The Success of Optical Variability in Uncovering AGNs in Low-stellar Mass Galaxies

TL;DR

This study demonstrates that optical variability analysis using random forest algorithms effectively uncovers active galactic nuclei (AGNs) in low-stellar-mass galaxies, confirmed by spectral and X-ray data.

Contribution

It introduces a machine learning approach to identify AGNs in low-mass galaxies using optical light curves, validated with spectral and X-ray observations.

Findings

87% of candidates confirmed as AGNs via spectra

High detection rate of broad Balmer lines in spectra

Two-thirds of candidates are Seyfert or Composite types

Abstract

We used random forest algorithms to classify all objects in a large portion of the sky, using optical light curves obtained, or built from images provided, by the Zwicky Transient Facility (ZTF). We compare different selection sets based on alerts or complete light curves derived from different photometric selection algorithms. The AGN candidates thus selected are cross-matched with objects in the NASA-Sloan Atlas (NSA) of local galaxies, with $M_*<2\times10^{10}M_\odot$. The AGN nature of these candidates is verified and characterized using archival optical spectra from SDSS. We further establish the fraction of candidates with counterparts in the eROSITA data release 1 catalog of X-ray sources. From an initial sample of 506 candidates, 415 have good-quality spectra. Among these 415 objects, we found significant broad Balmer lines in the spectra for $86\%$ (357) of the candidates. When considering BPT classifications, an additional 5 candidates were confirmed, resulting in $87\%$ (362) confirmed candidates. Specifically, broad Balmer lines were detected in $94\%$-$98\%$ of the AGN candidates selected from complete light curves and in $80\%$ of those selected from the less frequent ZTF alerts. The black hole masses estimated from the spectra range from $2.2\times10^6M_\odot$ to $4.2\times10^7M_\odot$, reaching lower values for the candidates selected using the more sensitive light curves. The black hole masses obtained cluster around $0.1\%$ of the stellar mass of the host from the NSA catalog. Two-thirds of the AGN candidates are classified as Seyfert or Composite by their narrow emission line ratios (BPT diagnostics) while the rest are star-forming. Almost all the candidates classified as Seyfert and over $50\%$ of those classified as star-forming have significant BELs. We found X-ray counterparts for $67\%$ of the candidates that fall in the footprint of the eROSITA-DE DR1.