Single-Peakedness and Total Unimodularity: New Polynomial-Time Algorithms for Multi-Winner Elections

TL;DR

This paper introduces a novel LP-based approach leveraging total unimodularity to efficiently solve multi-winner election problems with single-peaked preferences, simplifying algorithms and broadening applicability.

Contribution

The paper presents a new technique using IP formulations with totally unimodular LP relaxations for multi-winner voting rules under single-peaked preferences, enabling efficient solutions.

Findings

Efficient algorithms for PAV and Chamberlin--Courant rules with single-peaked preferences.

First technique to efficiently find optimal committees for PAV with single-peaked preferences.

Any standard IP solver can be used without special algorithms, exploiting total unimodularity.

Abstract

The winner determination problems of many attractive multi-winner voting rules are NP-complete. However, they often admit polynomial-time algorithms when restricting inputs to be single-peaked. Commonly, such algorithms employ dynamic programming along the underlying axis. We introduce a new technique: carefully chosen integer linear programming (IP) formulations for certain voting problems admit an LP relaxation which is totally unimodular if preferences are single-peaked, and which thus admits an integral optimal solution. This technique gives efficient algorithms for finding optimal committees under Proportional Approval Voting (PAV) and the Chamberlin--Courant rule with single-peaked preferences, as well as for certain OWA-based rules. For PAV, this is the first technique able to efficiently find an optimal committee when preferences are single-peaked. An advantage of our approach is that no special-purpose algorithm needs to be used to exploit structure in the input preferences: any standard IP solver will terminate in the first iteration if the input is single-peaked, and will continue to work otherwise.