This study experimentally simulated dehydration- and dehydroxylation-induced weakening of clay-rich gouges to investigate the potential of phyllosilicate minerals to generate abnormal pore pressure, triggering earthquakes. For the experiment, the gouge samples were subjected to sliding with heating at 9.6°C/min to 500°C under 80 MPa of confining pressure. In the undrained condition in which expulsed water was confined in the gouges, significant frictional strength reductions were found in the Na-montmorillonite gouge due to dehydration and in the kaolinite gouge due to dehydroxylation. However, frictional strength continued to increase in the fully drained condition that allowed released water to move outside the gouge. Frictional strengths under the two drained conditions were compared during dehydration and dehydroxylation reactions to estimate the pore pressure generated in the gouge zone under the undrained condition. The 6∼8 wt % of dehydroxylated water generated from the kaolinite caused a rapid increase in pore pressure to 68 MPa at 500°C. In contrast, in the Na-montmorillonite gouge, the pore pressure gradually increased to 72 MPa as the temperature rose to 500°C. Especially on a coseismic stage, these results suggest that dehydration/dehydroxylation could advance the frictional sliding acceleration induced by frictional heating, even if the fault is in a dry condition.
 The abrupt reduction of effective stress, due to the generation of high pore pressure, can create fault instability and increase the potential for an earthquake. This process has been considered to be an earthquake trigger, and this study looks at how such high pore pressure is generated in rock. Dehydration embrittlement or dehydration instability [e.g., Raleigh and Paterson, 1965; Kirby, 1995] relates to the dewatering reactions of minerals containing water (H2O) and/or having water as part of their constitutions (hydroxyl groups [−OH]). In terms of time, there are two possible processes of the pore pressure increase due to dewatering reactions. One is metamorphism in which the reaction proceeds slowly; the other involves rapid thermal decomposition induced by the frictional heating in a seismic event. The first case increases the potential for earthquake occurrence in a relatively wide range around the fault. In the second case, as an effect of thermal pressurization [e.g., Mase and Smith, 1987; Wibberley and Shimamoto, 2005], the dewatering reaction during fault sliding can further accelerate sliding. Many studies have suggested that dewatering reactions in serpentine can cause intermediate-depth earthquakes at subduction slabs [e.g., Ko et al., 1997; Wong et al., 1997; Peacock, 2001; Seno et al., 2001; Miller et al., 2003]. However, questions remain because the serpentine dehydration reaction induces a total volume decrease at more than 2 GPa of pressure (>60 km depth), which may theoretically limit the occurrence of earthquakes triggered by this process due to a negative Clapeyron curve [Wong et al., 1997; Dobson et al., 2002]. On the other hand, Jung et al.  tested antigorite dehydration under high pressure (1–6 GPa) and found brittle fractures delineated by fine-grained solid reaction products in the specimens. They concluded that dehydration embrittlement is a viable mechanism for triggering intraslab earthquakes at an intermediate depth. However, it is still unclear whether a high enough pore pressure can be generated to induce fault instability. Hirose and Bystricky  conducted high-velocity rotary shear tests on serpentine, monitoring the humidity around the serpentine to measure dehydration during shear. Their results indicated that coseismic shear heating induced serpentine dehydration and subsequent fluid pressurization in the impermeable fault zone which could promote further weakening. However, it is still unclear how much pore pressure is generated by dehydration and whether this type of dehydration makes a difference in the weakening process.
 Most research has focused on the experimental determination of clay strength and serpentinite gouges undergoing deformation under a constant temperature condition [e.g., Shimamoto, 1986; Moore et al., 1996]. Although these studies have reported increases and decreases in strength with increasing temperature, especially in regard to the temperature range for the reactions, they did not demonstrate a dynamic change in fault strength due to heating. Recently, studies have been conducted to observe the weakening dynamically induced by dehydration reactions in gypsum [Milsch and Scholz, 2005] and serpentinite [Hirose and Bystricky, 2007]. In these studies, they conducted sliding test with the specimens with increasing the temperature at a constant heating rate, and they found that strength started to decrease at the dehydration temperature for each mineral. The experimental method adopted by Milsch and Scholz  and Hirose and Bystricky  revealed how thermal decomposition affects the strength of a fault and what strength transitions occur during heating. Our idea underlying those studies was to estimate the pore pressure generated by heating minerals containing water or constitution water. Thus to address the question noted above, we tried to generate abnormal pore pressures on minerals containing water or having constitution water and examined their potential as earthquake triggers.
2. Experimental Procedure
2.1. Simulated Gouge Preparation
 Minerals including water or hydroxyls, such as clay minerals, zeolite, gypsum, serpentine, and amphibole, are commonly found in both the rock and fault rock. For this study, clay minerals are phyllosilicates that contain constitution water (−OH) in their crystals; in addition, some clays, such as montmorillonite, have interlayer water (H2O) between the clay sheets. Therefore, there are two types of clay dehydration, that of the interlayer water and that of the constitution water (usually called “dehydroxylation”). We used Na-montmorillonite-rich (from Aterazawa, Yamagata, Japan) and kaolinite-rich (from Kanpaku, Tochigi, Japan) powders to simulate fault gouges. Both powders were obtained from the Iwamoto Mineral Company (Tokyo, Japan). The average grain size, determined by a laser particle-size analyzer, was 0.43 μm for the Aterazawa Na-montmorillonite powder and 0.58 μm for the Kanpaku kaolinite powder. To compare a nonreactive material to the reactive clays, we also prepared a fine quartz gouge having grain sizes of 5 μm or slightly less (MIN-U-SIL 5®, U.S. Silica Company, Berkeley Springs, West Virginia).
 Thermogravimetry–differential thermal analysis (TG-DTA) with constant atmospheric heating of 10°C/min to 1000°C indicated that interlayer water in the Na-montmorillonite dehydrated at around 80°C, while dehydroxylation started at around 600°C and ended at approximately 700°C (Figure 4). The hydrate layers of montmorillonites are known to be quite sensitive to relative humidity [Sato et al., 1992]. As the starting condition, we created a one-layer hydrate Na-montmorillonite, having ∼10 wt % interlayer water, by placing it under 25°C and 40% relative humidity (RH) in a humidity control box. Impurity, identified as fine quartz by X-ray diffraction (XRD) analysis, comprised approximately 0.21 times the weight of the nonhydrated montmorillonite [Takahashi et al., 2007]. The dehydroxylation of our kaolinite (Al2[Si2O5][OH]2) started at approximately 450°C (see Figure 7 and Yeskis et al. ). For the Kanpaku kaolinite powder, the XRD analysis detected alunite (KAl3[SO4]2[OH]6) and quartz impurities and the following composition by weight percentages: 73 wt % kaolinite, 20 wt % alunite, and 7 wt % quartz, each with an error of ±4 wt %. The TG-DTA showed that the specimen lost 19–20% of its total weight (Figure 7). Less than 1 wt % of water was absorbed. Alunite is an insoluble mineral in water, acid, and bases unless calcined [Küçük et al., 2004; Küçük and Yildiz, 2006]. Both kaolinite and alunite include constitution water in their crystals. Moreover, the dehydroxylation reactions of both can progress rapidly at temperatures over 500°C.
 We conducted sliding deformation tests on the simulated clay gouges of Aterazawa Na-montmorillonite and Kanpaku kaolinite while elevating the temperature toward 500°C. Using this method, we were able to observe the dehydration of the interlayer water of the Na-montmorillonite and the dehydroxylation of the kaolinite. On the basis of the dehydration type and associated dehydration weakening, we could then compare strength changes in these two simulated faults.
2.2. Experimental Procedure
 For the experiments, we used a gas-medium, high-pressure, and high-temperature deformation triaxial testing machine located at the National Institute of Advanced Industrial Science and Technology (AIST), Japan [Masuda et al., 2002; Takahashi et al., 2007]. The machine included a cylindrical pressure vessel, a pressure generator system for confining pressure, and two servocontrolled pressure intensifiers that could apply a maximum of 200 MPa. Two strain gauge-type pressure transducers measured the pore pressures at both the upstream and downstream sides of the specimen to show the permeability of the specimen. Pressure measurements were accurate to 0.2 MPa. The axial load was applied with a hydraulically operated actuator and a control unit, permitting servocontrol via either displacement or load. The axial force could be measured both with an external load cell, installed in the load actuator, and with an internal load cell. The load measurements with the internal load cell were accurate to 0.5%, which influenced the error of ±0.01 in the friction coefficient. Piston displacement was measured with a linear variable differential transformer (LVDT) fitted to the load actuator outside the pressure vessel during the tests. Displacement measurements were accurate to 1 μm. We used argon gas for the confining pressure medium. The cylindrical internal furnace, which could reach 800°C (Figure 1a), was divided into upper and lower heating zones [Masuda et al., 2002]. Two thermocouples were inserted through the holes in the upper and lower pistons to points just above and below the specimen assembly. Each thermocouple measured the temperature; the temperature data were then used as feedback for independent operation of each thermocontroller. We filled a gap between the furnace bore and the copper jacket above the hot zones with a heat-insulating material to prevent convection of the argon gas.
 The specimen assembly and tungsten carbide spacers were placed into a 2-mm-thick copper sleeve, in contact with the upper and lower pistons (Figure 1a). Sliding deformation was applied on the thin zone of clay gouges (1.0 g) sandwiched between two alumina blocks with oblique surfaces at 30° to the cylindrical axis. That pair of the precut alumina blocks was sized 20 mm in diameter and 40 mm long. We prepared two kinds of alumina spacers to control the fluid transport environment around the gouge: a dense alumina block for the undrained condition (Figure 1b) and a porous block for the fully drained condition (Figure 1c). The porous alumina block initially had 17% porosity and 6.71 ± 0.91 × 10−16 m2 of permeability. The sliding surfaces of the alumina blocks were ground with #80 SiC abrasive. Prior to applying the confining pressure Pc, the pore pressure lines were vacuumed to approximately 5 kPa for 1 h. The lines were then cut from the vacuum pump. All experiments were carried out under the set of constant conditions of Pc = 80 MPa and vacuumed pore pressure. Although we attempted to monitor the pore pressure increase during the fully drained condition, the gauges for Pp (shown in Figure 1) did not respond. The total volume of the pore pressure line was much greater than the volume of dehydrated/dehydroxylated water, and thus the pore pressure increase due to release of that water to the vacuum-dried pore pressure line was lower than the accuracy of the pore pressure gauges (0.2 MPa). Axial loading by the hydraulically operated actuator produced the shear deformation in the gouge zone along the oblique surface of the alumina blocks (Figure 1a). The axial loading speed was a constant 0.1 μm/s, meaning a constant displacement velocity of 0.12 μm/s along the shear direction.
 We started to elevate the temperature around the specimen to a maximum of 500°C at a constant heating rate of 9.6°C/min after the initial elastic increase of the strength was finished. The temperature was decreased at a constant cooling rate of 3.2°C/min after maintaining a certain temperature for some displacement. Heating would irreversibly change the clay gouge both physically and chemically. To detect that effect on the gouge strength, we tried to cool the specimen toward room temperature. During heating, holding, and cooling, the differences between the temperatures monitored by the two thermocouples at the upper and lower sides of the specimen were less than 8°C, 5°C, and 20°C, respectively. We used the average of the upper and lower side temperatures as the temperature value. For reference, we also tested a normal deformation run (i.e., a test under room temperature using dense alumina blocks) for each gouge. Experimental products were analyzed by XRD and TG-DTA to deduce the dehydration and dehydroxylation reactions and volumes.
3.1. Dehydration of the Interlayer Water of Na-Montmorillonite Gouge
3.1.1. Frictional Strength Development due to Dehydration With Changing Temperature
Figure 2 presents the dehydration experiment results for the Aterazawa Na-montmorillonite gouge, plotted against the displacement d showing the friction alteration (Figure 2a) with changing temperature T (Figure 2b). The vertical axis in Figure 2a is the shear stress τ normalized by the normal stress σn, given as
where σdif is the differential stress (i.e., the axial stress minus Pc), and θ is the angle of the oblique surfaces of the alumina blocks to the axis (i.e., 30°). The internal load cell directly measured σdif. In the fully drained cases (DR), because Pp was constant at 0 MPa, (1) equaled the (effective) friction coefficient μDR. However, for the undrained cases (UD), (1) indicated the apparent friction coefficient μUDapp because we could not know the pore pressure during heating under the undrained condition. Heating toward 250°C, 350°C, and 500°C was started at d = 0.55–0.65 mm; each temperature was then held until d reached to 1.25–1.35 mm. Except for two runs at Tmax = 500°C, all runs ended owing to rupturing of the copper jacket before the temperature decreased to room temperature. Notably, as shown in Figures 2a and 2b, frictional strength decreased during the heating stage in runs under the undrained condition, but increased in runs under the fully drained condition. The apparent friction coefficient under the undrained condition μUDapp continued to decrease gradually as the temperature increased, having a value of 0.07 at the end of heating toward 500°C. In contrast, the friction coefficient at T = 500°C under the drained condition μUD was 0.54.
 Before heating started, the friction coefficient values were scattered (Figure 2a) in the range of 0.27 to 0.34 at d = 0.55 mm. For comparisons between the friction coefficients, μDR and μUDapp, we next tried to normalize each friction coefficient using the friction coefficient curve at room temperature μRT as a reference (black line in Figure 2). We drew a curve parallel to the μRT curve passing through point (d′, μ(d′)) where d′ represents displacement just before heating started (Figure 2c). We defined the friction curve μ′RT (d ≥ d′) as indicating the alteration of friction if the experiment were to progress without change in temperature. Thus μ′RT = μRT + Δμ∣d = d′ (d ≥ d′), where Δμ∣d = d′ is the difference in the friction coefficients between μDR and μRT at d′. For Figure 2d, we normalized the friction coefficient value μ at a certain T in the heating stage by that at room temperature μRT at a certain d and plotted μ∣(T, d)/μRT∣d against T. Thus we regarded the friction coefficient curve at room temperature as the standard to highlight frictional strength changes with increasing temperature for both the drained and undrained conditions. The results for the heating stage were grouped into two trends of μDR/μ′RT and μUDapp/μ′RT, where UD and DR represent the undrained and fully drained conditions, respectively. The trends in μDR/μ′RT and μUDapp/μ′RT began to differ at about 85°C. The decrease shown in the μURapp/μ′RT curve accelerated beyond approximately 300°C, while the increase shown in the μDR/μ′RT curve stabilized beyond approximately 300°C (Figure 2d). Thus the difference between μURapp/μ′RT and μDR/μ′RT appeared to be proportional to T in the range of 85°C to 500°C. There was a conspicuous difference (as much as seven times for the maximum temperature of 500°C) between μDR and μUDapp.
 The weakening of μUDapp in the heating stage could be considered an effect of interlayer dehydration of the Na-montmorillonite. On the other hand, μDR in the heating and holding stages showed continuous hardening (Figure 2a). In the cooling stage, μDR decreased but μUDapp increased. Following cooling toward room temperature, however, the values of μDR = 0.61 at d = 2.3 mm (blue) and μUDapp = 0.47 at d = 2.5 mm (red) did not agree with the standard friction of μ′RT = 0.40 (see Figure 2c) and 0.37, respectively. This suggests that heating produced some change in the dehydrated Na-montmorillonite clay, causing it to harden.
3.1.2. Dehydration of Na-Montmorillonite Gouge Implied by XRD Pattern and TG-DTA
 We tried to compare the water content of the experimental products with that of the Aterazawa Na-montmorillonite powder starting material, using XRD and TG-DTA. However, the adsorption properties of montmorillonite clays made it difficult to quantify the content change due to dehydration. For montmorillonite, the adsorption and dehydration reactions can be reversed, depending on the relative humidity of the air. Therefore, a dehydrated specimen could become rehydrated during the XRD analysis or TG-DTA, when the specimens were exposed to humid air. The thickness of the interlayer water in montmorillonite clays has a stepwise dependence on the relative humidity of the air [Sato et al., 1992]. The XRD and TG-DTA machines were kept in rooms where the usual humidity was 35–45 RH%; under that humidity, anhydrate montmorillonite (d001 ∼ 9.8 Å) tends to return to the one-layer hydrate state (d001 ∼ 12.4 Å). In preparing the specimens for XRD, we tried to make the specimen particles as randomly oriented as possible. Until preparation for the analyses, we kept the specimen material jacketed by a copper tube and encased in a plastic container to impede the adsorption reaction. While we could identify a rough dehydration trend in relation to Tmax during a run, we could not quantify the dehydration content exactly.
 When we took the Na-montmorillonite gouge from a jacketed specimen assembly, we found that the sheared gouge layer became more compacted under higher temperature conditions, suggesting greater cohesion among the particles in the gouge. While the Na-montmorillonite peaks remained at all temperature conditions, as shown by the XRD patterns (Figure 3a), the (001) peak shifted to larger 2θ at Tmax = 500°C under the undrained condition. The spacing d001 (Figure 3b) became approximately 10 Å, indicating unhydration, at Tmax = 500°C. At lower temperatures (<400°C), spacing decreased slightly with increasing temperature, but the value of d001 ∼ 12 Å indicated that the Na-montmorillonite specimens retained the one-layer hydrate state in the interlayer.
 The TG-DTA provided more quantitative results on the dehydration of the experimental products (Figure 4). The heating rate of 10°C/min almost equaled that of the deformation experiment. By 120°C, the TG data (Figure 4a) indicated that the water content in the starting material (7.4 wt %) was smaller than that in a specimen produced under the room temperature condition (RT-data, 9.8 wt %). The Na-montmorillonite powder may have absorbed moisture in our laboratory during the specimen preparation. The above results correspond to an XRD result shown in Figure 3b; that is, the spacing d001 = 11.8 Å in the starting material was narrower than that in the product of the room temperature experiment, d001 = 12.4 Å. Other TG results for heated specimens indicated a negative dependence of Tmax on the water content held by specimens. By 130°C at Tmax = 500°C, the TG data (Figure 4a) indicated interlayer water weight losses of 1.8 wt % and 3.6 wt % under the fully drained and undrained conditions, respectively. The DTA data (Figure 4b) by 130°C also indicated that the endothermic reaction was more inconspicuous with increasing Tmax. Assuming that the Na-montmorillonite gouge swelled by almost the same weight of water as in the air during sample preparation for TG-DTA, we could estimate that the shear experiment at 500°C dehydrated 6∼8 wt % of water content from the gouge. In addition, the dehydroxylation reaction of Na-montmorillonite appeared distinctly from approximately 600°C (Figure 4a). In the cases of no applied heat (i.e., the starting material and the product produced under room temperature conditions), dehydroxylation occurred in the temperature range of approximately 600°C to 720°C. On the other hand, the heated products reacted slowly, with dehydroxylation ending at approximately 800°C. The endothermic response of dehydroxylation (Figure 4b) also indicated gentler reactions by the heated products. Thus, once heated during deformation, the Na-montmorillonite became difficult to decompose.
3.2. Dehydroxylation of Kaolinite Gouge
3.2.1. Frictional Strength Development due to Dehydroxylation With Changing Temperature
 Here we present results for kaolinite-rich samples (Kampaku kaolinite) heated to 250°C, 450°C, and 500°C starting at d = 0.74–0.81 mm; each temperature was then held until d = 1.45–1.54 mm before cooling to room temperature (Figures 5a and 5b). Two runs at Tmax = 500°C ended before the temperature decreased to room temperature because the copper jackets ruptured. The friction coefficient values before heating started were also scattered (Figure 5a) in the range of 0.49 to 0.53 at d = 0.78 mm. To normalize each friction curve to that of kaolinite gouge at room temperature μRT (black line in Figure 5), we used the same method as described above for the Na-montmorillonite, illustrated in Figure 2c.
 Under both the fully drained and undrained conditions, all friction curves for the Kanpaku kaolinite gouge indicated conspicuous hardening (Figure 5a) starting at 80∼100°C and lasting to 450°C in the heating stage. From 450°C to just before reaching 500°C, all the friction curves for the kaolinite gouge decreased slightly; however, no reduction in friction was found when heating was stopped for the 450°C holding stage under the undrained condition (Figure 5a). This result demonstrates that the kaolinite gouge was not dehydroxylated until 450°C.
 The curves for μUDapp/μ′RT and μDR/μ′RT were almost the same until just before 500°C (Figure 5c). Irrespective of the drainage condition, the curves started to gradually weaken at 480°C. At 500°C a sudden and significant reduction occurred in frictional strength in the undrained condition; however, under the fully drained condition, the frictional curve showed continuous hardening following a slight decrease in strength. The friction coefficient difference just before and after 500°C under the undrained condition could be due to the increased pore pressure created by the dehydroxylated water. The TG-DTA showed that the constitution water in the kaolinite mineral was released explosively from approximately 480°C (Figure 7), which could also support the abrupt reduction of friction μUD, indicating increasing pore pressure. On the other hand, the μDR curves indicated continuous hardening at the temperature holding stage.
 Smaller and continuous stick slip was observed at Tmax = 250°C while holding at T = 250°C (Figure 5a). In the cooling stage, the stick slip disappeared. At higher Tmax, violent unstable sliding, indicated in Figure 5a, occurred during the cooling. The violent reduction of frictional strength due to the unstable sliding was clearly distinguishable from that caused by the dehydroxylation reaction. The unstable sliding in the cooling stage of the kaolinite deformation always emitted audible crackles. Almost irrespective of the differences in the drainage conditions and Tmax, the values of the friction coefficients after cooling ranged from 0.76 to 0.89, representing 1.29 to 1.38 times the standard friction μ′RT at approximately d = 2.5 mm and T = room temperature. Therefore, heating could have produced some change in the dehydrated kaolinite clay, causing it to harden.
3.2.2. Dehydroxylation of Kaolinite Gouge Demonstrated by XRD Pattern and TG-DTA
 The XRD pattern (Figure 6) revealed whether the kaolinite (or alunite) in the experimental product was decomposed, and the TG-DTA (Figure 7) result quantified the dehydroxylated water content released from the gouge. The XRD patterns of the experimental Kanpaku kaolinite gouge products (Figure 6) indicated that kaolinite and alunite did not decompose until Tmax = 450°C. At Tmax = 500°C, the (00h) peaks of kaolinite vanished (Figure 6). Thus, the kaolinite decomposed as it released the constitution water. In addition, several unknown new peaks were detected but were not consistent with the XRD patterns of several products of dehydroxylated alunite [Küçük and Gülaboğlu, 2002]. Additionally, the XRD pattern at Tmax = 500°C still showed the alunite peaks, indicating that alunite was not decomposed at that temperature. Thus, we could conclude that the constitution water in alunite did not contribute to the weakening of the gouge under 500°C heating. Thus, the impurity alunite in the Kanpaku kaolinite gouge had no effect on one of our main results, abrupt weakening at 500°C due to the kaolinite dehydroxylation under the undrained condition, although differences in purity could change the ultimate degree of weakening, meaning ultimate pore pressure increase at 500°C.
 The TG-DTA analysis (Figure 7) revealed the released water content from the kaolinite gouge. All TG data (Figure 7a) exhibited almost the same curve with the heating rate increase of 10°C/min, except for the data at Tmax = 500°C, which suggested a decrease in the constitution water of the experimental product. DTA showed that the endothermic reaction peaked at 540°C (Figure 7b), indicating that the peak intensity at Tmax = 500°C was weaker than that at lower Tmax. These results are compatible with the results from the frictional curve (Figure 5) and XRD (Figure 6), showing that the Kanpaku kaolinite gouge was inactive until T reached 480°C. The weight loss of the specimens at Tmax ≤ 450°C due to dehydroxylation ranged between 18.5 and 19.3 wt %, while that at Tmax = 500°C was 12.4 wt %. The slight weight reduction before the TG-DTA temperature reached 200°C was likely caused by the release of absorbed water from the specimen. Thus, the difference between the TG data for dehydroxylated and nondehydroxylated specimens corresponded to the 6∼8 wt % of dehydroxylated water released from the kaolinite gouge.
3.3. Heating Tests on Inactive Fine Quartz Gouge
 Finally, fine quartz gouge (5 μm or slightly less in diameter), which can be considered an inactive material in heating to 500°C, was also tested for frictional change with increasing temperature (Figure 8). The quartz gouge was sandwiched between dense alumina blocks in two experiments, and friction coefficient curves were drawn. One experiment involved the heating–holding–cooling test; for comparison, the other experiment was conducted under room temperature conditions. As expected, the increasing temperature never resulted in a decrease of the friction coefficient. Moreover, as temperature increased, the friction coefficient increased (Figure 8), but only in slight increments compared to the increases for the Na-montmorillonite under the fully drained condition (Figure 2) and for the kaolinite gouges (Figure 5).
 Under the fully drained condition, the friction coefficient of Na-montmorillonite gouge gradually increased with increasing T, while the kaolinite gouge, irrespective of drainage condition, also showed conspicuous strengthening before the dehydroxylation reaction occurred. These results, demonstrating the strengthening of clay gouges during heating, were not caused only by the thermal expansion of alumina blocks, pistons, and WC spacers, as illustrated in Figure 1. After cooling to T = room temperature, the friction coefficient of the heated fine quartz gouge returned to the level of standard friction μ′RT (Figure 8). Thus, the strengthening due to the thermal expansion of the inactive gouge and the piston assembly was not significant and could be reversed. On the other hand, as shown in Figures 2a and 5a, heated clay gouges retained their strength even after the temperature returned to room temperature. This irreversible strengthening was observed for all the clay gouges, irrespective of the undrained/fully drained conditions and whether or not a dehydration/dehydroxylation reaction occurred. Although this study could not clarify what mechanisms caused the hardening of the clay gouges during heating, the study revealed that heating irreversibly altered the clay mineral properties.
4.1. Estimating the Gouge Pore Pressure Generated by Dehydration and Dehydroxylation
 Following dehydration and dehydroxylation reactions, we obtained quite lower friction coefficient values under the undrained condition than under the fully drained condition. This difference between μUD and μDRapp can be explained by the generation of pore pressure in the narrow gouge zone under the undrained condition in which the dense alumina spacers confined the pore water generated from the clays. Therefore, comparing μUD and μDRapp at a certain displacement d and a certain high temperature T enabled us to estimate the pore pressure generation Pp caused by the clay's decomposition reaction. Figure 9 illustrates the relationship between the shear stress τ and (apparent) normal stress σn for both DR and UD conditions. Assuming that τDR is linearly proportional to σnDR with cohesion c, then Pp at a certain (d, T) can be calculated as
Assuming that c does not depend on d, T, or the undrained/fully drained conditions and is nearly equal to 0, then (2) becomes
Here we redefine the friction curve μRD as the reference for correcting the scattered curves of the experiments (see Figures 2a and 5a). The parameters μUD, μDRapp, and σnappUD are replaced by μ″UD, μ″appDR, and σ″nappUD, given as
Figures 2d and 5c give the values for μUDapp/μ′RT and μDR/μ′RT at a certain (d, T). Thus, (3) is rewritten as
Here Ppest is the estimated value of pore pressure by (5). We can then draw the alteration of pore pressure as a function of T, as shown in Figure 10a. The estimated pore pressure generation at T = 500°C, calculated by (5), corresponds to the quite high Ppest to Pc = 80 MPa and was 72 MPa for Na-montmorillonite gouge and 68 MPa for kaolinite gouge. Thus, in those narrow gouge zones under the undrained condition, the effective confining pressure (Pc − Ppest) was mostly canceled, leaving only about 10 MPa. On the basis of the TG-DTA tests, both the Na-montmorillonite and kaolinite gouges only released 6∼8 wt % of water. Further, Ppest for both gouge zones showed almost 70 MPa at T = 500°C. Therefore, only 6∼8 wt % of dehydrated/dehydroxylated water can potentially generate 70 MPa of pore pressure at a fault consisting of compacted and fine clay gouge.
 While a linear, gentle increase of Ppest was shown for the Na-montmorillonite gouge with T increases from 80°C, an abrupt increase in Ppest could occur in the kaolinite gouge starting at 480°C. This difference in the increase of Ppest implies that dehydration of Na-montmorillonite interlayer water produces a more gradual reaction than does the dehydroxylation reaction of kaolinite. The dehydroxylation reaction of kaolinite, which is a chemically irreversible decomposition reaction, would proceed abruptly once the specimen attains enough heat energy. On the other hand, the interlayer water in clays such as smectite and vermiculite is bound weakly on the mineral surface. The water moves reversibly between the interlayers of the clay and pore space, according to the P-T condition [Koster van Groos and Guggenheim, 1984; Wu et al., 1997]. The increase in Ppest means that the pore water density in the pore space is increasing if the pore volume is unchanging. The Ppest path of Na-montmorillonite dehydration, however, illustrated in Figure 10a is not isochoric, and the pore water density decreases with increasing Ppest and T. The gradual increase in Ppest with increasing T for the Na-montmorillonite gouge could mean that the ratio of pore volume to the total volume of the gouge zone would increase, with interdependence on Ppest and T. On the other hand, the density Ppest path of the kaolinite gouge becomes larger as the reaction progresses, meaning the ratio of pore volume to total volume decreases.
 For reference, Figure 10b also presents phase diagrams of dehydration/dehydroxylation reactions of Na-montmorillonite [Koster van Groos and Guggenheim, 1984] and kaolinite [Yeskis et al., 1985] reported by previous studies. These previous studies investigated dehydration/dehydroxylation reactions in high-pressure DTA systems using argon gas as a pressure medium. For Na-montmorillonite (SWy-1), Koster van Groos and Guggenheim  proposed that total pressure Ptotal was nearly equal to PH2O under high pressure. Using kaolinite (KGa-1), Yeskis et al.  controlled the conditions of Ptotalas Ptotal ≈ PH2O and Ptotal ≈ PAr. Those curves in Figure 10b are equivalent to the Clapeyron curves of Na-motmorillonite dehydration and kaolinite dehydroxylation. At Ptotal = 0 MPa in Figure 10b, corresponding to the initial Pp condition of our experiments, both the curves of Na-montmorillonite dehydration and kaolinite dehydroxylation are positive; Ppest in the gouge zone can increase under the undrained condition. The difference in the dehydration/dehydroxylation temperatures for those clays could reflect geographic differences in the original location of the specimens. However, the shapes of the P-T paths resulting from our experiments are quite different from those of previous studies. This discrepancy was probably caused by two differences in the experimental procedure. First, the experimental determination of Clapeyron curves is usually carried out at Pp = Pc. In our experiment, Pp was not equal to Pc. Second, the gouge zone was sheared continuously with a certain normal stress σn and shear stress τ during dehydration/dehydroxylation. The Ppest during the reactions could be increasing in counterpoise between the stress condition (σn, τ) and the poroelastic properties of the gouge. It is therefore difficult to directly compare our results with the Clapeyron curves of the clays.
4.2. Potential of Phyllosilicate Minerals as Earthquake Triggers
 High pore pressure generation is thought to be strongly associated with earthquake occurrence. Here, we consider how dehydration and dehydroxylation reactions may contribute to natural earthquakes. How high pore pressure can arise and how long it can be held in a fault result from the competition between the reaction velocity and the outflow rate of pore water to the neighboring rock. Numerical study of rock failures, combining the kinetics of dehydration and dehydroxylation reactions and fluid transport kinematics in rock as a porous media, would help clarify this issue and has been partially attempted in several past studies [e.g., Wong et al., 1997; Miller et al., 2003].
 Two possible cases of dehydration/dehydroxylation reactions can be considered: a gradual process of metamorphism and a drastic thermal decomposition induced by frictional heating during a coseismic event. In the first case, the pore pressure due to these dewatering reactions could increase slowly not only because of the slow metamorphosis process but also because the fault rocks would behave as if somewhat “drained” against the slow increase in pore pressure. Whether the pore pressure excess becomes enough to heighten the potential for earthquake occurrence would depend on both the hydraulic properties (e.g., permeability and storage capacity) of the fault rocks and the reaction kinetics controlling the volume change of the reacting materials as indicated by the Clapeyron curve. For example, with dehydroxylation of kaolinite at more than 4.7 MPa of Ptotal = PH2O [Yeskis et al., 1985] (also see Figure 10b) and of serpentine at more than 2 GPa of pressure [Ulmer and Trommsdorff, 1995], the Clapeyron curves are negative. Additionally, Na-montmorillonite is stable at lower temperatures (<100∼150°C) in nature, which correspond to depths lower than 5 km and are known as the illite–smectite transition temperature. Illitization advances the dehydration of the interlayer water from smectite, possibly increasing the pore pressure. However, the path of the pore pressure increase as illitization proceeds would not coincide with our experimental result (Figure 10a). Further study is needed to estimate the excess pore pressure due to the slow dewatering reaction during burial and subduction, such as experiments under a much slower heating rate condition.
 The heating rate of 9.6°C/min in our experiment is much faster than the temperature increase during burial and subduction; the latter occasionally becomes much slower than the temperature increase induced by coseismic shear heating, which increases within several seconds of the start of melting [e.g., McKenzie and Brune, 1972; Hirose and Shimamoto, 2005]. Such abrupt heating during seismic slip is thought not to give pore water enough time to escape to outside the thin fault gouge zone [Wibberley and Shimamoto, 2005], creating a “nearly undrained condition.” We suggest that, irrespective of the hydraulic properties of the gouge, one possible difference in the pore pressure generation during seismic slip is caused by the process of rock dewatering by dehydration or dehydroxylation. On the basis of our experiments, dehydration from swelling clays, such as smectite, would start from the earliest stage of seismic slip, and the pore pressure increase would be contiguous with the increase in temperature. On the other hand, dehydroxylation in materials having constitution water, such as clay minerals and serpentine, would start in the middle stage of the slip, as heating increases to the temperature needed for the reaction to occur. Once dehydroxylation starts, the pore pressure increase during the seismic slip would be rather abrupt, meaning a rather violent reduction of fault strength compared to the reaction induced by the dehydration reaction. Therefore, the difference in the dewatering reaction may induce the difference in the alteration of slip velocity. In either case, an important point is that significant weakening similar to the “thermal pressurization” mechanism during coseismic shear heating can occur in even dry, nonsaturated rock, as long as it consists of hydrous minerals and/or minerals having constitution water. These minerals, such as phyllosilicates, are very common and richly contained in fault gouge.
 This study conducted shear sliding tests on gouge at 0.12 μm/s with heating at 9.6°C/min to 500°C under 80 MPa of confining pressure condition to elucidate the potential of minerals including and/or constituted of water to trigger earthquakes due to high pore pressure generation caused by dehydration and dehydroxylation reactions. Frictional strength changes in the fault gouge during dehydration (expulsion of H2O) and dehydroxylation (expulsion of −OH) reactions were experimentally simulated using Na-montmorillonite-rich gouge and kaolinite-rich gouge, respectively.
 In the undrained conduction, the shear strengths of both clay gouges significantly weakened at the higher temperatures that induced the dehydration and dehydroxylation reactions of the clay minerals. In contrast, under the fully drained condition, the strengths almost continuously increased at those temperatures. From 80°C to 500°C, the Na-montmorillonite gouge gradually weakened as temperature increased under the undrained condition, but indicated gradual hardening under the fully drained condition. The frictional strength of the kaolinite indicated hardening until 480°C for both undrained/fully drained conditions. At higher temperatures, the strength decreased violently under the undrained condition. However, under the fully drained condition, the strength of the kaolinite gouge hardened continuously following a slight strength reduction.
 The difference in the frictional strengths under the fully drained and undrained conditions during dehydration and dehydroxylation reactions may reflect the generation of high pore pressure in the narrow gouge zone under the undrained condition. We were able to estimate the incremental changes in the pore pressure in response to the reactions by assuming zero cohesion of the clay specimens, irrespective of the deformation conditions. That calculation introduced quite high pore pressure of 72 MPa and 68 MPa for the Na-montmorillonite gouge and kaolinite gouge, respectively, at 500°C. Thus, these pressures mostly canceled the confining pressure (80 MPa). In addition, TG-DTA indicated that 6∼8 wt % of water was released from both the Na-montmorillonite and kaolinite gouge products until the temperature reached 500°C. Therefore, clay minerals compacted in impermeable rocks have the potential to generate abnormal pore pressure, expulsing only small amounts of water due to dehydration and dehydroxylation reactions. Especially during coseismic shear heating, dehydration/dehydroxylation of hydrous minerals and/or minerals having constitution water (i.e., phyllosilicates) could significantly weaken the fault strength, even if the fault is initially dry.
 We would like to thank Kelin Wang, Diane Moore, and an anonymous reviewer. We also thank Masaya Suzuki for helpful discussions regarding TG-DTA data. This study was partly supported by a Grant-in-Aid for Japan Society for the Promotion of Science (JSPS) fellows (18-40014; to M. Takahashi), financed by the Ministry of Education, Culture, Sports, Science and Technology (MEXT), Japan.