Simulations of the galaxy population constrained by observations from z=3 to the present day: implications for galactic winds and the fate of their ejecta

We apply Monte Carlo Markov Chain (MCMC) methods to large-scale semi-analytic simulations of galaxy formation in a Lambda cold dark matter cosmology in order to explore how star formation and feedback are constrained by the observed luminosity and stellar mass functions of galaxies. We build models jointly on the Millennium and Millennium-II dark matter-only simulations, applying fast sampling techniques which allow observed galaxy abundances over the ranges 7 < log M-star/M-circle dot < 12 and 0 <= z <= 3 to be used simultaneously as constraints in the MCMC analysis. When z = 0 constraints alone are imposed, we reproduce the results of previous modelling by Guo et al., but no single set of parameter values can reproduce observed galaxy abundances at all redshifts simultaneously, reflecting the fact that low-mass galaxies form too early and thus are overabundant at high redshift in this model. The data require the efficiency with which galactic wind ejecta are reaccreted to vary with redshift and halo mass quite differently than previously assumed. We propose a specific model in which reincorporation time-scales vary inversely with halo mass and are independent of redshift. This produces an evolving galaxy population which fits observed abundances as a function of stellar mass, B- and K-band luminosity at all redshifts simultaneously. It also produces a significant improvement in two other areas where previous models were deficient. It leads to present day dwarf galaxy populations which are younger, bluer, more strongly star forming and more weakly clustered on small scales than before, although the passive fraction of faint dwarfs remains too high.