Deep Horizon: A machine learning network that recovers accreting black hole parameters (van der Gucht et al.)

Vol. 636
15. Numerical methods and codes

Deep Horizon: A machine learning network that recovers accreting black hole parameters

by J. van der Gucht, J. Davelaar, Hendriks, et al. 2020, A&A, 636, A94 alt

In April 2019, the Event Horizon Telescope released the first ever image of the shadow of a black hole at the center of the M87 elliptical galaxy, dominating the Virgo cluster. These observations were made possible by eight ground-based radio telescopes spread all over the world. In this paper, van der Gucht et al. present a combination of two neural network algorithms that are able to recover the physical parameters of the accreting black hole based on the observed shadow. To train these networks, they used a set of general relativistic simulations of accretion disks around massive black holes, testing a grid of parameters. They discover that with the current resolution, only the mass of the black hole and the mass accretion rate can be safely recovered. To derive further parameters, such as the black hole spin, we would need to go beyond the Earth-based set of radio telescopes and include a space-based radio mission.