RSQSim is an earthquake cycle simulator that employs the rate-and-state friction (RSF) laws first introduced by Dieterich (1979) and later works by Ruina (1983), Tullis (1988) or Marone (1998) to model long-term earthquake catalogues in predefined boundary-element fault geometries. In the RSF constitutive law, 1τfrict=σμ0+aln⁡VV0+bln⁡θVDc, τfrict is the shear stress-resisting motion, σ is the normal stress, μ0 is the steady state friction coefficient, V is the slip speed, V0 is the reference slip speed, θ is the state variable, Dc is the characteristic slip distance, and a and b are the direct- and evolution effect parameters of the RSF law, respectively.

In RSF, the relationship between a and b coefficients (a-b) determines the fault slip behavior. (a-b)<0 implies velocity weakening behavior, which is observed in most seismic slip – i.e., stick-slip. Conversely, (a-b)>0 implies velocity strengthening, which is related to stable sliding – i.e., fault creep.

RSQSim works with discretized fault surfaces in the sense of Rice (1993) and the stresses are analyzed for each fault element throughout the whole computation. To model the seismic cycle, RSQSim considers three states – healing, nucleation and seismic rupture – that employ analytical equations to resolve the evolution of stresses, slip speed and the state variable at every state (Richards-Dinger and Dieterich, 2012).

Although RSQSim is not a fully dynamic rupture simulator, the code incorporates approximations to elasto-dynamics that allow to model the coseismic phase of the earthquake cycle realistically. These approximations include reduction of the direct effect constant of the RSF law a in slipping patches and the inclusion of a dynamic overshoot coefficient, which replicates the dynamic overshoot from fully dynamic rupture simulations (Richards-Dinger and Dieterich, 2012).

All these features make RSQSim a strong tool to model not only simulated seismic sequences over many seismic cycles, but also to statistically analyze fault rupture patterns and fault interaction in fault systems at the earthquake cycle scale.

We defined a set of fault plane geometric models that increase complexity in both fault connectivity and dip variability at depth, and along-strike trace sinuosity (Fig. 2). Fault trace sinuosity refers to the curvature of the fault traces at both the surface and base of the seismogenic thickness, which result in fault plane geometric roughness. We use a detailed fault mesh with 300 m-side triangular fault elements to better capture geometric fault trace complexities. In all models the seismogenic depth is set at 12 km, which is consistent with studies in the region (e.g., Lavecchia et al., 2016).

In terms of fault connectivity, we define three levels. The simplest one is based on the segmentation of the Mt. Vettore fault, as proposed in the Fault2SHA Central Apennines Database (Faure Walker et al., 2021; Fig. 1). The fault model consists of four fault surfaces completely disconnected from the top to the bottom of the seismogenic layer and with a constant dip of 60° (Fig. 2). This model assumes that the observed surface segmentation is preserved at depth. The second level assumes that the fault surfaces observed from the segmentation link at depth at 7 km, which is constistent with interpretations of the Mt. Vettore by Lavecchia et al. (2016). Like in the previous model, the dip is constant throughout the whole seismogenic layer at 60°. The third level has the same characteristics as the second one in terms of linkage at depth, but with a listric geometry, from 60 to 30° between 7 and 12 km. The listric geometry at these depths is also consistent with modeling data by Lavecchia et al. (2016). The three levels are referred to as D for Disconnected, C for Connected constant and L Connected listric, respectively, throughout the text.

In terms of along-strike trace sinuosity, we consider four levels to capture the complexity in the fault surface. The first level is defined by fully linear fault traces at top and bottom, which define fully planar fault surfaces. From the second to fourth levels the surface trace is based on the mapped fault trace at surface from the Fault2SHA Central Apennines Database (Fig. 1), with smoothing to remove details below resolution of the 300 m fault elements, while the bottom trace changes with different levels of complexity. The second level considers a straight trace at depth, the third one considers a smoothed fault trace at depth, and the fourth preserves the surface fault trace at depth. The four sinousity levels are defined as increasing numeric values for increasing sinuosity of both top and bottom fault trace. From no sinuosity to maximum sinuosity these four levels are 0, 0.3, 0.6 and 1, respectively. By combining all these levels, we define a matrix of 12 fault models that we explore in our study (Fig. 2).

RSQSim models are run for 100 000 simulated years, of which we analyze only the last 50 000 years to ensure that the initial stresses have evolved sufficiently from the initial conditions and that the earthquake cycle has stabilized. We also perform a magnitude completeness analysis to remove earthquakes below that magnitude following the maximum curvature approach, and we also remove events that involve less than 10 fault elements. This is done to ensure that the rupture process of the simulated events is well resolved and thus the analyzed magnitudes are those that fit the best the empirical relations (e.g., see Zielke and Mai, 2023).

The probabilities of surface rupture are obtained by filtering the catalogue data on the fault elements located at the surface analyzed with respect to the data registered throughout the whole fault. The likelihood of surface rupture is computed for the whole fault as the quotient between the number of events of magnitude m equal to a reference magnitude M (m=M) reaching the surface and the total number of events of m=M occurred in the fault (light grey and dark grey histograms, respectively in Fig. 4a). Surface ruptures are accounted regardless of the number of surface fault elements involved or their slip. Then, we compute the conditional probability of surface rupture for m=M at the whole fault by fitting a logistic regression to the discrete data points, as introduced by Youngs et al. (2003) 5P=ea+b×M1+ea+b×M, where P is the conditional probability of primary surface rupture, and a and b are the fit coefficients of the regression, not to be confused with the RSF law coefficients.

We also analyze the probability of surface rupture variations along strike. For each fault element at surface, we compute the ratio between the number of events of m≥M affecting that patch and the total number of events of m≥Min the whole fault. In Fig. 4b the dotted line is the number of events of m≥M in each patch at the surface along strike, while the red line represents the total number of events for m≥M in the whole fault (reported as a line for better visualization).

For the different surface rupture probability analyses, we use different magnitude thresholds. Regressions are computed for Mw≥4.0 to ensure all magnitudes are considered without biasing the regressions, including those with lower probability. Along-strike surface rupture probabilities are computed for Mw≥6.0 as these magnitudes have larger probability of surface rupture and, thus, ensure a better visualization of the spatial variability patterns along strike and the effect of fault segmentation.

We compare the regressions of surface rupture probability with the empirical curves for normal faults in the Great Basin by Youngs et al. (2003) and worldwide normal faults by Pizza et al. (2023), and the numerical curves for normal faults by Mammarella et al. (2024), considering the Thingbaijam et al. (2017) scaling equations and site-specific inputs for our study area: a seismogenic depth of 12±1 km, a dip angle of 60±5° and the hypocentral depth ratio of Italian normal faults (i.e., peak of hypocentral depths at around 67 % of the seismogenic depth). See Mammarella et al. (2024) for details on input parameters.

The analysis of the catalogues is performed for their entire 50 000 year length as well as for shorter time windows to capture potential variability in the earthquake rates (see Sect. 3.2). All probabilities computed in the paper refer to primary surface rupture on the principal fault; modeling secondary faulting or off-fault deformation is beyond of our scope.

We quantitatively compare the outputs of the simulations with surface geological observations of coseismic slip distribution of the largest main shock of the 2016 earthquake sequence (30 October Mw 6.5; Chiaraluce et al., 2017) and the cumulative throw for the past 18 kyr (approximately the age of the Mt. Vettore fault scarp in the Central Apennines; Puliti et al., 2020). To do this, we assign each field measurement to the nearest surface fault element in the model. If multiple measurements fall within the same element, their values are averaged. Because each fault model has a slightly different geometry and mesh configuration, the assignment of field data points to fault elements can vary, leading to small differences in the coseismic and cumulative throw profiles.

The coseismic slip distribution of the 2016 Mw 6.5 earthquake, obtained from the SURE 2.0 database (Nurminen et al., 2022), is compared against the net surface slip distributions along strike for all simulated earthquakes with magnitudes compatible with the 2016 coseismic magnitude. Given that the maximum magnitudes (Mmax) of the simulated catalogues do not exceed Mw 6.6 (see details in Sect. 3.1.3), we use the 2016 magnitude to select such magnitudes following the condition: Mmax-0.1≤M≤M2016+0.1.

The cumulative throw along strike is obtained from the measurements reported by Puliti et al. (2020) on the main fault and compared against the cumulative throw of all events of Mw≥ 5.5 that rupture the surface in the simulations. Because RSQSim computes the net slip of earthquakes in the dip direction on all fault elements, we convert them to fault throw considering the dip of the elements in the fault model (60°).

In both cases, the agreement between simulated values and observations is evaluated using the mean absolute error (MAE) – i.e., the average of the differences between model and observation.

Fault geometry plays a key role on the characteristics of the resulting simulated earthquake catalogues. The depth connectivity assumptions primarily influence the earthquake nucleation depth; while disconnected models nucleate earthquakes primarily following the depth-variable fault slip rate distribution prescribed to the models, the connected models primarily nucleate earthquakes at the depth where the fault segments merge. The sinuosity, on the other hand, favors shallower larger magnitude earthquake nucleations in models with depth connectivity.

Delogkos et al. (2023) recently investigated the impact of variable fault geometries and slip rates on simulated RSQSim earthquake catalogues and found outcomes consistent with those in our study. For one, they show how pre-imposing variable slip rate distributions, tapered toward the fault edges, enables a more realistic modeling of hypocenter depth distribution. For another, they demonstrate that increasing fault complexity (e.g., through the inclusion of antithetic structures or fault trace sinuosity) not only promotes shallower earthquake nucleations but also results in a more distributed hypocenter pattern across depth, rather than a single dominant nucleation level. Our models exhibit these same features. In addition, the depth distributions of more complex models (i.e., connected and with sinuosity) show better agreement with empirical distributions from Mammarella et al. (2024), compared to simpler configurations (Fig. 6).

In terms of earthquake catalogue statistics, we observe that the explored geometric features imply large differences in Mmax and MFD shape. Higher connectivity generally increases the Mmax due to the larger available rupture areas and a reduction in rupture segmentation barriers. In this context, there are models, especially the disconnected ones, that fail to reproduce magnitudes observed in the system, such as the Mt Vettore 2016 Mw 6.5 earthquake. This discrepancy indicates that the disconnected models underperform in terms of Mmax compared to their connected counterparts, and could serve as a criterion for assessing fault model plausibility in the region.

Sinuosity, on the other hand, tends to reduce the Mmax by acting as a frictional barrier that can inhibit slip and preclude full growth of large earthquakes over the entire fault (Dieterich and Richards-Dinger, 2010). This explains why smooth faults yield characteristic MFDs, while rough or geometrically more complex faults produce MFDs closer to a Gutenberg–Richter (e.g., Delogkos et al., 2023). Dieterich and Richards-Dinger (2010) already observed that fault roughness progressively shifts the characteristic peak to shorter inter-event times and increases the rates in the magnitude gap between the power-law and the characteristic domains of the MFD. However, fault roughness alone does not fully overcome the tendency towards characteristic behavior, as we also observe in our results. According to the authors, other model parameters such as fractal segmentation have a stronger impact on aiding this earthquake deficiency, and therefore is a feature that should be explored in future investigations.

One of the key outcomes of our analysis is that fault geometric assumptions are the primary control on surface rupture probabilities. Moreover, our findings are transferrable beyond the modeled case because they are independent of the earthquake rate along the catalogues.

Hypocenter depth is the key driver of surface rupture probability changes in our models. This observation is consistent with the conclusions on the numerical approach introduced by Mammarella et al. (2024). The authors identified the seismogenic depth as the main factor controlling the surface rupture probabilities, which controls the hypocenter depth. In line with this, the relatively short seismogenic depth in our models, combined with stable nucleation depths, explain why regressions show sharp increases in probability in a relatively narrow magnitude range between Mw 5 and 6. Shorter seismogenic depths nucleate shallower earthquakes, favoring the saturation of the seismogenic layer faster and thus allowing ruptures of a same magnitude to reach the surface easier than for deeper seismogenic depths.

Another important factor regarding nucleation is the location relative to the fault plane. The slip rate distribution, tapered to the whole fault, generates slip rate concentrations in the fault segment edges that lead to a significant number of earthquake nucleations in these regions (see Fig. S7), a phenomena that has been widely described in simulators (Shaw, 2019). However, the impact of these earthquakes is minimal in the surface rupture probabilities. First, most of these nucleations correspond to low magnitude earthquakes (Mw 4–5), which nucleate at consistent depths with depth distributions in the region (see Fig. S4) and rarely generate surface ruptures (Fig. 9a). For larger magnitudes Mw≥ 5.0, these anomalous nucleations are significantly reduced (Fig. S7). Second, even though local probability increases at segment boundaries (Fig. 10) could be linked to the few Mw≥ 6.0 earthquake nucleation at fault edges, these can also be explained by geometric complexities, such as fault bends and segment connection at depth. Such geometric controls on nucleation have been previously described in the Mt. Vettore fault. For instance, Lavecchia et al. (2016) documents that the 2016 Mw 6.0 Amatrice earthquake nucleated at an inter-segment zone, where two fault segments link at depth. Third, even if surface rupture probabilities along-strike are slightly affected by anomalous earthquake nucleation, in a full PFDHA application the fault displacement hazard is ultimately driven by the slip recorded in each site. Our models show that both coseismic and cumulative throw taper toward segment edges (Figs. 11 and 12) independently of nucleation location, suggesting that localized increases of surface rupture probability are unlikely to bias fault displacement hazard.

Fault geometry, especially trace sinuosity and segmentation, also has a critical impact on earthquake rupture behavior. Geometric roughness acts as a physical barrier that affects how ruptures propagate (e.g., Dieterich and Richards-Dinger, 2010; Zielke and Mai, 2025). For instance, introducing fault trace sinuosity in our models hinders lateral rupture propagation and instead favors along-dip rupture, which raises the probability of surface rupture in connected configurations. Likewise, fault connectivity favors segment-to-segment rupture propagation. When combined with no sinuosity (i.e., level 0) surface rupture probabilities are reduced as ruptures are able to propagate laterally, which is not observed for the disconnected counterparts that still preserve lateral geometric barriers (segmentation). Similarly, listric fault geometries generally reduce the probabilities of surface rupture in the regressions because the fault area available is larger, requiring larger magnitudes to reach the surface.

These modeling results are consistent with numerous field observations indicating that fault surface geometry strongly controls rupture propagation and slip patterns (e.g., Lettis, 2002; Rockwell and Klinger, 2013; Rodriguez Padilla et al., 2024). For example, the recent study by Rodriguez Padilla et al. (2024) showed that geometrical features, such as step-over width, are key locations for rupture arrest. This is consistent with the observed drop in surface rupture probabilities of Mw≥ 6.0 earthquakes across the step-over between the Mt. Porche and the Mt. Bove segments (Fig. 10). This agreement between simulation and observations reinforces the reliability of earthquake simulators for characterizing fault rupture behavior at the surface. In addition, earthquake simulators offer the opportunity to systematically quantify the influence of fault geometric features on rupture propagation, such as earthquake jump distance as a function of the fault slip. This is an issue that might be investigated in the future.

Our results also indicate that surface fault trace geometry has a stronger influence on magnitude-dependent rupture probability than the subsurface fault trace. In this sense, constraining fault traces at surface should be prioritized over subsurface traces. This finding supports recent conclusions by Zielke and Mai (2025), who demonstrated how models with a same surface fault trace produce long-term fault behavior results that are interchangeable, even if their subsurface geometry differs significantly. That said, fault segment connectivity at depth, especially when paired with surface fault trace sinuosity, remains a dominant factor controlling surface rupture probabilities in our models and therefore should not be neglected.

The strong influence of fault geometry in earthquake surface rupture statistics contrasts with the frequent lack of constraints on the subsurface geometric features of faults, often very expensive and difficult to image. In addition, frequent challenges in the identification of primary fault traces and uncertainties in fault trace location can also become a limitation for the correct characterization of fault geometries and, thus, for the implementation of the proposed approach. To tackle these issues in a hazard evaluation context, exploratory analyses represent the most suitable approach. For instance, as noted by Zielke and Mai (2025), exploring multiple realizations of fault geometries, including fault trace hypotheses, may be a practical solution to account for the epistemic uncertainties linked to the poor knowledge of subsurface fault geometries.

Our surface rupture probability regressions are generally closer to those derived from numerical approaches (Mammarella et al., 2024) than those from empirical earthquake data (Pizza et al., 2023; Youngs et al., 2003). On the one hand, simulation and numerical approaches generally show steeper slopes than the empirical regressions due to fault-specific modeling assumptions. That is, in both approximations the models are constrained to a single seismogenic depth (12 km), fault dimensions, and similar hypocenter depth distribution. This is further corroborated by the higher coincidence between the C0 and L0 regressions and the numerical regressions. The geometric considerations of the sinuosity level 0 models are closer to the ones by Mammarella et al. (2024), where fault sinuosity is not considered. Interestingly, such a coincidence between regressions is not observed for the D0 model probably because the strict fault segmentation limits earthquake lateral growth.

On the other hand, empirical regressions are based on mixed datasets of global earthquakes occurring on faults with different seismogenic depths, hypocentral distributions and geometries, which may smooth out the regressions. Modeling several fault systems would return regressions closer to the empirical ones, as they would result from a broader range of tectonic settings and parameters that do not change so much in a single fault system. To prove this, we tested how mixing fault models with different geometries and seismogenic depths influences the regressions (Fig. S8). We combine the catalogues from six different geometric models into a unified dataset to better approximate large-scale (multi-fault system) analyses in PFDHA. These models are selected to capture the broadest range of variability in regression behavior (Fig. 9), and we include two new models with seismogenic depths that differ from those assumed in our study. This analysis shows how mixing fault models with different seismogenic depth and geometric considerations generally smoothes out the regression slopes and shifts them toward larger magnitudes, similar to what is observed for empirical regressions (Fig. S8).

Generally speaking, the regressions of surface rupture we obtain are visibly off the empirical and numerical regressions available in literature. There are several modeling factors and assumptions that may contribute to this effect.

First, the geometric components explored in this study are a sample of the whole spectrum of potential geometrical complexities to be combined and explored. Here we focused on fault connectivity and trace-scale complexities, but smaller scale variability might also be explored. For instance, Zielke and Mai (2016) demonstrated how incorporating sub-patch geometrical roughness – i.e., roughness at spatial scale below fault element size – impacts earthquake behavior in earthquake rupture simulations. Similarly, exploring the impact of fractal roughness as in Allam et al. (2019) might provide new insights into fault surface rupture behavior.

Second, the model parameters, namely the uniformity in the initial stresses and rate-and-state coefficients (a-b), are considerable simplifications. Initial stresses are uniformly distributed throughout all elements in the simulations, while in reality stresses change at depth (e.g., normal stress increases as a function of depth and is also dependent on fluid pore pressure). This implies that stress conditions equivalent to 6 km depth are assigned to all fault elements, including near surface. Even though in RSQSim the initial stress conditions evolve throughout the simulation, these initial conditions affect how ruptures propagate (e.g., Liao et al., 2024), thus how they manifest at the surface.

We assumed uniform stresses to prevent shallow-dominated hypocenter nucleations caused by small stress values near surface in heterogeneous stress models. As identified by Liao et al. (2024), heterogeneous stresses would shift the hypocenter nucleations to unrealistically shallow depths, ultimately increasing the probability of surface rupture for all models. Even though depth-variable stresses might yield more realistic fault slip distributions, we decided to prioritize accuracy in nucleation depth given its impact in our analysis. Building on this idea, Hughes et al. (2024) used uniform initial stresses for RSQSim-based tsunami models to avoid shallow nucleation depths of heterogeneous stress models and to ensure more conservative (higher) wave heights for the hazard assessment, even if their earthquake slip patterns are less realistic.

Similarly to the initial stresses, RSF law coefficients a and b have been assumed uniform for all fault elements with velocity weakening conditions (a-b<0). In nature, however, these coefficients follow depth dependent profiles, with typical regions near surface under velocity strengthening conditions (a-b>0; e.g., Lapusta et al., 2000). While introducing such variability can improve some catalogue features like the depth distribution of earthquakes (especially when paired with depth-variable stresses; Liao et al., 2024) and rupture propagation, it would also add complexity to our analysis. Data on the depth variability of RSF parameters is generally not available in most regions, which would likely require adopting unvalidated assumptions in the model parametrization.

The roles of stress and frictional parameters are definitely topics to be explored in the future given the demonstrated impact they have shown in simulated catalogues (e.g., Delogkos et al., 2023; Gómez-Novell et al., 2025a; Liao et al., 2024). As an example, we have tested the influence of varying (a-b) in the regressions of our end-member models used for the benchmarking tests (see Table 1). We observe that these parameters have a strong influence in the probabilities of surface rupture (Fig. S9), but not so much in the shapes of the regressions. In general, more negative (a-b) decrease the probabilities of surface rupture by shifting the regressions toward larger magnitudes. In addition, the regression variability linked to the (a-b) variations depends on the fault geometric model; higher in the L geometry compared to the D (Fig. S9). This lower sensitivity to geometric changes in the disconnected model is consistent with the observations made in Sect. 3.2. Although the (a-b) parameters used in our study are the ones that better match magnitude-area empirical relations, we underscore the importance of constraining these physical parameters where earthquake simulation studies are conducted.

Despite the modeling factors, it is important to remark that part of the misfit between simulated and empirical regressions may also be linked to observational biases affecting empirical data. That is, surface ruptures with short rupture lengths, small displacements or those occurred in remote regions or historical times are more likely to be underreported or missing in databases. These omissions can ultimately lead to underestimated empirical regressions. Other causes for lower surface rupture probabilities in empirical models can be related to near-surface soil conditions. For instance, the presence of soft sediments, uncompressed rocks or loose materials can lead to fault offset attenuation and accommodation through warping or folding, ultimately decreasing the imprint or even recognition capabilities of surface ruptures.

A final limitation is that out study focuses solely on on-fault surface displacements, while distributed rupturing and off-fault deformation are important components of empirical PFDHA models. While the implementation of such components is beyond our scope, our work can serve as a basis for future work in this direction. For instance, the simulated surface displacements on the principal fault can be combined with empirical distributed faulting regressions (e.g., Visini et al., 2025) to develop probabilistic models of distributed fault ruptures. An example of that is the work by Daglish et al. (2025), where they use the simulated on-fault displacements to scale across-fault displacement based on empirical data on surface rupturing earthquakes.

We compare how well our models reproduce surface geological observations of the fault to test the suitability of our model parameters and future applicability to PFDHA.

This study demonstrates that earthquake cycle simulators like RSQSim are a valuable tool for advancing PFDHA, in line with Daglish et al. (2025). While the current models are implemented for a specific fault system, our methodological approach provides insights that are transferable to PFDHA applications in general.

First, our simulations produce coseismic slip patterns that are reasonably coherent with observed data on the 2016 Mw 6.5 Mt. Vettore earthquake.

Second, even though we have omitted long-term geomorphic processes and structural complexities at depth, the cumulative throw is partially captured by our models and its spatial trends are preserved with respect to observations (e.g., segment limits).

Third, the hypocenter depth distributions modeled match the regional observations in Italy and demonstrate its controlling role in the probability of surface rupture, findings that are in agreement with previous numerical approximations.

Earthquake cycle simulators can also help to overcome inherent completeness issues of earthquake databases for fault displacement hazard, enabling enrichment of earthquake databases, the systematic exploration of fault parameters like subsurface geometry and allowing fault-specific analyses. By extension, the explicit consideration of earthquake rupture physics and the high-resolution earthquake displacement data generated by the simulations allows the implementation of site-specific analysis with the displacement approach, one of the key challenges of PFDHA.

Consequently, as anticipated by Valentini et al. (2025b), simulation-based studies like the present one might contribute to the implementation of earthquake simulators into PFDHA in the future and to the overall enhancement of their capabilities. This enhancement potential has already been demonstrated for seismic hazard (e.g., Herrero-Barbero et al., 2023; Rafiei et al., 2022; Shaw et al., 2018) and tsunami hazard applications (e.g., Álvarez-Gómez et al., 2023; Hughes et al., 2024).

Despite the advantages, the use of earthquake simulators has limitations for PFDHA applications, some of which already described by Daglish et al. (2025). These include (i) less extensive statistical validation with respect to empirically-based seismic hazard models, (ii) rate-and-state parameters being calibrated to match magnitude-area scaling used for empirical approaches, preventing from a fully-independent analysis, and (iii) important simplifications in the physics of earthquake rupture propagation, compared to fully dynamic rupture simulations. In addition, the general lack of site-specific data on the fault systems, like in many numerical approaches, can become a challenge for the successful implementation and validation of the analyses proposed here. Having said that, to date these models offer the most computationally efficient solution to model earthquake cycles. These models provide a balance between reasonable earthquake rupture physics and the ability to generate near-surface displacements over several seismic cycles (e.g. Daglish et al., 2025), necessary for robust long-term fault statistics in PFDHA. In this line, the emergence of newer and improved earthquake cycle simulators such as MCQsim (Zielke and Mai, 2023), which allows the explicit incorporation of fault roughness and visco-elastic relaxation, or Tandem (Uphoff et al., 2022), which enables fully-dynamic multi-cycle rate-and-state friction, could further enhance the capability of simulators to reproduce more realistic earthquake rupture sequences in the future. Additionally, when fault data is scarce, earthquake simulators offer the opportunity to systematically explore epistemic uncertainties in many fault model parameters, making it a strong alternative to fully empirically-based approaches in PFDHA.