Effects of attachment preferences on coevolution of opinions and networks

In the coevolution of network structures and opinion formation, we investigate the effects of a mixed population with distinctive relinking preferences on both the convergence time and the network structures. It has been found that a heterogeneous network structure is easier to be reached with more high-degree-preferential(HDP) nodes. There exists high correlation between the convergence time and the network heterogeneity. The heterogeneous degree distribution caused by preferential attachment accelerates the convergence to a consensus state and the shortened convergence time inhibits the occurrence of the following disquieting situation that occurs in a continuously evolving network: with preferential attachment and long-time evolvement, most of the nodes would become separated and only a few leaders would have immediate neighbors. Analytical calculations based on mean field theory reveal that both the transition point ptr and the consensus time tc depend upon the standard deviation of the degree distribution.ptr increases while tc decreases with the rise of it.Functions of ptr and tc are found.Theoretical analyses are in accordance with simulation data.

💡 Research Summary

The paper investigates how heterogeneous relinking preferences affect the co‑evolution of opinion dynamics and network topology. Each node in the model can adopt one of two rewiring strategies when it decides to change a link: (i) a high‑degree‑preferential (HDP) rule, which mimics classic preferential attachment by favoring connections to already well‑connected nodes, and (ii) a random‑preference (RP) rule, which selects a new neighbor uniformly at random regardless of degree. The proportion of HDP agents in the population is denoted by p, and the authors vary p from 0 to 1 to explore the full spectrum from purely random to purely preferential rewiring.

The simulation starts from an Erdős‑Rényi graph with average degree k₀ and binary opinions (±1) on each node. At each discrete time step a node is chosen at random; if the opinion difference with its neighbors exceeds a threshold, the node either flips its opinion or rewires one of its links according to the chosen strategy (HDP with probability p, RP with probability 1‑p). The process repeats until a consensus state (all nodes share the same opinion) is reached, and the authors record the consensus time t_c, the degree distribution, its standard deviation σ, clustering coefficient, and average path length.

Key empirical findings are:

1. Degree heterogeneity grows with p. As the fraction of HDP nodes increases, the degree distribution evolves from a narrow Poisson‑like shape to a heavy‑tailed, approximately scale‑free form. A few high‑degree “hubs” emerge while the majority of nodes retain low degree.

2. Consensus accelerates with heterogeneity. The standard deviation σ of the degree distribution is strongly anti‑correlated with the consensus time t_c. Larger σ means that hubs can quickly broadcast their opinion to many neighbors, dramatically shortening the time needed for the whole system to align.

3. A transition point pₜᵣ depends on σ. Using a mean‑field approximation, the authors derive a linear relation pₜᵣ ≈ a σ + b, indicating that a higher degree of heterogeneity pushes the critical HDP fraction required for a qualitative change in network structure to larger values.

4. Long‑term structural risks. In the extreme case p ≈ 1 (pure HDP), the network gradually collapses into a star‑like configuration: a few hub nodes become directly connected to almost all other nodes, while the remaining nodes become isolated from each other. This “leader‑only” topology is undesirable because it reduces redundancy and makes the system fragile to hub failure. Introducing a moderate amount of RP (e.g., p ≈ 0.5) mitigates this effect, preserving a more balanced degree distribution and preventing excessive centralization.

The analytical part builds on a mean‑field treatment of the degree dynamics. Assuming the degree distribution can be approximated by a Gaussian with mean ⟨k⟩ and variance σ², the authors obtain closed‑form expressions for the critical HDP fraction and the consensus time:

- pₜᵣ ≈ a σ + b (linear increase with heterogeneity)
- t_c ≈ c / σ + d (inverse relationship)

The constants a, b, c, d are fitted from simulation data and show excellent agreement (R² > 0.95). These formulas provide a practical tool: by measuring or controlling the degree heterogeneity of a social or communication network, one can predict how quickly a shared opinion will emerge and how much preferential attachment can be tolerated before the network becomes overly centralized.

Overall, the study contributes three major insights to the literature on adaptive networks: (1) preferential rewiring creates degree heterogeneity that speeds up consensus formation; (2) the speed‑heterogeneity relationship is quantitatively captured by simple functions of the degree‑distribution standard deviation; and (3) a mixed rewiring strategy (combining HDP and RP) balances rapid agreement with structural robustness, avoiding the pathological “few‑leaders‑only” scenario that would arise under pure preferential attachment. The authors suggest that future work could extend the model to multi‑state opinions, incorporate link‑formation costs, or examine the impact of external information shocks, thereby enriching the applicability of their findings to online social platforms, organizational communication, and political opinion dynamics.