The origin and evolution of the mass-metallicity relation at high redshift using GalICS

The Galaxies in Cosmological Simulations (GalICS) semi-analytical model of hierarchical galaxy formation is used to investigate the effects of different galactic properties, including star formation rate (SFR) and outflows, on the shape of the mass-metallicity relation and to predict the relation for galaxies at redshift z = 2.27 and 3.54. Our version of GalICS has the chemical evolution implemented in great detail and is less heavily reliant on approximations, such as instantaneous recycling. We vary the model parameters controlling both the efficiency and redshift dependence of the SFR as well as the efficiency of supernova feedback. We find that the factors controlling the SFR influence the relation significantly at all redshifts and require a strong redshift dependence, proportional to 1 + z, in order to reproduce the observed relation at the low-mass end. Indeed, at any redshift, the predicted relation flattens out at the high-mass end resulting in a poorer agreement with observations in this regime. We also find that variation in the parameters associated with outflows has a minimal effect on the relation at high redshift but does serve to alter its shape in the more recent past. We thus conclude that the relation is one between the SFR and mass and that outflows are only important in shaping the relation at late times. When the relation is stratified by the SFR, it is apparent that the predicted galaxies with increasing stellar masses have higher SFRs, supporting the view that galaxy downsizing is the origin of the relation. Attempting to reproduce the observed relation, we vary the parameters controlling the efficiency of star formation and its redshift dependence and compare the predicted relations with those of Erb et al. at z = 2.27 and Maiolino et al. at z = 3.54 in order to find the best-fitting parameters. We succeed in fitting the relation at z = 3.54 reasonably well; however, we fail at z = 2.27, our relation lying on average below the observed one at the one standard deviation level. We do, however, predict the observed evolution between z = 3.54 and 0. Finally, we discuss the reasons for the above failure and the flattening at high masses, with regards to both the comparability of our predictions with observations and the possible lack of underlying physics. Several of these problems are common to many semi-analytic/hybrid models and so we discuss possible improvements and set the stage for future work by considering how the predictions and physics in these models can be made more robust in light of our results.