A predicted new population of UV-faint galaxies at z ≳ 4

We show that a bursty model of star formation explains several puzzling observations of highredshift galaxies.We begin by showing that because the observed star formation rate integrated over a Hubble time exceeds the observed stellar mass by an order of magnitude, the specific star formation rate requires a duty-cycle of ~10 per cent. We use the specific star formation rate to calibrate a merger-driven model of star formation regulated by supernova feedback, and reproduce the star formation rate density and stellar mass functions of galaxies at 4 ≲ z ≲ 7. The specific star formation rate is predicted not to evolve rapidly with either mass or redshift at z ≳ 4, consistent with observation. This is in contrast to expectations from hydrodynamical simulations where star formation closely follows accretion rate, and increases strongly towards high redshift. Bursty star formation explains the observation that there is not enough stellar mass at z ~ 2-4 to account for all star formation observed. A duty-cycle of ~10 per cent implies that there could be 10 times the number of known high-redshift galaxies at fixed stellar mass that have not yet been detected through UV selection. We therefore predict the possible existence of an undetected population of UV-faint galaxies that accounts for most of the stellar mass density at z ~ 4-8.