Issue |
A&A
Volume 698, May 2025
|
|
---|---|---|
Article Number | A62 | |
Number of page(s) | 8 | |
Section | Numerical methods and codes | |
DOI | https://doi.org/10.1051/0004-6361/202553786 | |
Published online | 06 June 2025 |
Deep learning inference with the Event Horizon Telescope
III. ZINGULARITY results from the 2017 observations and predictions for future array expansions
1
Department of Astrophysics, Institute for Mathematics, Astrophysics and Particle Physics (IMAPP),
Radboud University, PO Box 9010,
6500
GL
Nijmegen,
The Netherlands
2
Max-Planck-Institut für Radioastronomie,
Auf dem Hügel 69,
53121
Bonn,
Germany
3
Steward Observatory and Department of Astronomy, University of Arizona,
933 N. Cherry Ave.,
Tucson,
AZ
85721,
USA
4
Data Science Institute, University of Arizona,
1230 N. Cherry Ave.,
Tucson,
AZ
85721,
USA
5
Program in Applied Mathematics, University of Arizona,
617 N. Santa Rita Ave.,
Tucson,
AZ
85721,
USA
6
Department of Astrophysical Sciences,
Peyton Hall, Princeton University,
Princeton,
NJ
08544,
USA
7
Instituto de Astrofísica de Andalucía-CSIC,
Glorieta de la Astronomía s/n,
18008
Granada,
Spain
★ Corresponding author: M.Janssen@astro.ru.nl
Received:
16
January
2025
Accepted:
31
March
2025
Context. In the first two papers of this publication series, we present a comprehensive library of synthetic Event Horizon Telescope (EHT) observations and used this library to train and validate Bayesian neural networks for the parameter inference of accreting supermassive black hole systems. The considered models are ray-traced general relativistic magnetohydrodynamic (GRMHD) simulations of Sgr A* and M87*.
Aims. In this work, we infer the best-fitting accretion and black hole parameters from 2017 EHT data and predict improvements that will come with future upgrades of the array.
Methods. Compared to previous EHT analyses, we considered a substantially larger synthetic data library and the most complete set of information from the observational data. We made use of the Bayesian nature of the trained neural networks and apply bootstrapping of known systematics in the observational data to obtain parameter posteriors.
Results. Within a wide GRMHD parameter space, we find M87* to be best described by a spin between 0.5 and 0.94 with a retrograde MAD accretion flow and strong synchrotron emission from the jet. Sgr A* has a high spin of ∼0.8–0.9 and a prograde accretion flow beyond the standard MAD/SANE models with a comparatively weak jet emission, seen at a ∼ 20°–40° inclination and ∼106°–137° position angle. While previous EHT analyses could rule out specific regions in the model parameter space considered here, we are able to obtain narrow parameter posteriors with our ZINGULARITY framework without being impacted by the unknown foreground Faraday screens and data calibration biases. We further demonstrate that the Africa Millimeter Telescope extension to the EHT will reduce parameter inference errors by a factor of three for non-Kerr models, enabling more robust tests of general relativity.
Conclusions. Our results agree with multiwavelength constraints from the literature. It will be instructive to produce new GRMHD models with the inferred interpolated parameters for in depth model-data comparisons and to study their accretion rate plus jet power.
Key words: accretion, accretion disks / black hole physics / techniques: high angular resolution / techniques: interferometric / galaxies: active
© The Authors 2025
Open Access article, published by EDP Sciences, under the terms of the Creative Commons Attribution License (https://creativecommons.org/licenses/by/4.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.
This article is published in open access under the Subscribe to Open model. Subscribe to A&A to support open access publication.
Current usage metrics show cumulative count of Article Views (full-text article views including HTML views, PDF and ePub downloads, according to the available data) and Abstracts Views on Vision4Press platform.
Data correspond to usage on the plateform after 2015. The current usage metrics is available 48-96 hours after online publication and is updated daily on week days.
Initial download of the metrics may take a while.