How the Spatial Position of Individuals Affects Their Influence on Swarms: A Numerical Comparison of Two Popular Swarm Dynamics Models

Allison Kolpas, Michael Busch, Hong Li, Iain D. Couzin, Linda Petzold, Jeff MoehlisPLoS 2013 · DOI

Abstract

Schools of fish and flocks of birds are examples of self-organized animal groups that arise through social interactions among individuals. We numerically study two individual-based models, which recent empirical studies have suggested to explain self-organized group animal behavior: (i) a zone-based model where the group communication topology is determined by finite interacting zones of repulsion, attraction, and orientation among individuals; and (ii) a model where the communication topology is described by Delaunay triangulation, which is defined by each individual's Voronoi neighbors. The models include a tunable parameter that controls an individual's relative weighting of attraction and alignment. We perform computational experiments to investigate how effectively simulated groups transfer information in the form of velocity when an individual is perturbed. A cross-correlation function is used to measure the sensitivity of groups to sudden perturbations in the heading of individual members. The results show how relative weighting of attraction and alignment, location of the perturbed individual, population size, and the communication topology affect group structure and response to perturbation. We find that in the Delaunay-based model an individual who is perturbed is capable of triggering a cascade of responses, ultimately leading to the group changing direction. This phenomenon has been seen in self-organized animal groups in both experiments and nature.

Introduction

From schools of fish to flocks of birds, self-organized animal groups move with startling coordination without a central leader. While many models replicate these patterns of motion, the internal dynamics of information flow remain less understood. If a single fish at the edge of a school spots a predator and turns, how does that "velocity information" propagate through the rest of the group? Does it matter if that fish is at the front, the back, or the center? This study addresses these questions by comparing two different ways of modeling how individuals communicate with their neighbors: metric (distance-based) versus topological (neighbor-based) communication.

Methods

The researchers conducted numerical simulations of two individual-based models (IBMs):

  1. Zone-based Model (Metric): Individuals interact with anyone within fixed circular zones of repulsion, orientation, and attraction.
  2. Delaunay-based Model (Topological): Individuals interact only with their "Voronoi neighbors," defined by Delaunay triangulation, regardless of the exact distance.
Delaunay triangulation and Voronoi cells defining topological neighbors.

To test influence, a "perturbation experiment" was performed: in a steady-state swarm, the heading of a single individual was suddenly rotated by 90 degrees. A cross-correlation function then measured how much the rest of the group’s average heading changed in response over time. The study utilized synthetic data from 1,000 simulations per parameter set, varying population size ($N$ from 10 to 150) and the ratio of alignment-to-attraction ($r$).

Results and Key Insights

The most striking finding is that information propagates much more effectively in the Delaunay (topological) model. In this model, a single perturbed individual can trigger a "cascade" of responses, whereas in the zone-based model, the signal often gets dampened or averaged out by the high number of neighbors in a fixed radius.

Polarization, elongation, and fragmentation metrics across varying group sizes and alignment-to-attraction ratios.

Key observations include:

  • The "Front" Advantage: In topological swarms, individuals at the front have significantly higher influence on the group's direction than those at the rear, suggesting "leadership" is a function of spatial geometry.
  • Influence Decay: In the zone-based model, as group size ($N$) increases, the influence of any single individual drops to nearly zero. In contrast, individuals in the Delaunay model maintain significant influence even as the group grows.
  • Structural Dynamics: Both models show that increasing alignment weighting ($r$) leads to higher polarization. However, the zone-based model reaches a peak elongation and then becomes prone to fragmentation (splitting into subgroups), while Delaunay groups remain cohesive but more dispersed.
  • Neighbor Consistency: In the Delaunay model, individuals consistently interact with 4 to 6 neighbors regardless of density, whereas neighbor counts fluctuate wildly in the zone-based model, causing signals to be lost in dense metric groups.

Discussion

While these models provide deep insights, they involve simplifications such as 2D movement, constant speeds, and the absence of visual occlusion or individual heterogeneity. Future work could explore 3D environments and how "informed" individuals (e.g., those aware of a predator) interact with spatial positioning to exert even greater influence on the swarm.

Summary updated

Abstract

Schools of fish and flocks of birds are examples of self-organized animal groups that arise through social interactions among individuals. We numerically study two individual-based models, which recent empirical studies have suggested to explain self-organized group animal behavior: (i) a zone-based model where the group communication topology is determined by finite interacting zones of repulsion, attraction, and orientation among individuals; and (ii) a model where the communication topology is described by Delaunay triangulation, which is defined by each individual's Voronoi neighbors. The models include a tunable parameter that controls an individual's relative weighting of attraction and alignment. We perform computational experiments to investigate how effectively simulated groups transfer information in the form of velocity when an individual is perturbed. A cross-correlation function is used to measure the sensitivity of groups to sudden perturbations in the heading of individual members. The results show how relative weighting of attraction and alignment, location of the perturbed individual, population size, and the communication topology affect group structure and response to perturbation. We find that in the Delaunay-based model an individual who is perturbed is capable of triggering a cascade of responses, ultimately leading to the group changing direction. This phenomenon has been seen in self-organized animal groups in both experiments and nature.
BibTeX
@inproceedings{kolpas2013how,
  author = {Allison Kolpas and Michael Busch and Hong Li and Iain D. Couzin and Linda Petzold and Jeff Moehlis},
  title = {{How the Spatial Position of Individuals Affects Their Influence on Swarms: A Numerical Comparison of Two Popular Swarm Dynamics Models}},
  booktitle = {PLoS},
  year = {2013},
  doi = {10.1371/journal.pone.0058525},
}
Figure