Massively parallel joint end-to-end Bayesian analysis of past, present, and future CMB experiments

Descripción del proyecto

Facilitar el análisis simultáneo de datos de CMB pasados, presentes y futuros

El fondo cósmico de microondas (CMB, por sus siglas en inglés) es un «eco» de la primera luz emitida en el universo durante la Gran Explosión. La detección de este CMB requiere una instrumentación de una sensibilidad sin precedentes. Recientemente se ha utilizado un muestreador bayesiano de Gibbs para CMB de código abierto llamado Commander para derivar con éxito nuevos mapas de frecuencias de última generación basados en los datos de dos experimentos de CMB avanzados. Partiendo de este muestreador, el equipo del proyecto Commander, financiado por el CEI, pretende multiplicar por mil la potencia de procesamiento e integrar correcciones pioneras que aprovechen de forma óptima las sinergias entre los experimentos. Con esto se allanará el camino para integrar los datos de experimentos pasados, presentes y futuros y permitirá, potencialmente, la primera detección mundial de ondas gravitacionales primordiales.

Objetivo

Detecting inflationary gravitational waves from the Big Bang and mapping cosmic structure formation rank among the most important goals in modern cosmology, and detailed cosmic microwave background (CMB) measurements is a uniquely powerful probe of these effects. However, the predicted signatures are tiny, and their detections require unprecedented instrumental sensitivity and systematics control. In this project I propose to develop one single massively parallel end-to-end framework for the joint analysis of past, present and future CMB experiments, and use this to combine current data from WMAP, Planck LFI+HFI and others with forthcoming measurements from Simons Observatory (SO), all processed at the level of time ordered data (TOD). I will also prepare for the analysis of future CMB experiments, including LiteBIRD, CMB-S4, and a Voyage 2050 CMB spectral distortion probe. This framework will build on an Open Source Bayesian CMB Gibbs sampler called Commander that has already played a transformational role in the field for more than two decades, and that has recently been used successfully to derive new state-of-the-art frequency maps for both Planck LFI and WMAP. However, the existing code only scales well up to O(100) computing cores, and I propose in this project a new organization that will scale it up to O(100,000) cores, as required for next-generation experiments. I will also implement a wide range of ground-breaking TOD-level corrections for key systematic effects (non-linear ADCs, cosmic ray glitches, atmospheric fluctuations, detector cross-correlations etc.) that optimally exploit synergies between experiments. Once operational, I will use this global framework to establish a new state-of-the-art model of the microwave sky; shed new light on several hotly debated LCDM tensions; and, perhaps, make the world's first detection of primordial gravitational waves. This work represents a paradigm shift in the field of computational cosmology.