Uniform Capacitated Facility Location Problems with Penalties/Outliers

TL;DR

This paper introduces a novel LP-rounding framework that achieves constant-factor approximation algorithms for capacitated facility location problems with penalties and outliers, allowing slight capacity violations.

Contribution

It provides the first constant-factor approximation algorithms for CFLPO and CkFLPP with capacity violations using LP-rounding techniques.

Findings

Achieves constant-factor approximations with capacity violations.

First known algorithms for CFLPO and CkFLPP.

Solutions are easily integrable with other LP-based methods.

Abstract

In this paper, we present a framework to design approximation algorithms for capacitated facility location problems with penalties/outliers using LP-rounding. Primal-dual technique, which has been particularly successful in dealing with outliers and penalties, has not been very successful in dealing with capacities. On the other hand, no primal-dual solution has been able to break the hardness of capacitated facility location problem(CFLP). LP-rounding techniques had also not been very successful in dealing with capacities until a recent work by Grover et al. \cite{GroverGKP18}. Their constant factor approximation violating the capacities by a small factor ($1 + \epsilon$) is promising while dealing with capacities.Though LP-rounding has not been very promising while dealing with penalties and outliers, we successfully apply it to deal with them along with capacities. That is, our results are obtained by rounding the solution to the natural LP once again exhibiting the power of LP-rounding technique. Solutions obtained by LP-rounding are easy to integrate with other LP-based algorithms.In this paper, we apply our framework to obtain first constant factor approximations for capacitated facility location problem with outlier (CFLPO) and capacitated $k$-facility location problem with penalty(C$k$FLPP) for hard uniform capacities using LP-rounding. Our solutions incur slight violations in capacities, ($1 + \epsilon$) for the problems without cardinality($k$) constraint and ($2 + \epsilon$) for the problems with the cardinality constraint. For the outlier variant, we also incur a small loss ($1 + \epsilon$) in outliers. Due to the hardness of the underlying problems, the violations are inevitable. Thus we achieve the best possible by rounding the solution of natural LP for these problems. To the best of our knowledge, no results are known for CFLPO and C$k$FLPP.