Deep learning inference with the Event Horizon Telescope
M. Janssen, C.-k. Chan, J. Davelaar, I. Natarajan, H. Olivares, B. Ripperda, J. Röder, M. Rynge, M. Wielgus · 2025
Read Deep learning inference with the Event Horizon Telescope by M. Janssen, C.-k. Chan, J. Davelaar, I. Natarajan, H. Olivares, B. Ripperda, J. Röder, M. Rynge, M. Wielgus (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2506-13873.txt.
Introduction
Janssen and co-authors describe deep learning work with the Event Horizon Telescope, in a research paper from 2025. The telescope network images black holes in the galaxies M87 and the Milky Way. To judge which theory fits the data, the team needs a huge library of fake data made from models, built to look like real telescope output.
This first part reports a better data reduction pipeline and the library itself. Ray-traced simulations span many black hole and accretion settings. The signal path and calibration steps are folded in, so the fake data match real data closely. Near a million data sets cover past and planned observations.
Catalog record
| CREATOR | M. Janssen, C.-k. Chan, J. Davelaar, I. Natarajan, H. Olivares, B. Ripperda, J. Röder, M. Rynge, M. Wielgus |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 66 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2506-13873
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Deep learning inference with the Event Horizon Telescope M. Janssen, C.-k. Chan, J. Davelaar, I. Natarajan, H. Olivares, B. Ripperda, J. Röder, M. Rynge, M. Wielgus Abstract Context. In a series of publications, we describe a comprehensive comparison of Event Horizon Telescope (EHT) data with theoretical models of the observed Sagittarius A* (Sgr A ∗ ) and Messier 87* (M87 ∗ ) horizon-scale sources. Aims. In this first article, we report on improvements made to our observational data reduction pipeline and present the generation of observables derived from the EHT models. We make use of ray-traced general relativistic magnetohydrodynamic simulations that are based on different black hole spacetime metrics and accretion physics parameters. These broad classes of models provide a good representation of the primary targets observed by the EHT. Methods. We describe how we combined multiple frequency bands and polarization channels of the observational data to improve our fringe-finding sensitivity and stabilization of atmospheric phase fluctuations. To generate realistic synthetic data from our models, we took the signal path as well as the calibration process, and thereby the aforementioned improvements, into account. We could thus produce synthetic visibilities akin to calibrated EHT data and identify salient features for the discrimination of model parameters. Results. We have produced a library consisting of an unparalleled 962,000 synthetic Sgr A ∗ and M87 ∗ datasets. In terms of baseline coverage and noise properties, the library encompasses 2017 EHT measurements as well as future observations with an extended telescope array. Conclusions. We differentiate between robust visibility data products related to model features and data products that are strongly affected by data corruption effects. Parameter inference is mostly limited by intrinsic model variability, which highlights the importance of long-term monitoring observations with the EHT. In later papers in this series, we will show how a Bayesian neural network trained on our synthetic data is capable of dealing with the model variability and extracting physical parameters from EHT observations. With our calibration improvements, our newly reduced EHT datasets have a considerably better quality compared to previously analyzed data. S/N signal-to-noise ratio VLBI very long baseline interferometry GRMHD general relativistic magnetohydrodynamics GRRT general relativistic ray-tracing MHD magnetohydrodynamics as arcseconds AGNs Active Galactic Nuclei LLAGN low-luminosity AGN RIAF radiatively inefficient accretion flow Cen A Centaurus A Sgr A* Sagittarius A* M87 ∗ Messier 87* AGNs active galactic nuclei EHT Event Horizon Telescope BHC BlackHoleCam TANAMI Tracking Active Galactic Nuclei with Austral Milliarcsecond Interferometry SMBH supermassive black hole ALMA Atacama Large Millimeter/submillimeter Array APEX Atacama Pathfinder Experiment PV IRAM 30 m Telescope JCMT James Clerk Maxwell Telescope LMT Large Millimeter Telescope Alfonso Serrano SPT South Pole Telescope SMA Submillimeter Array SMT Submillimeter Telescope c speed of light pc parsec GR general relativity aips Astronomical Image Processing System casa Common Astronomy Software Applications Symba SYnthetic Measurement creator for long Baseline Arrays Rpicard Radboud PIpeline for the Calibration of high Angular Resolution Data hops Haystack Observatory Postprocessing System HBT Hanbury Brown and Twiss TOV Tolman-Oppenheimer-Volkoff MRI magnetorotational instability ADAF advection-dominated accretion flow ADIOS adiabatic inflow-outflow solution CDAF convection-dominated accretion flow BZ Blandford-Znajek BP Blandford-Payne EM electromagnetic BLR broad-line region NLR narrow-line region ISM interstellar medium eDF electron distribution function PIC particle-in-cell SANE standard and normal evolution MAD magnetically arrested disk JIVE Joint Institute for VLBI ERIC NRAO National Radio Astronomy Observatory \mu as microarcseconds Jy Jansky PA position angle SRMHD special relativistic magnetohydrodynamics I. Calibration improvements and a comprehensive synthetic data library † † orcid: 0000-0001-8685-6544 † † orcid: 0000-0001-6337-6126 † † orcid: 0000-0002-2685-2434 † † orcid: 0000-0001-8242-4373 † † orcid: 0000-0001-6833-7580 † † orcid: 0000-0002-7301-3908 † † orcid: 0000-0002-2426-927X † † orcid: 0000-0002-1779-7189 † † orcid: 0000-0002-8635-4242 1 Introduction Powered by accretion onto smbh , agn ( agn ) belong to the most luminous persistent sources in the known Universe ( Schmidt 1963 ; Lynden-Bell 1969 ) . With the eht ( eht ) vlbi ( vlbi ) array, we can resolve the innermost region of agn and study accretion onto compact objects, plasma physics, jet launching, and gravity in the strong field regime ( Event Horizon Telescope Collaboration et al. 2019a ; Event Horizon Telescope Collaboration et al. 2019b ; Kim et al. 2020 ; Psaltis et al. 2020 ; Event Horizon Telescope Collaboration et al. 2021 ; Kocherlakota et al. 2021 ; Janssen et al. 2021 ; Event Horizon Telescope Collaboration et al. 2022a ; Event Horizon Telescope Collaboration et al. 2022e ; Issaoun et al. 2022 ; Jorstad et al. 2023 ; Paraschos et al. 2024 ; Baczko et al. 2024 ) . However, interpreting EHT data poses significant challenges, such as the inherent variability of black hole accretion flows, atmospheric noise, and instrument-related data corruption effects. Addressing these issues requires both methodological advances and high-fidelity simulations of synthetic data. Currently, grmhd ( grmhd ) simulations ( Mizuno 2022 , e.g.,) are the most complete class of models available for a self-consistent description of agn ’ central activity. grmhd s naturally produce jets and, without (substantial) radiative cooling, a geometrically thick accretion flow. For parameters matching the sgra ( sgra ) and m87 ( m87 ) sources, grrt ( grrt ) calculations based on grmhd simulations predict (sub-) millimeter synchrotron emission to be produced in an (mostly) optically thin plasma surrounding the black hole ( Event Horizon Telescope Collaboration et al. 2019e ; Event Horizon Telescope Collaboration et al. 2022f ) . From grmhd - grrt models, full Stokes horizon-scale movies of the disk plus jet emission of sgra and m87 have been made. The geometrically thick and optically thin accretion scenario with accompanying jets is suitable for low-powered agn (e.g., Ho 2008 , and references therein) . The activity of luminous agn and X-ray binaries can also be simulated with grmhd - grrt models ( Dexter et al. 2021 ; Liska et al. 2021 ; Wielgus et al. 2022a , e.g.,) , but unlike low accretion rate sources, they generally require radiative cooling to be self-consistent. For the eht , grmhd - grrt simulations are used for the validation of image reconstructions ( Event Horizon Telescope Collaboration et al. 2019d ; Event Horizon Telescope Collaboration et al. 2022c ) and tests of gr ( gr ) together with the inference of MHD, accretion, and black hole parameters ( Event Horizon Telescope Collaboration et al. 2019e ; Event Horizon Telescope Collaboration et al. 2019f ; Psaltis et al. 2020 ; Kocherlakota et al. 2021 ; Event Horizon Telescope Collaboration et al. 2022f ; Event Horizon Telescope Collaboration et al. 2022e ) . In this first paper of a series, we present a comprehensive synthetic data library that is based on a wide range of grmhd - grrt models. Specifically, we used three broad classes of models with different spacetime metrics. The first class is based on the Kerr solution to Einstein’s field equations in GR without charge ( Kerr 1963 ) . The second class is based on the Kerr-Newman solution to GR, which describes a rotating, charged black hole ( Newman et al. 1965 ) . The third class is a dilaton black hole, which is a solution of the Einstein-Maxwell-dilaton-axion field equations ( García et al. 1995 ) . Beyond GR, the dilaton scalar