Paper detail

How robust are predictions of galaxy clustering?

We use the Millennium Simulation database to compare how different versions of the Durham and Munich semi-analytical galaxy formation models populate dark matter haloes with galaxies. The models follow the same physical processes but differ in how these are implemented. All of the models we consider use the Millennium N-body Simulation; however, the Durham and Munich groups use independent algorithms to construct halo merger histories. We compare the predicted halo occupation distributions (HODs) and correlation functions for galaxy samples defined by stellar mass, cold gas mass and star formation rate. The predictions for the HOD are remarkably similar for samples ranked by stellar mass. The predicted bias averaged over 5-25 Mpc/h is consistent between models to within 10%. At small pair separations there is a difference in the predicted clustering. This arises because the Durham models allow some satellite galaxies to merge with the central galaxy in a halo when they are still associated with resolved subhaloes. The agreement between the models is less good for samples defined by cold gas mass or star formation rate, with the small scale clustering differing by an order of magnitude, reflecting the uncertainty in the modelling of star formation. The model predictions in these cases are qualitatively similar, with a markedly shallower slope for the correlation function than is found for stellar mass selected samples and with the HOD displaying an asymmetric peak for central galaxies. We provide illustrative parametric fits to the HODs predicted by the models. Our results reveal the current limitations on how well we can predict galaxy bias in a fixed cosmology, which has implications for the interpretation of constraints on the physics of galaxy formation from galaxy clustering measurements and the ability of future galaxy surveys to measure dark energy.