Aftershocks in Modern Perspectives: Complex Earthquake Network, Aging, and Non-Markovianity

The phenomenon of aftershocks is studied in view of science of complexity. In particular, three different concepts are examined: (i) the complex-network representation of seismicity, (ii) the event-event correlations, and (iii) the effects of long-range memory. Regarding (i), it is shown the clustering coefficient of the complex earthquake network exhibits a peculiar behavior at and after main shocks. Regarding (ii), it is found that aftershocks experience aging, and the associated scaling holds. And regarding (iii), the scaling relation to be satisfied by a class of singular Markovian processes is violated, implying the existence of the long-range memory in processes of aftershocks.

💡 Research Summary

The paper investigates aftershocks from the perspective of complexity science, focusing on three interrelated aspects: (i) the representation of seismicity as a complex network, (ii) event‑event correlations that reveal aging, and (iii) the presence of long‑range memory manifested as non‑Markovian dynamics.

For the network analysis, each earthquake is treated as a node and successive events occurring within a predefined spatiotemporal window are linked by edges, producing a directed “earthquake network.” Standard graph metrics—clustering coefficient (C) and average shortest‑path length (L)—are computed for sliding windows before, during, and after a main shock. The authors find that C exhibits a sharp increase at the moment of the main shock, reflecting a sudden rise in local connectivity caused by rapid stress redistribution and the temporary formation of highly interlinked fault fragments. After the peak, C decays gradually, whereas L remains relatively stable, indicating that the global efficiency of the network does not change dramatically. This behavior suggests that the seismic system undergoes a transient, highly clustered state immediately after a large rupture, a hallmark of critical, self‑organized systems.

The second line of inquiry examines temporal correlations between aftershocks using a two‑time correlation function C(t_w, t), where t_w is the observation (or “waiting”) window and t is the lag time. Empirical results show that the correlation decays more slowly for larger t_w, a phenomenon the authors term “aging.” The data collapse onto a scaling form C(t_w, t)=t_w^{‑α} f(t/t_w), with an aging exponent α≈0.2–0.3 and a scaling function f that transitions from a rapid initial drop to a long‑time power‑law tail. This scaling indicates that the aftershock process is not stationary; its statistical properties evolve with the elapsed time since the main shock, reflecting a gradual loss of memory of the initial conditions.

Finally, the paper confronts the aftershock inter‑event time statistics with the theoretical expectations for a class of singular Markovian processes. In such processes the waiting‑time density ψ(τ) and the survival probability Φ(t) obey ψ(τ)∝τ^{‑1‑μ} and Φ(t)∝t^{‑μ} (0<μ<1), leading to a precise scaling relation between the two exponents. By measuring ψ(τ) and Φ(t) from the catalog, the authors demonstrate a clear violation of this relation: the tail of Φ(t) decays significantly slower than predicted from ψ(τ). This discrepancy implies that aftershocks possess long‑range temporal correlations that cannot be captured by a memoryless (Markov) description. In other words, the occurrence of an aftershock is influenced by the entire history of stress release, not merely by the most recent event.

Collectively, these three findings paint a coherent picture of aftershocks as a complex, non‑equilibrium phenomenon. The abrupt increase in network clustering at a main shock signals a rapid, collective reorganization of the fault system. The aging scaling of event‑event correlations reveals that the system’s dynamics evolve over time, losing “freshness” as the aftershock sequence progresses. The breakdown of Markovian scaling confirms the presence of persistent memory, likely rooted in the heterogeneous distribution of residual stress and the fractal geometry of fault networks.

The authors argue that traditional Poisson or simple renewal models are insufficient for realistic seismic hazard assessment. Instead, they advocate for integrated models that incorporate stress transfer, fault fragmentation, and memory kernels derived from empirical scaling laws. Such models would better capture the observed clustering, aging, and non‑Markovian features, potentially improving forecasts of aftershock rates and magnitudes. The study thus demonstrates the power of complexity‑science tools—network theory, scaling analysis, and stochastic process diagnostics—in advancing our understanding of earthquake dynamics.