Host: Gabriele Coppi
Title: Cosmoglobe - mapping the Universe from the Milky Way to the Big Bang
Abstract: The Cosmoglobe collaboration implemented the first integrated Bayesian end-to-end analysis pipeline for CMB experiments, with the long-term motivation of developing a common analysis platform that supports efficient global joint analysis of complementary radio, microwave, and sub-millimeter experiments. The main product of such a framework are samples of the full joint Bayesian posterior distribution of the unknown quantities P(ω|d), where ω represents the set of all free instrumental, astrophysical, and cosmological parameters, and d the observational data. In practice, given the dimensionality of the parameter space, on the order of billions, obtaining an effective exploration of the posterior distribution is non trivial. For this reason, the primary computational framework developed within Cosmoglobe, i.e. the Commander software, relies on an MCMC algorithm known as Gibbs sampling. Its main feature is the ability to draw samples from a high-dimensional joint posterior distribution through a sequence of simpler, univariate conditional distributions. Notable advantages of this approach are seamless end-to-end propagation of uncertainties; accurate modeling of both astrophysical and instrumental effects with the possibility of exploiting different instrumental characteristics for breaking critical degeneracies between the uncertain parameters; optimized computational costs with little or no need for intermediate human interaction between various analysis steps and datasets; and a complete overview of the entire analysis process within one single framework. The Cosmoglobe framework has been so far successfully applied to a joint reanalysis of Planck LFI and WMAP, and to an end-to-end reanalysis of COBE-DIRBE, jointly with COBE-FIRAS, GAIA, Planck HFI, and WISE observations, published as Cosmoglobe DR1 and DR2 respectively. The resulting sky maps showed several notable advantages with respect to the previously available ones, published by the single collaborations, including lower residuals, better consistency between different frequency bands, and a more complete and accurate noise characterization. The current and future efforts within Cosmoglobe are focused on expanding its sky modeling capabilities, by including in its analysis framework datasets such as Chipass, SO, AKARI, IRAS, eROSITA and FERMI, with the ambitious goal of obtaining one single consistent sky model extending from the radio to the X-rays regimes. Given the increasing data volume and angular resolution of current and future observational experiments, the computational challenge of performing an end-to-end bayesian analysis on such experiments jointly is huge. For this reason, a new iteration of the software Commander, namely Commander4, is being currently developed. It features a Python codebase with some C++ low-level optimized functions, and a completely rewritten MPI architecture which allows it to run on as many computing nodes as available, making analysis up to Healpix resolution of 32k time-feasible.