Galaxy Cluster Membership with Machine Learning

Galaxy clusters are the most massive gravitationally bound objects in the Universe, and contain hundreds or even thousands of galaxies. Because they are so massive and rare, cluster abundance as a function of mass and redshift is useful to constrain cosmological models. However, when using optical observations, galaxies that are actually in the fore- or background of the cluster look as though they belong in the cluster. These interloping galaxies add both scatter and bias to dynamical cluster mass estimates which introduce error to cosmological constraints. Therefore, it is imperative to develop methods that can accurately identify these interlopers. We use ~38,000 simulated clusters from the MultiDark N-body simulation consisting of dark matter only. For these clusters true membership is known. We then use this catalog to develop a data vector of engineered features containing galaxy and galaxy cluster properties, such as density, velocity along line of sight, and radius to cluster center. These features are used to train an artificial neural network, a deep machine learning algorithm. Our technique for determining cluster membership can correctly identify 60% of interloping galaxies and 85% of true members.