Abstract Cosmological simulations are essential for inferring cosmological and galaxy population properties based on forward-modelling, but this typically requires finding the population of (sub)haloes and galaxies that they contain. The properties of said populations vary depending on the algorithm used to find them, which is concerning as it may bias key statistics. We compare how the predicted (sub)halo mass functions, satellite radial distributions and correlation functions vary across algorithms in the dark-matter-only and hydrodynamical versions of the FLAMINGO simulations. We test three representative approaches to finding subhaloes: grouping particles in configuration- (Subfind), phase- (ROCKSTAR and VELOCIraptor) and history-space (HBT-HERONS). We also present HBT-HERONS, a new version of the HBT+ subhalo finder that improves the tracking of subhaloes. We find 10%-level differences in the M200c mass function, reflecting different field halo definitions and occasional miscentering. The bound mass functions can differ by 75% at the high mass end, even when using the maximum circular velocity as a mass proxy. The number of well-resolved subhaloes differs by up to 20% near R200c, reflecting differences in the assignment of mass to subhaloes and their identification. The predictions of different subhalo finders increasingly diverge towards the centres of the host haloes. The performance of most subhalo finders does not improve with the resolution of the simulation and is worse for hydrodynamical than for dark-matter-only simulations. We conclude that HBT-HERONS is the preferred choice of subhalo finder due to its low computational cost, self-consistently made and robust merger trees, and robust subhalo identification capabilities.
Moreno et al. (2025) studied this question.