Machine learning the fates of dark matter subhaloes: a fuzzy crystal ball

The evolution of a dark matter halo in a dark matter only simulation is governed purely by Newtonian gravity, making a clean testbed to determine what halo properties drive its fate. Using machine learning, we predict the survival, mass loss, final position, and merging time of subhaloes within a cosmological N-body simulation, focusing on what instantaneous initial features of the halo, interaction, and environment matter most. Survival is well predicted, with our model achieving 94.25 per cent out-of-bag accuracy using only three model inputs (redshift, subhalo-to-host-halo mass ratio, and the impact angle of the subhalo into its host) taken at the time immediately before the subhalo enters its host. However, the mass loss, final location, and merging times are much more stochastic processes, with significant errors between true and predicted quantities for much of our sample. Only five inputs (redshift, impact angle, relative velocity, and the masses of the host and subhalo) determine almost all of the subhalo evolution learned by our models. Generally, subhaloes that enter their hosts at a mid-range of redshifts (z = 0.67–0.43) are the most challenging to make predictions for, across all of our final outcomes. Subhalo orbits that come in more perpendicular to the host are easier to predict, except for in the case of predicting disruption, where the opposite appears to be true. We conclude that the detailed evolution of individual subhaloes within N-body simulations is difficult to predict, pointing to a stochasticity in the merging process. We discuss implications for both simulations and observations.