Cosmological simulations predict that AGN preferentially live in gas-rich, star-forming galaxies despite effective feedback
Negative feedback from active galactic nuclei (AGN) is the leading mechanism for the quenching of massive galaxies in the vast majority of modern galaxy evolution models. However, direct observational evidence that AGN feedback causes quenching on a population scale is lacking. Studies have shown that luminous AGN are preferentially located in gas-rich and star-forming galaxies, an observation that has sometimes been suggested to be in tension with a negative AGN feedback picture. We investigate three of the current cosmological simulations (IllustrisTNG, EAGLE and SIMBA) along with post-processed models for molecular hydrogen gas masses and perform similar tests to those used by observers. We find that the simulations predict: (i) no strong negative trends between AGN luminosity and molecular gas fraction or sSFR; (ii) both high-luminosity ($L_{bol}>10^{44}$ erg/s) and high-Eddington ratio (>1%) AGN are preferentially located in galaxies with high molecular gas fractions and sSFR; and (iii) that the gas-depleted and quenched fractions of AGN host galaxies are lower than a control sample of non-active galaxies. These three findings are in qualitative agreement with observational samples at $z=0$ and $z=2$ and show that such results are not in tension with the presence of strong AGN feedback, which all simulations we employ require to produce realistic massive galaxies. However, we also find quantifiable differences between predictions from the simulations, which could allow us to observationally test the different subgrid feedback models.