2026MNRAS.545f1862D

The fraction of 'dark pixels' in the Ly$\alpha$ and other Lyman-series forests at $z\sim 5$-6 provides a powerful constraint on the end of the reionization process. Any spectral region showing transmission must be highly ionized, while dark regions could be ionized or neutral, thus the dark pixel fraction provides a (nearly) model independent upper limit to the volume-filling fraction of the neutral intergalactic medium, modulo choices in binning scale and dark pixel definition. Here, we provide updated measurements of the 3.3 comoving Mpc dark pixel fraction at $z=4.85$-6.25 in the Ly$\alpha$, Ly$\beta$, and Ly$\gamma$ forests of 34 deep $5.8 \lesssim z\lesssim 6.6$ quasar spectra from the (enlarged) XQR-30 sample. Using the negative pixel method to measure the dark pixel fraction, we derive fiducial $1\sigma$ upper limits on the volume-average neutral hydrogen fraction of $\langle x_{\rm HI} \rangle \le \lbrace 0.030+0.048,0.095+0.037,0.191+0.056,0.199+0.087\rbrace$ at $\bar{z}=\lbrace 5.481,5.654,5.831,6.043\rbrace$ from the optimally sensitive combination of the Ly$\beta$ and Ly$\gamma$ forests. We further demonstrate an alternative method that treats the forest flux as a mixture of dark and transparent regions, where the latter are modelled using a physically motivated parametric form for the intrinsic opacity distribution. The resulting model-dependent upper limits on $\langle x_{\rm HI} \rangle$ are similar to those derived from our fiducial model-independent analysis. We confirm that the bulk of reionization must be finished at $z>6$, while leaving room for an extended 'soft landing' to the reionization history down to $z\sim 5.4$ suggested by Ly$\alpha$ forest opacity fluctuations.