Paper

Worst-case analysis of restarted primal-dual hybrid gradient on totally unimodular linear programs

Hinder, Oliver

2023

Overview

Recently, there has been increasing interest in using matrix-free methods to solve linear programs due to their ability to attack extreme-scale problems. Each iteration of these methods is a matrix-vector product, so they benefit from a low memory footprint and are parallelized easily. Restarted primal-dual hybrid gradient (PDHG) is a matrix-free method with good practical performance. Prior analysis showed that it converges linearly on linear programs where the linear convergence constant depends on the Hoffman constant of the KKT system. We refine this analysis of restarted PDHG for totally unimodular linear programs. In particular, we show that restarted PDHG finds an \\(\\epsilon\\)-optimal solution in \\(O( H m_1^{2.5} \\sqrt{\\textbf{nnz}(A)} \\log(H m_2 /\\epsilon) )\\) matrix-vector multiplies where \\(m_1\\) is the number of constraints, \\(m_2\\) the number of variables, \\(\\textbf{nnz}(A)\\) is the number of nonzeros in the constraint matrix, \\(H\\) is the largest absolute coefficient in the right hand side or objective vector, and \\(\\epsilon\\) is the distance to optimality of the point outputted by PDHG.

• Publisher: Cornell University Library, arXiv.org
• Subject: Convergence / Hierarchies / Linear programming / Mathematical analysis / Matrices (mathematics) / Optimization