ROGER: Reconstructing Orbits of Galaxies in Extreme Regions using machine learning techniques

VALERIA COENDA; CRISTIAN A. VEGA-MARTÍNEZ; MARTÍN DE LOS RÍOS; HERNÁN MURIEL; SOFIA A. CORA; HÉCTOR J. MARTÍNEZ; ANDRÉS N. RUÍZ

MONTHLY NOTICES OF THE ROYAL ASTRONOMICAL SOCIETY

WILEY-BLACKWELL PUBLISHING, INC

Lugar: Londres; Año: 2021 vol. 510 p. 1934 - 1944

0035-8711

We present the ROGER (Reconstructing Orbits of Galaxies in Extreme Regions) code, which uses three different machine learning techniques to classify galaxies in, and around, clusters, according to their projected phase-space position. We use a sample of 34 massive, M200>1015h−1M⊙, galaxy clusters in the MultiDark Planck 2 (MDLP2) simulation at redshift zero. We select all galaxies with stellar mass M⋆≥108.5h−1M⊙, as computed by the semi-analytic model of galaxy formation SAG, that are located in, and in the vicinity of, the clusters and classify them according to their orbits. We train ROGER to retrieve the original classification of the galaxies out of their projected phase-space positions. For each galaxy, ROGER gives as output the probability of being a cluster galaxy, a galaxy that has recently fallen into a cluster, a backsplash galaxy, an infalling galaxy, or an interloper. We discuss the performance of the machine learning methods and potential uses of our code. Among the different methods explored, we find the K-Nearest Neighbours algorithm achieves the best performance.