Laboratory Experiments on Rock Friction Focused on Understanding Earthquake Mechanics
USGS Award Number: 1434-HQ-97-GR-03034
Terry E. Tullis and David L. Goldsby
Tele: (401) 863-3829
Program Element: II
Key Words: Laboratory studies, Fault dynamics, Source characteristics
We have worked on two problems during the course of this grant period:
Velocity dependence of friction of granite gouge at large displacements
It is important to determine the velocity dependence of friction in fault zones, since velocity weakening behavior promotes unstable earthquake sliding, whereas velocity strengthening results in fault creep. Laboratory experiments on initially bare rock surfaces and on layers of simulated gouge show opposite velocity dependence, and hence it is important to understand why this occurs and which behavior is more likely to be representative of natural fault zones. One possibility we are investigating is that the localized slip characteristic of gouge generated from initially bare surfaces, and which is accompanied by velocity weakening behavior, may also occur for simulated gouge if the gouge is sheared to sufficiently large displacements. If this were the case, then gouge of granitic composition would be expected to slip unstably via earthquakes, unlike the situation for gouge in most laboratory experiments that do not attain very large shear strains.
Previous studies of granite gouge in our laboratory have shown a complex sequence of frictional resistance, its velocity dependence, and gouge textures as a function of slip to displacements of 400 mm (Beeler et al., 1996). The friction coefficient decreases from its initial values of about 0.7 to about 0.6 after about 50 mm of slip. This decrease is accompanied by a switch from velocity strengthening to velocity weakening behavior and a localization of slip on a boundary-parallel shear located near the lower gouge-rock interface. At this displacement, the frictional resistance, velocity dependence and localized slip resemble those seen for initially bare surfaces of granite at all displacements. However, with further slip the friction increases, the velocity dependence becomes positive (velocity strengthening), and slip migrates to a series of anastomosing surfaces that work their way up into the gouge with continuing displacement. Our explanation for this is that the very fine-grained gouge in the localized slip region is slip strengthening, as shown by the increase in friction coefficient, and consequently the weaker, less-deformed gouge in the upper part of the sample accommodates the deformation until it too slip strengthens. Eventually the entire gouge layer would become highly deformed and strengthened. However, after our largest previously attained displacements of abut 400 mm, the texture of the gouge has not reached steady state, since relatively undeformed gouge remains.
In our current work, we have sheared gouge to displacements of several meters to determine whether the entire gouge layer will attain the fine-grained character seen both in the more highly deformed part of a simulated gouge layer and in gouge generated from initially bare surfaces. In particular, we want to determine whether, if this does occur, slip will localize on a single boundary-parallel slip surface, accompanied by a return to velocity weakening behavior. We have done experiments to several meters of slip. These tests show a return to velocity weakening behavior and localized slip after 1 to 3 meters of slip. However, gouge layers in nearly all of these experiments have experienced so much thinning that at the end of the experiment the gouge is no thicker than that generated from initially bare surfaces; thus, they cannot be characterized as having a local slip surface within a wide fine-grained layer of gouge as is characteristic of many natural faults. Nevertheless, in our most recent experiment we reduced the differential normal stress across the fault surface enough that thinning was minimized. At the end of the experiment, at a displacement of 1.7 m, the velocity dependence became negative, followed by small stick-slip events. This occurred while the gouge was about 0.7 mm thick, much thicker than the gouge generated from initially bare surfaces. This suggests that the eventual localization we anticipated has indeed occurred. Figure 1 shows friction and its velocity dependence for this experiment, superimposed on the lower-displacement data of Beeler et al. (1996) for both gouge and bare surfaces. Thus, we believe that we have demonstrated that gouge will eventually relocalize after enough slip and return to velocity weakening behavior, although additional experiments are required to confirm our result.
Figure 1. Friction and velocity dependence with displacement for simulated gouge layers and initially bare surfaces. Our recent data for a initially 1 mm thick layer is shown in black. The data of Beeler et al. (1996) for initially bare surfaces (blue) and 1 mm thick layers of gouge (red) are shown for comparison. After 1.6 m of slip the gouge in our recent experiment returns to velocity weakening behavior, a trend that began at about 1.3 m of slip. No data exist at further displacement due to a jacket leak soon after the last data point. The sample was undergoing small-scale stick slip prior to the leak, confirming its transition to unstable behavior.
Friction at intermediate slip rates relevant to dynamic resistance during earthquakes
Few laboratory data exist on rocks that are relevant to understanding the frictional resistance of faults during dynamic earthquake slip. A number of mechanisms have been suggested that might cause dynamic resistance to be less than that found in quasi-static friction experiments. These include shear melting, the transient increase of pore fluid pressures due to shear-heating-induced thermal pressurization, acoustic fluidization, and normal separation of the fault during dynamic slip. Previous laboratory experiments have primarily been conducted at combinations of slip rates, normal stresses, and displacements that are too low to result in significant frictional heating. The experiments of Tsutsumi and Shimamoto (1997) that produced considerable shear melting are an exception to this, but, perhaps due to the fact that their samples were unconfined and melt rapidly escaped from their samples, the frictional resistance was not greatly reduced. We are investigating reductions in friction that result from slip at intermediate velocities, elevated normal stresses and slip displacements of several meters. Although we cannot slip at seismic slip rates of 1 m/s, we here report data on quartzite at slip rates of 3.2 mm/s, a velocity that results in appreciable frictional heating at large displacements.
Last year we reported on reductions in apparent friction coefficient from 0.7 to 0.4 that we inferred were produced by elevated pore pressures due to thermal pressurization of water in the pore space of a fully-saturated, low permeability rock. We inferred pore fluid pressurization to be the dominent weakening mechanism because of the agreement we showed between the observed reduction in shear stress and that predicted by a theoretical analysis of this mechanism (Figure 2).
However, we have now repeated the experiment several time without any water in the pore spaces of the rock and we see a nearly identical weakening (Figure 3). Thus, a different mechanism must be operating in the dry rocks.
Figure 3. Comparison of frictional resistance of a wet sample (blue) with two dry samples (red and green), all at an effective normal stress of 25 MPa. The fact that both wet and dry samples show a similar reduction in apparent friction as a function of displacement suggests that the same process operates in both cases and thus cannot be due to thermal pressurization of pore fluids.
A subsequent experiment on a dry sample at a normal stress of 112 MPa showed an even more dramatic reduction in frictional resistance as shown in Figure 4. The coefficient of friction was reduced to 0.14 after 1.6 m of slip at a velocity of 3.2 m/s.
Figure 4. Comparison of weakening in dry samples of quartzite as a function of normal stress.
The reductions in friction with displacement shown in Figure 4 likely result from various degrees of shear melting on the surface. We do not yet have measurements of fault-surface temperature, nor very accurate calculations of the expected temperatures; the reduction in frictional resistance with slip precludes simple calculations assuming constant heat input. However, estimates for the temperature in the 28 MPa experiment after 3 m of slip range from 430 to 650 deg. C and for the 112 MPa experiment range from 460 to 1825 deg. C. These ranges presumably represent lower and upper bounds on the average surface temperature in each experiment, because they are estimated by assuming that the coefficient of friction during the entire slip was either the lowest value observed or the highest.
We have textural observations suggesting that local melting may have occurred in the 28 MPa experiment as shown in Figure 5. This apparent melting may have resulted from locally high transient "flash" temperatures at highly stressed asperity contacts.
Figure 5. Scanning electron micrographs of the sliding surface in a 28 MPa normal stress experiment. The cuspate boundaries of the smooth areas suggest an origin due to surface tension and suggest that local melting occurred on the sliding surface. Such textures are not representative of the entire surface, suggesting that the temperatures were only locally high enough to cause melting.
Much more dramatic wholesale melting appears to have occurred in the 112 MPa normal stress sample as shown in Figure 6 and Figure 7. The surface is greatly altered from the starting appearance (Figure 6), and in detail the dark transparent material is seen to intrude into the white powdered gouge (Figure 7).A layer of the clear brownish "glass" covers much of the sliding surface. The brown color may come from small amounts of magnetite in the quartzite.
The considerable reduction in shear stress shown at 25-28 MPa normal stress, but especially at 112 MPa normal stress (Figure 4) correlates well with the apparent amounts of melt seen in the images of Figure 5, Figure 6 and Figure 7. We are in the process of further study of the brown material using transmission electron microscopy (TEM) to establish whether it is truly a glass, and are making better calculations of the temperatures we expect for the experiments.
Although these results are still preliminary, it is apparent that very large reductions in shear stress can result from frictional heating at slip rates 2.5 orders of magnitude lower than typical seismic slip rates. This suggests that dynamic shear stresses could be very low during seismic slip due to shear melting. Shear heating-induced melting during earthquakes may be more common than usually believed, suggesting that products of such melting, i.e., pseudotachylites, often may be overlooked in studies of exhumed fault zones. If these large reductions in shear stress are chareactersitc of earthqaukes it would seem to imply that dynamic stress drops may be nearly complete and that accelerations could be quite large.
Our laboratory experiments measuring the frictional behavior of rocks show two interesting results. First, a sheared layer of ground-up granite undergoes a complex sequence of behavior. After enough sliding, slip occurs on a very sharp surface within the powdered layer and faster slip speeds lower the frictional resistance. Consequently faults containing granite should exhibit earthquakes. Second, fast sliding results in a large weakening of a fault surface accompanied by considerable melting. Thus, slip during earthquakes may occur with very low resistance and consequently strong shaking should be expected.