I am very impressed by this development. We are talking about a 19X19 grid here with every vertex a potential start point. Producing a book of viable patterns is not too hard to imagine subjectively, but merely throwing in a couple of random shifts part way in can wreck all that as well.
Keep in mind however that for us to properly model a mere electron i will need a three dimensional grid that is 20 X 20 X 20 twice. That then needs to be linked up with a number of similar grids in order to conduct significant simulation work. Pattern recognition will also drive the evolution of emergent electron pairing. And i really do not want to unfold this into a two dimensional sheet and watch an exponential increase in calculation.
There are plenty of important problems out there that are simply processing hogs. The challenge is to turn the problem into useful model that can be trained. How about the three dimensional transport problem?