The Luminosity Function of UV-Selected Unobscured Quasars at z≈5: A Deep Learning Selection with Gemini/GMOS Confirmation

The number density of quasars during the epoch of reionization shows significant discrepancies across different wavelength regimes, leaving key uncertainties in our understanding of early supermassive black hole growth and the role of AGN feedback. Recent JWST observations have intensified this tension by revealing up to two orders of magnitude more faint AGN than predicted from extrapolations of UV-based quasar luminosity functions, commonly referred to as Luminous Red Dots (LRDs). In this talk, I present a quasar sample at z≈5 constructed from the Subaru/HSC Deep and UltraDeep surveys, extending down to M₁₄₅₀≈−22 AB, where the bright end of X-ray–based constraints begins to overlap. Instead of traditional color-selection techniques, we employ a deep neural network–based classification approach while minimizing the use of photometric flags to maximize the effective survey area. This strategy enables a search over a total area of 30.65 deg², approximately twice as large as in previous studies at similar depths. As a result, we identify 76 quasar candidates in the redshift range z=4.5-5.5. Follow-up Gemini/GMOS spectroscopy totaling 31 hours confirms the validity of our selection method and reveals 10 high-redshift quasars spanning z = 4.42–5.81 with a wide range of Lyα luminosities. These results provide a robust UV-based determination of the luminosity function for faint, unobscured AGN during the epoch of reionization and establish a critical observational bridge to test the connection between UV-selected quasars and the LRD population discovered by JWST through future near-infrared observations.