A Comprehensive Bayesian re-analysis of the SARAS2 data from the Epoch of Reionization

Abstract

We present a Bayesian re-analysis of the sky-averaged 21-cm experimental data from SARAS2 using nested sampling implemented with , spectrally smooth foreground modelling implemented with , detailed systematic modelling and rapid signal emulation with . Our analysis differs from previous analysis of the SARAS2 data through the use of a full Bayesian framework and separate modelling of the foreground and non-smooth systematics present in the data. We use the most up-to-date global signal models including Lyman-𝛼 and CMB heating and parameterised by astrophysical parameters such as star formation efficiency, X-ray heating efficiency, minimal virial circular velocity, CMB optical depth and the low energy cutoff of the X-ray spectral energy distribution. We also consider models with an excess radio background above the CMB produced via radio emission from early galaxies and parameterised by a radio production efficiency. A non-smooth systematic is identified and modelled as both a frequency damped sinusoid introduced by the electronics and separately from the sky. The latter is modulated by the total efficiency of the antenna and marginally favoured by the data. We consider three models for the noise in the data with different frequency dependencies. We find that the SARAS2 constraints on individual astrophysical parameters are extremely weak however we identify classes of disfavoured signals. Specifically, we weakly disfavour standard astrophysical models with high Lyman-𝛼 fluxes and generally weak heating and more confidently disfavour exotic models with high Lyman-𝛼 fluxes, low X-ray efficiencies, high radio production efficiencies in early galaxies and high CMB optical depths. We intend to follow this work up with a similar analysis of the recently published SARAS3 data.

Early-Universe Model Building

“…Each of its three generations, SARAS 1-3, was located in a different part of India. Measurements by SARAS 2 [1022] in the frequency range 110-200 MHz were used to mildly constrain different astrophysical and cosmological models [1023][1024][1025]. The third generation, SARAS 3, observed in the frequency range 43.75-87.5 MHz during the first quarter of 2020 [1026,1027].…”

Accurate modelling of the Lyman-$α$ coupling for the 21-cm signal, observability with NenuFAR and SKA

“…From the upper limits or a detection of the signal, the parameters of astrophysical models can be inferred using either a classical MCMC approach (e.g. Greig & Mesinger 2015 or with methods involving some aspects of machine learning (Shimabukuro & Semelin 2017;Gillet et al 2019;Schmit & Pritchard 2018;Jennings et al 2019;Doussot et al 2019;Cohen et al 2020;Hortúa et al 2020;Bevins et al 2022;Zhao et al 2022;Bye et al 2022;Abdurashidova et al 2022). In all cases, the modelling of the signal is a fundamental step of the inference process: either at each step of the MCMC approach or for building a learning sample in supervised learning based methods.…”

Marginal Bayesian Statistics Using Masked Autoregressive Flows and Kernel Density Estimators with Examples in Cosmology

“…Bayesian inference is a cornerstone of modern cosmology and astrophysics. It is frequently employed to derive parameter constraints on key signal parameters from data sets such as the Dark Energy Survey (DES, [1,2]), Planck [3], REACH [4], and SARAS2 [5], among others. Often, experiments are sensitive to different aspects of the same physics, and by combining constraints across probes we can improve our understanding of the Universe or reveal tensions between different experiments.…”