Repulsion-based p-dispersion with distance constraints in non-convex polygons

Zhengguan Dai, Kathleen Xu, Melkior Ornik

Research output: Contribution to journalArticlepeer-review


Motivated by the question of optimal facility placement, the classical p-dispersion problem seeks to place a fixed number of equally sized non-overlapping circles of maximal possible radius into a subset of the plane. While exact solutions to this problem may be found for placement into particular sets, the problem is provably NP-complete for general sets, and existing work is largely restricted to geometrically simple sets. This paper makes two contributions to the theory of p-dispersion. First, we propose a computationally feasible suboptimal approach to the p-dispersion problem for all non-convex polygons. The proposed method, motivated by the mechanics of the p-body problem, considers circle centers as continuously moving objects in the plane and assigns repulsive forces between different circles, as well as circles and polygon boundaries, with magnitudes inversely proportional to the corresponding distances. Additionally, following the motivating application of optimal facility placement, we consider existence of additional hard upper or lower distance bounds on pairs of circle centers, and adapt the proposed method to provide a p-dispersion solution that provably respects such constraints. We validate our proposed method by comparing it with previous exact and approximate methods for p-dispersion. The method quickly produces near-optimal results for a number of containers.

Original languageEnglish (US)
Pages (from-to)75-91
Number of pages17
JournalAnnals of Operations Research
Issue number1-2
StatePublished - Dec 2021


  • Dispersion
  • Location
  • Packing
  • Strategic planning

ASJC Scopus subject areas

  • Decision Sciences(all)
  • Management Science and Operations Research


Dive into the research topics of 'Repulsion-based p-dispersion with distance constraints in non-convex polygons'. Together they form a unique fingerprint.

Cite this