Quantifying the Role of 3D Fault Geometry Complexities on Slow and Fast Earthquakes J. Cheng Abstract \AtEveryCite Key Points • 3D simulations of two parallel faults show that interacting faults can generate slow slip events (SSEs), earthquakes, or complex sequences, while a single planar fault produces only earthquakes. • We quantify fault interaction strength to understand how geometrical parameters control the occurrence and proportion of slow slip events within slip sequences. • The observed moment-duration scaling for SSEs depends strongly on the slip rate threshold used to detect events, suggesting instrumental sensitivity influences real-world observations. CRediT Conceptualization: H. S. Bhat Methodology: H. S. Bhat, J. Cheng Software: J. Cheng, B. Lecampion, P. Dubernet Investigation: J. Cheng, H. S. Bhat Writing - original draft: J. Cheng, H. S. Bhat Writing - review & editing: H. S. Bhat, J. Cheng, B. Lecampion, M. Almakari Supervision: H. S. Bhat, B. Lecampion, M. Almakari Funding acquisition: H. S. Bhat Abstract Traditional models of slow slip events (SSEs) often oversimplify fault geometry, yet imaging studies show that real subduction faults are segmented and complex. We investigate how fault interactions influence slip behavior using 3D quasi-dynamic earthquake sequence simulations of two parallel faults with uniform rate-weakening friction, accelerated with hierarchical matrices. Our results identify four slip regimes—periodic earthquakes, coexisting SSEs and earthquakes, only SSEs, and complex sequences—while a single planar fault under the same conditions produces only earthquakes. We quantify fault interaction using the maximum Coulomb stress induced on a target fault by unit, spatially uniform stress drop on a neighboring fault. Because the source stress drop is normalized, the metric depends only on geometry and is independent of friction and nucleation length, and it can be extended to arbitrary fault configurations. The occurrence of SSEs is confined to an intermediate range of interaction strength. We also reproduce the observed moment-duration scaling and show that it depends on event detection thresholds. These results demonstrate that complex fault geometry can naturally generate both slow and fast earthquakes through evolving traction heterogeneities. Plain Language Summary Slow slip events (SSEs) are slow fault movements that release stress gradually, with smaller stress drops and little to no seismic wave radiation, but they still change the stress on the fault and can influence where and when earthquakes occur. Traditional models often assume simple fault geometry, but real faults are complex and can interact with each other. Using 3D simulations of two parallel faults, we show that their basic interactions can generate a wide range of slip behaviors, including periodic earthquakes, slow slip events, and mixed or complex earthquake sequences. A single flat fault under the same conditions produces only earthquakes, highlighting the importance of fault geometry and interaction. We also find that the transition between slip behaviors depends on how strongly the faults interact each other. These results suggest that the complex geometry of real faults can naturally produce both slow and fast earthquakes observed in subduction zones. 1 Introduction It has long been recognized that earthquake-related slip accounts for only a fraction of the overall slip budgets within plate tectonics. As continuous geodetic networks have improved, researchers have discovered slow slip events (SSEs) in various tectonic environments, for example in subduction zone: Cascadia subduction zone ( 22 ; 16 ) , in continental strike-slip fault systems e.g., Haiyuan fault ( 26 ) , San Andreas fault ( 54 ) . These events involve episodic, slow shear motion along faults (a few orders of magnitude faster than plate motion velocity) with no or minimal seismic activity. These events can range from small to large magnitudes, sometimes comparable to earthquakes of the same magnitude. Although SSEs generate little seismic radiation, large-magnitude events can still cause significant stress perturbations and accumulated slip on the fault, affecting its behavior. They are closely linked spatially and temporally with low-frequency earthquakes (LFEs), very low-frequency earthquakes (VLFEs), and tremors, exhibiting a lower frequency of seismic radiation compared to regular earthquakes of same magnitudes. Therefore, seismological instruments can indirectly detect slow slip events by tracking the migration of tremors ( 25 ; 59 ; 41 ) , repeating earthquakes ( 28 ; 65 ) , or LFEs ( 8 ; 19 ) , which improve detection capabilities for slow slip events. Slow slip events are ubiquitous in subduction zones and exhibit a diverse range of spatiotemporal complexities. Sometimes, they can be observed in shallow depths or below the seismogenic zone. In Nankai Trough, short-term SSEs are discovered with a duration spanning from days to weeks and 3-6 months recurrence time ( 45 ; 23 ) and long-term SSEs are observed in deep areas with around a 1-year duration and 6-year recurrence time ( 47 ) . Shallow VLFEs and tremor, megathrust earthquake, long-term SSEs, and short-term SSEs are observed from trough to the deep ( 46 ) . This pattern of depth-dependent SSEs is also seen in the Mexican subduction zone ( 18 ) . In Hikurangi, shallow SSEs are accompanied by microearthquakes, and deep SSEs are long-term with a duration of 2-3 months and a recurrence interval of 2 years with no tremors ( 68 ) . Slow slip events (SSEs) have a complex relationship with earthquakes in space and time. In the San Andreas fault, slow and fast rupture can coexist on the same section of the fault ( 60 ; 66 ) . Tremor signals were observed 18 months prior to the 2004 Mw 6.0 Parkfield earthquake. There are also examples of SSEs that can occur before earthquakes ( 9 ; 40 ) . The 1999 Mw7.6 Izmit earthquake was preceded by a 44-minute slow slip ( 8 ) . A slow slip event in Guerrero triggered the 2014 Mw 7.3 Papanoa earthquake ( 49 ) . In the Cascadia subduction zone, GPS observations suggest that the merging of slow slip event fronts, potentially leading to a major earthquake, maybe a possible mechanism for earthquake occurrence ( 7 ) . Earthquakes can also trigger SSEs. The 2016 Mw7.8 Kaikoura earthquake triggered a slow slip on the southern Hikurangi subduction zone ( 70 ) . The 2017 Chiapas earthquake in Mexico triggered a slow slip event on the southern San Andreas Fault, located 3000 km away from the earthquake’s epicenter ( 64 ) . SSEs can also occur periodically without earthquakes, like in Cascadia ( 52 ) and Hikurangi subduction zone ( 69 ) . The relationship between earthquakes and SSEs is still unclear and needs more studies and investigation. Those slow phenomena significantly influence fault behavior by altering the stress field and having intricate relationships with earthquakes ( 3 ; 9 ; 46 ) . Understanding slow slip events is crucial for gaining insights into earthquake mechanisms. There are several explanations for the mechanism of SSEs. SSEs can emerge from the transition of rate and state friction stability from velocity-weakening to velocity-strengthening ( 36 ; 55 ) . Heterogeneous frictional properties, such as varying the proportions of velocity-weakening and velocity-strengthening patches, can produce stable, slow, or dynamic slip events along the fault ( 62 ; 39 ; 44 ) . Moreover, fault width plays a role in limiting rupture nucleation and stabilizing the faults ( 37 ) . Mechanisms like dilatant strengthening ( 58 ; 38 ) or frictional restrengthening at high slip speeds ( 29 ; 61 ; 24 ) can modulate fault stabilization and instigate slow slip events. Additionally, thermal instabilities resulting from shear heating and temperature fluctuations can trigger SSEs ( 71 ) . Furthermore, the brittle-ductile (frictional and viscous deformation) transition ( 43 ; 2 ) , viscoelastic materials ( 72 ) and the presence of fluids ( 5 ; 15 ; 6 ; 20 ) are factors that also contribute to the occurrence of slow slip events. For a long time, the geometry of subduction interfaces was assumed to be simple, especially when studying slow slip events through other physical mechanisms such as heterogeneous friction or considering fluid as mentioned before. In reality, fault systems possess complex three-dimensional structures. For example, recent imaging of the Ecuadorian subduction zone by ( 11 ) reveals that earthquakes occur across multi-fault segments and subparallel planes, challenging this simplistic view. This complexity emphasizes the need to account for complex fault geometries when studying SSEs. Observations from Hikurangi, including drilling data and seismic reflection images, further highlight the role of material and geometric complexity in promoting slow slip events ( 4 ; 30 ) . In Cascadia, downdip variability of SSEs suggests a strong geometric influence ( 21 ; 42 ) . Advanced simulations, such as those performed by 35 for Cascadia and 48 for Guerrero, indicate that non-planar fault geometries are pivotal in understanding SSEs. Fundamental complexities in fault geometry are also considered in SSE modeling, including investigations into precursory slow slip influenced by fault roughness ( 10 ; 63 ) and the emergence of slow slip events from two parallel faults under spatial uniform rate-weakening friction in 2D models ( 53 ) . Recent work by 1 have highlighted the role of fault zone architecture in producing the complete slip spectrum. Laboratory experiments by ( 31 ) further underscore the complexity, showing both fast and slow ruptures with varying fault slip on rough faults. The effect of stress interaction between multiple faults on slow slip events in 3D models is still unresolved. In this study, we investigate a step-over configuration of two parallel faults under spatially uniform, rate-weakening frictional conditions to explore the interplay between geometrical and frictional parameters in three dimensions. Three-dimensional simulations of fault slip are computationally intensive, and modeling slow slip events (SSEs) is even more demanding because weakly rate-weakening friction produces large nucleation sizes and requires high temporal resolution. To make such simulations feasible, we use a 3D quasi-dynamic earthquake sequence model based on the boundary element method accelerated with hierarchical matrices ( 13 ; 34 ) . Our analysis identifies four distinct slip regimes: (1) only SSEs, (2) SSE-dominant behavior, (3) earthquake-dominant behavior, and (4) only earthquakes. Under identical parameters, a single planar fault produces only earthquakes (Figure. S1), highlighting the critical role of fault geometry in governing slip behavior. We introduce a fault interaction metric, inspired by fracture mechanics, based on stress perturbations, fault width, length, overlap distance, and spacing, and use it to link geometric complexity with slip behavior through the ratio of SSE to total moment release. Moreover, our simulations reproduce the observed moment-duration scaling across different step-over geometries. The complex spatiotemporal slip patterns arise naturally from evolving traction heterogeneities, or “traction asperities” generated by fault interactions. 2 Method We used the recently developed numerical approach, FASTDASH which integrates a 3D quasi-dynamic earthquake cycle modelling using boundary element methods accelerated by hierarchical matrices ( 13 ) . This technique significantly optimizes computational efficiency, reducing complexity from O(N^{2}) to O(N\log{N}) , where N represents the number of discretized fault elements. Such efficiency is crucial for solving complex 3D fault systems effectively. We analyse a fault system with two overlapping faults subjected to a far-field constant stress rate loading. Both faults are governed by laboratory-derived rate and state friction (RSF) law with aging state evolution. Friction is spatial uniform rate weakening. This model includes radiation damping to approximate inertial effect and neglects any wave propagation effect ( 50 ) . Both shear and normal traction can vary with slip due to elastic interaction between two faults. With this approach, we can calculate key information on faults including maximum slip rate, moment rate, duration, and the distributions of stress and slip during multiple earthquake cycles. Linear stability analysis of rate and state friction brings out two important length scales namely the process zone size, L_{b} , and the nucleation length, L_{nuc} . L_{b} represents the region where the strength breakdown occurs, and where traction and slip change rapidly. Numerical methods require a sufficient number of grid points within this zone to accurately capture these variations. In this study, we use a grid size of \Delta s=L_{b}/3 . On the other hand, L_{nuc} is the critical length necessary for slip instability to occur under idealized conditions. For faults that governed by rate and state friction with ageing law, these parameters are defined as follows: ( 32 ; 56 ; 67 ) , \displaystyle L_{b} \displaystyle=\frac{\mu D_{c}}{b\sigma_{n}} (1) \displaystyle L_{\text{nuc}} \displaystyle=\begin{cases}2.7548\,L_{b},&0\leq\dfrac{a}{b}\leq 0.3781,\\[4.0pt] \dfrac{2L_{b}}{\pi\,(1-a/b)^{2}},&\dfrac{a}{b}\to 1~.\end{cases} (2) where \mu is shear modulus, D_{c} is the rate and state characteristic slip distance, a, b are the rate and state friction parameters to represent direct effect and evolution effect. For rate weakening friction, 01.2 ), faults predominantly host earthquakes. The low \Lambda values result in periodic earthquakes (Figure S2), while high \Lambda values lead to complex earthquakes (Figure S3). At the left boundary, \Lambda\approx 0.03 , the primary controlling factor is the distance between two faults ( D ). Since \Lambda is roughly proportional to 1/{\sqrt{D}} , the SSE ratio drops rapidly as \Lambda approaches this boundary. At the right boundary, \Lambda\approx 1.2 , fault width, W , is the main controlling factor. With \Lambda being roughly proportional to \sqrt{W} , the SSE ratio decreases gradually as it nears this boundary. However, when \Lambda falls within an intermediate range ( 0.03<\Lambda<1.2 ), both earthquakes and SSEs can coexist. Within a more specific range ( 0.05<\Lambda<0.75 ), SSEs dominate, shown in Fig 3(b). This is because SSEs are highly sensitive to minor stress perturbations, as noted by 46 . A proper range of interaction is essential to favor slow slip events. If the interaction is too weak, the stress perturbation is insufficient to generate SSEs. Conversely, if the interaction is too strong, SSEs cannot be sustained and will grow into earthquakes. We also tested different friction parameters (see in Fig 3b with different colors). For high values of a/b , a slightly wider range of geometrical parameters needs to be considered to identify SSE-dominated regimes. Despite these variations, geometry remains the primary controlling factor. The SSE ratio \chi_{sse} can potentially be inferred from these geometrical parameters. This framework provides a mechanical interpretation of how different types of slip events are distributed based on fault geometry. For a given \Lambda , various combinations of geometrical parameters can yield the same value. Figure 3(c) illustrates \Lambda as a function of the overlap and the distance between two faults, expressed as a fraction of the fault length. The value of \Lambda decreases rapidly with increasing distance between the faults. \Lambda increases with greater overlap, but when the distance between the two faults is larger, the effect of overlap becomes less significant. Figure 3(c) demonstrates these possible combinations. Most of our simulation results, which feature the coexistence of SSEs and EQs, cover possible combinations where 0.03<\Lambda<1.2 . When the geometrical configuration approaches the lower and upper \Lambda threshold boundaries ( \Lambda=0.03,\Lambda=1.2 ), the SSE ratio gradually decreases and transitions to only earthquakes. This provides a comprehensive view of how different geometrical configurations influence fault interactions and the coexistence of SSEs and EQs. Additionally, it helps us select appropriate geometrical parameters based on the SSE ratio. This finding doesn’t change with different friction properties, see the circles with different color in Fig. 3(b). This study highlights the critical role of geometrical parameters in understanding and predicting complex slip behavior emerged from fault interaction. Future research should focus on refining these metrics and exploring additional realistic geometrical parameters to further enhance our understanding of the interplay between slow and fast earthquakes in complex fault systems. 51 analyzed normal fault systems by varying along-strike and across-strike spacing separately, focusing on rupture periodicity, synchronicity, and event statistics. In contrast, our study emphasizes the conditions that generate slow slip events. While their analysis treated spacing and overlap distance independently, we introduce a unified fault interaction metric that systematically quantifies how geometric parameters influence both slow and fast slip behaviors. This metric captures the mutual stress interactions between faults, showing that faults “see” each other not merely through geometric distance, but through the stress fields they impose on one another. 4.2 Traction asperities Notably, our study is based on faults with spatially uniform rate-weakening friction conditions. Previous studies ( 33 ; 66 ) used the single planar fault and demonstrated that slow slip events emerge in small asperities with rate-weakening friction and earthquakes occur in large asperities with rate-weakening friction, within the context of friction asperities. Those friction asperities are spatially and temporally stable. Here, we introduce the concept of traction asperities, which naturally arise from multiple fault systems due to stress interactions and previous events. These asperities can also determine the nucleation and arrest of both slow and fast earthquakes. Figure 4 shows the distribution of the shear-to-normal traction ratio on Fault 2 before and after selected events (E–H) in Model 2 (Figure 1). Owing to stress interactions with neighboring faults and previous ruptures, the traction field exhibits spatial and temporal heterogeneity. For example, the left edge, which overlaps with Fault 1, develops a low traction-ratio patch as a result of fault interaction. Nucleation occurs at regions with a high ratio of shear-to-normal traction asperity (HRA). When the HRA is narrow, nucleation cannot complete within a single asperity, potentially leading to aseismic events (e.g., events F and G). Conversely, regions with a low ratio of shear-to-normal traction (LRA) possess higher strength and act as barriers to arrest events nucleated from the HRA. For example, after nucleation, event E prefers to propagate towards the right HRA and arrests at the left LRA, resulting in a partial rupture. Same for event F and G, rupture arrests at LRA. The heterogeneity of traction asperities complicates nucleation processes. The coalescence of slip fronts and relatively homogeneous traction fields may lead to full rupture earthquakes, as observed in event H. Figure 4 : Traction field evolution during four events (E, F, G, and H) on fault 2 in Model 2. The colormap indicates the ratio of shear to normal traction. The first column displays the traction field before the nucleation of each event, with red stars marking earthquake hypocenters and blue stars indicating SSE hypocenters. The second column shows the traction field after each event, with black contour lines representing the slip distribution. The duration of each event and the inter-event times are noted. 4.3 Comparison with single planar fault with heterogeneous friction properties In this work, we focus on a fault system with two parallel faults and examine the stress interaction between them, leading to a spatio-temporal complex stress field that causes slow slip events and earthquakes with partial and full ruptures. This complexity arises from the dynamic interaction of the faults and their rupture history, rather than from fixed fault parameters. In contrast, models focusing on friction heterogeneity, such as those by 17 and 27 , suggest that variations in frictional properties can significantly influence seismic behavior. However, frictional heterogeneity is static and once set and does not evolve over time and space, requiring parameter adjustments, such as changes in the sizes, densities, and distances of asperities, to match the observed complexity. Our model demonstrates that simple geometric configurations, like two parallel faults, can naturally produce a wide range of moment-duration laws and complex nucleation patterns, closely mimicking natural seismic and aseismic slip events. Geometric features can be more readily measured using methods like surface rupture tracing and seismic imaging, offering a practical advantage over frictional measurements. Thus, while both geometric complexity and frictional heterogeneity contribute to understanding seismic events, geometric complexity provides a more robust and dynamically evolving framework that can be more readily linked to observations. Future research should integrate these mechanisms to refine predictive models and enhance our understanding of complex slip events. 5 Conclusion In this study, we investigate the emergence of complex slow and fast seismic events driven by elastic interactions and heterogeneous stress fields in a three-dimensional fault system composed of two parallel planar faults. Using a 3D quasi-dynamic earthquake cycle model based on the boundary element method accelerated by hierarchical matrices ( 13 ) , we explore how fault geometry and frictional properties control slip dynamics. For a single isolated fault under spatially uniform rate-weakening friction, only regular earthquakes occur when the fault length exceeds the nucleation length. However, when fault interactions are included, stress coupling between neighboring faults generates spatiotemporally complex slip behaviors, including slow slip events (SSEs) and partial or full earthquake ruptures. We quantify geometric complexity using a fault interaction metric defined as the maximum Coulomb stress change induced on a target fault by unit, spatially uniform stress drop on a neighboring fault. Slip complexity is characterized using the SSE moment release ratio, defined as the fraction of total seismic moment released during slow slip events. SSEs occur only within an intermediate range of interaction strength—too weak or too strong coupling promotes earthquake-dominated behavior—and this pattern remains robust across frictional parameters. Our results also reproduce the moment-duration scaling of SSEs, showing a predominantly linear trend that depends strongly on the slip rate threshold used to identify events. These findings highlight how 3D fault interactions can naturally generate the coexistence of earthquakes and SSEs, bridging the gap between seismic and aseismic slip behaviors. Conflict of Interest The authors declare no conflicts of interest relevant to this study. Open Research Section The datasets generated and analyzed during this study, and codes used to analyze, are available at 12 . Appendix A The definition of fault interaction metric The Muskhelisvili-Kolosov complex potentials for a shear crack of length 2a are given by ( 57 ) \displaystyle\Phi^{\prime}(z) \displaystyle=\dfrac{1}{2i}\left[\dfrac{z}{\left(z^{2}-a^{2}\right)^{1/2}}-1\right] (A.1) \displaystyle\Phi^{\prime\prime}(z) \displaystyle=\dfrac{1}{2i}\left[\dfrac{-a^{2}}{\left(z^{2}-a^{2}\right)^{3/2}}\right] (A.2) \displaystyle\Psi^{\prime}(z) \displaystyle=-2\Phi^{\prime}(z)-z\Phi^{\prime\prime}(z) (A.3) The stress field is then given by \sigma_{ij}(z)=\sigma_{ij}^{0}+\Delta\sigma_{ij}(z) where \displaystyle\Delta\sigma_{22}(z) \displaystyle=2y\Im{\left[\Phi^{\prime\prime}(z)\right]} (A.4) \displaystyle\Delta\sigma_{12}(z) \displaystyle=-2\Im{\left[\Phi^{\prime}(z)\right]}-2y\Re{\left[\Phi^{\prime\prime}(z)\right]} (A.5) where \Re and \Im correspond to the real and imaginary parts of their arguments respectively. See figure S5. Length 2a is the minimum value of fault length L_{f} and fault width W . The complex coordinate of fault 2 is denoted as z . Fault 1 spans from (-L_{f}/2,0) to (L_{f}/2,0) . Fault 2, on the other hand, extends from (x_{1},D) to (x_{2},D) , where x_{1}=(2L-1)\cdot(L_{f}/2) , x_{2}=x_{1}+L_{f} and L is the overlap. We define the metric \Lambda as the maximum of \Delta\sigma_{12}+f_{s}\Delta\sigma_{22} on fault 2 (figure S6). This metric, which is a function of all geometrical parameters, quantifies the strength of the stress interaction between the two faults. \displaystyle\Lambda \displaystyle=\textrm{max}\left\{-2\Im{\left[\Phi^{\prime}(z)\right]}-2y\Re{\left[\Phi^{\prime\prime}(z)\right]}+2f_{s}y\Im{\left[\Phi^{\prime\prime}(z)\right]}\right\} (A.6) where f_{s}=0.6 is assumed. Supplementary Information for Quantifying the Role of 3D Fault Geometry Complexities on Slow and Fast Earthquakes J. Cheng 1,∗,¶ , H. S. Bhat 1 , M. Almakari 1 , B. Lecampion 2 , P. Dubernet 1 1. Laboratoire de Géologie, Ecole Normale Superieure, CNRS-UMR 8538, PSL Research University, Paris, France 2. Geo-Energy Lab - Gaznat Chair on Geo-Energy, Swiss Federal Institute of Technology in Lausanne, EPFL-ENAC-IIC-GEL, Station 18, Lausanne CH-1015, Switzerland * Currently at Division of Geological and Planetary Sciences, California Institute of Technology. \lx@paragraphsign Corresponding author: jcheng95@caltech.edu Introduction This Supplementary Information provides additional numerical results and simulation parameters supporting the main manuscript. The document is organized as follows: Supplementary Table: Table S1 lists the full set of 65 simulations performed, parametrized by a/b , W/L_{\mathrm{nuc}} , D/L_{\mathrm{nuc}} , and L/L_{f} . Supplementary Figures: Figure S1 shows that a single-fault model produces strictly periodic earthquakes ( a/b=0.8 , W/L_{\mathrm{nuc}}=2 ), with a recurrence interval of 3.73 years, serving as the reference baseline for the two-fault stepover simulations. Figure S2 illustrates the periodic earthquake regime ( \lambda=0.0183 ) for both faults in a stepover configuration with a/b=0.8 , W/L_{\mathrm{nuc}}=2 , D/L_{\mathrm{nuc}}=5 , and L/L_{f}=0.5 , with time normalized by the single-fault recurrence interval. Figure S3 illustrates the complex earthquake regime ( \lambda=1.9341 ) for a/b=0.4 , W/L_{\mathrm{nuc}}=8 , D/L_{\mathrm{nuc}}=0.1 , and L/L_{f}=0.5 , demonstrating how a wider fault with smaller stepover distance produces irregular recurrence on both fault segments. Figure S4 shows the slip-rate evolution on the fault plane during the coalescence of two slow-slip events, illustrating the spatial merging process. Figure S5 presents the shear ( \sigma_{12} ) and normal ( \sigma_{22} ) stress fields produced by a 2D crack subjected to a unit stress drop, used to validate the stress transfer kernel. Figure S6 shows the resulting Coulomb stress field \sigma_{12}+\mu\sigma_{22} for a friction coefficient \mu=0.6 , highlighting the stress loading and shadowing regions around the fault tips. Table S1 : Simulation parameters. ID a/b W/L_{\mathrm{nuc}} D/L_{\mathrm{nuc}} L/L_{f} 1 0.4 1.5 0.1 0.5 2 0.4 2.5 0.1 0.0 3 0.4 2.5 0.1 0.5 4 0.4 2.5 0.1 1.0 5 0.4 2.0 0.1 0.0 6 0.4 2.0 0.1 0.5 7 0.4 2.0 0.1 1.0 8 0.4 2.0 0.2 0.5 9 0.4 2.0 0.5 0.5 10 0.4 2.0 1.0 0.5 11 0.4 3.0 0.1 0.5 12 0.4 4.0 0.1 0.5 13 0.4 4.0 0.5 0.2 14 0.4 4.0 5.0 0.5 15 0.4 6.0 0.1 0.5 16 0.4 6.0 0.1 0.2 17 0.4 6.0 0.1 1.0 18 0.4 8.0 0.1 0.5 19 0.6 1.5 0.1 0.5 20 0.6 1.8 1.5 0.8 21 0.6 1.8 2.0 0.8 22 0.6 2.0 0.1 0.0 23 0.6 2.0 0.1 0.5 24 0.6 2.0 0.1 1.0 25 0.6 2.0 0.2 0.5 26 0.6 2.0 0.5 0.5 27 0.6 2.0 1.5 0.0 28 0.6 2.0 1.5 0.2 29 0.6 2.0 1.5 1.0 30 0.6 2.0 1.0 0.2 31 0.6 2.0 1.0 0.5 32 0.6 2.0 1.0 0.8 33 0.6 3.0 0.1 0.0 34 0.6 3.0 0.1 0.5 35 0.6 3.0 0.1 0.8 36 0.6 3.0 0.1 1.0 37 0.6 4.0 0.1 0.5 38 0.6 4.0 0.5 0.2 39 0.6 4.0 5.0 0.5 40 0.6 4.0 5.0 0.8 41 0.6 4.0 5.0 0.2 42 0.6 6.0 0.1 0.5 43 0.6 8.0 0.1 0.5 44 0.8 1.5 0.1 0.5 45 0.8 1.8 0.1 0.5 46 0.8 1.8 0.1 0.8 47 0.8 1.8 0.1 1.0 48 0.8 1.8 1.5 0.5 49 0.8 1.8 2.0 0.5 50 0.8 2.5 0.1 0.0 51 0.8 2.5 0.1 0.5 52 0.8 2.5 0.1 1.0 53 0.8 2.75 0.1 0.5 54 0.8 2.0 5.0 0.5 55 0.8 2.0 0.07 0.5 56 0.8 2.0 0.1 0.5 57 0.8 2.0 0.2 0.5 58 0.8 2.0 0.5 0.5 59 0.8 2.0 1.0 0.5 60 0.8 3.0 0.1 0.5 61 0.8 4.0 0.1 0.5 62 0.8 4.0 0.5 0.2 63 0.8 4.0 5.0 0.5 64 0.8 6.0 0.1 0.5 65 0.8 8.0 0.1 0.5 Figure S1 : A single-fault model exhibits periodic earthquakes ( a/b=0.8 , W/L_{\mathrm{nuc}}=2 ) with a recurrence interval of 3.73 years. Figure S2 : Periodic earthquake regime with \lambda=0.0183 for the simulation with a/b=0.8 , W/L_{\mathrm{nuc}}=2 , D/L_{\mathrm{nuc}}=5 , and L/L_{f}=0.5 . Figure S3 : Complex earthquake regime with \lambda=1.9341 for the simulation with a/b=0.4 , W/L_{\mathrm{nuc}}=8 , D/L_{\mathrm{nuc}}=0.1 , and L/L_{f}=0.5 . Figure S4 : Slip-rate evolution on the fault plane illustrating the coalescence of two slow-slip events. Figure S5 : Shear ( \sigma_{12} ) and normal ( \sigma_{22} ) stress fields produced by a 2D crack subjected to a unit stress drop. Figure S6 : Coulomb stress field \sigma_{12}+\mu\sigma_{22} assuming a friction coefficient of \mu=0.6 .