Comparison of cosmological parameter inference methods applied to supernovae lightcurves fitted with SALT2

TL;DR

This paper compares chi-square and Bayesian hierarchical methods for inferring cosmological parameters from supernovae lightcurves, highlighting biases and convergence with larger datasets.

Contribution

It provides a direct comparison of two inference methods applied to supernova data, revealing biases and the conditions under which their estimates converge.

Findings

Both methods exhibit small biases in parameter recovery.

Bayesian hierarchical method yields slightly more accurate results.

Discrepancies in matter density estimates can reach about 2 sigma for SNLS3 data.

Abstract

We present a comparison of two methods for cosmological parameter inference from supernovae Ia lightcurves fitted with the SALT2 technique. The standard chi-square methodology and the recently proposed Bayesian hierarchical method (BHM) are each applied to identical sets of simulations based on the 3-year data release from the Supernova Legacy Survey (SNLS3), and also data from the Sloan Digital Sky Survey (SDSS), the Low Redshift sample and the Hubble Space Telescope (HST), assuming a concordance LCDM cosmology. For both methods, we find that the recovered values of the cosmological parameters, and the global nuisance parameters controlling the stretch and colour corrections to the supernovae lightcurves, suffer from small biasses. The magnitude of the biasses is similar in both cases, with the BHM yielding slightly more accurate results, in particular for cosmological parameters when applied to just the SNLS3 single survey data sets. Most notably, in this case, the biasses in the recovered matter density $\Omega_{\rm m,0}$ are in opposite directions for the two methods. For any given realisation of the SNLS3-type data, this can result in a $\sim 2 \sigma$ discrepancy in the estimated value of $\Omega_{\rm m,0}$ between the two methods, which we find to be the case for real SNLS3 data. As more higher and lower redshift SNIa samples are included, however, the cosmological parameter estimates of the two methods converge.