Compound earthquakes on a bimaterial interface and implications for rupture mechanics

Rubin and Ampuero [2007] simulated slip-weakening ruptures on a 2-D (line fault) bimaterial interface and observed differences in the timescales for the two edges to experience their peak stress after being slowed by barriers. The barrier on the "negative" side reached its peak stress when the P-wave stopping phase arrives from the opposite end, which takes ~20 ms for a 100 m event. This may be long enough for a potential secondary rupture to be observed as a distinct subevent. In contrast, the same timescale for a barrier at the "positive" front is nearly instantaneous (really the distance from the stopped rupture edge to the barrier divided by the shear wave speed), possibly making a secondary event there indistinguishable from the main rupture. Rubin and Gillard [2000] observed that of a family of 72 similar earthquakes along the San Andreas fault in Northern California, 5 were identified as compound and in all cases the second event was located on the negative (NW) side of the main event. Based on their simulations, Rubin and Ampuero interpreted this as being due to the above-mentioned asymmetry in the dynamic stressing-rate history on the two sides of a rupture on a bimaterial interface. To test this hypothesis for the asymmetric distribution of subevents within compound earthquakes, we search more systematically for secondary arrivals within 0.15 s of the first P arrival for microearthquakes on the San Andreas. We take advantage of similarity between waveforms of adjacent events and deconvolve the first 0.64 s following the P arrival of a target event using a nearby Empirical Green's Function (EGF). We use the iterative deconvolution method described in Kikuchi & Kanamori [1982]. When the EGF is a simple earthquake and the target is compound, the deconvolution is expected to show two spikes, corresponding to the main and secondary events. Due to the existence of noise, a second spike is considered robust only when the difference between the waveforms of the target event and the aligned and scaled EGF is similar enough (cross-correlation coefficient higher than 0.6) to the EGF at multiple stations. The azimuthal consistency of delays between the main and secondary arrivals is more convincing evidence that the target is a compound event. Using these criteria we temporarily identified ~70 compound events out of ~8200 in our catalog. Future work will include improving the quality of the inter-event delay time by using Monte Carlo simulations to allow the amplitudes and arrival times of both spikes (as opposed to just the second spike) to vary. Accurate relative locations and times can improve our understanding of the triggering mechanism of the subevents and perhaps the longer-timescale aftershock asymmetry observed in this region as well. For example, it has been proposed that the deficit of longer-timescale aftershocks in the SE (positive) direction could be due to triggering by propagation of a tensile stress pulse down the fault as the mainshock is stopped.