An artificial neural network approach for ranking quenching parameters in central galaxies

Abstract

We present a novel technique for ranking the relative importance of galaxy properties in the process of quenching star formation. Specifically, we develop an artificial neural network (ANN) approach for pattern recognition and apply it to a population of over 400,000 central galaxies taken from the Sloan Digital Sky Survey Data Release 7. We utilise a variety of physical galaxy properties for training the pattern recognition algorithm to recognise star forming and passive systems, for a 'training set' of ∼100,000 galaxies. We then apply the ANN model to a 'verification set' of ∼100,000 different galaxies, randomly chosen from the remaining sample. The success rate of each parameter singly, and in conjunction with other parameters, is taken as an indication of how important the parameters are to the process(es) of central galaxy quenching. We find that central velocity dispersion, bulge mass and B/T are excellent predictors of the passive state of the system, indicating that properties related to the central mass of the galaxy are most closely linked to the cessation of star formation. Larger scale galaxy properties (total or disk stellar masses), or those linked to environment (halo masses or δ 5 ) perform significantly less well. Our results are plausibly explained by AGN feedback driving the quenching of central galaxies, although we discuss other possibilities as well.

The impact of galactic properties and environment on the quenching of central and satellite galaxies: a comparison between SDSS, Illustris and L-Galaxies

“…Teimoorinia et al (2016) found strong evidence from an ANN analysis that central velocity dispersion is the most predictive, and hence most tightly constraining, observable for central galaxy quenching out of the following list of variables: stellar, halo, bulge and disk mass; local galaxy density and galactic structure (B/T ). Moreover central velocity dispersion is found to be tightly correlated with surface mass density within 1 kpc.…”

“…A fibre correction is applied based on galaxy colour and magnitude outside the aperture. This is the same sample of SFRs used in many recent quenching papers (e.g., Woo et al 2013;Bluck et al 2014;Woo et al 2015;Teimoorinia et al 2016). All of the results and conclusions of this work are recovered qualitatively even if we use photometric SFRs from SED fitting, or construct the analogous red fraction instead of the quenched fraction from star formation rates.…”

“…Full details on the observational data and measurements used in this paper are given in the prior works of this series, Bluck et al (2014) and Teimoorinia et al (2016). In addition to the SDSS observations, we also compare to the Illustris simulation (Vogelsberger et al 2014a,b) and to the latest version of the Munich model of galaxy formation (L-Galaxies, Henriques et al 2015).…”

“…For a smaller list of variables (not including bulge or halo mass) Wake et al (2012) established that central velocity dispersion outperforms stellar mass, morphology and environment in constraining the quenching of a general population of local galaxies. Recently, Teimoorinia, Bluck & Ellison (2016) found strong evidence from an ANN technique that central velocity dispersion is the best single variable for parameterizing the quenching of centrals, improving upon even bulge mass. Additionally, Cheung et al (2012), Fang et al (2013) and Woo et al (2015) find strong evidence for the central stellar mass density within 1 kpc being a particularly tight correlator to the quenched fraction.…”

“…The primary motivation for this paper is to expand on the work of Wake et al (2012), Bluck et al (2014) and Teimoorinia et al (2016) by investigating the impact on the quenched fraction of central and satellites galaxies from varying galaxy and environmental properties at fixed central velocity dispersion. For centrals, this allows us to look for additional dependencies of quenching, whilst controlling for the parameter which matters most statistically.…”