Constraining the giant radio galaxy population with machine learning and Bayesian inference

Large-scale sky surveys at low frequencies, such as the LOFAR Two-metre Sky Survey (LoTSS), allow for the detection and characterisation of unprecedented numbers of giant radio galaxies (GRGs, or `giants', of at least $l_ p,GRG Mpc $ long). This, in turn, enables us to study giants in a cosmological context. A tantalising prospect of such studies is a measurement of the contribution of giants to cosmic magnetogenesis. However, this measurement requires en masse radio--optical association for well-resolved radio galaxies and a statistical framework to infer GRG population properties. By automating the creation of radio--optical catalogues, we aim to significantly expand the census of known giants. With the resulting sample and a forward model that takes into account selection effects, we aim to constrain their intrinsic length distribution, number density, and lobe volume-filling fraction (VFF) in the Cosmic Web. We combined five existing codes into a single machine learning (ML)--driven pipeline that automates radio source component association and optical host identification for well-resolved radio sources. We created a radio--optical catalogue for the entire LoTSS Data Release 2 (DR2) footprint and subsequently selected all sources that qualify as possible giants. We combined the list of ML pipeline GRG candidates with an existing list of LoTSS DR2 crowd-sourced GRG candidates and visually confirmed or rejected all members of the merged sample. To infer intrinsic GRG properties from GRG observations, we developed further a population-based forward model and constrained its parameters using Bayesian inference. Roughly half of all GRG candidates that our ML pipeline identifies indeed turn out to be giants upon visual inspection, whereas the success rate is 1 in 11 for the previous best giant-finding ML technique in the literature. We confirm $5,647$ previously unknown giants from the crowd-sourced LoTSS DR2 catalogue and $2,597$ previously unknown giants from the ML pipeline. Our confirmations and discoveries bring the total number of known giants to at least $11,585$. Our intrinsic GRG population forward model provides a good fit to the data. The posterior indicates that the projected lengths of giants are consistent with a curved power law probability density function whose initial tail index $ p,GRG changes by $ 0.3$ over the interval up to $l_ p Mpc $. We predict a comoving GRG number density $n_ GRG Mpc $, close to a recent estimate of the number density of luminous non-giant radio galaxies. With the projected length distribution, number density, and additional assumptions, we derive a present-day GRG lobe VFF $ V GRG-CW (z=0) = 1.4 $ in clusters and filaments of the Cosmic Web. We present a state-of-the-art ML-accelerated pipeline for finding giants, whose complex morphologies, arcminute extents, and radio-emitting surroundings pose challenges. Our data analysis suggests that giants are more common than previously thought. More work is needed to make GRG lobe VFF estimates reliable, but tentative results imply that it is possible that magnetic fields once contained in giants pervade a significant ($ gtrsim 10$) fraction of today's Cosmic Web.