[Paper Review] Single-Peakedness and Total Unimodularity: Efficiently Solve Voting Problems Without Even Trying.

Dominik Peters|arXiv (Cornell University)|Sep 12, 2016

Game Theory and Voting Systems40 references3 citations

TL;DR

The paper introduces a novel integer programming approach that leverages total unimodularity to efficiently solve single-peaked approval and ranked voting problems. By formulating PAV, Chamberlin-Courrant, and OWA-based committee elections as IP models, it guarantees integral solutions when preferences are single-peaked—enabling standard solvers to solve these NP-hard problems in polynomial time without custom algorithms.

ABSTRACT

Many NP-hard winner determination problems 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 fast algorithms for finding optimal committees under the PAV and Chamberlin-Courant voting rules under single-peaked preferences, as well as for certain OWA-based rules. Under single-crossing preferences, Young scores can also be calculated. An advantage of this technique 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.

Motivation & Objective

• To develop a general-purpose method for solving NP-hard voting problems efficiently when preferences are single-peaked.
• To eliminate the need for custom dynamic programming algorithms by using integer programming with total unimodularity.
• To enable standard IP solvers to detect and exploit single-peaked structure automatically, solving in one iteration when applicable.
• To extend the approach to single-crossing preferences for calculating Young scores.
• To unify the solution of diverse voting rules—PAV, Chamberlin-Courant, OWA-based rules—under a single theoretical framework.

Proposed method

• Formulate PAV, Chamberlin-Courant, and OWA-based committee elections as integer linear programs (IPs).
• Design the IP constraints such that the constraint matrix is totally unimodular when preferences are single-peaked.
• Leverage the property that totally unimodular LP relaxations yield integral optimal solutions, ensuring exactness without branch-and-bound.
• Use standard IP solvers that detect integrality in the first simplex iteration when the constraint matrix is totally unimodular.
• Apply the same framework to Young scores under single-crossing preferences by constructing a compatible IP formulation.
• Allow the solver to proceed normally when preferences are not single-peaked, maintaining correctness while gaining efficiency only when structure is present.

Experimental results

Research questions

• RQ1 Can total unimodularity be exploited in integer programming formulations of voting rules to achieve polynomial-time solutions under single-peaked preferences?
• RQ2 Does a general IP-based approach outperform or replace custom dynamic programming algorithms for single-peaked voting problems?
• RQ3 Can the same IP formulation framework handle multiple voting rules—including PAV, Chamberlin-Courant, and OWA-based rules—under a unified structure?
• RQ4 To what extent can standard IP solvers automatically detect and exploit single-peakedness without specialized code?
• RQ5 Can this approach be extended to other domain restrictions, such as single-crossing preferences, for additional voting rules like Young scores?

Key findings

• The proposed IP formulations yield integral optimal solutions when preferences are single-peaked due to total unimodularity of the constraint matrix.
• Standard IP solvers solve the problem in a single simplex iteration when inputs are single-peaked, achieving polynomial-time performance.
• The method avoids the need for custom dynamic programming or problem-specific algorithms by relying on off-the-shelf solvers.
• The approach extends to single-crossing preferences, enabling efficient computation of Young scores.
• The framework unifies the solution of multiple voting rules—PAV, Chamberlin-Courant, OWA-based rules—under a single, reusable IP formulation.
• The technique maintains correctness and efficiency across all input types, with performance gains only when structural constraints like single-peakedness are present.