Orateur
Description
Spatial variations of the polarized dust spectral energy distribution (SED) are a major concern for the search for primordial Cosmic Microwave Background (CMB) $B$-modes, as they could be misinterpreted as a cosmological signal. These variations are however difficult to characterize, since only a few noisy polarized channels trace dust in currently available surveys. For this purpose, we propose a new Bayesian method to characterize these SED variations from noisy Planck data at 217 and 353 GHz. We formulate this task as an inverse problem, where noise and CMB are additive contaminants. We solve it using a recently introduced approach, in which Scattering Transform (ST) statistics provide a low-dimensional representation of the signal. Cross-ST statistics between frequencies allows to capture inter-frequency information. We perform the inference in the ST space, where we recover a posterior distribution over multifrequency models of the polarized dust emission. We validate this approach on a single simulated Planck patch with caracteristics matching the Chamaeleon-Musca region, from which we infer pixel-wise maps of the SED parameters with calibrated uncertainties. With typical SED variations expected from current dust emission models, we detect the overall decorrelation between frequencies, as well as spatial variations of the effective spectral index, but not of the polarization angle. These results pave the way for mapping the polarized dust SED properties in Planck data or in upcoming surveys.