On the evolution of the bi-modal distribution of galaxies in the plane of specific star formation rate versus stellar mass

We have compared the observed distribution of galaxies in the plane of specific star formation rate versus stellar mass with the predictions of the Garching semi-analytic model at redshifts 0, 1 and 2. The goal is to test whether the implementation of radio-mode active galactic nucleus feedback, which, in the simulation, is responsible for terminating the formation of stars in high-mass galaxies, provides an adequate match to current high-redshift observations. The fraction of quenched galaxies as a function of stellar mass in the models is in good agreement with data at z = 0 and 1. By z = 2, there are too few quenched galaxies with low stellar masses in the models. At z = 2, the population of galaxies with no ongoing star formation is clearly separated from the 'main sequence' of star-forming galaxies in the data. This is not found in the models, because z = 2 galaxies with stellar masses less than similar to 10(11) M-circle dot are predicted to host black holes with relatively low masses (less than 10(8) M-circle dot). The current implementation of radio-mode feedback from such black holes reduces the cooling rates from the surrounding halo, but does not generate sufficient energy to stop star formation entirely. We suggest that the models may be brought into better agreement with the data if black hole growth is triggered by disc instabilities in addition to major mergers, and if feedback mechanisms associated with the formation of galactic bulges act to quench star formation in galaxies.