Inferring galaxy dark halo properties from visible matter with machine learning

• DOI: 10.1093/mnras/stac2449
• Publication Date: 2022-09-02T04:13:10Z
• Authors: Rodrigo von Marttens; Luciano Casarini; Nicola R Napolitano; Sirui Wu; Valeria Amaro; Rui Li; Crescenzo Tortora; Askery Canabarro; Yang Wang
• Keywords: Velocity dispersion; Stellar mass

ABSTRACT Next-generation surveys will provide photometric and spectroscopic data of millions to billions galaxies with unprecedented precision. This offers a unique chance improve our understanding the galaxy evolution unresolved nature dark matter (DM). At scales, density distribution DM is strongly affected by feedback processes, which are difficult fully account for in classical techniques derive masses. We explore capability supervised machine learning (ML) algorithms predict content from ‘luminous’ observational-like parameters, using TNG100 simulation. In particular, we use (magnitudes different bands), structural (the stellar half-mass radius three baryonic masses), kinematic (1D velocity dispersion maximum rotation velocity) parameters total mass, radius, mass inside one two radii. adopt coefficient determination, R2, as metric evaluate accuracy these predictions. find that all observational quantities together (photometry, structural, kinematics), reach high (up R2 ∼ 0.98). first test shows ML tools promising real galaxies. The next steps be implement realism training sets, closely selecting samples accurately reproduce typical observed scaling relations. so-trained pipelines suitable collected Rubin/Large Synoptic Survey Telescope (LSST), Euclid, Chinese Space (CSST), 4-metre Multi-Object Spectrograph (4MOST), Dark Energy Spectroscopic Instrument (DESI), e.g. properties their central fractions.