Briefly, the current thinking, although not all agree, is that the sliding in an earthquake starts days aheads of time, starting imperceptibly slowly and accelerating for weeks until it achieves rates comparable with seismic waves - km/s crack speeds and m/s particle velocities. Two problems keep us from finding the "nucleation zone" - (1) it is probably very small, perhaps 1-10 m, (2) the slow precursory slip is tiny, much of the time microns per second or less. Some seismologists don't think such a phase even exists. The theory is called rate-and-state friction, and it has been associated with Jim's name.