Two-population Bayesian hierarchical model of Type Ia supernovae

The currently used standardization of Type Ia supernovae results in Hubble residuals whose physical origin is unaccounted for. Here, we present a complete physical interpretation of the Hubble residuals based on a novel Bayesian hierarchical model of Type Ia supernovae in which latent variables describing intrinsic and extrinsic (dust related) supernova properties originate from two supernova populations. Fitting the model to SALT2 light-curve parameters of supernovae in the Hubble flow we find strong (4σ) evidence for the presence of two overlapping, but distinct, populations differentiated primarily by their mean SALT2 shape parameter (stretch) x₁. Supernovae from the population with predominantly slow decliners (higher average x₁) are found to be intrinsically bluer (mean SALT2 colour c ≈ -0.11) and twice as reddened by dust (mean reddening E(B - V) ≈ 0.10) than those from the other population which is dominated by fast decliners (lower average x₁) with c ≈ -0.04 and E(B - V) ≈ 0.05. The inferred extinction coefficient R_B in both supernova populations follows a broad distribution (scatter 0.9) with a mean of 4.1, which coincides closely with the value associated with mean extinction law in the Milky Way. We also find that the supernova data favour a peaked (two-tailed) distribution of selective extinction E(B - V) over the commonly adopted exponential model. Our approach provides a complete explanation of the distribution of supernova light-curve parameters in terms of extinction properties and the above-mentioned differences between the two populations, without the need for introducing any intrinsic scatter.