Earthquake rupture dynamics from Graph Neural Networks

Dunyu Liu, & Thorsten Becker

In Preparation May 12, 2025, SCEC Contribution #14197

Large earthquakes control seismic hazard and arise from tectonic loading of a complex fault system consisting of heterogeneous material parameters, geometry, and stress state. All of those are subject to uncertainties, and their interactions and sensitivities for the nonlinear dynamic rupture problem are incompletely understood. Here, we apply Graph Neural Networks (GNNs) to approximate rupture dynamics learned from more computationally intensive, high-fidelity computations. We built a new GNN-based Simulator (GNS) capable of predicting earthquake rupture dynamics. Given only a minimum input – the hypocenter location– our GNS can simulate rate weakening friction rupture dynamics in 2-D, from nucleation to propagation and termination. The GNS can generalize to unseen hypocenter locations, fault geometries of different size, and pre-stress state. The GNS achieves significant speedup, rolling out predictions in seconds on a single GPU, compared to minutes for a highly parallelized high fidelity software running on 8 CPU cores. Our GNNs allows for more efficient estimates of the mapping from pre-earthquake stress state, as might be inferred from geodetic locking inversions to expected rupture dynamics. More broadly, the GNS may enable new kinds of parameter space exploration, provide surrogates for Bayesian model inference, uncertainty quantification, and optimal experimental design. Moreover, the GNS provides new perspectives to explore the physics of rupture dynamics.

Citation
Liu, D., & Becker, T. (2025). Earthquake rupture dynamics from Graph Neural Networks. ESS Open Archive, (in preparation).