Opinion Dynamics with Set-Based Confidence: Convergence Criteria and Periodic Solutions

TL;DR

This paper extends the opinion dynamics model to include set-based confidence, allowing for non-convex and unbounded confidence sets, and analyzes convergence and periodic behaviors in this more general setting.

Contribution

Introduces the SCOD model with set-based confidence, generalizing the HK model to non-convex and unbounded confidence sets, and studies its convergence and oscillation properties.

Findings

Solutions can converge to non-equilibrium points or oscillate periodically.

Symmetric confidence sets with zero in interior ensure convergence to equilibrium.

Stubborn agents influence convergence and can induce oscillations.

Abstract

This paper introduces a new multidimensional extension of the Hegselmann-Krause (HK) opinion dynamics model, where opinion proximity is not determined by a norm or metric. Instead, each agent trusts opinions within the Minkowski sum $\xi+\mathcal{O}$, where $\xi$ is the agent's current opinion and $\mathcal{O}$ is the confidence set defining acceptable deviations. During each iteration, agents update their opinions by simultaneously averaging the trusted opinions. Unlike traditional HK systems, where $\mathcal{O}$ is a ball in some norm, our model allows the confidence set to be non-convex and even unbounded. We demonstrate that the new model, referred to as SCOD (Set-based Confidence Opinion Dynamics), can exhibit properties absent in the conventional HK model. Some solutions may converge to non-equilibrium points in the state space, while others oscillate periodically. These ``pathologies'' disappear if the set $\mathcal{O}$ is symmetric and contains zero in its interior: similar to the usual HK model, SCOD then converges in a finite number of iterations to one of the equilibrium points. The latter property is also preserved if one agent is "stubborn" and resists changing their opinion, yet still influences the others; however, two stubborn agents can lead to oscillations.