What drives the HI content of central galaxies -- A comparison between hydrodynamic simulations and observations using Random Forest

Abstract

We investigate the driving mechanisms for the HI gas content in star-forming central galaxies at low redshift, by examining the HI-to-stelalr mass ratio ($M_{\rm HI}/M_\ast$) in both the state-of-the-art hydrodynamic simulations, IllustrisTNG (TNG) and EAGLE, and the xGASS sample. We quantify the correlations of $M_{\rm HI}/M_\ast$ with a variety of galaxy properties using the random forest regression technique, and we make comparisons between the two simulations, as well as between the simulations and xGASS. Gas-phase metallicity is found to be most important in both simulations, but is ranked mildly for xGASS, suggesting that metals and gas driven by feedback effects in real galaxies is not as tightly coupled as in the simulations. Beyond that, the accretion rate of supermassive black holes is the most important feature in TNG, while specific star formation rate is the top ranked in EAGLE. This result can be understood from the fact that the HI gas is regulated mainly by thermal-mode AGN feedback in TNG and by stellar feedback in EAGLE. Although neither simulation can fully reproduce the feature importance obtained for real galaxies in the xGASS, EAGLE performs better than TNG in the sense that the observationally top-ranked property, $u-r$, is also highly ranked in EAGLE. This result implies that stellar feedback plays a more dominant role than AGN feedback in driving the HI gas content of low-redshift galaxies.