The Role of Clay in Limiting Frictional Healing in Fault Gouges

Frictional healing is fundamental to the seismic cycle and plays a role in the energy balance, dynamics, and recurrence interval of earthquakes and slow slip events. Although the healing behavior of quartz has been studied extensively, the role of clay content is less understood. We tested synthetic mixtures of quartz and smectite (10%–100% smectite) in a double‐direct shear configuration to measure frictional healing. We show that the magnitude and rate of healing decreases systematically with higher clay content (from 0.008/decade at 10% smectite to 0.002/decade at 100% smectite). Healing scales with both the magnitude of stress relaxation during holds and layer‐normal compaction of the gouge. We suggest this reflects the alignment of clay minerals, leading to saturation of the real area of contact that limits restrengthening during holds. The low healing rates of clay‐rich faults—together with rate‐neutral to rate‐strengthening friction—should promote frequent, small failures or stable sliding.


Materials and Methods
We analyze data extracted from friction experiments originally described by Saffer and Marone (2003) who reported on the velocity-dependence of friction.These experiments were conducted on synthetic quartz-clay gouges in a double-direct shear configuration in a servo-controlled biaxial apparatus at room temperature (∼25°C) and humidity (26%-49%) (Figure 1a).Synthetic gouges were produced by mixing end-members of granular quartz and smectite powder in proportions ranging from 10 to 100 wt.% smectite.The quartz powder was Natural Grain product F-110 supplied by U.S. Silica Co. and is >99% SiO 2 with an initial median particle size of 127 μm and 99% of grain diameters in the 53-212 μm range.The smectite powder was Ca-montmorillonite SM1502A, supplied by GSA Resources Inc., and was dominantly smectite with trace amounts of zeolite, feldspar, and unaltered volcanic glass and a mean grain size of 60 μm with 80% of grain diameters in the 3-142 μm range (Ikari et al., 2007;Saffer & Marone, 2003;Saffer et al., 2001).
During each experiment, the sample was sheared at a loading rate of 10 μm/s under a constant normal stress of 25 MPa.An unload-reload cycle after the initiation of sliding promoted comminution and development of a steady-state shear fabric within the gouge samples, which provided a reproducible initial state.The cycle was followed by (a) a series of velocity steps with rates from 0.1 to 200 μm/s (used to determine frictional rate-dependence, as reported by Saffer and Marone (2003)) and (b) a series of slide-hold-slide tests with hold times ranging from 3 to 3,000 s (Figure 1b).
During holds, the vertical (shear) ram was held stationary while the normal stress was held constant.When holding the loading point stationary, the shear stress decayed with time, indicating frictional creep relaxation.Shear and normal stress were measured with load cells (resolution of <0.01 MPa), and shear and normal displacements were measured at the respective load points with displacement transducers (resolution of 0.1 μm).We calculated the apparent coefficient of friction as μ = τ/σ n where τ is shear stress and σ n is normal stress.Shear strain was calculated as γ = δ i /h i where δ is displacement and h i is the layer thickness for shear increment i.

Frictional healing (
∆ ) was measured from slide-hold-slide tests and is defined as the difference between the peak friction upon re-shear and the steady-state friction (   ) prior to the hold (Figure 1b).The rate of frictional healing with hold time (  hold ) is the healing rate (   ; units of  ∆∕ decade): where t c is the cutoff time in seconds (Dieterich, 1972).Creep relaxation (∆μ c ) during the hold period, when the load point displacement is held constant, is defined as the difference between the initial friction (μ i ) and the minimum friction during the hold.The rate (β c ) and cutoff time (t c,c ) for creep relaxation were fit following Equation 1.

Results
Both our frictional healing (∆μ) and creep relaxation (∆μ c ) data define a log-linear increase or decrease, respectively, with time over the range of hold times we investigated (c.f.Equation 1; Table 1; Figures 1c and 1d) (Dieterich, 1972).The absolute magnitude of healing (∆μ) decreases with increasing clay content, and healing rate (β) decreases systematically but nonlinearly with increasing clay content, from β = ∼0.008for 10% smectite, to β = ∼0.002for 100% smectite (Figures 1e and 1f).Inset: Representative slide-hold-slide test where μ i is the friction before the hold, ∆μ c is decay in apparent frictional strength due to relaxation during the hold, and ∆μ is the increase in friction upon re-shearing.Note that the mechanical data are presented in terms of shear strain, while the inset shows the evolution in time.(c) Frictional strengthening (∆μ) and (d) creep (shear stress) relaxation (∆μ c ) from slide-hold-slide test as a function of t hold .Model fits are listed in Table 1.Frictional healing rate (β) as a function of (e) clay content) and (f) steady-state friction coefficient (μ ss ) for samples from this study (colored points) and previous studies (gray points).Data from previous studies include: synthetic quartz and smectite end-members (Carpenter, Ikari, & Marone, 2016;Marone & Saffer, 2015), IODP drill core from the Nankai (Carpenter, Ikari, & Marone, 2016;Ikari et al., 2012), Costa Rica (Carpenter, Ikari, & Marone, 2016), and Hikurangi subduction zones (Shreedharan et al., 2022), and samples from the San Andreas Fault system (Carpenter, Ikari, & Marone, 2016).

10.1029/2023GL104984
4 of 9 The healing rates we report and their systematics with both clay content and friction coefficient are consistent with those for natural fault zone and protolith samples that span a wide range of clay content and clay species, as well as previous work on synthetic gouges with end-member compositions (Figures 1e and 1f).These samples include synthetic quartz and smectite gouges (Carpenter, Ikari, & Marone, 2016;Marone & Saffer, 2015), as well as natural fault gouges predominantly composed of quartz and clay from a variety of settings (e.g., San Gregorio Fault, Costa Rica, Nankai, and Hikurangi subduction zones, and the San Andreas Fault Observatory at Depth drill site) (Carpenter, Ikari, & Marone, 2016;Ikari et al., 2012;Shreedharan et al., 2022).It is important to note that these studies were conducted with different controls on pore water content, from room humidity (our study), 100% humidity (Carpenter, Ikari, & Marone, 2016), to water saturated or pressurized (Ikari et al., 2012;Shreedharan et al., 2022).Despite the sensitivity of smectite frictional behavior to hydration state (Ikari et al., 2007), the results are generally consistent.
Creep relaxation rate (β c ) also decreases with increasing clay content, with the synthetic quartz-clay mixtures again overlapping with rates reported for other synthetic and natural samples (Figure 2a).This mutual decrease in healing and creep relaxation rate with clay content leads to a positive correlation between these rates (Figure 2b).These observations are also consistent with previously reported rates and scaling with healing rate for natural fault gouges (e.g., Carpenter, Ikari, & Marone, 2016).Creep relaxation is the result of shear creep during the hold, which results in a reduction in shear stress while the load point is held stationary.Creep relaxation during the hold period is accompanied by a decrease in the thickness of the gouge layer (∆h) due to layer-normal compaction (Figure 2c).We observe a clear correlation between compaction and the magnitude of restrengthening and creep relaxation.Compaction, healing, and creep relaxation are all largest for quartz-rich gouges, and smallest for clayrich gouges (Figures 1 and 2).
To varying degrees, rate-and-state frictional (RSF) parameters are also linked to healing rates (Figure 3).Healing rate is positively correlated with both the direct effect (a) and the evolution effect (b), and negatively correlated with the critical slip distance (D c ).The velocity dependence (a-b) is weakly but positively correlated with β at lower velocities, and negatively correlated at higher velocities.This difference follows the observation that clayrich samples transition from velocity-neutral to velocity-weakening behavior at lower velocities (e.g., Ikari & Kopf, 2017;Saffer & Marone, 2003).The higher velocity trend follows the pattern seen in natural fault gouges (Carpenter, Ikari, & Marone, 2016;Ikari et al., 2016).

Clay Content and Frictional Healing
Frictional healing can be achieved through increasing contact quality (i.e., strength) or contact quantity (i.e., total real area of contact with atomic-scale interactions).We suggest that the reduction in healing we observe with increasing clay content is driven by (a) a reduction in the number of quartz-quartz contacts with decreased abundance of quartz together with enhanced shear and compaction creep of quartz-rich gouges (c.f. Figure 2c) and (b) the saturation of contact area in clay-dominated gouges as a function of fabric development (Figures 4a and 4b).
We discuss each of these ideas below.
Quartz-quartz contacts are known to strengthen and grow via numerous fluid-assisted mechanisms (Frye & Marone, 2002;Q. Li et al., 2011).We hypothesize that introducing smectite limits the efficacy of these mechanisms by reducing the number of available quartz contacts, thereby decreasing the overall contact quality in the gouge.Additionally, compaction during hold periods increases the contact area and thus contributes to frictional healing (e.g., Carpenter, Ikari, & Marone, 2016;Ryan et al., 2018).Compaction is also correlated with creep relaxation, likely due to a combination of elastic deformation, stress rearrangement, and shear-enhanced compaction (e.g., Crawford et al., 2008;Sleep, 1995).The amount of compaction and creep relaxation are linked to the steady-state frictional strength (and thus to quartz content; e.g., Figures 2b and 2c); in general, this is consistent with the idea that higher strength gouges will have more stored elastic strain energy to drive creep.
The presence of clay can also saturate the real area of contact through the alignment of clay grains (Figures 4a  and 4b).Clay particle alignment develops quickly during shear, even at low clay contents (e.g., Haines et al., 2013;Kenigsberg et al., 2019).This alignment, together with the formation of throughgoing shear planes, is reflected in our experiments by the pronounced peak in frictional strength with increasing shear followed by a decay to a steady-state residual friction value (Figure 1b, Saffer & Marone, 2003).We suggest that because of contact area saturation, little evolution of friction with increasing hold time will occur (i.e., there is no additional capacity to increase the real area or quality of contacts), leading to decreasing magnitude of healing and healing rates with increased clay content.A similar model has been suggested for calcite-shale mixtures (Ruggieri et al., 2021).
The relationships between β and individual RSF parameters are also consistent with an explanation rooted in the evolution and saturation of real contact area.The direct (a) and evolution (b) parameters broadly relate to the dilation or compaction of the gouge and the evolution of contact area following a change in velocity, respec tively.Because dilation is minimal in platy materials where grains do not need to roll past or over one-another, a is smaller in the clay-rich gouges.Then, when the contact area is saturated or nearly so, there is little evolution of friction with displacement (or strain) and b is also smaller (Saffer & Marone, 2003).Similarly, D c represents the slip distance needed for this evolution to take place and is commonly understood as an average asperity size (Dieterich & Kilgore, 1994).In this sense, if a small β indicates that the real area of contact is nearly saturated, this would imply that the average asperity has grown nearly to its limit (i.e., D c would be large).Additionally, the transition to stress-independent shear strength at ∼25-40 MPa observed in clay-rich gouges (e.g., Carpenter et al., 2012) and decreasing magnitude of the b parameter (Saffer & Marone, 2003) are also considered a reflection of real contact area saturation under boundary conditions of increasing effective normal stress rather than hold time.
The nonlinear decrease in healing rate with clay content further supports this interpretation.This pattern is similar to that of decreasing friction coefficient with clay content reported for in previous studies (e.g., Crawford et al., 2008;Kopf & Brown, 2003;Logan & Rauenzahn, 1987;Numelin et al., 2007), and is also evident from the nearly linear correlation between β and μ (i.e., healing rate is directly proportional to steady state friction coefficient; Figure 1f, Carpenter, Ikari, & Marone, 2016;Ikari et al., 2016).The drop in healing rates a clay fraction of ∼30%-40% can be linked to the switch from a granular framework to a clay-matrix based on idealized grain packing arrangements for quartz-clay mixtures (Crawford et al., 2008).A clay-matrix supported microstructure is more susceptible to contact area saturation from the alignment of clays, likely explaining the low healing rates at clay contents above 40%.
We suggest that while quartz-quartz contacts may grow with time, the ability of clay-clay contacts to grow is inhibited by their alignment together with their low stiffness (Vanorio et al., 2003), thus limiting the total area available for healing via both mechanical and chemically assisted mechanisms.While this model matches our observations, clay-rich materials tested under different experimental conditions have demonstrated negative healing rates (Orellana et al., 2018;Scuderi & Carpenter, 2022) and, more commonly, negative b values (e.g., Bedford et al., 2021;Carpenter et al., 2015;Ikari et al., 2009;Sánchez-Roa et al., Scuderi & Collettini, 2018).
Negative values for these parameters cannot be explained by contact area saturation and suggest the operation of other healing processes (e.g., adsorbed water effects that can produce adhesive and repulsive forces; Israelachvili, 1992).

Implications for Tectonic Faults
The rates of frictional healing and tectonic loading are one control on the recurrence interval and stress drop of earthquakes (Figure 4c) (e.g., Carpenter et al., 2012;Marone, 1998a;Marone et al., 1995;Shreedharan et al., 2023).For a given tectonic loading rate, lower healing rates (as expected in clay-rich fault gouges; c.f. Figure 1) should lead to more frequent and smaller failures, meaning lower stress drops at shorter intervals.In the limit of zero or near-zero healing, as might be expected for faults composed exclusively of smectite, slip should be accommodated by creep, or by frequent very small slip events (e.g., Shreedharan et al., 2023).In this context, our results demonstrating the systematic behavior of synthetic gouges provide a potential framework for understanding the behavior of natural fault zones (e.g., the example of the Calaveras and SAF CDZ shown in Figure 4c).
More broadly, fault locking and slip behavior are governed by the full suite of rate-and-state frictional properties, and our data support the idea that these properties are linked (Figure 3).In addition to the low healing rates, wet (or water saturated) clay-rich gouges are also characterized by low absolute frictional strength and rate-strengthening friction (with the exception of very low velocities) (e.g., Ikari & Kopf, 2017;Ikari et al., 2009Ikari et al., , 2011;;Phillips et al., 2020;Shreedharan et al., 2022Shreedharan et al., , 2023)).
Because contact area evolution is linked to the state variable in rate-and-state friction (Rabinowicz, 1951(Rabinowicz, , 1958)), the hypothesized saturation of contact area in clay-rich gouges provides a link between: (a) low μ, (b) low β, (c) longer D c , and (d) rate-strengthening friction (driven in part by small values of (b) reported here and documented by others (e.g., Carpenter et al., 2012;Carpenter, Ikari, & Marone, 2016;Ikari et al., 2016).We posit that these linked behaviors work in tandem to promote small, frequent, and/or slow earthquakes or aseismic slip on clay-rich faults, whereas quartz-rich (or clay-poor) faults should both heal more rapidly and exhibit a greater tendency for rate-weakening friction (e.g., Ikari et al., 2007;Marone et al., 1995) thus promoting larger and more infrequent stick-slip events.

Conclusions
Slide-hold-slide experiments on synthetic quartz-smectite gouges from 0% to 100% smectite reveal a systematic reduction in frictional healing rates with clay content.The reduction in frictional healing magnitudes and rates is accompanied by decreases in frictional strength, creep relaxation, and compaction during hold periods.These observations are attributed to the saturation of the real contact area within the clay-rich gouges resulting from the platy nature of aligned clay grains, which limits the potential for contact growth or evolution.These relationships indicate that clay content contributes to the overall behavior of fault zones by influencing multiple mechanical properties, and the clear systematic relationship between clay content and frictional behavior provides a framework for understanding the behavior of natural faults.

Figure 1 .
Figure 1.Mechanical data and frictional healing rates.(a) Schematic of double direct shear testing configuration with sample highlighted in red.(b) Mechanical data from friction experiments on quartz-smectite mixtures.Inset: Representative slide-hold-slide test where μ i is the friction before the hold, ∆μ c is decay in apparent frictional strength due to relaxation during the hold, and ∆μ is the increase in friction upon re-shearing.Note that the mechanical data are presented in terms of shear strain, while the inset shows the evolution in time.(c) Frictional strengthening (∆μ) and (d) creep (shear stress) relaxation (∆μ c ) from slide-hold-slide test as a function of t hold .Model fits are listed in Table1.Frictional healing rate (β) as a function of (e) clay content) and (f) steady-state friction coefficient (μ ss ) for samples from this study (colored points) and previous studies (gray points).Data from previous studies include: synthetic quartz and smectite end-members(Carpenter, Ikari,  & Marone, 2016;Marone & Saffer, 2015), IODP drill core from the Nankai(Carpenter, Ikari, & Marone, 2016;Ikari et al., 2012), Costa Rica(Carpenter, Ikari, &  Marone, 2016), and Hikurangi subduction zones(Shreedharan et al., 2022), and samples from the San Andreas Fault system(Carpenter, Ikari, & Marone, 2016).

Figure 3 .
Figure 3. Rate-and-state parameters (a) b, (b) a, (c) D c , and (d) (a-b) versus healing rate.Rate-and-state frictional parameters from Saffer and Marone (2003) and Ikari et al. (2016).Rate steps from Saffer and Marone (2003) with upstep velocities of 10 and 100 μm/s shown with open and closed symbols, respectively.

Figure 4 .
Figure 4. Conceptual model for the reduction in healing rates in clay-rich gouges.(a) Cross-section view of quartz (red) and clay (blue) grains within the gouge before (t 0 ) and after (t 1 ) a hold.(b) Plan view of quartz-quartz and clay-clay contacts within the gouge increasing with time.In quartz-rich gouges, quartz-quartz contacts progressively grow with time.In clay-rich gouges, the contact area between the aligned clay grains does not significantly change.(c) Schematic example illustrating differences in expected evolution of stress (loading) and strength (healing) for clay-rich and clay-poor faults.Frictional healing rates reported for the Calaveras Fault (clay-poor; 0.008/decade) and San Andreas Fault Central Deforming Zone (clay-dominated; SAF CDZ) (0.0005/decade) (using data fromCarpenter, Ikari, and Marone (2016) and assuming 50 MPa effective normal stress) relative to estimated tectonic loading rate, for v = 3 cm/yr and G = 30 GPa.Stars indicate fault failure when stress overcomes fault strength.

Table 1 Friction
Experiments and Slide-Hold-Slide Parameters Figure 2. Creep relaxation rate and compaction.Creep relaxation rate (a) as a function of clay content for samples from this study and previous studies; and (b) versus healing rate.β c scales with β and both decrease with increasing clay content.(c) Frictional restrengthening versus layer-normal compaction for each individual slide-hold-slide test.Note the strong positive correlation, indicating that healing is greater when creep compaction, which is equal to layer-normal strain, is larger.