Exploring the Model Dependence of MCMC-Based 21 cm Power Spectrum Parameter Constraints

TL;DR

This work tests how MCMC-based inferences of Epoch of Reionization (EoR) parameters from the 21 cm power spectrum depend on the semi-numerical modeling choice within 21cmFAST. By generating paired light-cones with two bubble-finding algorithms and performing in-domain and out-of-domain recoveries of $\zeta$ and $T_{vir}^{min}$ with 21CMMC, the authors demonstrate strong model dependence, particularly for $\zeta$, caused by degeneracy with the bubble-finding scheme. The study further shows that adding instrumental noise via 21cmSense does not remove this bias, indicating that current analyses may be significantly affected by algorithmic choices. The results motivate development of model-independent analysis strategies and cross-validation across multiple reionization codes to robustly interpret 21 cm EoR data.

Abstract

Detection and analysis of the cosmic 21 cm signal of neutral hydrogen has long been considered the most promising route towards exploration of the Epoch of Reionization (EoR). 21CMMC, a Markov Chain Monte Carlo sampler of the semi-numerical simulation code 21cmFAST, has already been used in conjunction with published upper limits on the 21 cm signal from the Murchison Widefield Array (MWA), the LOw Frequency ARray (LOFAR), and the Hydrogen Epoch of Reionization Array (HERA) to constrain the astrophysics of the EoR. Here we investigate the extent to which analysis of the EoR performed using 21CMMC is dependent on the underlying semi-numerical model. We used 21cmFAST to simulate two datasets of 21 cm light-cones which differ only in the algorithm used to identify ionized regions (the so-called "bubble-finding" algorithm). We then tested 21CMMC's ability to return key astrophysical parameters when using the different bubble-finding algorithms. We find that the performance of 21CMMC depends sensitively on the agreement between the astrophysical model of our mock data and the model used for sampling. This result has important implications for the analysis of the 21 cm signal performed using 21CMMC and further motivates investigation into model-independent analysis techniques for 21 cm EoR data.

Figures (8)

• Figure 1: Brightness temperature maps showing a slice along one spatial axis of a light-cone pair. The two light-cones differ only by whether they were simulated using bubble-finding algorithm 1 (top) or 2 (bottom). The underlying density field along with all astrophysical parameter values are equal between these two light-cones.
• Figure 2: Comparison of the 21 cm PS of a bubble-finding algorithm 1 and 2 light-cone pair. This light-cone pair corresponds to that used in Figure \ref{['fig:lightcones']}. Both light-cones were simulated with the same density field.
• Figure 3: Returned $\zeta$ values for conditions 1$\rightarrow$1 (top left), 2$\rightarrow$2 (bottom right), 1$\rightarrow$2 (top right), and 2$\rightarrow$1 (bottom left). The black line in each subplot represents a perfect $\zeta$ recovery (i.e. y=x), and the solid colored line in each subplot represents the line of best fit for our recovered $\zeta$ values.
• Figure 4: Returned $T_{vir}^{min}$ values for conditions 1$\rightarrow$1 (top left), 2$\rightarrow$2 (bottom right), 1$\rightarrow$2 (top right), and 2$\rightarrow$1 (bottom left). The black line in each subplot represents a perfect $T_{vir}^{min}$ recovery (i.e. y=x), and the solid colored line in each subplot represents the line of best fit for our recovered $T_{vir}^{min}$ values. One anomalous return in condition 1$\rightarrow$2 with no asymmetric error was omitted from the calculation of our line of best fit.
• Figure 5: Comparison of the 21 cm PS for a light-cone simulated using bubble-finding algorithm 1 (red), bubble-finding algorithm 2 with the median parameters of a run of 21CMMC with an expanded $\zeta$ range (Run B, yellow), bubble-finding algorithm 2 varying only $T_{vir}^{min}$ (Run T, blue), and bubble-finding algorithm 2 varying only $\zeta$ with an expanded range (Run Z, green).