The sensitivity of stellar feedback to IMF averaging versus IMF sampling in galaxy formation simulations

Abstract: Galaxy formation simulations frequently use Initial Mass Function (IMF) averaged feedback prescriptions, where star particles are assumed to represent single stellar populations that fully sample the IMF. This approximation breaks down at high mass resolution, where stochastic variations in stellar populations become important. We discuss various schemes to populate star particles with stellar masses explicitly sampled from the IMF. We use Monte Carlo numerical experiments to examine the ability of the schemes to reproduce an input IMF in an unbiased manner while conserving mass. We present our preferred scheme which can easily be added to pre-existing star formation prescriptions. We then carry out a series of high resolution isolated simulations of dwarf galaxies with supernovae, photoionization and photoelectric heating to compare the differences between using IMF averaged feedback and explicitly sampling the IMF. We find that if supernovae are the only form of feedback, triggering individual supernovae from IMF averaged rates gives identical results to IMF sampling. However, we find that photoionization is more effective at regulating star formation when IMF averaged rates are used, creating more, smaller H II regions than the rare, bright sources produced by IMF sampling. We note that the increased efficiency of the IMF averaged feedback versus IMF sampling is not necessarily a general trend and may be reversed depending on feedback channel, resolution and other details. However, IMF sampling is always the more physically motivated approach. We conservatively suggest that it should be used for star particles less massive than $\sim500\,\mathrm{M_\odot}$.