Tracing the Cosmic Origins: Machine Learning Reconstruction of the Primordial Density Field from EoR Observations
arXiv:2609.05412
Abstract
Reconstructing the initial conditions of the Universe from late-time tracers would unlock cosmological information buried by non-linear structure formation and astrophysics. We reconstruct the initial density field at from simulated 21-cm and CO(1-0) line-intensity maps at generated with LIMFAST. Using a three-dimensional U-Net, we reconstruct the initial conditions and evaluate its impact on cosmological parameter constraints. The two tracers probe complementary environments: 21-cm emission traces neutral, low-density regions of the intergalactic medium, while CO traces overdense, star-forming regions. To emulate realistic observations, we model instrumental effects for SKA1-Low- and COMAP-ERA-like surveys, including finite angular resolution and thermal noise. We assess reconstruction performance through the cross-correlation coefficient between reconstructed and true initial density fields, . In the noiseless case, combining both tracers delivers the most accurate recovery across ionisation states, with 0.90 for 0.75 Mpc. With observational effects, small-scale information is degraded, but combining tracers still achieves 0.70 for 0.3 Mpc. To quantify information gain, we perform simulation-based inference of cosmological parameters from power-spectrum summaries before and after reconstruction. In both noiseless and noisy settings, reconstruction tightens parameter constraints: uncertainties on and improve by , with smaller but consistent gains for other parameters. This is further confirmed using Kullback-Leibler divergence diagnostics for an ensemble of observations. These results indicate that joint analysis of future 21-cm and CO surveys, combined with such reconstruction, can partially recover otherwise inaccessible cosmological information.
18 pages, 13 figures. Submitted to the Monthly Notices of the Royal Astronomical Society (MNRAS)