Probabilistic selection of high-redshift quasars

Abstract

High redshift quasars (HZQs) with redshifts of z >~ 6 are so rare that any photometrically-selected sample of sources with HZQ-like colours is likely to be dominated by Galactic stars and brown dwarfs scattered from the stellar locus. It is impractical to reobserve all such candidates, so an alternative approach was developed in which Bayesian model comparison techniques are used to calculate the probability that a candidate is a HZQ, P_q, by combining models of the quasar and star populations with the photometric measurements of the object. This method was motivated specifically by the large number of HZQ candidates identified by cross-matching the UKIRT Infrared Deep Sky Survey (UKIDSS) Large Area Survey (LAS) to the Sloan Digital Sky Survey (SDSS): in the ~1900 deg^2 covered by the LAS in the UKIDSS Seventh Data Release (DR7) there are ~10^3 real astronomical point-sources with the measured colours of the target quasars, of which only ~10 are expected to be HZQs. Applying Bayesian model comparison to the sample reveals that most sources with HZQ-like colours have P_q <~ 0.1 and can be confidently rejected without the need for any further observations. In the case of the UKIDSS DR7 LAS, there were just 88 candidates with P_q >= 0.1; these object were prioritized for reobservation by ranking according to P_q (and their likely redshift, which was also inferred from the photometric data). Most candidates were rejected after one or two (moderate depth) photometric measurements by recalculating P_q using the new data. That left seven confirmed HZQs, three of which were previously identified in the SDSS and four of which were new UKIDSS discoveries. The high efficiency of this Bayesian selection method suggests that it could usefully be extended to other HZQ surveys (e.g. searches by Pan-STARRS or VISTA) as well as to other searches for rare objects.Comment: submitted to MNRAS; 20 pages, 13 figure

Photometric brown-dwarf classification

“…This could be achieved most rigorously by using Bayesian model comparison (Mortlock et al 2012). A formalism for doing so is detailed in the Appendix, which also includes the series of approximations and assumptions which lead from the full Bayesian result to the much simpler χ 2 -based classification we actually use.…”

“…This is justified at high signal-to-noise ratios (S/N). At low S/N, the correct approach is to work with fluxes (e.g., Mortlock et al 2012) where, except in the low-photon regime (e.g., X-ray astronomy), the errors are Gaussian. A few T dwarfs in our sample have low S/N in the SDSS i and z bands; for these we use |asinh| magnitudes and errors (Lupton et al 1999) as the resultant χ 2 value is close to that which would have been calculated from the fluxes.…”

The VMC Survey

“…A series of papers have described the K excess (KX) method (Warren et al 2000) using UKIDSS JK bands and the SDSS g band (e.g. Maddox et al 2012;Mortlock et al 2012). Findlay et al (2012) based their selection on single epoch VISTA ZY J data.…”

Discovery of 16 New z ∼ 5.5 Quasars: Filling in the Redshift Gap of Quasar Color Selection

“…Our new quasar sample and selection method could provide a new training sample for future probabilistic selection. Besides, an extension of the Bayesian model from Mortlock et al (2012) to include more NIR colors will also be expected to be useful for z ∼ 5.5 quasar selection. Additionally, in the future, the variability (e.g., LSST) will also play an important role on high redshift quasar selection.…”

“…In this work, we only use a color box to select z∼5.5 quasar candidates. Recently, several modern techniques, probabilistic selections based on the Bayesian model or extreme deconvolution method, have been explored to search for quasars at different redshift ranges (e.g., Mortlock et al 2012;DiPompeo et al 2015). The extreme deconvolution method in DiPompeo et al (2015) required a sample of quasars used as a training set.…”