From THESAN-ZOOM to JWST: Predicting ionizing photon escape and the rise of UV-bright reionization sources

Abstract: Understanding the sources and evolution of cosmic reionization remains a central challenge in astrophysics, with the escape of ionizing Lyman-continuum (LyC) photons from early galaxies representing a major uncertainty. In this work, we use more than 35,000 galaxy realisations from the THESAN-ZOOM cosmological radiation-hydrodynamic simulations to identify indirect diagnostics of the LyC photon escape fraction ($f_\mathrm{esc}$) and the LyC photon escape rate ($\dot{N}\mathrm{ion,esc}$) across the redshift range $z=3-16$. We train random forest regression models using these diagnostics to predict both quantities. We present four models: two trained with the full set of simulation-derived indicators to predict $f\mathrm{esc}$ and $\dot{N}\mathrm{ion,esc}$, and two restricted to observables accessible to JWST photometric surveys. We find the 10-to-100$\,$Myr star-formation rate ratio ($\mathrm{SFR}{10} / \mathrm{SFR}{100}$) and the gas-to-stellar mass ratio ($M\mathrm{gas} / M_*$) to be the strongest diagnostics of $f_\mathrm{esc}$, suggesting a strong relationship between ionizing photon escape and gas clearing through bursty star formation. In contrast, rest-frame UV ($1500 \, Å$) absolute magnitude ($M_\mathrm{UV}$) dominates $\dot{N}\mathrm{ion,esc}$ prediction. Motivated by the strong predictive power of $M\mathrm{UV}$, we combine observed UV luminosity functions with derived $\dot{N}\mathrm{ion,esc} - M\mathrm{UV}$ relations to construct histories of reionization. These are consistent with observational constraints, avoiding the recently reported crisis in the ionizing photon budget. Our analysis suggests that the bulk of reionization occurred rapidly after $z \approx 8$, driven by UV-bright galaxies, with the $M_\mathrm{UV} < -17$ populations providing the dominant contribution.