StarNet: An In Situ Surrogate Model of Primordial Star Formation and Feedback for Astrophysical Simulations

Abstract This work introduces StarNet , an in situ surrogate model for primordial star formation and feedback within cosmological simulations. Traditional simulations of primordial star formation and supernovae require extreme resolution to resolve both time and spatial scales of star formation. Our framework combines deep learning models and empirical feedback parameterizations from high-resolution simulations to predict and model heterogeneous metal enrichment from Population (Pop) III stars. By coupling StarNet with the Enzo simulation suite, we model the spatially and temporally varying enrichment patterns of Pop III associations without the computational burden of resolving their fine-scale dynamics. Comparisons with high-resolution ground-truth simulations demonstrate the framework’s ability to replicate key features of Pop III feedback: volume fraction enrichment and spatial distribution of enriched halos and galaxies.