Stretch to stretch, dust to dust: low-value local $H_{0}$ measurement from two-population modelling of type Ia supernovae

We revisit the local Hubble constant measurement from type Ia supernovae calibrated with Cepheids (SH0ES) by remodelling the supernova data using two supernova populations emerging from the observed bimodal distribution of the SALT2 stretch parameter. Our analysis accounts for population differences in both intrinsic properties (related to possible initial conditions, including supernova progenitor channels) and host-galaxy extinction (expected from well-known environmental differences associated observationally with the two populations). Based on a two-population Bayesian hierarchical modelling of the SALT2 light-curve parameters from the Pantheon+ compilation, we simultaneously constrain intrinsic and extrinsic properties of both supernova populations, match probabilistically the calibration supernovae with the corresponding population in the Hubble flow, and derive the Hubble constant. The difference between the two supernova populations is primarily driven by their mean absolute magnitudes and total-to-selective extinction coefficients. This is related but not equivalent to the traditional mass-step correction (including its broadening for reddened supernovae). The mean extinction coefficient of the supernova population used to propagate distances from the calibration galaxies to the Hubble flow is found to be consistent with the Milky Way-like interstellar dust model with R_B~4 and substantially higher than the extinction model assumed in the SH0ES measurement. Allowing for possible differences between reddening in the calibration galaxies and the corresponding population in the Hubble flow, we obtain H_0=70.59+/-1.15 km/s/Mpc. For the most conservative choice assuming equal prior distributions, we find H_0=71.45+/-1.03 km/s/Mpc. Our reanalysis of type Ia supernovae results in a reduction of the discrepancy with the Planck H_0 by at least 30 and up to 50 per cent (3.5-2.2sigma).