The specific star formation rate of high redshift galaxies: the case for two modes of star formation

We study the specific star formation rate (SSFR) and its evolution at z greater than or similar to 4, in models of galaxy formation, where the star formation is driven by cold accretion flows. We show that constant star formation and feedback efficiencies cannot reproduce the observed trend of SSFR with stellar mass and its observed lack of evolution at z > 4. Model galaxies with log(M-*) less than or similar to 9.5M(circle dot) show systematically lower SSFRs by orders of magnitudes, while massive galaxies with M-* greater than or similar to 5 x 10(10) M-circle dot have up to an order of magnitude larger SSFRs, compared to recent observations by Stark et al. To recover these observations we apply an empirical star formation efficiency in galaxies that scales with the host halo velocity dispersion as proportional to 1/sigma(3) during galaxy mergers. We find that this modification needs to be of stochastic nature to reproduce the observations, i.e. only applied during mergers and not during accretion driven star formation phases. Our choice of star formation efficiency during mergers allows us to capture both, the boost in star formation at low masses and the quenching at high masses, and at the same time produce a constant SSFR-stellar mass relation at z greater than or similar to 4 under the assumption that most of the observed galaxies are in a merger-triggered star formation phase. Our results suggest that observed high-z low-mass galaxies with high SSFRs are likely to be frequently interacting systems, which experienced bursts in their star formation rate and efficiency (mode 1), in contrast to low redshift z less than or similar to 3 galaxies which are cold accretion-regulated star forming systems with lower star formation efficiencies (mode 2).