We incorporate a model for black hole growth during galaxy mergers into the semi-analytical galaxy formation model based on Lambda-CDM proposed by Baugh et al. (2005). Our black hole model has one free parameter, which we set by matching the observed zeropoint of the local correlation between black hole mass and bulge luminosity. We present predictions for the evolution with redshift of the relationships between black hole mass and bulge properties. Our simulations reproduce the evolution of the optical luminosity function of quasars. We study the demographics of the black hole population and address the issue of how black holes acquire their mass. We find that the direct accretion of cold gas during starbursts is an important growth mechanism for lower mass black holes and at high redshift. On the other hand, the re-assembly of pre-existing black hole mass into larger units via merging dominates the growth of more massive black holes at low redshift. This prediction could be tested by future gravitational wave experiments. As redshift decreases, progressively less massive black holes have the highest fractional growth rates, in line with recent claims of "downsizing" in quasar activity.

Recent observations of the distant Universe suggest that much of the stellar mass of bright galaxies was already in place at $z>1$. This presents a challenge for models of galaxy formation because massive halos are assembled late in hierarchical cosmologies such as cold dark matter (CDM). In this paper, we discuss a new implementation of the Durham semi-analytic model in which feedback due to active galactic nuclei (AGN) is assumed to quench cooling flows in massive halos. This mechanism naturally creates a break in the local galaxy luminosity function at bright magnitudes. The model is implemented within the Millennium N-body simulation; the accurate dark matter merger trees and large number of realizations of the galaxy formation process that the simulation provides results in highly accurate statistics. After adjusting the values of the physical parameters in the model by reference to the properties of local galaxies, we use it to investigate the evolution of the K-band luminosity and galaxy stellar mass functions. We also calculate the volume averaged star formation rate density of the Universe as a function of redshift and the way in which this is apportioned amongst galaxies of different mass. The model robustly predicts a substantial population of massive galaxies out to redshift $z\sim 5$ and a star formation rate density which rises with increasing redshift in objects of all masses. Although observational data on these properties have been cited as evidence for ``anti-hierarchical'' galaxy formation, we find that when AGN feedback is taken into account, the fundamentally hierachical CDM model provides a very good match to these observations.