The radio luminosity function from the low-frequency 3CRR, 6CE and 7CRS complete samples

We measure the radio luminosity function (RLF) of steep-spectrum radio sour ces using three redshift surveys of flux-limited samples selected at low (1 51 and 178 MHz) radio frequency, low-frequency source counts and the local RLF. The redshift surveys used are the new 7C Redshift Survey (7CRS) and th e blighter 3CRR and 6CE surveys totalling 356 sources with virtually comple te redshift z information, This yields unprecedented coverage of the radio luminosity versus z plane for steep-spectrum sources, and hence the most ac curate measurements of the steep-spectrum RLF yet made, We find that a simp le dual-population model for the RLF fits the data well, requiring differen tial density evolution (with z) for the two populations. The low-luminosity population can be associated with radio galaxies with weak emission lines, and includes sources with both FRI and FRII radio structures; its comoving space density rho rises by about one dex between z - 0 and 1 but cannot ye t be meaningfully constrained at higher redshifts. The high-luminosity popu lation can be associated with radio galaxies and quasars with strong emissi on lines, and consists almost exclusively of sources with FRII radio struct ure; its rho rises by nearly three dex between z similar to 0 and 2, These results mirror the situation seen in X-ray and optically selected samples o f AGN where: (i) low-luminosity objects exhibit a gradual rise in rho with z that crudely matches the rises seen in the rates of global star formation and galaxy mergers; and (ii) the density of high-luminosity objects rises much more dramatically. The integrated radio luminosity density of the comb ination of the two populations is controlled by the value of rho at the low -luminosity end of the RLF of the high-luminosity population a quantity whi ch has been directly measured at z - 1 by the 7CRS, We argue that robust de termination of this quantity at higher redshifts requires a new redshift su rvey based on a large (similar to 1000 source) sample about five times fain ter than the 7CRS.