Predicting the spectrum of UGC 2885, Rubin's Galaxy with machine learning

TL;DR

This study tests machine learning predictions of galaxy spectra on UGC 2885, comparing them with actual observations, and explores the potential of ML in identifying active galactic nuclei in massive, isolated disk galaxies.

Contribution

It demonstrates the viability of using ML-predicted spectra to identify AGN activity in a unique, massive, isolated galaxy, with validation against multiple spectroscopic observations.

Findings

ML predictions qualitatively match observed spectra

ML suggests slightly higher AGN-like line ratios

Spectroscopic data confirms AGN activity in the galaxy

Abstract

Wu & Peek (2020) predict SDSS-quality spectra based on Pan-STARRS broad-band \textit{grizy} images using machine learning (ML). In this letter, we test their prediction for a unique object, UGC 2885 ("Rubin's galaxy"), the largest and most massive, isolated disk galaxy in the local Universe ($D<100$ Mpc). After obtaining the ML predicted spectrum, we compare it to all existing spectroscopic information that is comparable to an SDSS spectrum of the central region: two archival spectra, one extracted from the VIRUS-P observations of this galaxy, and a new, targeted MMT/Binospec observation. Agreement is qualitatively good, though the ML prediction prefers line ratios slightly more towards those of an active galactic nucleus (AGN), compared to archival and VIRUS-P observed values. The MMT/Binospec nuclear spectrum unequivocally shows strong emission lines except H$\beta$, the ratios of which are consistent with AGN activity. The ML approach to galaxy spectra may be a viable way to identify AGN supplementing NIR colors. How such a massive disk galaxy ($M^* = 10^{11}$ M$_\odot$), which uncharacteristically shows no sign of interaction or mergers, manages to fuel its central AGN remains to be investigated.