Identifying FeLoBAL Quasars in SDSS DR7Q with the Convolutional Neural Network

The Fe Low-ionization Broad Absorption Line Quasar (FeLoBALQ) is one of the rarest types of all quasars. Quasars blow out the surrounding violently, forming extreme outflows from which low ionized elements e.g. Fe provide the absorbing feature in FeLoBALQ spectra. Carrying high kinetic energy, the outflows of FeLoBALQ may possibly be enough for powering the M - \sigma_* relationship between the supermassive black hole mass M and the host-galaxy bulge velocity dispersion \sigma_* . On the other hand, evidence has been found for the co-existence of FeLoBALQ with hosts' starburst or recent major merger. However, the FeLoBALQ sample collected so far is not large enough to stand for these theories statistically. This research focuses on digging out hidden FeLoBALQs from large quasar surveys, forming a FeLoBALQ catalog large enough for statistical and physical analyze. Adopting Convolutional Neural Network (CNN) method, 160 FeLoBALQs are newly identified from totally 50931 quasars in the SDSS (Sloan Digital Sky Survey) DR7Q (Data Release 7 Quasar catalog) in the redshift range of 0.8 < z < 2.125 , with previous identified FeLoBALQ spectra as training sample. The FeLoBALQs' color are found redder than normal quasars, and previously identified FeLoBALQs are lightly redder than newly identified ones; these differences are more obvious on bluer end than on redder end, and nearly disappear in mid-infrared band. The proportion of FeLoBALQs out of all quasars given is 0.43\% , higher than previous prediction, but may still be underestimated. Further researches may expand this method to larger samples e.g. SDSS DR16Q (Data Release 16 Quasar catalog) for larger FeLoBALQ sample, which may help to answer the questions of the relationship between FeLoBALQ and host galaxy star formation, FeLoBALQ and galaxy major merger, and the co-evolution of galaxies and central supermassive black holes.