Phase-transformation-induced lubrication of earthquake sliding

Abstract

Frictional failure is not possible at depth in Earth, hence earthquakes deeper than 30–50 km cannot initiate by overcoming dry friction. Moreover, the frequency distribution of earthquakes with depth is bimodal, suggesting another change of mechanism at about 350 km. Here I suggest that the change at 30–50 km is from overcoming dry friction to reduction of effective stress by dehydration embrittlement and that the change at 350 km is due to desiccation of slabs and initiation by phase-transformation-induced faulting. High-speed friction experiments at low pressure indicate that exceeding dry friction provokes shear heating that leads to endothermic reactions and pronounced weakening. Higher-pressure studies show nanocrystalline gouge accompanying dehydration and the highest pressure experiments initiate by exothermic polymorphic phase transformation. Here I discuss the characteristic nanostructures of experiments on high-speed friction and high-pressure faulting and show that all simulated earthquake systems yield very weak transformation-induced lubrication, most commonly nanometric gouge or melt. I also show that phase-transformation-induced faulting of olivine to spinel can propagate into material previously transformed to spinel, apparently by triggering melting analogous to high-speed friction studies at low pressure. These experiments taken as a whole suggest that earthquakes at all depths slide at low frictional resistance by a self-healing pulse mechanism with rapid strength recovery.

This article is part of the themed issue ‘Faulting, friction and weakening: from slow to fast motion’.

1. Preamble

This paper explores the hypothesis that all earthquakes of significant size slide on a lubricated surface that is generated by phase transformation in the process zone as a slip patch expands. Supporting this hypothesis is the observation that this is the case in all experiments simulating earthquake slip, whether they be quasi-static experiments at high pressure or high-speed friction experiments at low pressure. In these experiments, the transformation-induced lubricant is most commonly a nanocrystalline solid of extremely low viscosity but, depending on rock type, could be a melt or phase peculiar to a specific chemistry such as a hydrous silica gel in quartz. All of these experiments display a pronounced stress drop due to lubrication, in contrast to seismically inferred small stress drops, suggesting that earthquakes slide by rapidly healing slip pulses reminiscent of dislocations in crystals.

2. Introduction and earthquake initiation

Virtually all earthquakes at shallow depths occur on pre-existing faults and it is the consensus view that sliding initiates when static friction is overcome locally, either by stress rising to that level or by fluid pore pressure rising sufficiently that the effective stress exceeds static friction. Laboratory experiments show that the level of static friction rises rapidly with increasing normal stress on a simulated fault or a potential fault in unbroken rock. Because both pressure and temperature increase with depth in Earth and the flow stress of rocks decreases exponentially with temperature, at some depth there must be a crossover where the level of stress needed to overcome friction is higher than that required to initiate flow. That crossover in Earth depends on rock type but is approximately 30–50 km at maximum for dry rocks and correlates with the maximum focal depth of earthquakes on the top surface of subducting slabs of lithosphere. At greater depths, rocks become ductile and one might expect there would not be any more earthquakes. However, we have known for almost 100 years that earthquakes within cold subducting slabs of lithosphere are to a first approximation seismically indistinguishable from shallow ones and extend almost to 700 km, coinciding with the bottom of the mantle transition zone (MTZ), where they stop abruptly; no earthquakes have been recorded from the lower mantle (figure ;1). Taken together, these arguments suggest that earthquakes must initiate differently above and below approximately 50 km but the physics by which they slide could be the same.

Figure 1.

Depth distribution of earthquakes. Semilog plot of the number of earthquakes worldwide per year of magnitude ≥5 in concentric spherical shells of 10 km thickness. Exponential decline between 40 km and 300 km suggests exhaustion of the mechanism; rise in frequency again below approximately 350 km suggests two different physical mechanisms of initiation. (modified after ref. [1]). (Online version in colour.)

The mechanism(s) by which earthquakes could initiate at depths greater than approximately 50 km was a major paradox in geophysics from the time of discovery of deep earthquakes (early 1920s) until failure induced by dehydration of serpentinite was discovered in the 1960s [2] and transformation-induced faulting was discovered during the olivine–spinel transformation in the late 1980s ([3–5] and references therein). Both of these failure mechanisms involve a phase transformation that, in the former case, initiates faulting by generation of a high-pressure pore fluid and, in the latter case, generates an extraordinarily weak nanocrystalline product phase that initiates failure and lubricates sliding without the presence of a fluid. A third mechanism has been proposed for deep earthquakes [6,7] but could apply to any depth ≥50 km, in which a hypothetical pre-existing shear zone is subject to runaway shear heating that could raise the temperature to melting, at which time great weakening would occur and an earthquake might be generated. There is no experimental evidence for such instability but shear melting can be forced upon rocks or other materials by subjecting them to boundary conditions that result in extreme shear heating. Note that this potential earthquake mechanism also involves a phase transformation (melting). We will see examples of shear melting later in this paper. A possible exception to these systematics is a theoretical model of intermediate-depth earthquakes [8] that invokes recrystallization of olivine but does not have experimental confirmation.

The pattern of earthquake frequency with depth (figure 1) also has long suggested (even prior to the plate tectonic revolution [9]) that deeper earthquakes must initiate differently than shallow earthquakes but the dividing line appears to be at approximately 350 km rather than the 50 km referred to above. If, however, one attributes the earthquakes between 50 and 350 km to the dehydration embrittlement mechanism [10–13], the only fundamental difference between those earthquakes and the shallower ones could be that most of the shallow ones (hypocentres less than approx. 50 km) initiate by stress overcoming static friction and that the deeper ones initiate by reduction of the effective stress by fluid production from phase transformation. The exponential decline in earthquake frequency from the surface to more than 300 km suggests that whatever is enabling them is being exhausted. One explanation for that could be decreasing water content of the subducting slab, hence exhaustion of the critical component—hydrous phases that can dehydrate, providing the necessary pore pressure to trigger faulting by lowering the effective normal stress. It is known that earthquakes in subducting lithosphere occur progressively further into the slab with increasing depth, which is consistent with dehydration reactions being triggered by warming due to heat flow into the slab as subduction proceeds. These earthquakes generally describe two zones, one progressing into the subducting slab through the oceanic crust and the other initiating at approximately 40 km within the subducting lithospheric mantle [11,14,15]. Indeed, that is the mechanism now favoured by the majority of those who work on intermediate-depth earthquakes (50–300 km). Moreover, recent evidence shows that slabs are ‘dry’ below approximately 400 km [16,17], strongly supporting slab desiccation as the explanation for the exponential decline in earthquake frequency. An important contribution to the conclusion that deep slabs are ‘dry’ is the discovery of metastable olivine in five cold subducting slabs [17–21] and imaging of the metastable wedge beneath Japan [17]. The metastable persistence of olivine into the MTZ is only possible if the water content of deep cold slabs is below approximately 80 ppm [22].

The deeper peak in earthquake frequency shown in figure 1 is completely contained in the MTZ. The phase-transformation-induced failure mechanism discovered in my laboratory in 1989 during the olivine → spinel transition in Mg₂GeO₄ [3] and subsequent confirmation of the same failure mechanism during the olivine → wadsleyite transformation in olivine of mantle composition, (Mg,Fe)₂SiO₄ [4], can operate only during exothermic polymorphic phase transformations with a volume change (see reviews [10,23] and references therein). This mechanism therefore is a perfect match for the needed additional earthquake mechanism; it operates during the exothermic olivine reactions to its two high-pressure polymorphs, wadsleyite and ringwoodite, in the MTZ and cannot operate in the lower mantle because if ringwoodite were to be carried into the lower mantle, the resulting phase transformation is both an endothermic and a disproportionation reaction, neither of which can support the instability [23]. Moreover, even if olivine were to be carried metastably all the way into the lower mantle, the reaction would then be exothermic but still would be a disproportionation reaction that, because of the diffusive step in the separation of chemical components from one phase to two, is too slow to enable the thermal runaway necessary for the instability [23,24]. In areas of subducting slabs that are too cold and dry for olivine to react to its polymorphs, the kinetics of dissolution of pyroxenes (enstatite and diopside) into garnet are also likely to be inhibited, leaving them as metastable phases as well as olivine [25]. Whether or not the pyroxenes could also participate in phase-transformation-induced faulting remains unknown; the metastable enstatite → akimotoite reaction is exothermic and has a volume change but enstatite rarely dominates the mineralogy of mantle rocks, hence it is unlikely that its breakdown could independently trigger faulting.

Briefly, the mechanism for phase-transformation-induced faulting in the laboratory is the following [3,10,23]: by inference, when metastable olivine under stress in subducting lithosphere warms to a temperature at which nucleation of the stable phase (wadsleyite in the upper MTZ, ringwoodite in the lower MTZ) is just possible but grain growth is not yet possible, runaway nucleation occurs locally, driven by the exothermic nature of the reaction and stress concentrations caused by the approximately 6% reduction in specific volume. The result is growth of lenses of the stable phase that nucleate on grain boundaries in olivine or interphase boundaries with pyroxene and grow normal to the maximum principle stress (figure 2a,b). The lenses assume the shape of fluid-filled cracks because the nanocrystalline filling within them (grain size approximately 10 nm) has a very low viscosity; they orient preferentially normal to maximum compression because their contents are more dense than the olivine from which they form. The result is very strong compressive stresses at the lens tips. Because of their orientation, shape and compressive stresses at their tips, we call the lenses Mode I anticracks, following [26]. The instability that leads to self-organization of the anticracks and results in compressive failure by faulting requires an exothermic polymorphic transformation (figure 2c; left side). Because the reaction that marks the base of the upper mantle (ringwoodite → bridgmanite + ferropericlase) is both endothermic and produces two phases from one (disproportionation), the instability would not be supported in the lower mantle and earthquakes nucleated by this mechanism are not possible in the lower mantle (figure 2c; right side). Thus, the exothermic polymorphic phase transformation defining the top of the MTZ would enable this mechanism in cold subducting slabs and the endothermic disproportionation reaction at the bottom of the MTZ would terminate operation of this mechanism. It therefore fulfils the two major requirements for the deeper branch of the bimodal distribution of earthquake frequency with depth (figure 1). An additional, related, mechanism has been discovered in Mg₂GeO4 but not confirmed in silicate olivine [27,28]. This mechanism shows only rare anticracks but abundant dislocation pileups in olivine that have been transformed to nanocrystalline spinel due to the high internal stresses within the pileups. This alternative mechanism for generating nanocrystalline spinel is also self-organizing and leads to macroscopic faulting. This latter mechanism requires high stresses in olivine that produce pile-ups prior to phase transformation. Whether this alternate phase-transition-induced instability could also function in Earth depends on how high stresses get during subduction.

Figure 2.

Description, imaging and thermodynamics of Mode I anticracks. (a) Model of anticrack. During an exothermic polymorphic phase transformation under nonhydrostatic stress, rapid nucleation of nanometric crystals of a more dense phase in a less dense phase of the same composition (ΔV < 0) occurs preferentially on planes perpendicular to the maximum compressive stress and yields a crack-shaped lens. The lens takes the shape of a crack because its nanometric contents have a very low viscosity, hence exert vanishingly small tangential tractions on the surface. As a consequence, lenses have very strong compressive stresses at their tips through which they communicate with other anticracks and self-organize to nucleate faults. (b) Back-scattered scanning electron microscope image of Mg₂GeO₄ spinel-filled anticracks (light grey) growing on Mg₂GeO₄ olivine grain boundaries (medium grey). Very bright, rounded objects are pyroxene crystals. Bright objects on right are charging artefacts around holes in surface. (c) Thermodynamic explanation for why rapid nucleation occurs in metastable olivine under conditions where the kinetics of nucleation are difficult (left side of diagram). Because the reaction is exothermic, each nucleus induces a small increase of temperature that increases the local temperature and leads to runaway nucleation. Endothermic reactions (right side of diagram) do not lead to similar runaway. (modified after ref. [23]). (Online version in colour.)

Given the thickness of subducting slabs (approx. 100 km) and the confinement of parts of the slab cold enough to inhibit breakdown of olivine in the slab's centre, it has been argued that this mechanism is limited in its ability to accommodate large earthquakes that involve slippage along tens of kilometres of fault surface and cut across the slab. It was suggested by [29] that deep earthquakes initiated by any mechanism could potentially ‘burn’ their way through large distances by runaway shear heating that leads to melting that lubricates sliding. If an earthquake initiated by transformation-induced faulting could induce shear melting when it encounters parts of the slab that have already transformed to one of the high-pressure phases, that could provide the environment described in [29], thereby circumventing the potential problem of insufficient metastable olivine to support long-distance propagation.

We have tested that possibility in the laboratory. We prepared a specimen of Mg₂GeO₄ (the system in which transformation-induced faulting was originally discovered) in which the upper half was the olivine phase and the lower half had been previously transformed to the spinel phase. We then performed a deformation experiment at a confining pressure of 1 GPa and temperature of 1200 K, the faulting ‘window’ in the Mg₂GeO₄ system [30]. Optical examination of the specimen afterward (figure 3) showed that faulting had originated in the upper half of the specimen, as expected, and the fault propagated through the (horizontal) interface into the spinel half and out the side of the specimen a significant distance away, cutting through the platinum capsule wall in the process. However, there is no fault zone visible between the offset of the interface with olivine and the offset of the specimen edge either in plane-polarized light (figure 3c) or when viewed between crossed polarizers (figure 3d). The spinel crystal structure is isometric (cubic), hence it is optically isotropic and appears black when viewed between crossed polarizers. As seen in figure 3, however, rather than appearing totally black, the spinel portion of the specimen appears like a ‘starry night’, with very small bright spots scattered throughout. The bright spots are small birefringent pyroxenes included among the olivine in the starting material that are not involved in the phase transformation. We noted that there is a narrow band along the line where the fault had to be that took a much better polish than the rest of the spinel specimen (figure 3c) and appeared much blacker between crossed polarizers (figure 3d). The latter effect is a result of the birefringent pyroxenes being fewer and smaller than elsewhere in the spinel. An interpretation that fits all of these observations is that the fault zone through the spinel was partially melted during sliding, reducing the number and size of pyroxenes and leaving a trace consisting partially of glass that is easily polished and optically isotropic. This process is closely analogous to that envisioned by [29], although our expectation is that the fault zone in nature would be probably thinner than they assumed, perhaps on the scale of a few centimetres, and not heated to the liquidus.

Figure 3.

Propagation of faulting initiated in olivine into material previously transformed to spinel. (a) Cartoon of a composite olivine/spinel specimen after deformation and faulting; fault initiated upper right in olivine, offset interface with spinel near centre and offset left side of specimen and Pt capsule (black). (b) Optical image of central 2/3 of a composite specimen composed of Mg₂GeO₄ olivine (top) and Mg₂GeO₄ spinel (bottom) viewed between crossed polarizers. (c) Same as (b) but with interface between materials indicated with dotted line; fault indicated by dashed line. Black areas on both sides of specimen (labelled Pt) are the walls of the platinum capsule. Bright speckles in lower portion of images are birefringent pyroxene crystals among the spinel that is optically isotropic and therefore black. Note great reduction of number and size of pyroxene ‘speckles’ within fault zone in spinel. (d) Detail of lower portion of (b) but in plane-polarized light. Offset along the fault is illustrated by offset of interface top right and disrupted Pt area at lower left. The variable ‘graininess’ of the specimen image is primarily due to scattering of light due to surface topography; the previously produced fine-grained spinel tends to pluck more easily than coarser-grained olivine during polishing. Note that the fault zone polished much better than its surroundings. (e) Crossed polarizers. Fault zone is much less ‘speckled’ by birefringent pyroxene crystals than remainder of the spinel polycrystal, and those that are present are smaller than outside the fault zone. See discussion in text. (Online version in colour.)

Thus, modern laboratory experiments provide explanation for how and why earthquakes could initiate differently above and below approximately 50 km. Nevertheless, seismologists have long pointed out that seismic signals from earthquakes at all depths are highly similar, suggesting that earthquakes may slide by the same physics regardless of the way they begin. The simplest way to address this question would be to examine the products of earthquake sliding (fault gouge). For real earthquakes, that is very difficult (although careful examination of faults has revealed tantalizing suggestions [31–34] but it is straightforward for simulated earthquakes in the laboratory [31,35]).

As a consequence, here we look at two case studies with an eye to discover similarities and differences between high-pressure faulting analogues and low-pressure faulting analogues, with an eye to understand the physics of sliding in the latter based on understanding of the physics of sliding in the former (see also [3]). Firstly, we examine at very high magnification the gouge from a representative high-pressure faulting experiment because it simulates the special conditions necessary to capture the physics that allows faulting to occur at great depth and the gouge can be shown to exhibit extremely low frictional resistance despite extreme normal stress and lack of a fluid phase. Secondly, we examine, at the same high magnification, the sliding surface of a high-speed friction experiment on dolomite because it simulates the true speed of earthquake sliding necessary to capture the physics of shallow earthquakes (less than 50 km).

3. Earthquake sliding

(a). Mechanism of sliding during high pressure faulting

High-pressure phase-transformation-induced failure was discovered in Mg₂GeO₄ during the transformation from olivine to spinel [3] and confirmed in the olivine → wadsleyite transformation the following year [4]. Here, I use experiments in this system to put an upper bound on the sliding resistance of the gouge produced during faulting in this system and to determine the mechanism of that sliding [31]. Such faulting produces an extremely thin through-going fault zone (figures 4 and 5) and is accompanied by a large and rapid stress drop and abundant acoustic emissions (figure 5a). The implication of those observations is that the nanocrystalline gouge that fills the fault zone (figure 5c) has an extremely low viscosity, yielding a very small effective friction coefficient, despite the very great normal stress across the fault.

Figure 4.

Transformation-induced fault in Mg₂GeO₄ olivine. SEM back-scattered image shows a faulted specimen with a crack running on the lower side of the fault zone. Smooth grey areas are olivine (ol); bright areas are MgGeO₃ pyroxene (pyx); mottled grey areas are a mixture of olivine and spinel (ol + sp). Note small pyroxene cut and offset 3 µm by fault zone that is below the resolution of this image. ‘FIB Foil’ denotes location where a foil for TEM was cut by a focused ion beam. All open cracks formed during decompression of the specimen after termination of the experiment. (modified after ref. [31]).

Figure 5.

Combination of SEM and TEM images of a transformation-induced fault. (a) Stress-strain curve (yellow online; light grey in print) of a representative faulted Mg₂GeO₄ olivine specimen showing acoustic emissions (blue online; dark grey in print) coincident with large stress drop. (b) Copy of upper left corner of figure 4 showing location where FIB foil was cut perpendicular to the surface. (c) TEM image of the FIB foil. The upper half of this image has moved toward the viewer and the lower half has moved away from the viewer. The open decompression-induced crack noted in figure 4 runs along the lower side of the spinel-filled fault zone (outlined by white dashed lines). Note that fault zone is only 70 nm thick. (d) Detail of the fault zone. Figure modified after figures in ref. [24]; see text and ref. [31] for discussion. (Online version in colour.)

Examination of the fault zone in the transmission electron microscope (TEM) (figure 5d) shows that the individual crystals of the gouge are equant in shape and the largest grains are approximately 15 nm diameter. Detailed examination shows complete lack of porosity. The combination of these observations restricts the mechanism of flow to strain accumulation by grain-boundary sliding but does not elucidate whether the rate-controlling mechanism is by a dislocation-controlled or diffusion-controlled mechanism.

To quantify this effect, we note that frictional sliding gives rise to shear heating (ΔT) that can be described by: ΔT = μσ_nd/ρ_ch [36], where μ is the coefficient of friction, σ_n is the normal stress across the fault, d is the slip distance, ρ_c is the specific heat capacity and h is the thickness of the fault zone. For the experiments involved here, all parameters in this equation are known from the material involved and the experimental results except for ΔT and µ. Because the phase relations for Mg₂GeO₄ are known, along with the specific conditions of the experiment shown in figures 4 and 5, we can estimate an upper bound on ΔT [24]. The faulting occurs rapidly (figure 5a) and Mg₂GeO₄ is a poor thermal conductor, hence we assume adiabatic shear heating and use the knowledge that the conditions of faulting are at 1.3 GPa confining pressure, 1200 K, which is 230 K below the olivine–spinel phase boundary. As a consequence, if ΔT in the fault zone were to exceed 230 K during sliding, the spinel structure of the Mg2GeO4 would become unstable relative to that of olivine, the instability-enabling reaction would stop and sliding would be impossible because of loss of lubrication. Indeed, microscopic examination of the fault and its environs at the highest resolution confirm that there is no sign of reversion to olivine. Therefore, 230 K provides an upper bound on ΔT and we conservatively assume that the temperature rise just reached this maximum amount. The measured differential stress at failure was 1.5 GPa on top of a confining pressure of 1.3 GPa. We therefore conservatively set σ_n at 1.5 GPa. The observed fault thickness is h = 70 nm; the fault offset from figure 4 is d = 3 µm, and the specific heat capacity, ρ_c, is 2–3 MPa K⁻¹ (we use 3 MPa K⁻¹). Using these parameters, we solve the shear heating equation for μ, yielding μ < 0.01 [31].

(b). Mechanism of sliding during high-speed friction experiments

Over the last 20 years, high-speed friction experiments have become abundant and, remarkably, large numbers of experimentalists report essentially the same result, independent of rock type studied ([35] and references therein). In particular, all experiments reported by [35] record that at the end of each experiment, the phase(s) in the sliding surface were not there when the experiment began; in all cases, the sliding surface consisted of the product(s) of a phase transformation triggered by extreme shear heating in the starting material within second(s) of experiment initiation. In most cases, the sliding surface consists of oxides produced by decomposition of the starting material (hydrous phases or carbonates), but in one case, a gabbro, there were no volatile-bearing phases to be broken down and the sliding surface melted, yielding a glass upon experiment termination.

To provide examples of the phenomena of decomposition and melting during high-speed friction, I show in figure 6 experimental results on both dolomite and granite. Experimental data (figure 6a) and images (figure 6b,c) of a high-speed dolomite experiment [31] show that the sliding surface at the end of the experiment consisted of a nanometric gouge. The experiment slid 0.6 s at a rate of 0.6 m s⁻¹ and yielded a sliding surface with grain size ≤25 nm (figure 6b,c), consisting of only magnesium and calcium oxides [31]. Thus, the temperature on the sliding surface was raised to more than 1000 K by shear heating during a very small fraction of that time (≤0.002 s estimated from figure 6a) and the decomposition reaction MgCa(CO₃)₂→ MgO + CaO + 2CO₂ ran to completion by 0.006 s. Away from the sliding surface, the temperature also was raised above the decomposition temperature for dolomite but beyond 200 nm of the surface, small relicts of dolomite still remained at the end of the experiment (figure 6c,d). On the other hand, a similar experiment run on Sierra White Granite had no endothermic reaction to capture the heating spike, allowing the temperature to run all the way to melting in approximately the same time (figure 6e).

Figure 6.

High-speed friction experiments on dolomite and granite. (a) Plot of friction and velocity against slip distance in high-speed friction experiment. One half of a specimen of dolomite was rotated while pressed together with another half of the specimen with a normal stress of 28 MPa (a large normal stress by high-speed friction standards). Experiment slid 0.6 m at a sliding speed of 0.6 m s⁻¹. Note that the coefficient of friction rose to 0.8 before sliding began and then dropped to 0.3 in 0.006 s, trended down to about 0.2 and then rose back to about 0.4 as sliding ceased. (b) High-resolution SEM image of sliding surface showing an average grain size of approximately 10 nm. (c) TEM Cross section of the sliding surface (bottom) shows nanocrystalline solid of grain size less than 25 nm in layer about 200 nm thick at the surface, with some dolomite relics further away. Chemical analysis of the surface [24] showed surface composed of MgO and CaO. (d) Further back from the surface (approx. 400 nm), nanometric oxides can be seen ‘budding off’ of a dolomite remnant (black). (e) Sliding surface of Sierra White Granite specimen showing glass with tiny quench crystals within it (white dots). Panels modified after various figures of Ref. [31]. (Online version in colour.)

4. Discussion and Conclusion

Fifty years ago, two of the major paradoxes in geophysics concerned earthquakes. Paradox no. 1: At shallow depths, the consensus was (and remains) that earthquakes initiate by the shear stress on the fault overcoming static friction. However, the stress drops universally observed by seismology and slow friction experiments are small and the normal stress at depths where earthquakes start is such that, if Byerlee friction applies [31], extreme shear heating should ensue. The lack of a heat-flow anomaly along the San Andreas or other large faults indicated that there was something wrong with this scenario. Paradox no. 2: At depths greater than about 50–70 km in Earth, the normal stress required to overcome friction on any fault or potential fault is higher than the flow stress of all common rocks, hence earthquakes should not occur at greater depths. However, earthquakes were known to occur to depths approaching 700 km and no mechanism to initiate or slide them was known. Paradox no. 2 was overcome when high-pressure experiments demonstrated phase transformations under stress could enable faulting and generate acoustic emissions [2,3,28,37]. Paradox no. 1 was overcome when it became clear from friction experiments conducted at earthquake sliding rates that almost immediately after friction is overcome and sliding begins, extensive shear heating begins and great weakening ensues (see review in [35] and references therein), vastly reducing the total heat production and explaining the lack of thermal aureoles around large mature faults. Nevertheless, the physical mechanism(s) of earthquake sliding remain obscure.

The earthquake simulation experiments summarized here, whether at high pressure or at high speed, are fundamentally dependent on phase transformation; in all cases, the sliding surfaces consist of new phases that are generated early in the experiments—either nanometric oxides or melt, both of which radically reduce the frictional resistance (or, in special circumstances like quartz, the lubricating material could be different [38] but the basic process is the same). Under low pressures where failure is initiated by overcoming of static friction, shear heating immediately ensues and the temperature rises until it encounters a subsolidus endothermic reaction to a nanometric solid [35,39] or to the melting point. At high pressures, an exothermic polymorphic phase transformation is required to initiate faulting by providing a reaction-induced thermal runaway resulting in a nanocrystalline product that lubricates sliding from the moment of fault initiation. If faulting generated and propagated in this way encounters a region where the polymorphic transformation has already occurred, shear heating analogous to that occurring in high-speed experiments at low pressure can be initiated that drives the temperature to melting and fault propagation continues if the ambient stresses are sufficient. That is apparently what happened in figure 3.

These similarities strongly suggest that, even though earthquakes at low pressure and high pressure must initiate by different physical mechanisms, their sliding mechanisms must be very similar. In particular, the weak, rounded, fully dense, nanometric transformation products in high-speed experiments strongly suggest that, like the fault gouge at high pressure, those materials provide lubrication as the gouge flows by grain boundary sliding. The mechanism underlying this profound weakening in nanometric solids is not yet fully understood. In materials science there is an hypothesis called the ‘reverse Hall-Petch effect’ [40] that is very controversial because some experimentalists find nanometric materials to be very weak and others observe no such weakening. Apparently, simulated earthquake experiments find the conditions for great weakening automatically; perhaps that is because the grain size they achieve is so small (≤15–25 nm).

Another change in our understanding of earthquakes at all depths is that both transformation-induced faulting and high-speed friction experiments show a large stress drop accompanying initiation of sliding due to the extreme weakness of the gouge material. However, seismically measured stress drops are usually small. This inconsistency suggests that, rather than recording the dynamics of earthquakes, seismically measured stress drops record the average stress drop on a fault after sliding has ended locally. An initially large stress drop that is followed closely in time by a restrengthening strongly suggests that earthquakes propagate as pulses, much like dislocations in crystals ([41] and references therein).

Figure 7 shows a conceptual model of pulsed earthquake sliding adapted from [41]. At a given point along an existing fault, a new slip pulse arrives rapidly from the left. The front end of the pulse carries a stress that is at or above the frictional shear strength of the fault and, as it passes, the stress drops precipitously (analogous to the left side of figure 6) to less, perhaps much less, than the ambient shear stress at the point in question. Almost instantly, the strength of the fault begins to recover and, as the pulse passes or shortly thereafter, the strength of the fault returns to essentially the same value as before the pulse arrived and the local shear stress recovers to a value somewhat smaller than the local stress prior to slipping. The mechanism of restrengthening also remains unknown but from our ongoing, unpublished, experimental results, we suggest that it probably is a result of cooling or grain growth of the gouge material.

Figure 7.

Pulse model for earthquake sliding. From the nature of the data in this paper, we conclude that earthquakes slide by self-healing pulses analogous to slip dislocations in crystals [41]. The leading edge of the pulse arriving from the left carries a stress at or above the frictional strength of the fault zone. It triggers frictional failure that, in turn, leads to a rapid cascade of damage and heating, culminating in a temperature spike that rises to the lowest-temperature endothermic phase transformation available that causes a precipitous drop in friction coefficient and thus a large drop in shear stress to below, perhaps way below, the ambient stress at this point (Δσ_p). Shortly after the large stress drop, self-healing begins and defines the trailing edge of the slip pulse, leading to recovery of a large fraction of the frictional strength and a small drop of the local shear stress. It is not yet clear how rapid healing is and therefore how wide the slip pulse is. Our current experimental work suggests that the mechanism of recovery is grain growth of the gouge or cooling, whichever comes first. Figure modified after ref. [41]. (Online version in colour.)

Chester et al. [32] and references therein, and Green et al. [31] provide some observations from the Punchbowl Fault, a deeply eroded ancestral branch of the San Andreas Fault of California showing an extremely thin fault zone across which tens of km of slip appear to have occurred [32], suggesting that nanometric gouge may have been important in fault lubrication. Close examination of the fault zones of other deeply eroded faults [33,34] yield further insight into the connection between laboratory experiments and natural earthquakes.

Data accessibility

The only original data presented in this paper are the images shown in figure 3. That figure is the sum total of the data; the specimen was lost while preparing it for SEM shortly after those images were made.

Competing interests

I declare I have no competing interests.

Funding

This paper is partially based on work supported by the US National Science Foundation under grants no. 1247951 (PIs, H.W.G. and Z. Reches) and no. 1015624 to H.W.G.