BayeSN $\times$ Dovekie: Joint Photometric Cross-calibration and SED Modelling of Type Ia Supernovae

Abstract

We present a new framework for BayeSN, the hierarchical Bayesian SED model for type Ia supernovae (SNe Ia), incorporating cross-calibration of samples observed across heterogeneous telescopes. This framework is the first to parametrise the filter wavelength and zero-point offsets commonly used in SN~Ia cosmology within SN SED model training, enabling additional constraint on cross-calibration from SNe beyond the standard stellar-based cross-calibration pipeline. We apply this framework to train a new G26 BayeSN model on the same SED model training sample used in recent cosmological analyses, an order-of-magnitude increase over previous BayeSN training samples, and include a novel training methodology to leverage high-redshift SNe Ia in BayeSN training. We present the G26 model and apply it to the DES-SN5YR sample to assess performance, finding a 12 per cent reduction in $σ_{\rm NMAD}$ scatter when compared with SALT3$.$Dovekie; 0.164 mag compared with 0.185 mag for a sample of likely SNe Ia at $z < 0.7$, without bias corrections. We additionally present constraints on cross-calibration wavelength and zero-point shifts from our framework when using the latest `Dovekie&#39; calibration constraints as a prior. This work is a key step towards a full end-to-end cosmological analysis with BayeSN; the new G26 model is incorporated within the public BayeSN code.