How Low Can We Go: Minimum Spectroscopic Requirements for SN Subtype Classification

After first light for the Rubin Observatory, the Legacy Survey of Space and Time will discover millions of transient events and hundreds of supernovae (SNe) each night. As a result, spectrographs around the world will have to make difficult decisions about which transients will get resource intensive spectroscopic follow-ups. Our work aims to discover the minimum spectral resolving power, R = Δlambda/lambda at which spectral classification is still possible for specific SNe and SN subtypes, including subtypes of Core Collapse SNe: IIP, IIL; SN Ibc: IIb, Ib, Ic, Ic-broad; and interacting SNe: Ibn and IIn. We developed a rigorous method to degrade spectra to simulate future low resolution observations from existing high resolution data. We have applied existing classifier models to the synthetic low-resolution spectra, including Williamson (2019) — a Machine Learning classifier based on Principal Component Analysis (PCA) and a Support Vector Machine (SVM) — and DASH (Muthukrishna et al. 2019), a convolutional neural network-based classifier. Finally, we are now developing a new, transformer-based neural network classifier (Vaswani et al. 2017) specifically tuned to early lightcurve and low-resolution spectral joint classification. These results will inform future spectrograph design and help alleviate the stress that Rubin will cause on existing facilities.