Description
Precision inference from 21 cm intensity mapping surveys requires robust separation of signal, noise, and foregrounds across frequency channels and redshift bins. In this work, I present a Gibbs sampling framework that operates mode-by-mode in harmonic space, enabling efficient inference of redshift-binned sky maps while rigorously propagating uncertainties into power spectrum estimation — with a focus on all-sky SPHEREx-like surveys spanning broad redshift coverage.
Such surveys produce harmonic-space data across many redshift bins, with signal and foreground components exhibiting strong cross-bin correlations driven by large-scale cosmic structure. I introduce a Gibbs sampling approach that operates mode-by-mode in harmonic space by sampling spherical harmonic modes independently at fixed multipole, enabling efficient sampling of redshift-binned sky maps while rigorously propagating uncertainties into power spectrum estimation.
The formalism is general and modular, designed to accommodate future extensions such as cross-correlation with SPHEREx infrared maps, inclusion of large-scale diffuse and point-source foregrounds, and realistic noise mitigation. In doing so, it provides a flexible path toward unbiased recovery of cosmological signals from the cosmic dawn and epoch of reionisation, even in the presence of complex systematics.
These results demonstrate the feasibility of full-sky Bayesian methods for next-generation intensity mapping experiments, offering a principled foundation for cosmological parameter estimation in the low-frequency regime.
