NETLIB suite. The method of Peña [Peña, J., 1998, {Computing the distance to infeasibility: theoretical and practical issues}. Technical report, Center for Applied Mathematics, Cornell University.] was used to compute the bounds on the distance to ill-posedness ρ(d) of a given problem instance with data d, and the condition number was computed as C(d) = ∥d∥/ρ(d). We discuss the efficient implementation of Peña’s method and compare the tightness of the estimates on C(d) computed by Peña’s method to that computed by the method employed by Ordóñez and Freund [Ordóñez, F. and Freund, R.M., 2003, {Computational experience and the explanatory value of condition measures for linear optimization}. SIAM Journal on Optimization, 14, 307–333.]. While Peña’s method is generally much cheaper, the bounds provided are generally not as tight as those computed by Ordóñez and Freund. As a by-product, we use the computational results to study the correlation between log C(d) and the number of interior-point method iterations taken to solve a LP problem instance. Our computational findings on the preprocessed problem instances from NETLIB suite are consistent with those reported by Ordóñez and Freund.