The exotic fraction among unassociated Fermi sources

TL;DR

This study uses machine learning to identify potential counterparts for unassociated gamma-ray sources, suggesting that less than 20% may be exotic or unknown types, thus constraining the search for dark matter and novel emitters.

Contribution

It applies an unsupervised classification algorithm to associate unconfirmed Fermi sources with potential counterparts, refining the understanding of their nature.

Findings

Potential associations for 119 out of 128 sources in the studied region.

Less than 20% of high-latitude unassociated sources may be exotic or unknown.

Identification of faint or obscured counterparts remains challenging.

Abstract

Revealing the nature of unassociated high-energy (> 100 MeV) gamma-ray sources remains a challenge 35 years after their discovery. Of the 934 gamma-ray sources at high Galactic latitude (|b| > 15 degrees) in the First Fermi-LAT catalogue (1FGL), 316 have no obvious associations at other wavelengths. In this paper, we apply the K-means unsupervised classification algorithm to isolate potential counterparts for 18 unassociated Fermi sources contained within a 3000 square degree `overlap region' of the sky intensively covered in radio and optical wavelengths. Combining our results with previous works, we reach potential associations for 119 out of the 128 Fermi sources within said region. If these associations are correct, we estimate that less than 20% of all remaining unassociated 1FGL sources at high Galactic latitude (|b| > 15 degrees) might host `exotic' counterparts distinct from known classes of gamma-ray emitters. Potentially even these outliers could be explained by high-redshift/dust-obscured analogues of the associated sample or by intrinsically faint radio objects. Although such estimate leaves some room for novel discoveries, it severely restricts the possibility of detecting dark matter subhaloes and other unconventional types of gamma-ray emitters in the 1FGL. In closing, we argue that the identification of Fermi sources at the low end of the flux density distribution will be a complex process that might only be achieved through a clever combination of refined classification algorithms, multi-wavelength efforts, and dedicated optical spectroscopy.