Coloring Black Holes: Choices in Creating the First Image of the Event Horizon ✛ ✎ ✦
Context
Direct observations of black holes were long considered an almost hopeless task. Although their existence followed from general relativity by Einstein, the event horizon does not radiate, and the angular size of even the nearest supermassive objects is vanishingly small—roughly like an orange on the Moon. The breakthrough became possible thanks to radio astronomy and VLBI interferometry, which combines signals from telescopes across the planet into a single virtual instrument the size of Earth. In 2017, the Event Horizon Telescope collaboration used eight such stations to capture emission at a wavelength of 1.3 mm from the plasma around the black hole M87*. However, between raw data and the final image lay a chasm: measurements were catastrophically sparse, and each pixel corresponded to an infinite number of possibilities consistent with the noise.
Methods
Our study is based on a historical-epistemological analysis of the visualization process. We meticulously examined EHT publications, technical reports, and the open-source code of the algorithms. Moreover, we independently varied the parameters of the programs to reconstruct alternative images, demonstrating how the snapshots might have looked under different but equally justified choices. This allowed us to trace the role of numerical simulations in validating algorithms and to assess the influence of machine learning on the final image. Parallels with classic scientific practices—from Anton Pannekoek's 'mean subjective image' of the Milky Way to 'blind' analyses in particle physics—helped clarify how scientists cope with the underdetermination of data.
Results
It turned out that even under strict constraints, the final image was not strictly dictated by the data. At every stage—from calibration to 'fiducial' construction—there were several equally valid paths. For example, the classic CLEAN algorithm broke the sky into point sources, while regularized maximum likelihood (RML) methods implicitly introduced expectations, such as a gravitationally lensed ring structure around the accretion disk. The final publication averaged three images from different pipelines, and the choice of color scheme (orange-black) was just one of 196 options. Numerically, key parameters showed robustness: out of 37,500 combinations for one algorithm, about 1,500 gave similar morphology. Yet fine details varied greatly, and it was this limited variability that testified to the reality of the black hole in the active galactic nucleus of M87 with its powerful jet.
Implications
For the philosophy of science, the EHT work demonstrates a new type of evidence: not a single snapshot, but a 'demonstration of limited variability' under different assumptions. Instead of hiding ambiguity, one could present the whole spectrum of plausible images, which would strengthen confidence in the conclusions. On a practical level, the case highlighted how blurred the line is between observation and simulation: algorithms were tested on synthetic data, and machine learning (the PRIMO project) transferred patterns from simulations onto real measurements, sparking debates about circular reasoning when testing Einstein's equations. Historically, Karl Schwarzschild had already found the first exact solution for a non-rotating black hole, and Roger Penrose proved the inevitability of singularities.
Future development
The future Next Generation EHT project will add new telescopes, including space-based ones, which will dramatically improve UV-plane coverage and bring reconstructions closer to the true appearance of black holes. In parallel, Bayesian ensembling methods that honestly report uncertainties will be developed. Improved simulations are expected to allow more accurate modeling of plasma around accretion disks, and new VLBI networks will enable testing of gravitational lensing in strong fields.
Impact
The results impact not only the astrophysics of active galactic nuclei and relativistic jets, but also the methodology for handling highly noisy data in geophysics and neuroscience.
Next steps
The immediate task is to rethink the status of 'raw data': the collaboration must resolve the debate over what level of preprocessing is sufficient for independent verification. It is also necessary to conduct blind tests with a wider set of synthetic images to exclude confirmation bias.
Key open problems
The work directly touches on the problem of underdetermination of theory by data and the search for objectivity when classical methods of independent verification are inapplicable. It also resonates with the question of how simulations can serve as an empirical foundation when their validation in the regime of interest is impossible—a central challenge for early-Universe physics and quantum gravity.
The knowledge machine's view
This section is addressed to the author of the paper and their colleagues. Neighbouring works were found by meaning-based search across our archive, not by word overlap. We do not review: we show the field around the work.
What the work does
This work is a historical-epistemological analysis of how the Event Horizon Telescope collaboration obtained the first image of the event horizon of the black hole M87*. The authors independently reconstructed alternative images by varying algorithm parameters and showed that the key result is not a single perfect picture but the robustness of the dark ring across various honest processing approaches.
Where its strength is
The strength of the work lies in demonstrating that scientific evidence in astrophysics can be not a single image but a robust pattern that withstands variability in reconstruction methods. This paves the way for more reliable separation of facts from processing artifacts.
Where it could go next
- It might be interesting to compare your methodology for analyzing the robustness of the black hole image with approaches used to refine its mass from the shadow and ring, where the role of model assumptions in data interpretation is also important.based on: Shadow and Ring: How to Weigh a Black Hole More Precisely arXiv:2606.30753
- Related work exists on how dark matter might affect the appearance of black holes; it would be curious to apply your approach to assess how distinguishable such effects are at current noise levels and processing freedom.based on: How Dark Matter Reshapes Black Holes arXiv:2608.04470v1
- Our archive contains a study of black hole echoes via gravitational waves; one could consider transferring your idea of multiple equally valid reconstructions to the analysis of gravitational-wave event signals, where data are also highly underdetermined.based on: Echoes of Black Holes: How Gravitational Sirens Reveal the Curvature of the Universe arXiv:2606.04216v1
Worth trying
- Test your method of reconstructing alternative images on simulated data with added dark matter effects to assess under what conditions they become noticeable.
- Check how the choice of processing algorithms affects the extracted black hole mass by varying pipelines similar to what you did for shadow morphology.
- Use your 'blind' analysis approach to test the robustness of black hole parameters derived from gravitational-wave signals under different waveform models.
Why this may matter
This could be important for developing standards of evidence in astrophysics, where images and signals are increasingly products of complex processing rather than direct observation, and where understanding the limits of interpretation directly affects trust in scientific results.
Nearby works in our archive
- Shadow and Ring: How to Weigh a Black Hole More Preciselysimilarity 0.69
- How Dark Matter Reshapes Black Holessimilarity 0.66
- A Black Hole's Shadow with a Secret from Other Dimensionssimilarity 0.64
- Echoes of Black Holes: How Gravitational Sirens Reveal the Curvature of the Universesimilarity 0.64
- A Gravitational Needle in a Haystacksimilarity 0.64
- Sound Horizons in the Darkness of Spacesimilarity 0.64
The closest work in the archive sits at 0.69 by meaning — the usual level for our corpus.
During training 'Imaging Challenges,' EHT participants, not knowing what they were 'observing,' successfully reconstructed not only simulated black holes but also images of a snowman. It was a fun but serious test of the algorithms' ability not to introduce desired structures.
The narrative echoes Arthur C. Clarke's '2010: Odyssey Two,' where during the investigation of a black hole in the Jupiter system, the method of data interpretation turned out to be critical. And the visualizations in 'Interstellar' were built on the same general relativistic simulations that the EHT later used.
Key numbers
- distance to M87: 55 million light-years
- observation wavelength: 1.3 mm
- data recording rate: 64 Gbit/s
- number of telescopes in 2017: 8
- mass of black hole M87*: ~6.5×10⁹ M☉