Modelling the selection of galaxy groups with end to end simulations

TL;DR

This paper models the selection process of galaxy groups in the X-GAP sample using simulations, neural networks, and observed data to understand biases, completeness, and purity, aiding comparisons with hydrodynamical models.

Contribution

It introduces a forward modelling approach combining simulations and neural networks to accurately reproduce the selection function of galaxy groups in X-ray and optical surveys.

Findings

50% completeness at 450 km/s velocity dispersion

93% purity level in the parent sample

Velocity dispersion measurement errors dominate for small groups

Abstract

Feedback from supernovae and AGN shapes galaxy formation and evolution, yet its impact remains unclear. Galaxy groups offer a crucial probe, as their binding energy is comparable to that available from their central AGN. The XMM-Newton Group AGN Project (X-GAP) is a sample of 49 groups selected in X-ray (ROSAT) and optical (SDSS) bands, providing a benchmark for hydrodynamical simulations. In sight of such a comparison, understanding selection effects is essential. We aim to model the selection function of X-GAP by forward modelling the detection process in the X-ray and optical bands. Using the Uchuu simulation, we build a halo light cone, predict X-ray group properties with a neural network trained on hydro simulations, and assign galaxies matching observed properties. We compare the selected sample to the parent population. Our method provides a sample that matches the observed distribution of X-ray luminosity and velocity dispersion. The 50% completeness is reached at a velocity dispersion of 450 km/s in the X-GAP redshift range. The selection is driven by X-ray flux, with secondary dependence on velocity dispersion and redshift. We estimate a 93% purity level in the X-GAP parent sample. We calibrate the velocity dispersion-halo mass relation. We find a normalisation and slope in agreement with the literature, and an intrinsic scatter of about 0.06 dex. The measured velocity dispersion is accurate within 10% only for rich systems with more than about 20 members, while the velocity dispersion for groups with less than 10 members is biased at more than 20%. The X-ray follow-up refines the optical selection, enhancing purity but reducing completeness. In an SDSS-like setup, velocity dispersion measurement errors dominate over intrinsic scatter. Our selection model will enable the comparisons of thermodynamic properties and gas fractions between X-GAP groups and hydro simulations.