Oral Presentation

Field-level Reconstruction from Foreground-Contaminated 21-cm Maps

Author(s): Shu-Fan Chen (Columbia University)

Presenter: Shu-Fan Chen (Columbia University)

Upcoming 21-cm experiments will open a new window into the high-redshift universe by mapping spatial fluctuations of the 21-cm signal across a wide range of redshifts. However, bright foregrounds and the limitations of radio interferometry obscure a large fraction of Fourier modes in the so-called “foreground wedge,” hiding much of the cosmological information. In this talk, I will present two complementary approaches we developed to recover these missing modes and reconstruct the full 21-cm density field. The first is a field-level inference framework based on effective field theory, which uses a fully differentiable forward model to connect initial conditions with observed, foreground-filtered maps. The second is a diffusion-based deep generative model trained on semi-numerical simulations across a broad parameter space. Both approaches enable physically motivated reconstruction of lost modes, improving cosmological constraints and facilitating cross-correlations between different experiments, even when their accessible modes have little overlap.