Revealing new high redshift quasar populations through Gaussian mixture model selection

Abstract

We present a novel method to identify candidate high redshift quasars (HzQs; ($z\gtrsim5.5$), which are unique probes of supermassive black hole growth in the early Universe, from large area optical/infrared photometric surveys. Using Gaussian Mixture Models to construct likelihoods and incorporate informed priors based on population statistics, our method uses a Bayesian framework to assign posterior probabilities that differentiate between HzQs and contaminating sources. We additionally include deep radio data to obtain informed priors. Using existing HzQ data in the literature, we set a posterior threshold that accepts ${\sim}90\%$ of known HzQs while rejecting $>99\%$ of contaminants such as dwarf stars or lower redshift galaxies. Running the probability selection on test samples of simulated HzQs and contaminants, we find that the efficacy of the probability method is higher than traditional colour cuts, decreasing the fraction of accepted contaminants by 86% while retaining a similar fraction of HzQs. As a test, we apply our method to the Pan-STARRS Data Release 1 (PS1) source catalogue within the HETDEX Spring field area on the sky, covering 400 sq. deg. and coinciding with deep radio data from the LOFAR Two-metre Sky Survey Data Release 1 (LoTSS DR1). From an initial sample of ${\sim}5\times10^5$ sources in PS1, our selection shortlists 251 candidate HzQs, which are further reduced to 63 after visual inspection. Shallow spectroscopic follow-up of 13 high probability HzQs resulted in the confirmation of a previously undiscovered quasar at $z=5.66$ with photometric colours $i-z = 1.4$, lying outside the typically probed regions when selecting HzQs based on colours. This discovery demonstrates the efficacy of our probabilistic HzQ selection method in selecting more complete HzQ samples, which holds promise when employed on large existing and upcoming photometric data sets.