Ah, we use the ML methods to understand what happens to a large collection of galaxies (maybe 100k - 1 million), all co-evolving together. Trying to use these m

Ah, we use the ML methods to understand what happens to a large collection of galaxies (maybe 100k - 1 million), all co-evolving together. Trying to use these methods v.s. direct simulation on individual galaxies is a no-go, too many nonlinearities exist. Your comment about a numerical gravitational simulation being 'wrong' because of chaos may be true on an individual resolution element level, yes, but our simulations are made up of many billions of individual particles. Like traditional CFD simulations we can also demonstrate some level of convergence.