We've gone through the reasonable assumptions of the next San Andreas earthquake, that's what those folks burning supercomputer oil are doing. Now, let's be unreasonable...
Why? Let's digress.... I've been in the seismic nuclear biz for 30 years. During that time, we always asked the question - "What sort of ground motions are we going to get?". This has been answered using Uniform Hazard Spectra, which is a probabilistic way to get these answers. There is a big problem when you get to very low probabilities (which is explained in the paper), in that you are essentially starting to divide by zero. This results in huge, unrealistic accelerations, etc. We ran into that with the Dam Safety program, where we had to anchor down concrete gravity dams! (rather Quebecish that).
So, these guys had the bright idea of determining the physical limits of ground motion, which is something I have advocated over the years.
The researchers devised a numerical model of a fault that made anything realistic look like a damp squib. They stressed up a thrust fault to the max, set the critical displacement to a tiny value, and made the dynamic friction drop to zero. Then they let it go!
-to be continued.