Determinations of the UV luminosity function of AGN at high redshifts are important for constraining the AGN contribution to reionization and understanding the growth of supermassive black holes. Recent inferences of the luminosity function suffer from inconsistencies arising from inhomogeneous selection and analysis of data. We address this problem by constructing a sample of more than 80,000 colour-selected AGN from redshift z = 0 to 7.5 using multiple data sets homogenised to identical cosmologies, intrinsic AGN spectra, and magnitude systems. Using this sample, we derive the AGN UV luminosity function from redshift z = 0 to 7.5. The luminosity function has a double power law form at all redshifts. The break magnitude M * shows a steep brightening from M * ∼ −24 at z = 0.7 to M * ∼ −29 at z = 6. The faint-end slope β significantly steepens from −1.7 at z < 2.2 to −2.4 at z 6. In spite of this steepening, the contribution of AGN to the hydrogen photoionization rate at z ∼ 6 is subdominant (< 3%), although it can be non-negligible (∼ 10%) if these luminosity functions hold down to M 1450 = −18. Under reasonable assumptions, AGN can reionize He II by redshift z = 2.9. At low redshifts (z < 0.5), AGN can produce about half of the hydrogen photoionization rate inferred from the statistics of H I absorption lines in the IGM. Our analysis also reveals important systematic errors in the data, which need to be addressed and incorporated in the AGN selection function in future in order to improve our results. We make various fitting functions, codes, and data publicly available.

High-redshift quasi-stellar object (QSO) spectra show large spatial fluctuations in the Ly α opacity of the intergalactic medium on surprisingly large scales at $z$ ≳ 5.5. We present a radiative transfer simulation of cosmic reionization driven by galaxies that reproduces this large scatter and the rapid evolution of the Ly α opacity distribution at 5 &lt; $z$ &lt; 6. The simulation also reproduces the low Thomson scattering optical depth reported by the latest cosmic microwave background (CMB) measurement and is consistent with the observed short near-zones and strong red damping wings in the highest redshift QSOs. It also matches the rapid disappearance of observed Ly α emission by galaxies at $z$ ≳ 6. Reionization is complete at $z$ = 5.3 in our model, and 50 per cent of the volume of the Universe is ionized at $z$ = 7. Agreement with the Ly α forest data in such a late reionization model requires a rapid evolution of the ionizing emissivity of galaxies that peaks at $z$ ∼ 6.8. The late end of reionization results in a large scatter in the photoionization rate and the neutral hydrogen fraction at redshifts as low as $z$ ≲ 5.5 with large residual neutral ‘islands’ that can produce very long Gunn–Peterson troughs resembling those seen in the data.

We have re-analyzed the data in which Bowman et al. (2018) identified a feature that could be due to cosmological 21-cm line absorption in the intergalactic medium at redshift z ∼ 17. If we use exactly their procedures then we find almost identical results, but the fits imply either non-physical properties for the ionosphere or unexpected structure in the spectrum of foreground emission (or both). Furthermore we find that making reasonable changes to the analysis process, e.g., altering the description of the foregrounds or changing the range of frequencies included in the analysis, gives markedly different results for the properties of the absorption profile. We can in fact get what appears to be a satisfactory fit to the data without any absorption feature if there is a periodic feature with an amplitude of ∼ 0.05 K present in the data. We believe that this calls into question the interpretation of these data as an unambiguous detection of the cosmological 21-cm absorption signature.