Physical processes leading to laboratory rupture have a nonlinear character which justifies optimism that earthquakes are predictable. However, nonlinear effects at the source of a smaller earthquake may be an insufficient basis for the prediction of an earthquake with rupture lengths tens to hundreds of kilometers. Thus, a strategy for predicting such major or great events must depend on knowing the state of readiness of the entire length of the suspect fault. The two most promising avenues are: (1) to search for variations in shear and dilatational strains within the fault zone and in the bordering regions that signify the onset of a Stuart instability; and (2) to follow in detail the seismicity in response to the temporal variations of shear and normal strains as a measure of what parts of a long fault are at the point of failure.
An observational strategy based on these physical models requires detailed and frequent to continuous observation of strain within the fault and border zones. Focal mechanisms and accurate locations of earthquakes M ≧ 2 are needed to relate seismicity to a given configuration of the strain field. With the exception of observations such as levels in water wells, which carry information on the strain field, a purely empirical search for earthquake precursors needs to be deemphasized.