Using Machine Learning to Link Black Hole Accretion Flows with Spatially Resolved Polarimetric Observables
Richard Qiu,Angelo Ricarte,Ramesh Narayan,George N. Wong,Andrew Chael,Daniel Palumbo
DOI: https://doi.org/10.1093/mnras/stad466
2023-02-10
Abstract:We introduce a new library of 535,194 model images of the supermassive black holes and Event Horizon Telescope (EHT) targets Sgr A* and M87*, computed by performing general relativistic radiative transfer calculations on general relativistic magnetohydrodynamics simulations. Then, to infer underlying black hole and accretion flow parameters (spin, inclination, ion-to-electron temperature ratio, and magnetic field polarity), we train a random forest machine learning model on various hand-picked polarimetric observables computed from each image. Our random forest is capable of making meaningful predictions of spin, inclination, and the ion-to-electron temperature ratio, but has more difficulty inferring magnetic field polarity. To disentangle how physical parameters are encoded in different observables, we apply two different metrics to rank the importance of each observable at inferring each physical parameter. Details of the spatially resolved linear polarization morphology stand out as important discriminators between models. Bearing in mind the theoretical limitations and incompleteness of our image library, for the real M87* data, our machinery favours high-spin retrograde models with large ion-to-electron temperature ratios. Due to the time-variable nature of these targets, repeated polarimetric imaging will further improve model inference as the EHT and next-generation (EHT) continue to develop and monitor their targets.
High Energy Astrophysical Phenomena,Astrophysics of Galaxies,Instrumentation and Methods for Astrophysics