Photometric Selection of type 1 Quasars in the XMM-LSS Field with Machine Learning and the Disk-Corona Connection

Abstract

We present photometric selection of type 1 quasars in the $\approx5.3~{\rm deg}^{2}$ XMM-Large Scale Structure (XMM-LSS) survey field with machine learning. We constructed our training and \hbox{blind-test} samples using spectroscopically identified SDSS quasars, galaxies, and stars. We utilized the XGBoost machine learning method to select a total of 1\,591 quasars. We assessed the classification performance based on the blind-test sample, and the outcome was favorable, demonstrating high reliability ($\approx99.9\%$) and good completeness ($\approx87.5\%$). We used XGBoost to estimate photometric redshifts of our selected quasars. The estimated photometric redshifts span a range from 0.41 to 3.75. The outlier fraction of these photometric redshift estimates is $\approx17\%$ and the normalized median absolute deviation ($\sigma_{\rm NMAD}$) is $\approx0.07$. To study the quasar disk-corona connection, we constructed a subsample of 1\,016 quasars with HSC $i<22.5$ after excluding radio-loud and potentially X-ray-absorbed quasars. The relation between the optical-to-X-ray power-law slope parameter ($\alpha_{\rm OX}$) and the 2500 Angstrom monochromatic luminosity ($L_{2500}$) for this subsample is $\alpha_{\rm OX}=(-0.156\pm0.007)~{\rm log}~{L_{\rm 2500}}+(3.175\pm0.211)$ with a dispersion of 0.159. We found this correlation in good agreement with the correlations in previous studies. We explored several factors which may bias the $\alpha_{\rm OX}$-$L_{\rm 2500}$ relation and found that their effects are not significant. We discussed possible evolution of the $\alpha_{\rm OX}$-$L_{\rm 2500}$ relation with respect to $L_{\rm 2500}$ or redshift.