Luca Dal Zilio

Seismic hazard assessments rely on scaling laws that convert mapped fault length into the magnitude a fault can produce. Yet earthquakes of similar rupture length differ substantially in magnitude. We compiled a global catalog of magnitude 6+ strike-slip earthquakes and found that slowly loaded faults host systematically larger earthquakes than equally long ruptures on fast plate-boundary faults. Physics-based earthquake-cycle simulations explain this offset. Longer interseismic intervals allow greater healing, and colder lithosphere sustains a thicker seismogenic layer, increasing stress drop, slip, and downdip rupture area. For equivalent rupture lengths, slowly deforming Mongolia yields seismic moments up to six times larger than the San Andreas Fault. Loading rate is a measurable control on maximum earthquake size that belongs in hazard-assessment scaling laws.

Improving our understanding of induced and natural earthquakes benefits from controlled experiments in insitu laboratories. To investigate the processes during an Mw∼0 earthquake, we performed the “ Mzero ” experiments in a densely instrumented testbed of the Bedretto Underground Laboratory. These multi‐day hydraulic stimulation experiments went the opposite direction to typical induced seismicity research in that they were designed to enhance seismic rupture. In the first experiment, MzeroA, the rock mass was preconditioned by injecting water for 4 days at a pressure just below fracture reactivation pressure followed by a hydraulic stimulation above reactivation pressure. This strategy aimed at increasing the fractures area near critical stress conditions for shear failure, thus facilitating larger ruptures. During MzeroA, an event with Mw–0.54 was induced, followed by a distinct aftershock sequence. This event was one magnitude unit larger than any preceding event. In the second experiment, MzeroB, stimulation was initiated directly, without preconditioning. Compared to MzeroA, MzeroB exhibited higher seismicity rates, a larger seismicity cloud, and a pronounced migrating seismicity front. Combining seismological and hydromechanical observations, we discuss mechanisms that may have influenced the contrasting seismicity responses. In addition to fluid preconditiong, stress transfer from the Mw–0.54 mainshock, and natural seasonal pressure changes, may have contributed to modulating the seismogenic response. However, the relative importance of these processes remains uncertain given experimental limitations. Our results highlight both the potential and challenge of designing hydraulic stimulations to enhance or suppress seismic rupture, with implications for earthquake physics and induced seismicity.

Back‐propagating earthquake rupture (BPR), where slip migrates opposite to the main rupture direction, is observed in several earthquakes but its origin remains poorly understood, particularly the role of fluids. We use dynamic simulations of frictional sliding coupled with fluids to explore the mechanisms of BPR. Our simulations show that BPR occurrence and rupture mode are controlled by the ratio of initial shear stress to effective normal stress : low produces pulse‐like rupture with BPR, whereas high promotes crack‐like rupture without BPR. Fluids modulate this ratio by reducing . Localized patches of differential pore pressure act as asperities or barriers, inducing slip rate and shear stress perturbations that generate a secondary backward‐propagating front. BPR emerges from incomplete stress release, temporal healing and stress perturbations. These findings highlight the role of fluids in rupture dynamics and provide an energy framework for BPR in earthquakes.

SUMMARY: Earthquake fault slip arises from nonlinear coupling among frictional evolution, elastic loading and pore-pressure changes. When pore pressure evolves dynamically, the resulting hydromechanical rate-and-state models can be stiff and strongly coupled, making parameter inversion computationally demanding. Here we develop a physics-informed neural network (PINN) solver for a coupled spring-slider system that combines rate-and-state friction with pore-pressure/porosity evolution. The network approximates the time-dependent state variables and is trained by enforcing the governing differential equations together with initial conditions and, for inverse problems, observational constraints. To improve training stability, we employ adaptive inverse-residual weighting and a two-stage optimization schedule, in which Adam is followed by limited-memory Broyden–Fletcher–Goldfarb–Shanno (L-BFGS) optimization. In forward simulations, PINN predictions closely match a Runge–Kutta reference solution across steady sliding and slow-slip transients, with normalized mean squared error below 0.08 and Pearson correlation coefficient above 0.975 for block velocity and frictional shear stress in the cases tested. In inverse experiments, the framework recovers the applied normal stress from noisy shear-stress observations; uncertainty increases with noise amplitude, but the ensemble mean remains stable, and at the highest noise level considered ($q = 1$) the inferred normal stress deviates by less than $\sim$1 per cent from the reference value. These results suggest that PINNs provide a differentiable alternative for forward modelling and parameter inversion in coupled hydromechanical rate-and-state fault models.

A key challenge in induced seismicity is that fluid injection perturbs stress and pore pressure on faults with heterogeneous properties, leading to complex earthquake nucleation, migration and magnitude. The BedrettoLab (Bedretto Underground Laboratory for Geosciences and Geoenergies), located in the Swiss Alps, provides a unique natural testbed to study how these coupled hydro-mechanical processes interact with fault heterogeneity under controlled injection conditions, with direct access to well-characterized and densely instrumented fault zones. Previous characterization of the target MC fault zone at BedrettoLab show that layers of frictional velocity-strengthening (VS) fault gouge are embedded within velocity-weakening (VW) granitic bare rock, forming a strongly heterogeneous frictional architecture. However, how this frictional partitioning controls fault slip behavior and the magnitude of induced seismicity remains unclear. In this study, we use the newly developed 3D hydro-mechanical model HydroMech3D to explore the interplay between frictional heterogeneity and seismicity in fluid injection simulations, governed by rate- and state-dependent friction. We simulate injection scenarios using parameters and conditions derived from the ongoing FEAR (Fault Activation and Earthquake Rupture) experiments at Bedretto. Our simulations investigate how the spatial distribution of VS and VW patches control seismicity magnitude. By systematically changing the partition of VS and VW patches, we explore its influence on event size distributions and maximum magnitudes. Further simulations are conducted under various hydro-mechanical pre-conditioning conditions, by pre-determining the pressurized patch on the fault via the injection protocol prior to the main injection. These simulations allow us to understand whether fault pre-conditioning may influence the maximum magnitude of induced seismicity. Our results emphasize the critical role of frictional heterogeneity and injection strategy in fault dynamics, providing new insights into the hydro-mechanical behavior of complex fault zones during fluid injection and improving seismic risk assessment and mitigation strategies.

Producing a probabilistic tsunami hazard assessment (PTHA) at an inundation level for a large-scale region is computationally demanding. The main reasons are the vast number of scenarios and the high spatial resolution in coastal areas required. Various methodologies have been proposed to overcome these challenges. However, they are limited to local scales or specific sites. In addition, the scarcity of high quality and accuracy of digital elevation model (DEM) that create this task is even more expensive. Here, we applied a two-stage framework to perform a large-scale inundation PTHA and utilised open sources DEM from BATNAS [1] and DeltaDTM [2]. As a pilot study area, we focused on the southern coast of Java Island, Indonesia, covering over 1,000 km length of coastal area. This region has high potential tsunami from the Java megathrust earthquake with high density population at several locations.At the first stage, we focused on simulating offshore tsunami propagation in a low-resolution configuration model using the JAGURS code [3]. Further, tsunami elevation timeseries at 10 m isobath were extracted and used as boundary conditions for high-resolution inundation modelling at the second stage utilising the SFINCS code [4]. We generated a synthetic earthquake event catalogue by adopting a space-time Epidemic-Type Aftershock Sequence (ETAS) model [5] and coupled it with heterogenous earthquake slip models [6].This is a proposed modular framework where we could strategically adjust the configuration as needed to suit a range of risk-based applications and the facilities availability. For example, users might apply other hydrodynamic software for the simulations, consider different tsunamigenic sources, and refine the Stage 2 results by incorporating a better quality of DEM without redo the whole processes. Finally, it enables us to progressively develop a national, regional, or even at a global level in parallel processes. References: [1] BATNAS: https://tanahair.indonesia.go.id/portal-web/; [2] DeltaDTM: https://doi.org/10.4121/21997565; [3] JAGURS: https://github.com/jagurs-admin/jagurs/; [4] SFINCS: https://github.com/Deltares/SFINCS/tree/v2.1.1_Dollerup_release/; [5] Petrillo, G., & Zhuang, J. (2024). Bayesian earthquake forecasting approach based on the Epidemic Type Aftershock Sequence model. Earth, Planets and Space, 76(1), 78; [6] RPTHA to generate random slip models: https://github.com/GeoscienceAustralia/ptha

Fluid injection associated with geoenergy applications such as geothermal energy, CO2 sequestration, and hydraulic fracturing can alter fault stability through a combination of coupled hydro-mechanical processes. Laboratory experiments and underground observatories have provided valuable constraints on fault friction and near-fault pressure evolution, yet translating these observations to field-scale behavior requires physics-based numerical models that can resolve fault slip under realistic geometrical and mechanical conditions.A major limitation of existing modeling approaches is the high computational cost of fully coupled three-dimensional simulations. As a result, many studies rely on one-dimensional fault representations or simplified elastic and hydraulic coupling. While such models have provided important insights into key physical mechanisms, they are not well suited to support the design, interpretation, and long-term forecasting of modern injection experiments equipped with dense monitoring systems. These experimental settings increasingly demand three-dimensional models capable of capturing realistic fault geometry, spatially variable frictional and hydraulic properties, and stress interactions beyond reduced-dimensional assumptions.Here we present HydroMech3D, a physics-based numerical framework designed to efficiently simulate fluid-driven fault slip over earthquake-cycle timescales in three dimensions. The model employs a quasi-dynamic Boundary Element Method, discretizing only the fault surface embedded in elastic medium, thereby avoiding volumetric meshing. Fault slip is governed by rate-and-state friction and coupled to pore-pressure diffusion along the fault. Computational efficiency is achieved through a C++ implementation accelerated by hierarchical matrix from the Bigwham Library, enabling large-scale simulations with realistic fault geometry.This framework allows systematic investigation of fault-scale heterogeneity, including asperities with contrasting frictional and hydraulic properties, and provides a platform to explore how three-dimensional fault structure influences aseismic slip, stress transfer, and earthquake nucleation during fluid injection. Benchmarking against established earthquake-cycle test cases validates the mechanical solver and establishes a baseline for ongoing fully coupled simulations. HydroMech3D offers a computationally efficient open-source tool to support experiment design, interpretation of near-fault observations, and assessment of induced seismicity in geoenergy applications.

The largest strike-slip earthquakes ever recorded (M > 8) occurred in Mongolia in the 20th century, far from any plate boundaries. Rupture length-magnitude data indicate that the magnitude of these intraplate Mongolian events is, on average, ~0.5 magnitude larger than that of typical interplate earthquakes. The physical mechanisms that allow for such extra-large events remain mostly unresolved, largely due to the long return time of such events, hence the limited observational data. To address this, we employed a dual approach—numerical simulations with the PyQuake3D boundary element code (Tang et al., 2025) and theoretical analyses using the Rate and State Friction (Aging) Law on the spring slider—focusing on the role of slow plate rates. Our findings show that lower plate rates result in higher slip and greater stress drop, driven by enhanced fault restrengthening (healing). This healing, quantified by the state parameter, increases linearly with the inverse plate rate, in agreement with both analytical spring-slider models and 3D simulations. Critically, however, the observed GNSS plate rates of 1–3 mm/yr are insufficient to account for the ~0.5-unit magnitude excess relative to typical interplate earthquakes. We rigorously examine two scientific hypotheses: First, plate rates may be at residual levels (