Voter model with non-Poissonian interevent intervals

Recent analysis of social communications among humans has revealed that the interval between interactions for a pair of individuals and for an individual often follows a long-tail distribution. We investigate the effect of such a non-Poissonian nature of human behavior on dynamics of opinion formation. We use a variant of the voter model and numerically compare the time to consensus of all the voters with different distributions of interevent intervals and different networks. Compared with the exponential distribution of interevent intervals (i.e., the standard voter model), the power-law distribution of interevent intervals slows down consensus on the ring. This is because of the memory effect; in the power-law case, the expected time until the next update event on a link is large if the link has not had an update event for a long time. On the complete graph, the consensus time in the power-law case is close to that in the exponential case. Regular graphs bridge these two results such that the slowing down of the consensus in the power-law case as compared to the exponential case is less pronounced as the degree increases.

💡 Research Summary

The paper investigates how non‑Poissonian timing of human interactions influences the dynamics of opinion formation in the classic voter model. Empirical studies have shown that the waiting times between successive communications—whether between a pair of individuals or for an individual’s activity—often follow heavy‑tailed distributions rather than the exponential distribution assumed in most theoretical models. To explore the consequences of this discrepancy, the authors modify the voter model by assigning each network edge an independent renewal process that governs when the edge becomes “active” and can transmit an opinion. Two families of inter‑event interval (IEI) distributions are examined: (i) the standard exponential distribution (λe^{‑λτ}), which yields a memoryless Poisson process, and (ii) a power‑law distribution P(τ) ∝ τ^{‑α} with exponent α>2, which produces long‑tailed waiting times and a pronounced memory effect—edges that have not fired for a long time are expected to wait even longer before the next update.

The authors keep the mean waiting time ⟨τ⟩ identical for both distributions so that any observed differences can be attributed solely to the shape of the tail and the associated memory. Simulations are performed on three canonical network topologies: a one‑dimensional ring (each node connected to its two nearest neighbors), a complete graph (all‑to‑all connectivity), and regular random graphs with fixed degree k (k‑regular). System size is varied (N≈10³–10⁴) and each configuration is averaged over thousands of independent runs to obtain reliable estimates of the consensus time T_c, defined as the first time at which all nodes share the same opinion.

Results reveal a striking dependence on network structure. On the ring, the power‑law IEI dramatically slows down consensus: T_c is typically two to three times larger than in the exponential case. The authors attribute this slowdown to a “bottleneck” effect. Because each edge must fire to propagate an opinion across the lattice, a single edge that experiences an unusually long waiting time can block the flow of information for an extended period, and the heavy tail makes such extreme delays relatively common. Consequently, the system spends a long time in a fragmented state where local domains of opposite opinions coexist.

In contrast, on the complete graph the difference between the two IEI families is negligible. Each node interacts with O(N) neighbors, so the inactivity of any particular edge is quickly compensated by many other active edges. The collective dynamics thus average out the memory effect, and the consensus time remains close to that of the standard voter model. This finding underscores that the impact of non‑Poissonian timing is not universal but is mediated by the degree of redundancy in the interaction pattern.

Regular graphs provide a bridge between the two extremes. As the degree k increases, the slowdown induced by the power‑law IEI diminishes smoothly. For low k (e.g., k=4), the behavior resembles the ring, with a noticeable increase in T_c. For higher k (e.g., k≥16), the consensus time converges to the complete‑graph value, indicating that sufficient connectivity dilutes the effect of long waiting times. The authors also explore the role of the power‑law exponent α: larger α values thin the tail, making the distribution approach exponential behavior, and consequently the consensus time decreases toward the Poisson baseline.

The discussion highlights several broader implications. First, in social systems where interactions are sparse and locally constrained (e.g., offline communities, geographically limited networks), heavy‑tailed inter‑event times can substantially delay collective agreement, potentially affecting the spread of norms, technologies, or misinformation. Second, in highly connected digital platforms, the same heavy‑tailed timing may have little practical effect on macroscopic opinion dynamics, justifying the continued use of Poisson approximations in many analytical studies. Finally, the work suggests that models of epidemic spreading, information diffusion, or coordinated action should incorporate realistic timing mechanisms when the underlying contact network is low‑dimensional or otherwise limited in redundancy.

Overall, the paper provides a clear, quantitative demonstration that the temporal statistics of human activity—specifically, the presence of long‑tailed, memoryful inter‑event intervals—can alter the speed of consensus formation in a manner that is highly sensitive to network topology. This insight opens avenues for future research on how to mitigate undesirable delays (e.g., in public health campaigns) or exploit them (e.g., to sustain diversity of opinion) by engineering the structure of interaction networks or by influencing the timing patterns of communication.