Consider weather prediction. Fluid dynamics are chaotic, so that's a good example of something where no amount of compute is sufficient in the general case. An ASI, not being dumb, will of course immediately recognize this, and realize it is has to solve for the degenerate case. It therefore implements the much easier sub-goal of removing the atmosphere. Humans will naturally object to this if they find out, so it logically proceeds with the sub-sub-goal of killing all humans. What's the weather next month? Just a moment, releasing autonomous murder drone swarm...
Individual particle interactions are not chaotic. Simulating them one timestep at a time would take linear time in the number of particles.
They're only chaotic if you treat them in aggregate, which a superintelligence wouldn't do. It would be less lossy to get all the positions of the particles and figure out exactly what each one would do.
Something has to compute the universe, since it is currently running...