Laboratory experiments on rock friction focused on understanding earthquake mechanics
Grant Number: 1434-HQ-97-GR-03034
Terry E. Tullis
Phone: (401) 863-3829
PROGRAM ELEMENT III: Understanding Earthquake Processes
Key Words: Laboratory studies, Fault dynamics, Thermophysical Modeling
We have studied two topics during the grant period reported on here:
We studied the role of deformation on permeability and its anisotropy in fault zones. Permeability is a critical parameter affecting the mechanics of fault zones. The models of Byerlee (1990) and of Rice (1992) both call upon high pore pressure in the fault zone to allow the effective normal stress, and hence the frictional shear resistance, to be low. However, the models differ substantially in their assumptions of the permeability needed to create or maintain that high pore pressure. Neither model focuses on the possible anisotropy of permeability in the fault zone. Based on previous studies (Antonellini et al., 1994a,b) we considered that anisotropy is a real possibility, and that it could be a significant factor in determining what magnitudes and gradients of pore pressures might exist. For example, if the permeability of the San Andreas fault zone is considerably greater parallel to the fault than perpendicular to it, this might support Rice’s (1992) model in which fluid flows up through the fault from depth, but maintains a higher pressure than in the surrounding rock. We have developed the capability to measure not only permeability as a function of large displacements and changes in fault zone texture, but also the difference between permeability perpendicular and parallel to our experimental fault zones. Developing an understanding of the evolution of permeability and its anisotropy with fault displacement is an important element in formulating more realistic models of the mechanics of earthquakes. Fluid flow prior to earthquakes, with the electrical and magnetic changes it can induce, could result in important precursory signals to an impending earthquake (Uyeda et al., 1993).
We have measured permeability and its anisotropy as a function of displacement both in quartz gouge and mica gouge (Zhang and Tullis, 1997; Zhang et al., 1997). Because the reduction of permeability with increasing effective pressure is an important effect in Rice’s (1992) model, we have measured the magnitude of this effect as well. This was done for mica gouge as a function of shear displacement, since clay minerals can be important in fault zones.
Some of our results for permeability and its anisotropy are summarized in Figures 1 and 2.
Figure 1. Permeability as a function of displacement for flow parallel and perpendicular to a 1 mm layer of quartz gouge and perpendicular to an initially bare sandstone surface. The permeability perpendicular to the fault decreases more rapidly than that parallel to the fault. Petrographic examination of the sample shows slip localized in a narrow comminuted zone. The calculated k1 and k2 permeabilities are the isotropic values in the less and more deformed layers, respectively, required to produce the observed macroscopic anisotropy, given the observed thickness of the layers.
Figure 2. Permeability as a function of displacement for flow parallel and perpendicular to a 1 mm layer of muscovite gouge. The initial permeability reduction is faster as a function of shear displacement for specimens sheared at higher normal stress. The initial permeability anisotropy of about an order of magnitude is due to initial alignment of the basal planes parallel to the fault. Shearing reduces this preferred orientation and results in a smaller anisotropy at high strain. The strain is much more homogeneous than in the quartz gouge. The scattering of data at normal stresses of 100-125 MPa is caused by hysteresis of permeability during pressure cycling.
The permeability in both quartz and mica decreases by about 3 orders of magnitude both parallel and perpendicular to the fault zone as a result of sliding 200 mm. The quartz gouge is isotropic in its permeability before shearing, but develops an anisotropy of about one order of magnitude as a consequence of the development of a fine-grained localized shear zone. Thus in the quartz gouge the anisotropy of permeability in the shear zone as a whole is a consequence of the development of a layered structure, with lower permeability in the more comminuted shear zone. Zhang and Tullis (1997) show the relationship between the thickness of the layers, their permeabilities, and the permeability anisotropy. The permeability anisotropy resulting from this geometric situation will be about one order of magnitude less than the permeability contrast between the shear zone and the less deformed gouge (Figure 1), although the details depend on the exact thickness of the shear zone. In the muscovite gouge the permeability contrast begins at about one order of magnitude due to the preferred orientation of basal planes parallel to the fault zone produced by sample preparation and initial pressurization. Interestingly, as a result of shearing, the permeability contrast is reduced to about half an order of magnitude (Figure 2). This is due to the development of structures in the gouge layer in which the micas become oriented at a high angle to the shear zone (Scruggs and Tullis, 1996). Thus in these two materials we have found that permeability anisotropy can be produced by development of an inhomogeneous fault zone structure by localization and/or by alignment of flow channels producing a contrast in tortuosity. Although neither of these effects is surprising, the magnitude of anisotropy that results from them is not as great as one might anticipate. These two causes of anisotropy could occur together and combine to increase the overall anisotropy.
The large reduction of permeability that occurs as a result of shearing is a consequence of the great reduction in the particle and pore sizes from the initial high-porosity simulated gouge to a much more compact shear-produced gouge (Zhang and Tullis, 1997; Zhang et al., 1997). We have also measured a reduction in permeability due to shearing by starting with bare surfaces of sandstone and producing a fine-grained fault gouge. It is clear that the highly deformed localization zones in natural fault zones (Chester et al., 1993) will be less permeable than the surrounding less deformed gouge. Unless the country rock is very permeable, the initial process of faulting may well increase the permeability due to the production of fractures and the associated dilatancy. However, continued shearing will quite commonly reduce the permeability of the fault zone to even lower than that of the country rock. What emerges from our experiments and these obvious considerations is that the permeability structure of fault zones may be quite complex. Thus it is not surprising that observations from the oil industry show that the permeability of faults can be more or less than that of the surrounding country rock (Magara, 1978). Detailed studies of fault zone structures and textures like that of Chester et al. (1993) can give at least a qualitative model of fault zone permeability structure. Our experiments help quantitatively constrain what the permeabilities and their anisotropies can be.
Rice (1992) has produced an interesting and important model suggesting that pore fluid pressure may be close to lithostatic over the depth of the seismogenic part of the San Andreas Fault. This high pore pressure could be responsible for the apparent weakness of the San Andreas in a strong crust. However, the model depends on preventing the upward flow of water out of the fault zone by assuming that increasing depth and thus increasing overburden pressure rapidly reduces the permeability. Thus, starting down from the earth’s surface, the computed pore pressure gradient in the model follows the hydrostatic one until a modest effective confining stress is attained, at which point the permeability becomes low. This effective pressure is then maintained with increasing depth, so that the pore pressure parallels the lithostatic gradient. If the permeability reduces rapidly from the surface with increasing effective pressure, then the pore pressure gradient begins to parallel the lithostatic one at shallow depth, with only a small upward flux of water required to produce this effect. If increasing effective pressure is less efficient in reducing permeability, then either much larger fluxes of water are required to keep the pore pressure close to lithostatic, or the pore pressure will follow a gradient closer to the hydrostatic one and the fault will not be weak. Rice (1992) assumed that effective pressure was very efficient at reducing permeability. In the relation k = k0 exp(-g snneff) between permeability k and effective normal stress snneff where k0 is a constant and g is the sensitivity to effective normal stress, Rice used g = 0.2 MPa-1. This is representative of values for crystalline rock like granite, in which the permeability is reduced rapidly by increasing effective normal stress since the permeability is primarily due to flow through thin cracks that close easily with increasing pressure. David et al. (1994) showed that for porous rock like most sandstones effective pressure is much less efficient at reducing permeability (g ranges from 0.0014 to 0.012 MPa-1), so that Rice’s (1992) model would have trouble without very large fluxes of water. The effect of pressure on permeability of clay (Morrow et al., 1984) is intermediate between that of crystalline rock and porous rock such as sandstone (g ranges from 0.012 to 0.055 MPa-1), so that it is not clear whether the model might work or not if fault zones contain large amounts of clay gouge. What we have found (Figure 3) is that the effect of extensively shearing muscovite gouge is to change it from behaving like clay with g = 0.032 MPa-1 to behaving like sandstone with g = 0.0065 MPa-1. This means that although the model of Rice (1992) cannot be rejected, it is increasingly difficult to be confident that it accurately describes the pore pressure in the San Andreas fault.
Figure 3. Efficiency of effective normal stress in changing permeability for a 1 mm layer of muscovite gouge after various shear displacements. After progressively higher shearing there is less and less change in permeability as the effective normal stress is changed. This apparently results from disruption by shearing of the strong initial preferred orientation of the muscovite flakes.
Earthquake slip speeds are on the order of 1 m/sec. At these high velocities significant shear heating can occur, and a variety of processes may operate that might greatly lower the shear resistance. If such dynamic weakening occurs, it could cause dynamic stress drops much larger than static drops, resulting in seismic slip propagating as self-healing slip pulses rather than as a conventional crack (Brune, 1970, 1976; Heaton, 1990; Beroza and Spudich, 1988; Cochard and Madariaga, 1994; Perrin et al., 1995; Beeler and Tullis, 1996). This has implications for the magnitude of peak seismic accelerations expected during earthquakes and thus for the amplitude of strong ground shaking. It also is important for the state of stress on the fault following an earthquake. This stress state can affect the behavior of the fault in the time interval leading up to the next earthquake, and in turn can affect the nature of premonitory accelerating slip or fluid flow that might be helpful in predicting the next earthquake. Furthermore, if dynamic weakening is large enough, then it is easier for small earthquakes to grow into large ones, and earthquake recurrence intervals should be much less regular. This is because once the rupture begins, the stress concentration at the tip of the slip zone could overcome the static friction and dynamic slip could occur with very little resistance. This fits in with the views of many concerning the unpredictability of earthquakes (e.g., Geller et al., 1997).
It has been recognized for some time that slip at high speed can cause a variety of processes to occur that could drastically alter frictional strength (Brune et al. 1992; Lachenbruch, 1980; Mase and Smith, 1984, 1987). If the normal stress, slip velocity, and displacement are high enough and no fluids are present, frictional melting can occur. This has been investigated in exploratory experiments (Spray, 1987), and more recently some preliminary mechanical data have been obtained (Tsutsumi and Shimamoto, 1997). If pore fluids are present, pore pressure may rise due to thermal expansion of the pore fluid, depending on the compressibility and thermal expansion of the fluid and of the rock matrix. However, high slip speeds may also result in dilatancy of the fault zone (Marone and Kilgore, 1993; Beeler et al. 1996; Segall and Rice, 1995; Beeler and Tullis, 1997), and this will tend to reduce the pore pressure. At the moment we are woefully lacking in experimental measurements to check the applicability of the theoretical analyses (Lachenbruch, 1980; Mase and Smith, 1984, 1987; Segall and Rice, 1995) and to determine the roles played by critical parameters such as pore and fluid compressibility, fault zone and wall rock permeability, fault zone width, and velocity dependence of dilatancy.
We have begun a series of experiments at intermediate slip velocities and low rock permeabilities to evaluate the effect that dynamic slip velocity is likely to have on stresses during earthquakes. Although we are unable to conduct experiments at speeds of 1 m/sec typical of dynamic earthquake slip, we can slide for large distances at velocities of up to 5 mm/s, about three orders of magnitude faster than most laboratory friction experiments. We can therefore attain high enough frictional temperatures to cause interesting effects. We have done several experiments at 25-50 MPa normal stress with a jacketed sample. In some cases our measurements of dynamic changes in sample length, apparently due primarily to thermal expansion, indirectly suggest that temperatures were elevated by 100-130° C, in agreement with our expectations from finite element modeling and earlier experiments where the temperatures were measured with thermisters (Blanpied et al., 1998). For thermal pressurization of pore fluid to occur it is necessary that the permeability be low enough that the pressure source from the heating is not counteracted by too much fluid flow. In the experiment illustrated in Figures 4 and 5 we used a sample of Cheshire quartzite that we measured to have a permeability of 10-19 m2. We can measure the change in thickness of the sample during the experiments and so determine how much dilatancy is occurring and what effect that should have on the pore pressure.
In our must successful experiment to date (Figure 4) we have measured a significant drop in shear resistance that appears to be due to thermal pressurization of the pore fluid.
Figure 4. Results of sliding at 3.2 mm/s for nearly 3 meters in a water-saturated sample of Cheshire quartzite. The externally imposed normal stress was 29 MPa and the initial pore pressure was 4 MPa. On the abscissa the shear stress is normalized by the externally applied normal stress. Due to the low permeability of the sample, we have no direct measurement of the pore pressure during the high-speed part of the experiment, so using the externally imposed normal stress for normalization seems best.
After a run-in of about 6 mm at slip speeds of 1 and 10 mm/s, during which the resistance was m0 0 as shown in Figure 4, the speed was abruptly raised from zero to 3.162 mm/s. This results in a rapid increase in shear stress over the first 80 mm of slip, probably due to an intrinsic dependence of friction on velocity and temperature (Blanpied et. al., 1998). After about 80 mm of slip, the resistance begins a large decrease over the next meter of slip. There is little displacement normal to the fault zone during this interval; the small oscillations have a period of one revolution of the sample. At about 1.1 m of slip the sample undergoes an abrupt compaction across the fault, followed by a larger dilation and an increase in the amplitude of the periodic oscillations. The first part of the abrupt compaction is accompanied by a decrease in the shear stress, and the subsequent dilation is accompanied by an increase in the shear stress. This relation between the fault normal displacement and the shear stress is what would be expected if the shear stress were responding to changes in pore fluid pressure resulting from the compaction and dilation. The LVDT measuring the fault -normal displacement failed after 1.5 m so no further information is available. After the rapid sliding the sample was allowed to cool off. Subsequent slow sliding at speeds of 1 and 10 mm/s showed that the shear resistance had returned to approximately its initial value, shown as mf on Figure 4, supporting the idea that thermal pressurization of the pore fluid was responsible for the significant weakening during the first meter of slip.
The magnitude of the reduction in shear stress and the rate at which it decreases with displacement matches theoretical predictions, as is shown in Figure 5.
Figure 5. Comparison of the predicted and observed shear stress reduction as a function of time and shear displacement due to thermal pressurization of water by shear heating. This is Figure 3c of Mase and Smith (1984), modified to correspond to our easily attained slip rate of 3.2 mm/s, with our experimental results superimposed onto it.
In Figure 5 the theoretical calculation of Mase and Smith (1984) represents the situation for which sliding occurs at an interface between two rocks, rather than shearing of a gouge layer. Our experiments began with initially bare surfaces of rock; during sliding a layer of gouge 0-25 mm thick was developed. The figure illustrates that our experimental results match what would be expected for a permeability of 10-19 m2, the permeability of Cheshire quartzite. This figure is drawn for the case of zero dilatancy and corresponds to a bulk compressibility 10-9 Pa-1. This is more compressible than bulk Cheshire quartzite, but may be appropriate for the compressibility of the partially gouge-filled sliding interface. Because the shear stress increased initially after fast sliding was initiated, it is not clear exactly what value of initial shear stress s0 0 should be used to normalize the results shown in Figure 4 to allow them to be plotted on Figure 5. We have taken 0.8 as the value of s0, which is the magnitude that the shear stress would have been at zero displacement if we extrapolate the falling curve of Figure 4 back to the start of rapid sliding.
More experiments are needed to investigate the range of situations in which thermal pressurization of pore fluids can cause dynamic weakening during earthquake slip. Our results to date suggest that thermal pressurization caused the large reduction in shear stress shown in Figures 4 and 5 and that we are able to investigate this phenomena at intermediate displacement rates in our rotary shear apparatus.
We find that the ability of fluid to flow along and across fault zones filled with powdered quartz and mica is reduced by a factor of about 1000 after shearing 30-200 mm. Flow along the fault is up to 10 times easier than across it due either to alignment of mica or to layers of finely ground quartz that are produced by the shearing. We also find that frictional resistance on a fault surface is reduced dramatically as a result of sliding at 3 mm per second if fluid is contained in the fault zone. This is due to frictional heating causing the fluid to expand and thus its pressure to increase, thereby reducing the effective normal stress on the fault.
Beeler, N. M., and T. E. Tullis, The roles of time and displacement in velocity-dependent volumetric strain of fault zones, J. Geophys. Res., 102, 22595-22909, 1997.
Blanpied, M. L., J. D. Weeks, and T. E. Tullis, The effects of displacement, sliding rate and shear heating on the friction constitutive behavior of granite, J. Geophys. Res., in press, 1998.
Costello, S. W., and T.E. Tullis, Investigation of a proposed earthquake prediction method that envisions foreshock triggering by free oscillations, Eos. Trans. Am. Geophys, Union, Fall Meeting Suppl., 78, 490, 1997.
Scruggs, V. J., and T. E. Tullis, Strain localization in fault gouge at large displacements and correlation with slip instability, Tectonophysics, in press, 1998.
Scruggs, V.J., T.E. Tullis, S. Zhang, and J. D. Fitzgerald, The frictional behavior of granitic gouge under high pressure, Eos. Trans. Am. Geophys, Union, Fall Meeting Suppl., 78, 732, 1997.
Tullis, T.E., and D.L. Goldsby, Shear heating induced pressurization of pore fluid as a dynamic fault weakening mechanism, Eos. Trans. Am. Geophys, Union, Fall Meeting Suppl., 78, 472, 1997.
Worthington, C., N. M. Beeler, and T.E. Tullis, Stress-dilatancy relationships during frictional sliding, Eos. Trans. Am. Geophys, Union, Fall Meeting Suppl., 78, 475, 1997.
Zhang, S., and T. E. Tullis, The effect of fault slip on permeability and its anisotropy in an artificial quartz gouge, Tectonophysics, in press, 1998.
Zhang, S., T.E. Tullis, and V.J. Scruggs, Effects of fault slip and fluid pressure on permeability and its anisotropy in artificial fault gouges, Eos. Trans. Am. Geophys, Union, Fall Meeting Suppl., 78, 732, 1997.