Highly Variable Quasar Candidates Selected from 4XMM-DR13 with Machine Learning

Abstract

We present a sample of 12 quasar candidates with highly variable soft X-ray emission, selected from the fourth XMM-Newton Serendipitous Source Catalog (4XMM-DR13), using random forest (RF). Optical to mid-IR photometric data for the 4XMM-DR13 sources were obtained by correlating the sample with the Sloan Digital Sky Survey (SDSS) DR18 photometric catalog and the AllWISE database. By further cross matching with known spectral catalogs from the SDSS and LAMOST surveys, we compiled a training data set containing stars, galaxies, and quasars. The RF algorithm was trained to classify the XMM–Wide-field Infrared Survey Explorer–SDSS sample. We then refined the quasar candidate selection by applying Gaia proper motion data to eliminate stellar contaminants. As a result, 52,486 quasar candidates were classified, with 8410 of them matching known quasars in SIMBAD. The quasar candidates exhibit systematically lower X-ray fluxes compared to quasars in the training set, suggesting that the classifier is effective in identifying fainter quasars. From this quasar candidate sample, we constructed a subset of 12 sources that have shown variations in their soft X-ray flux by a factor of 10 over ∼20 yr in the XMM-Newton survey. These highly variable quasar candidates extend the quasar sample characterized by extreme soft X-ray variability to the optically faint end, with magnitudes around r ∼ 22. Notably, none of these 12 sources were detected in ROSAT observations. Given the flux sensitivity of ROSAT, the result indicates that quasars exhibiting more than 2 orders of magnitude of variation are extremely rare.

Keywords

• Quasars
• X-ray quasars