Ah got it, you simulate individually per atom, I thought you were describing the entire population. Indeed, you can always use that vector as the parameter to a multinomial distribution and recover the real distribution.

Regarding the Pareto distribution, random growth model are probably the reason why power laws are everywhere.