Probing the Diversity of Type Ia Supernova Light Curves in the Open Supernova Catalog

TL;DR

This study uses neural networks to analyze Type Ia supernova light curves, revealing underlying diversity and natural sub-classes without relying on pre-existing labels, and correlates these with physical and environmental factors.

Contribution

It introduces a data-driven, autoencoder-based approach to identify supernova sub-classes and disentangle redshift effects from light curve diversity.

Findings

Three latent variables explain 95% of the variance in supernova light curves.

The method naturally separates 91T and 91bg supernova subclasses.

One latent variable correlates strongly with redshift, enabling de-redshifting.

Abstract

The ever-growing sample of observed supernovae enhances our capacity for comprehensive supernova population studies, providing a richer dataset for understanding the diverse characteristics of Type Ia supernovae and possibly that of their progenitors. Here, we present a data-driven analysis of observed Type Ia supernova photometric light curves collected in the Open Supernova Catalog. Where available, we add the environmental information from the host galaxy. We focus on identifying sub-classes of Type Ia supernovae without imposing the pre-defined sub-classes found in the literature to date. To do so, we employ an implicit-rank minimizing autoencoder neural network for developing low-dimensional data representations, providing a compact representation of the supernova light curve diversity. When we analyze light curves alone, we find that one of our resulting latent variables is strongly correlated with redshift, allowing us to approximately ``de-redshift'' the other latent variables describing each event. After doing so, we find that three of our latent variables account for $\sim$95\% of the variance in our sample, and provide a natural separation between 91T and 91bg thermonuclear supernovae. Of note, the 02cx subclass is not unambiguously delineated from the 91bg sample in our results, nor do either the over-luminous 91T or the under-luminous 91bg/02cx samples form a clearly distinct population from the broader sample of ``other'' SN Ia events. We identify the physical characteristics of supernova light curves which best distinguish SNe 91T from SNe 91bg \& 02cx, and discuss prospects for future refinements and applications to other classes of supernovae as well as other transients.