The strengths of crustal faults inferred from borehole-derived heat-flow measurements and maximum stress orientations are lower than those determined from laboratory measurements. Because long-term changes of fault strength cannot be directly monitored using geophysical techniques, the causes of fault weakening are not well understood. We provide laboratory evidence that supports the view that long-term weakening of the frictional strength of faults is caused by microfracturing at asperity contacts, which is a result of crack growth at subcritical stress levels. We conducted triaxial compression tests on mylonite samples at successively higher temperatures from room temperature to 600°C to accelerate reaction processes so that they were observable at laboratory time-scales. Our results suggest that long-term reductions of fault strength are related to chemical reactions that take place in the presence of water. In the presence of water, frictional strength decreased as temperature increased, whereas it changed little in the absence of water. Thus, the presence of fluids has an important influence on changes of fault strength, and it is not only high fluid pressure that is important.
 Frictional strengths of crustal faults are generally weaker than the stress magnitudes of the surrounding crust and lower than the frictional strengths of rocks measured in the laboratory [Byerlee, 1978; Zoback et al., 1987; Saffer et al., 2003]. Fault zones are sites of repeated frictional movements over hundreds of years, suggesting that they exhibit persistent weakness relative to the surrounding crust. Although, there are conflicting viewpoints on the weak-fault hypothesis [e.g.,Scholz, 2006], we aim here to address it in terms of crustal geomechanics. The weakness of faults has major implications for the mechanics of earthquakes and the forces that move lithospheric plates. The frictional strength of a fault is a critical factor in earthquake generation, as is the regional state of stress, which causes energy to be stored in the form of elastic strain.
 When a fault starts to slip, some of the stored energy is used to overcome the initial resistance, the frictional strength, which must be overcome before shear failure occurs. Many explanations have been proposed for fault weakness: high fluid pressures [Rice, 1992; Tobin and Saffer, 2009]; intrinsically low coefficients of friction of fault-zone materials [Moore and Rymer, 2007; Collettini et al., 2009; Lockner et al., 2011]; low-stress deformation mechanisms caused by solution-transport reactions [Bos and Spiers, 2001]; and dynamic weakening mechanisms such as shear heating [Hirose and Shimamoto, 2005], thermal pressurization [Wibberley and Shimamoto, 2005], and acoustic fluidization [Melosh, 1996]. However, all of these require very specific fault-zone conditions, for example, an abundance of small amounts of particular materials. Thus, there must be a more general explanation for the weakening of the resistance to shear movement of a fault. Earthquake researchers are faced with the problem of explaining why fault zones are substantially weaker than the surrounding crust [Zoback et al., 2009]. Here we propose that the long-term mechanism that weakens the frictional strength of faults is microfracturing of asperity contacts, and that it is caused by crack growth at subcritical stress levels. This long-term mechanism has not previously been considered in detail because direct evidence for it cannot be obtained by field observation. We show laboratory evidence to support this new model.
 The overall shear strength at an interface is determined by the area of contact and the strength of the contact. In crustal faults, areas where fault surfaces are in contact are referred to as asperities [Scholz, 2002]. Fracture at asperities is the essential process by which the frictional strength of a fault is overcome to allow slip to occur. We propose that microfracturing at asperity contacts is responsible for long-term changes of fault strength. Such microfracturing is caused by subcritical crack growth, which is assisted by the interaction of water molecules with the strained apexes of cracks. For the specific case of vitreous silica exposed to water, strength is reduced in the presence of water by stress corrosion in response to the chemical reaction [Anderson and Grew, 1977; Freiman, 1984]
 This model has implications for the long-term strength behavior of a wide variety of brittle materials [Michalske and Freiman, 1982]. The occurrence of reactions between aqueous fluids and minerals at depth is universally accepted; softening of rocks in fault zones by such reactions may be important in faulting processes [Wintsch et al., 1995]. Experimental evidence shows that subcritical crack growth assisted by stress corrosion is responsible for changes of rock strength in a brittle regime [Brace and Jones, 1971; Olsson, 1991; Masuda, 2001]. Because most geological processes have durations of hundreds to a thousand of years, we need to accelerate these chemical reactions so that they can be observed in the laboratory. The rate of a chemical reaction is related to temperature as expressed by the Arrhenius equation:
where A is a constant, H is activation energy, R is the gas constant, and T is absolute temperature. We accelerated the process of crack propagation assisted by chemical reaction at crack apexes by increasing the temperature of the reaction of equation (1) [Paterson, 1990]. If the mechanism we propose for the long-term weakening of fault strength is valid, we should observe an effect of water on long-term frictional strength. To test our hypothesis, we measured the changes of frictional strength on a rupture surface with increasing temperature under wet and dry conditions.
 We carried out a series of conventional triaxial compression tests on fault rock (mylonite) samples at a constant strain rate of 5.5 × 10−6 s−1 in an Ar gas medium. The mylonite samples were from an exposed brittle–ductile transition zone in the Hatagawa fault zone in northeastern Japan [Shigematsu et al., 2009]. We analyzed the stress–strain relationships and frictional behavior of rupture surfaces formed during the tests. Cylindrical mylonite samples were placed inside a 2-mm-thick graphite sleeve that acted as a buffer to prevent the surrounding 0.25-mm-thick annealed copper jacket from rupturing (Figure 1a) [Masuda et al., 2002]. The mylonite we used was particularly useful because it has a distinct foliation that allowed us to identify the orientation of the stress field under which it was formed. We cut the cylindrical mylonite samples so that their long axes were oriented at an angle of 30° to the foliations. During the compression tests, rupture surfaces formed along the foliations (Figures 1b and 1c).
 The rupture surfaces we created experimentally are comparable to natural faults in the deep crust: we used sample materials that resemble fault rocks (rather than pure materials such as quartz sand) and applied shear stress under high-pressure and high-temperature conditions to create rupture surfaces with the same orientation as the local stress field under which the samples were formed.
 When load was applied to the samples during compression testing, the axial stress increased steadily to a peak, whereupon rupture occurred along a fracture surface oriented at 30° to the loading direction. Strain softening behaviors were observed in all experiments. The peak stress corresponds to the fracture strength of the sample (Figure 1d). The flat part of the stress–strain curve after rupture provides a measure of the frictional strength of the rupture surface (Figure 1d).
 Before measuring the effects of water on frictional strength at high temperatures, we measured the effects of temperature alone on dry rock strength (Figure 2). From room temperature to 600°C, fracture strength changed little. But at 800°C, fracture strength decreased dramatically. Therefore, we measured the effects of water on frictional strength only in the range from room temperature to 600°C. To ensure that effective pressure conditions were the same for our wet and dry tests, we measured frictional strength under dry conditions at a confining pressure of 130 MPa, and under wet conditions at a confining pressure of 200 MPa with pore-water pressure of 70 MPa.
3. Results and Discussion
 Under dry conditions, frictional strength changed little from room temperature to 600°C (Figure 3a), although stick–slip behavior (intermittent slip) was observed at room temperature and at 200°C. Under wet conditions, we observed no stick–slip behavior, but frictional strength decreased as temperature increased, suggesting that frictional strength at the fault plane was lowered by chemical processes associated with the presence of water, rather than by physical processes governed by the effective pressure law (Figure 3b). There may be greater variability of fracture strength with increasing temperature in our small experimental data set, because fracture is a more dynamic process than steady sliding. However, similar trends of fracture strength (peak differential stress) decreasing with increasing temperature were recognized. The response of differential stress to increasing axial strain at 600°C for the wet experiment was different from the response at lower temperatures. At 600°C, the stress drop at fracture during the wet experiment was smaller than the drop at lower temperatures and there was no elastic rebound (Figure 3b). Because there was little observable difference between the behaviors of fracture surfaces formed during the four wet experiments, and because there was no evidence of melting in the run at 600°C, we included the data obtained in the wet run at 600°C in our discussion of frictional strength.
 For each experimental run, we calculated friction (i.e., shear stress τ normalized by normal stress σn) on the rupture surface according to the geometry of our experimental apparatus as follows [Takahashi et al., 2009]:
where σdiff is the differential stress, θ is the angle between the ruptured surface and the loading direction (30° in this study), and Pc is the effective confining pressure. We first calculated the average differential stress and its error by taking the difference between its maximum value and the average value in the flat portion of the stress–strain curve where axial strain was greater than 4.5%. For the cases where stick–slip was observed, we used the peak stress immediately before slip occurred.
 A comparison of friction as a function of temperature for the wet and dry experimental runs (Figure 4) showed that raising the temperature accelerated the effect of water on chemical reactions at the rupture surface. We assumed that increasing temperature in the laboratory mimics long-term behavior in the field. Without water, friction changed little with increasing temperature. However, in the presence of water, friction decreased dramatically as temperature increased from room temperature to 600°C (Figure 4, wet). This result suggests that the frictional strength at the fault surface was lowered by chemical processes associated with the presence of water. Our results also suggest that friction coefficients are higher for wet faults than for dry faults at room temperature and 200°C. However, the magnitude of the errors for these data makes it difficult to consider the implications of these results.
 We propose that the mechanism that weakens the frictional strength of faults in the long-term is microfracturing at asperity contacts, and that the microfractures are caused by subcritical crack growth. Subcritical crack growth is assisted by stress corrosion, which is the interaction between water molecules and strained crack apexes. This model is supported by our experimental measurements of the effect of water on friction at the rupture surface of fault rocks at high pressures and temperatures. The long-term weakening of faults cannot be detected in situ by geophysical and geological techniques, but our laboratory results show that chemical reactions play an important role in this process. High fluid pressure may weaken rocks, but it does not fully explain the long-term lowering of the frictional strength of a fault. Fluids have an important role, but it is not only high fluid pressure that is important. The effect of the mere presence of water on chemical reactions at fault rupture surfaces appears to be important.
 The model we propose has important implications in that it explains the weakening of faults by simple mechanisms that cannot be examined by traditional geophysical data such as crustal stress measurements and earthquake source mechanisms, or by geological observations such as evidence of heating or the presence of clay minerals. Our experimental results also provide key parameters for use in computer simulations of earthquake processes that can be used to substantially improve methods of earthquake hazard evaluation.