Quantum mechanics

Anisotropy

Algorithm

Bayesian probability

Biology

Statistics

Computational physics

Taxonomy (biology)

Astrophysics

Absorption (acoustics)

Selection (genetic algorithm)

Calibration

Computer science

Botany

Statistical physics

Physics

Mathematics

Optics

Data set

Artificial intelligence

Systematics

Cosmic microwave background

Model selection

Machine learning

Monthly Notices of the Royal Astronomical Society

Cosmic Dawn, when the first stars and proto-galaxies began to form, is commonly expected to be accompanied by an absorption signature at radio frequencies. This feature arises as Lyman-$\alpha$ photons emitted by these first luminous objects couple the 21 cm excitation temperature of intergalactic hydrogen gas to its kinetic temperature, driving it into absorption relative to the CMB. The detailed properties of this absorption profile encode powerful information about the physics of Cosmic Dawn. Recently, Bowman et al. analysed data from the EDGES low-band radio antenna and found an unexpectedly deep absorption profile centred at 78 MHz, which could be a detection of this signature. Their specific analysis fit their measurements using a polynomial foreground model, a flattened Gaussian absorption profile and a white noise model; we argue that a more accurate model, that includes a detailed noise model and accounting for the effects of plausible calibration errors, is essential for describing the EDGES data set. We perform a Bayesian evidence-based comparison of models of the EDGES low-band data set and find that those incorporating these additional components are decisively preferred. The subset of the best fitting models of the data that include a global signal favour an amplitude consistent with standard cosmological assumptions (A < 209 mK). However, there is not strong evidence to favour models of the data including a global 21 cm signal over those without one. Ultimately, we find that the derivation of robust constraints on astrophysics from the data is limited by the presence of systematics.

Testing for calibration systematics in the EDGES low-band data using Bayesian model selection