Modelling competition for space: Emergent inefficiency and inequality due to spatial self-organization among a group of crowd-avoiding agents

Competition for a limited resource is the hallmark of many complex systems, and often, that resource turns out to be the physical space itself. In this work, we study a novel model designed to elucidate the dynamics and emergence in complex adaptive systems in which agents compete for some spatially spread resource. Specifically, in the model, the dynamics result from the agents trying to position themselves in the quest to avoid physical crowding experienced locally. We characterize in detail the dependence of the emergent behavior of the model on the population density of the system and the individual-level agent traits such as the extent of space an agent considers as her neighborhood, the limit of occupation density one tolerates within that neighborhood, and the information accessibility of the agents about neighborhood occupancy. We show that inefficiency in utilizing physical space shows transition at a specific density and peaks at another distinct density. Furthermore, we demonstrate that the variation of inefficiency relative to the information accessible to the agents exhibits contrasting behavior above and below this second density. We also look into the inequality of resource sharing in the model and show that although inefficiency can be a non-monotonic function of information depending upon the parameters of the model, inequality, in general, decreases with information. Our study sheds light on the role of competition, spatial constraints, and agent traits within complex adaptive systems, offering insights into their emergent behaviors.

💡 Research Summary

The paper introduces a minimalist yet insightful model to explore how agents compete for a spatially distributed, limited resource—physical space—when each agent seeks to avoid local crowding. The authors place N agents on a one‑dimensional lattice of L sites (periodic boundaries) and define four key parameters: (i) population density ρ = N/L, (ii) neighborhood size z (the number of sites considered for local density evaluation), (iii) tolerance threshold τ (the maximum allowable local occupancy density for an agent to be a “winner”), and (iv) information radius r (the range within which an agent can observe vacancies and relocate). At each Monte‑Carlo step a randomly chosen losing agent moves to a randomly selected vacant site inside its information radius; winners stay put. An agent’s payoff for a round is 1 if the local occupancy ν ≤ τ, otherwise 0, and cumulative payoff is the sum over rounds.

The dynamics generate absorbing states when all agents are winners (no further moves). The authors identify two characteristic densities. Below a “transition density” the system almost always reaches an absorbing configuration, yielding low global inefficiency η (defined as the normalized deviation of the number of losers from its theoretical minimum). At a higher “peak sustainable density” no configuration allows all agents to win; the system remains dynamically active and η reaches a maximum. Analytically, the peak density can be approximated as ρ_peak ≈ 1/(z·τ + 1), reflecting the maximal packing compatible with the tolerance rule.

A central focus is the role of information. The authors introduce a standardized information radius rs = r/(z/2). When rs < 1, agents have less information than the size of their decision‑making neighborhood, leading to a mismatch between perception and action. Simulations reveal a non‑monotonic relationship between η and rs. For densities below the transition point, an intermediate rs (≈0.5–1) minimizes η, indicating that limited but sufficient information can curb unnecessary relocations and improve space utilization. Conversely, for densities above the peak sustainable density, increasing rs consistently raises η; agents with broader knowledge move more often but cannot find vacant spots, producing a “information overload” that degrades efficiency.

Inequality is quantified via the dispersion of cumulative payoffs across agents. Across all parameter regimes, higher information (larger rs) reduces inequality, because agents are better able to locate low‑density regions and thus the payoff distribution flattens. Notably, the trends for inefficiency and inequality diverge: more information can simultaneously lower inequality while raising inefficiency in crowded regimes.

The authors validate their analytical predictions with extensive Monte‑Carlo simulations (up to 10⁵ steps, multiple initial conditions) and explore variations in (z, τ, r). They also discuss extensions to higher dimensions (e.g., square lattices with von Neumann neighborhoods) and note that the qualitative behavior—transition density, peak inefficiency, and the opposite effects of information—persists.

In summary, the study demonstrates that local crowd‑avoidance rules generate rich macroscopic phenomena: a sharp transition from fully coordinated to perpetually shifting states, a density‑dependent peak in space‑use inefficiency, and a decoupling of efficiency from payoff inequality. The counter‑intuitive finding that more information can worsen global efficiency under high density challenges conventional wisdom and has implications for the design of public‑space allocation, traffic management, and any system where agents compete for limited spatial resources. Future work could incorporate heterogeneous agents, dynamic movement costs, or adaptive tolerance thresholds to further bridge the model with real‑world complexities.