Deep learning from 21-cm tomography of the cosmic dawn and reionization. Issue 1 (14th January 2019)

• Record Type: Journal Article
• Authors: Gillet, Nicolas; Mesinger, Andrei; Greig, Bradley; Liu, Adrian; Ucci, Graziano
• Is Part Of: Monthly notices of the Royal Astronomical Society. Volume 484:Issue 1(2019)
• Page Start: 282
• Page End: 293
• Publication Date: 2019-01-14
• Subjects: galaxies: high-redshift -- intergalactic medium -- cosmology: theory -- dark ages, reionization, first stars -- diffuse radiation -- early Universe; Astronomy -- Periodicals; Periodicals; 520.5
• DOI: 10.1093/mnras/stz010
• Languages: English
• ISSNs: 0035-8711

Abstract

ABSTRACT: The 21-cm power spectrum (PS) has been shown to be a powerful discriminant of reionization and cosmic dawn astrophysical parameters. However, the 21-cm tomographic signal is highly non-Gaussian. Therefore there is additional information which is wasted if only the PS is used for parameter recovery. Here we showcase astrophysical parameter recovery directly from 21-cm images, using deep learning with convolutional neural networks (CNN). Using a data base of 2D images taken from 10 000 21-cm light-cones (each generated from different cosmological initial conditions), we show that a CNN is able to recover parameters describing the first galaxies: (i) T vir , their minimum host halo virial temperatures (or masses) capable of hosting efficient star formation; (ii) $\rm { \zeta \ }$, their typical ionizing efficiencies; (iii) L X /SFR , their typical soft-band X-ray luminosity to star formation rate; and (iv) E 0 , the minimum X-ray energy capable of escaping the galaxy into the IGM. For most of their allowed ranges, log T vir and log L X /SFR are recovered with $\lt 1 {{\ \rm per\ cent}}$ uncertainty, while $\rm { \zeta \ }$ and E 0 are recovered with ${\sim } 10{{\ \rm per\ cent}}$ uncertainty. Our results are roughly comparable to the accuracy obtained from Monte Carlo Markov Chain sampling of the PS with 21CMMC for the two mock observations analysed previously, although we caution that we do not yet include noise and foreground contaminants in this proof-of-concept study.