Improved LP-based Approximation Algorithms for Facility Location with Hard Capacities

TL;DR

This paper introduces improved LP-based approximation algorithms for the Capacitated Facility Location problem, significantly reducing the approximation ratio and providing the first such guarantees for related restricted versions.

Contribution

It presents a new LP-based approximation algorithm with a ratio of approximately 9.09 for CFL, and a 4-approximation for CFL with cardinality facility costs, improving longstanding results.

Findings

New LP-based approximation ratio of ~9.09 for CFL

First LP-based 4-approximation for CFL-CFC

Significant improvement over previous ratios

Abstract

The Capacitated Facility Location (CFL), a long-standing classic problem with intriguing approximability and literature dated back to the 90s, is considered. Following the open question posted in [Williamson and Shmoys, 2011] and the notable work due to [An et al., FOCS~2014], we present an LP-based approximation algorithm with a guarantee of $(10+\sqrt{67})/2 \approx 9.0927$, a significant improvement upon the previous LP-based ratio of $288$ due to An et al. in 2014. Our contribution for this part is a simple and elegant rounding algorithm that brings clear insights for the MFN relaxation and the CFL problem. For CFL with cardinality facility cost (CFL-CFC), we present an LP-based $4$-approximation algorithm, which improves upon the decades-old ratio of 5 due to Levi et al. that ages up since 2004. Prior to our work, it was not clear whether or not LP-based methods can be used to provide a guarantee better than 5 for the CFL problem, even for restricted versions of this problem, for which natural LPs are already known to have small integrality gaps. Our rounding algorithm provides the first affirmative answer on the case with cadinality facility cost.