Correcting Type Ia Supernova Distances for Selection Biases and Contamination in Photometrically Identified Samples

TL;DR

This paper introduces the BBC method, a new technique to create bias-corrected Hubble Diagrams from photometric SN Ia data, enabling precise cosmological parameter estimation despite contamination and selection biases.

Contribution

The paper presents the BBC method, which combines bias correction, machine learning, and Monte Carlo simulations to improve cosmological inferences from photometric SN Ia samples.

Findings

BBC accurately recovers nuisance parameters within 1%

Small bias in dark energy parameter w (~0.006)

Method validated on large simulated datasets

Abstract

We present a new technique to create a bin-averaged Hubble Diagram (HD) from photometrically identified SN~Ia data. The resulting HD is corrected for selection biases and contamination from core collapse (CC) SNe, and can be used to infer cosmological parameters. This method, called "BBC" (BEAMS with Bias Corrections), includes two fitting stages. The first BBC fitting stage uses a posterior distribution that includes multiple SN likelihoods, a Monte Carlo simulation to bias-correct the fitted SALT-II parameters, and CC probabilities determined from a machine learning technique. The BBC fit determines 1) a bin-averaged HD (average distance vs. redshift), and 2) the nuisance parameters alpha and beta, which multiply the stretch and color (respectively) to standardize the SN brightness. In the second stage, the bin-averaged HD is fit to a cosmological model where priors can be imposed. We perform high precision tests of the BBC method by simulating large (150,000 event) data samples corresponding to the Dark Energy Survey Supernova Program. Our tests include three models of intrinsic scatter, each with two different CC rates. In the BBC fit, the SALT-II nuisance parameters alpha and beta are recovered to within 1% of their true values. In the cosmology fit, we determine the dark energy equation of state parameter w using a fixed value of Omega_matter as a prior: averaging over all six tests based on 6 x 150,000 = 900,000 SNe, there is a small w-bias of 0.006 +- 0.002. Finally, the BBC fitting code is publicly available in the SNANA package.