Classification of a New X-ray Catalog of Likely Counterparts to 4FGL-DR4 Unassociated Gamma-ray Sources Using a Neural Network

TL;DR

This work addresses the challenge of classifying unassociated gamma-ray sources from the 4FGL-DR4 catalog by leveraging Swift-XRT/UVOT multiwavelength data to identify plausible low-energy counterparts. An MLP classifier is trained on X-ray flux, photon index, and $V$-band magnitude to output a blazar probability $P_ ext{bzr}$, enabling discrimination between likely blazars, pulsars, and ambiguous sources. For 213 singlet X-ray sources with X-ray and optical data, 173 are classified as likely blazars ($P_ ext{bzr}>0.99$) and 6 as likely pulsars ($P_ ext{bzr}<0.01$), with 34 ambiguous; including 70 multiplet X-ray sources increases those counts to 227 and 16, respectively. The majority of classifications agree with prior studies for sources previously analyzed, supporting the validity of the NNC approach for classifying unknown and newly detected gamma-ray sources and expanding their low-energy counterpart catalog.

Abstract

Our survey of the fourth $\mathit{Fermi}$ Large Area Telescope catalog (4FGL) unassociated gamma-ray source regions using the X-Ray Telescope (XRT) and Ultraviolet/Optical Telescope (UVOT) aboard the Neil Gehrels $\mathit{Swift}$ Observatory ($\mathit{Swift}$) provides new XRT and UVOT source detections and localizations to help identify potential low-energy counterparts to unassociated $\mathit{Fermi}$ gamma-ray sources. We present a catalog of 218 singlet and 70 multiplet $\mathit{Swift}$ X-ray sources detected within the positional uncertainty ellipses of 244 unassociated $\mathit{Fermi}$ gamma-ray sources from the 4FGL-DR4 catalog, 144 of which are not previously cataloged by Kerby et al. (2021b). For each X-ray source, we derive its X-ray flux and photon index, then use simultaneous UVOT observations with optical survey data to estimate its $V$-band magnitude. We use these parameters as inputs for a multi-layer perceptron (MLP) neural network classifier (NNC) trained to classify sources as blazars, pulsars, or ambiguous gamma-ray sources. For the 213 singlet sources with X-ray and optical data, we classify 173 as likely blazars ($P_\mathrm{bzr} > 0.99$) and 6 as likely pulsars ($P_\mathrm{bzr} < 0.01$), with 34 sources yielding ambiguous results. Including 70 multiplet X-ray sources, we increase the number of $P_\mathrm{bzr} > 0.99$ to 227 and $P_\mathrm{bzr} < 0.01$ to 16. For the subset of these classifications that have been previously studied, a large majority agree with prior classifications, supporting the validity of using this NNC to classify the unknown and newly detected gamma-ray sources.

Figures (1)

• Figure 1: Normalized histogram of the galactic latitudes for unassociated gamma-ray sources. The samples of all unassociated gamma-ray sources (UGSs) are shown in green, and the 4FGL targets with overlapping XRT sources are in black with cross-hatches.