Microstructural controls on seismicity distribution in simulated fault gouges

In order to make seismic hazard estimates, it is necessary to assume some distribution of the number of earthquakes of a certain magnitude, i.e. a Gutenberg-Richter distribution. This is true for both natural seismicity as well as induced seismicity, but in both cases the number of historical earthquakes at the tail end of the distribution (i.e. the largest ones) is limited and often the maximum possible magnitude is unknown. In contrast, in a laboratory setting the maximum size of an unstable slip event (“stick-slip” or laboratory earthquake) is controlled by the size of the sample and the imposed stress. In our rotary shear apparatus, we can theoretically achieve unlimited fault displacement which allows us to produce earthquake-like distributions with thousands to tens of thousands event.

In this presentation, I will present results from experiments on simulated fault gouges which show unstable frictional behaviour at room temperature conditions. The results show that the event size distribution can change spontaneously, without any changes in the boundary conditions. Observations of fault gouge material after the experiment, suggest that wear of the granular material generates alternative surfaces for slip, which changes the macroscopic behaviour. Interestingly, the change in event size distribution is reversable, presumably because the fine-grained layers become disturbed with ongoing shear.

In an attempt to simulate the macroscopic behaviour, we have, for the first time, measured the rate-and-state frictional (RSF) properties on single grain contacts. Using these values in a numerical model for seismic slip (so-called “seismic cycle simulator”), we obtain maximum stress drops that are comparable to those obtained in the experiments, but with some differences. The differences are most likely due to the fact that the grains in our simulated fault are affected wear in previous slip events, which should change their RSF parameters. In addition, the normal stress at each individual grain contact is unknown in the experiment and could vary significantly from event to event.  This latter difference between model and experiment can be overcome by using a discrete element method with contact-scale RSF properties to simulate slip.