Source-linked AI summary
Concerns about Modelling of the EDGES Data
Richard Hills, Girish Kulkarni, P. Daniel Meerburg, Ewald Puchwein
TL;DR
Whether Bowman et al.2's fitted 21-cm absorption profile uniquely and physically represents the cosmological signal is uncertain. Re-examining their fits and testing alternative models, the authors find that different signal profiles can fit the data and that some fits require unphysical parameters.
Problem
Whether Bowman et al.2's fitted 21-cm absorption profile uniquely and physically represents the cosmological signal remains unresolved.
Source: pp. 1–2Method
The authors re-examined Bowman et al.2's fits and tested nonlinear and alternative foreground and signal formulations against the released data.
Source: pp. 1–2Results
Different signal profiles fit the data, while fits require unphysical foreground or ionosphere parameters under some formulations.
Source: pp. 1–2Takeaways & Limitations
The analysis does not establish that the feature is absent, but the evidence falls short of the level required to invoke new physics to explain it.
Source: p. 2Nine parameters are fitted to a ~50 MHz spectrum with little structure shorter than ~10 MHz, so the number of independent data points may not be much larger than the parameter count.
Source: p. 2
Abstract
from arXiv · showhide
It is predicted that the spectrum of radio emission from the whole sky should show a dip arising from the action of the light from the first stars on the hydrogen atoms in the surrounding gas, which causes the 21-cm line to appear in absorption against the cosmic microwave background. Bowman et al. 2018 identified a broad flat-bottomed absorption profile centred at 78 MHz, which could be this feature, although the depth of the profile is much larger than expected. We have examined the modelling process they used and find that their data implies unphysical parameters for the foreground emission and also that their solution is not unique in the sense that we found other simple formulations for the signal that are different in shape but that also fit their data. We argue that this calls into question the interpretation of these data as an unambiguous detection of the cosmological 21-cm absorption signature.
Methods
The authors tested the presented and supplementary models with least-squares fitting and estimated parameter posterior distributions using Polychord. Their analysis used the EDGES Data Release data for Figure 1 of Bowman et al. (2018).
- The authors used least-squares fitting to test the models presented in the text and supplementary material.Source: Methods · p. 3
- They derived posterior distributions of model parameters by implementing a likelihood function in the multi-nested sampler Polychord.Source: Methods · p. 3
- Their analysis used the “Data for Figure 1 of Bowman et al. (2018)” from the EDGES Data Release.Source: Data Availability · p. 3
Author Information
The paper lists Girish Kulkarni, P. Daniel Meerburg, and Ewald Puchwein as authors, with affiliations at the University of Cambridge and the University of Groningen. E.P. initiated the study by highlighting the role of foreground parameters, and all authors contributed to the analysis and writing.
Author Information
- The listed affiliations include University of Cambridge institutes and University of Groningen institutes in the UK and the Netherlands.Source: Author Information · p. 4
- The listed authors are Girish Kulkarni, P. Daniel Meerburg, and Ewald Puchwein.Source: Author Information · p. 4
- E.P. initiated the study by highlighting the importance of foreground parameters, and all authors participated in the detailed analysis and writing.Source: Author Information · p. 4
Supplementary Information
The supplementary material documents EDGES fits using multiple foreground and signal formulations, including alternatives to the Bowman et al.2 profile.
Supplementary Information
- Detailed derivations, fits, and parameter values are available in the supplementary material.Source: Supplementary Information · p. 4
- Figure 1 displays fits using Bowman et al.2's linearized foreground, a nonlinear foreground with unrestricted or physically plausible parameters, and the PowLin model.Panels also show residuals for foreground-only fits and fits that include Bowman et al.2's profile.Source: Figure 1 · p. 5
- Figure 2 shows a good fit using the Bowman et al.2 profile with four signal parameters and five foreground parameters, alongside alternative nine-parameter fits using two Gaussians or a cosine of about 0.06 K.The two Gaussian features have equal width and amplitude.Source: Figure 2 · p. 6