Strong asperities nucleate earthquakes on laboratory faults Open Access

Faults are structurally complex with abundant heterogeneity from the microscopic scale up to the kilometer scale (Chester et al., 1993; Faulkner et al., 2003). Fault motion or slip is regulated by frictional processes with some faults exhibiting both seismic and aseismic slip (Miyazaki et al., 2011; Thomas et al., 2014; Caballero et al., 2021). Fault slip can be simplified to three mechanical states: a locked state where no slip occurs, a stable sliding state dominated by slow slip, and an unstable sliding state dominated by fast slip. Identifying factors that govern the transition between these states is key to advancing the mechanics of fast and slow earthquakes and can inform other frictionally governed geologic processes, including landslides and glacial flow. Two potentially important factors relate to heterogeneity: roughness and mineralogy.

Fault roughness is well documented from the km- to µm-scale (Bistacchi et al., 2011; Candela et al., 2012). Spatial distributions of normal stress, σ_n, are thought to be controlled by fault roughness (Candela et al., 2011; Fang and Dunham, 2013; Cattania and Segall, 2021). Kilometer-scale roughness has been linked to megathrust and shallow subduction zone earthquakes (Bilek and Lay, 2002; Kirkpatrick et al., 2020), earthquake swarms (Cochran et al., 2023), and nucleation of secondary rupture fronts (Xu et al., 2024). Large-scale roughness has also been linked to fault locking and earthquake nucleation (Lee et al., 2024). Though there is consensus that roughness is important, the role of roughness on frictional strength and stability is not clear. There are conflicting interpretations from laboratory experiments on whether roughness promotes frictional instabilities (Eijsink et al., 2022; Goebel et al., 2023) or inhibits them (Fryer et al., 2022; Xu et al., 2023). Other experiments document transitional stability regimes but with conflicting interpretations on when roughness enhances or inhibits frictional instabilities (Harbord et al., 2017; Morad et al., 2022).

Experiments investigating mineralogical heterogeneities often combine frictionally strong, velocity weakening materials (µ decreases as velocity, V, increases) with frictionally weak, velocity strengthening materials (µ increases as V increases) (Crawford et al., 2008; Collettini et al., 2009; Tembe et al., 2010; Moore and Lockner, 2011; Tesei et al., 2014; Hirauchi et al., 2023). These investigations show decreases in µ and increases in stability with increasing phyllosilicate content or when fabrics are present. More complex faults with distinct heterogeneous patches show decreased stability on bimaterial faults compared to homogeneous faults (Bedford et al., 2022). In other experiments, long-term strengthening was documented and attributed to mixing of bimaterial patches (Arts et al., 2024).

As natural faults exhibit abundant heterogeneity, experiments that incorporate complex, multi-source heterogeneity can advance descriptions of fault mechanics and help scale results from the lab to nature. We investigated the frictional properties of rough, bimaterial laboratory faults to explore the role of complex heterogeneity on frictional stability. The sliding surfaces of our experimental faults were engineered to allow direct links between slip-dependent mechanical behaviors and geometries of rough surfaces. We present mechanical, microstructural, and seismic source data from experiments to explore stability in the context of heterogeneity and facilitate links between lab and natural earthquakes. Ultimately, we present a framework for transitions from stable to unstable sliding during fault creep that could nucleate seismic and aseismic slip events in nature, including earthquakes, low-frequency earthquakes, slow slip events, and tremor.

Experiments were performed using the Tullis Rotary Shear Apparatus (see Supplemental Material¹) at 25 MPa confining stress, 30 MPa σ_n, room temperature, and room humidity. We prepared annular samples of Frederick diabase with flat-topped macroscopic asperities (Fig. 1), minimizing changes in stress state from inclined asperity geometries (see Patton, 1966, p. 50). The diabase is velocity weakening with a smooth, reference friction coefficient, µ_ref, of 0.7 at the σ_n of our experiments. Between and level with the tops of asperities, we compacted a velocity strengthening talc gouge with a µ_ref of 0.1. Coupling a talc gouge with the rough diabase has two benefits. At high σ_n, the gouge compacts relative to the initial sample geometry (Fig. 1C), forcing asperities to ride over one another during encounters and allowing us to explore roughness and mineralogy simultaneously. The contrasting mechanical behavior also allows us to better distinguish the mechanical effects of each material.

To vary roughness, we used four geometries with amplitude to wavelength ratios, R, of 0.007–0.003, representing the higher end of natural fault R that ranges from 0.01 to 0.0001 (Power and Tullis, 1991). We calculate R using asperity height divided by the circumferential length of an asperity and the clockwise adjacent gap and report the average for all lower sample asperities (Fig. 1C). Experiments were conducted with symmetric, R_S, or asymmetric roughness, R_A. In R_S experiments, the upper samples included seven asperities while the lower samples hosted seven, five, or three, corresponding to average R_S of 0.007, 0.005, or 0.003. While the number of asperities decreased, the locations remained uniform, ensuring that asperities enter and exit contact synchronously (Figs. 1B and 1C). This allows us to define a mating index where 0 signifies no asperities are in contact and 1 indicates all lower block asperities are fully mated with upper block asperities. The synchronicity and symmetry simplify identifying the mechanical effects of roughness. For the R_A experiment, asperity locations were randomly permuted within 50° sectors while maintaining an average R_A of 0.007. To limit V-dependent changes in wear rates (Boneh et al., 2013), a constant V of 5 μm/s was typically maintained. Experiments were conducted with displacements, d, of 30–170 mm, or 1.5–7 asperity encounters, respectively.

When asperities interact on rough surfaces, rapid increases in µ trigger a shift from stable to unstable sliding with roughness controlling the peak µ and stick-slip events (StSE) characteristics (Fig. 2). Stable sliding with a low µ dominated the first 5 mm of d for all experiments when there was no diabase-diabase contact (Fig. 2C). As the asperity mating index approached 1, representing fully mated asperities, µ increased approaching diabase µ and the fault transitioned from stable to unstable sliding characterized by frequent StSE. The peak µ ranged from 0.78 to 0.6, decreasing with R_S. The increase in µ coincided with dilation at the sliding surface, with the magnitude of dilation increasing at higher R_S. As asperities unmated, µ decreased, approaching the initial value, and stable sliding resumed. This history repeated during a second encounter, though the µ peaks were reduced, after which there were no significant changes in µ with d or mating index. Quasi-stable sliding was maintained for the remainder of d with occasional instabilities at d up to 150 mm; µ always exceeded talc µ_ref in all experiments. The amplitude of dilational events related to mating asperities decreased with d and decreasing R_S. By the seventh encounter, the amplitude of dilation was significantly diminished.

StSE, with d of 2–9 µm, occurred in all geometries, predominately during the first two asperity encounters (Fig. 2A). When R_S was low, StSE were most frequent as the mating index increased and asperities entered contact. With larger R_S, StSE were more frequent while the mating index decreased (Fig. 2D).

Representative microstructures from four experiments are shown in Figure 3. Pulverization and beveling of the trailing and leading asperity edges was observed in all experiments. In low R_S experiments (Figs. 3A and 3C), extensive, penetrating fracturing and pulverization occurred behind the asperity leading edges, with these regions elongating and connecting on one of the sample blocks. No widespread fracturing or pulverization was observed in high R_S experiments (Figs. 3B and 3D). The overall degree of fracturing and pulverization did not vary significantly between low and high d experiments in either geometry.

When asperities interact on rough, bimaterial faults, the mechanical behavior of the fault reflects the mineralogy and abundance of asperities; outside of asperity encounters, the mechanical behavior is an intermediate between the two materials. In our experiments, talc comprises 71%–80% of the sliding surface while diabase comprises 29%–20% of the sliding surface. If µ is calculated using µ = µ_talcP_talc + µ_diabaseP_diabase where P is the area percent, we would expect values of 0.29–0.23, averaging at 0.26. Instead, during the first asperity encounter when diabase juxtaposed diabase (strong contacts) µ increased dramatically, approaching peak diabase µ_ref, though strong contact area was only 29%–12% of the sliding surface. In nominally flat experiments, the true contact area is significantly lower than the apparent area, which may explain the agreement between peak diabase µ_ref and µ during early asperity encounters.

After the first encounter, µ decreased, approaching the aggregate estimates. Dramatic, though reduced, increases in µ repeated during the second encounter. The reduction in µ likely reflects a combination of mechanical damage that occurred during the first contact and smearing of talc along asperity surfaces. After two encounters, µ values remain low for the duration of sliding, reflecting intense mechanical damage that renders asperities mechanically insignificant (Fig. 3). In low R_S experiments, long-term µ is consistent with aggregate estimates. For the high R_S experiment, long-term µ was 0.5, exceeding the aggregate estimate of 0.29. This may reflect mixing of pulverized diabase into the talc gouge due to the extensive damage at trailing and leading edges of asperities (Fig. 3). Though more fracturing and pulverization occurs in lower R_S experiments, there are fewer asperities, and upper sample asperities experience extended periods of no contact likely resulting in less mixing and a lower µ.

We calculated the seismic moment, M₀, and cumulative moment for StSE using M₀ = GAd_e where G is the rigidity or shear modulus, A is the sliding surface area, and d_e is the StSE d. StSE are defined as drops in µ of 0.005 or more during 0.5 µm of d. We used a G of 24 GPa, calculated from a proportional average of the values of G of 22 GPa for talc (Bailey and Holloway, 2000) and 30 GPa for diabase (Weijermars, 1997), an A of 729 mm² based on the sample dimensions assuming the entire interface slips when instabilities occur, and measured d for d_e. Shear stress drop, Δτ, was independently measured during StSE.

Seismic moment and stress drop increase with decreasing R_S, suggesting that roughness increases stability (Fig. 4). StSE presumably reflect cataclastic failure at critically stressed microscopic asperity contacts located on strong contacts. Inhomogeneous σ_n distributions have been documented in experiments on granite (Barbery et al., 2023), and strong asperities likely serve as stress localizers with concentrated effective normal stress, σ_e, during encounters. Since decreasing the number of asperities lowers the total strong contact area, if stresses are localized on strong asperities, fewer asperities result in higher σ_e at strong contacts. Larger σ_e and shear stresses at microscopic asperities prior to failure would explain larger Δτ and M₀ in experiments with fewer asperities since critical stiffness increases with σ_n. With reduced strong contact area, there may also be fewer asperities to serve as barriers to arrest or slow slip when asperities fail.

Concentrated stresses could also explain the different timing of StSE. With larger R_S, increased strong contact area may result in sufficiently distributed initial stresses preventing early StSE. As illustrated in Figure 2D, microscopic asperities tend to fail as strong contact area is reduced during unmating. In contrast, when R_S is low and strong contact area is reduced, StSE are more prevalent as asperities enter contact, suggesting stresses are sufficiently concentrated to induce failure. Mechanical damage due to early StSE may alleviate stress concentrations during unmating and explain why instabilities do not recommence. The late StSE in experiment 404 (Fig. 4A) likely reflect a lack of mechanical damage; in this experiment asperities were initially mated with no mechanical damage prior to unmating, and StSE began as the mating index neared 0.

Cumulative moment was smaller for high R_S and similar for intermediate to low R_S (Fig. 4A), reflecting a balance between asperity abundance and longevity. When asperities are more numerous, stress is sufficiently distributed to minimize damage and instabilities, resulting in low cumulative moments and occasional StSE after the first encounter. With fewer asperities, stress is more localized, resulting in numerous instabilities during the first 2–3 encounters, after which StSE frequency, M₀ and Δτ decrease. The maximum cumulative moment occurred in the asymmetric experiment and likely reflects a similar balance between abundance and longevity, enhanced by the complex asperity mating history during sliding.

This work demonstrates the complex mechanical behavior of heterogeneous faults. Stable sliding dominates when frictionally weak materials juxtapose strong or weak materials. When strong asperities interact, the mechanical and frictional stability behavior alters dramatically, approaching that of the strong asperities. The average, peak µ decreased as R_S decreased. Assuming µ = τ/σ_n, with µ equal to the peak diabase µ_ref of 0.79 and a constant τ, σ_n would increase by 120% and 140% as R_S decreases, in overall agreement with the decreased dilation and increased damage observed with decreasing R_S. Encounters of strong, velocity weakening asperities may promote the nucleation of seismic and aseismic events on nominally creeping faults. Whether failure occurs seismically or aseismically may depend on the size and rheology of asperities, with increased σ_e on larger asperities nucleating earthquakes, and smaller asperities with lower σ_e nucleating slow slip or tremor. Similar asperity interactions may also contribute to landslide initiation. While strong asperities can initiate rapid and dramatic transitions from stable to unstable sliding, asperities undergo extensive damage during encounters and become mechanically obsolete following multiple contacts. This suggests earthquake nucleation may be limited on mature, highly damaged faults, in contrast with natural observations, indicating the significance of damage recovery processes during interseismic periods.

We thank John Bedford and an anonymous reviewer for comments that improved this manuscript. This work was funded by the National Science Foundation (grant EAR-2052897).