Astro Seminar Series - VIA ZOOM

Arthur is currently a research fellow at the Royal Observatory of Edinburgh (and about to move to The Oskar Klein Centre in Stockholm in March). He graduated from the Federal University of Rio Grande do Sul with a master's from the University of Sao Paulo and a PhD from University College London in 2019. Over the years, Arthur has worked as a post-doctoral researcher at UCL and Imperial College, focused on developing new statistical analysis tools for the Euclid Space Telescope. His primary research focuses are bayesian hierarchical models, field-level inference, primordial non-gaussianities, galaxy clustering, cosmic shear, and neutrino cosmology.

Title: Almanac:Generic Field Level Inference for Full-Sky Cosmological Fields and Angular Power Spectra

Abstract: With the advent of Euclid, LSST, Simons Observatory and other upcoming Stage-IV Surveys, we will soon map cosmic structure at an unprecedented precision over a large portion of the sky, such that sky curvature becomes influential. In this talk, I will present Almanac: a generic Bayesian solution for inferring the full-sky underlying cosmological fields and their power spectra from noisy partial sky observations. The crux to inferring these science-ready data products is to develop a Monte Carlo sampler that can handle the high resolution of upcoming data maps. A further challenge is that cosmological structures often have power spectra spanning many orders of magnitude. Thus, Almanac handles strongly scale-dependent signal-to-noise cases for spin-0 and spin-2 cosmological fields. This talk will show different applications of Almanac, focusing on the challenging applications to upcoming Weak Lensing data – presented in Loureiro et al. 2022 (ArXiv:2210.13260). Using a Euclid-like survey as a test study, we jointly infer all-sky E-mode and B-mode tomographic auto- and cross-power spectra from the masked sky and potentially parity-violating EB-mode power spectra. In this test, we probe scales up to a maximum multipole of 2048 – a Hamiltonian Monte Carlo with a total of ~16.8 Million parameters. The main output and natural outcome is the set of samples of the posterior, which does not suffer from leakage of power from E to B unless reduced to point estimates.