Read the full paper here.

This paper investigates recent observational reports of damping wing–like absorption features at the edges of Lyα dark gaps in high-redshift quasar spectra. These features have been interpreted as evidence for large, mostly neutral hydrogen regions (“neutral islands”) persisting down to $z \lesssim 5.5$, well after the expected end of reionization. Using the stacking procedure developed in earlier observational studies, we test whether fully coupled cosmological simulations can reproduce these signals.

The analysis uses synthetic Lyα and Lyβ spectra generated from two state-of-the-art reionization simulations: CROC and THESAN. Both simulations reproduce the observed abundance of long dark gaps, but when Lyβ-defined gap edges are stacked following the observational method, neither simulation shows any damping wing–like suppression on the Lyα side. This holds even when realistic noise is added.

Examining the simulated gaps clarifies the discrepancy. In both CROC and THESAN, most dark gaps are not neutral islands—they arise from long path lengths through moderately overdense, fully ionized filaments rather than through mostly neutral regions. To test whether neutral islands would produce the observed effect, we manually insert artificial fully neutral segments into the spectra. A clear damping wing only emerges when neutral islands account for $\gtrsim 90\%$ of all dark gaps, far higher than in the simulations.

The paper also shows that any mostly neutral region at $z \sim 5.9$ should be short-lived. A simple photon-conservation argument implies ionization-front speeds of $v_I \sim 2.4 \times 10^4\ \mathrm{km\ s^{-1}}$, sufficient to ionize a 10 cMpc region in only $\sim 60$ Myr. Neutral islands should therefore not survive to $z < 5.5$ unless the local photoionization rate is strongly suppressed. CROC shows no such suppression near gap edges, and THESAN shows only a modest $\sim 20\%$ decrease—too small to extend island lifetimes.

These results suggest that current fully coupled simulations may underestimate large-scale fluctuations in the ionizing background—potentially due to limitations in modeling rare sources, escape-fraction variations, or Lyman-limit systems—and may therefore miss a population of long-lived, partially neutral regions. If the observational detections are robust, they pose a significant challenge to existing reionization models.