Reconstructing an image from an array of telescopes scattered across continents is fundamentally an ill-posed mathematical exercise. When radio astronomers combine signals across thousands of kilometres to synthesise an Earth-sized instrument, the resulting baseline coverage gathers only a vanishingly small fraction of the spatial frequencies required to form a complete picture. The gap between the recorded data and the final visual representation must inevitably be bridged by computational assumptions, regularisation terms, and calibration models.
This structural limitation came to the fore following the initial images of the supermassive black hole at the core of the galaxy Messier 87, designated M87*. Those images, generated by the Event Horizon Telescope Collaboration using millimetre very-long baseline interferometry (VLBI), depicted a luminous, gravitationally lensed and redshifted emission ring encircling a dark central shadow. Yet because the baseline measurements were inherently sparse, differing data analysis routines and imaging heuristics soon produced divergent interpretations of the target’s fine-scale features. Several independent re-analyses even called the fundamental ring morphology into question, arguing that the circular feature might be an artefact of prescriptive geometric priors or sensitive calibration adjustments rather than a physical reality.
Now, a study published in Nature demonstrates that the ring-like morphology of M87* persists when analysed through an independent imaging pipeline specifically engineered to discard prescriptive geometric assumptions and rely on observables immune to calibration drift.
The fragility of sparse interferometric reconstruction
In conventional radio astronomy, an interferometer does not record photographic pixels directly. Instead, each pair of antennas measures a single spatial frequency component of the radio brightness distribution on the sky, a quantity known as a complex visibility. The distance and orientation between any two stations define a baseline vector. As the Earth rotates, these baselines trace arcs across the spatial frequency domain (the Fourier plane).
For an array attempting to achieve microarcsecond angular resolution at millimetre wavelengths, the physical gaps between observing stations are vast, leaving large swaths of the Fourier plane entirely empty. Transforming these sparsely sampled Fourier components back into a spatial image requires solving an underconstrained inverse problem: an infinite variety of different spatial brightness profiles can match the observed visibilities equally well.
Standard imaging workflows traditionally resolve this ambiguity through two main routes: maximum entropy and regularised maximum likelihood methods, or iterative deconvolution techniques such as CLEAN. Both families depend heavily on hyperparameters that control image smoothness, compactness, or adherence to expected geometries. Furthermore, high-frequency millimetre signals are notoriously sensitive to atmospheric turbulence and receiver instability, which introduce station-dependent gain and phase errors. If an algorithm attempts to calibrate these instrument errors iteratively while simultaneously reconstructing an image, it runs the risk of circular reinforcement: the calibration solutions can warp to fit the assumed source structure, inadvertently baking the expected outcome into the final map.
Calibration-independent observables and prior-free imaging
To resolve whether the ring structure in M87* was an objective feature of the astrophysical emission or a product of algorithmic conditioning, the authors implemented a distinct computational framework. Rather than tuning instrument calibration alongside spatial models, the new algorithm operates directly on interferometric observables that are mathematically immune to station-specific calibration errors.
In radio interferometry, forming specific algebraic combinations of visibilities across closed loops of three or four stations cancels out station-dependent phase and amplitude corruptions. Phase closures around a closed triangle of baselines eliminate antenna-specific phase delays, while amplitude closure ratios across quadrangles of stations neutralise station gain fluctuations. By formulating the image reconstruction directly around these invariant quantities, the new pipeline removes the need to iteratively solve for volatile calibration parameters.
Equally critical is the elimination of prescriptive morphological priors. Rather than nudging the solver toward compact annular geometries, crescent shapes, or centrally darkened templates, the new algorithm explores the parameter space without imposing predefined structural expectations. Despite the absence of these geometric constraints and the complete exclusion of calibration-tuning stages, the independent reconstruction unequivocally reproduces the characteristic ring-like morphology of the emission surrounding the central shadow.
What the reconstruction confirms and its physical bounds
The reconstructed image yields an angular diameter for the shadow of 42.2 +/- 1.5 microarcseconds. When interpreted alongside nominal scaling parameters, this physical scale corresponds to a central black hole mass of 6.5 +/- 0.5 billion solar masses. The reconstruction also reveals that the system is viewed at an inclination angle of approximately 20 degrees relative to the approaching plasma jet.
These parameters provide solid confirmation that the ring feature identified in earlier reconstructions is an authentic physical structure rather than an artefact of algorithmic bias or baseline sparsity. The diameter, mass estimation, and viewing geometry align closely with the gravitational lensing and relativistic deflection predicted for a supermassive compact object at the galactic nucleus of Messier 87.
Nevertheless, it is important to delineate what this analytical milestone does and does not establish. The method confirms the macroscopic ring morphology, but it does not resolve persistent uncertainties regarding sub-ring structural details or dynamic temporal fluctuations in the accretion flow. Because the underlying physical data remain fundamentally constrained by the sparse baseline sampling of the global array, the observable-based method cannot conjure missing high-frequency information. Finer physical questions—such as the precise internal magnetic topology of the inner accretion disk or the detailed boundary conditions where the relativistic jet launches—remain beyond reach without denser baseline coverage or shorter observing wavelengths. Furthermore, the derived mass remains tied to nominal scaling assumptions and depends on the adopted distance to the host galaxy.
Sources
- Independent interferometric observable confirms black hole shadow in M87*, Samuel Lai et al., arXiv; published in Lai, S., Thyagarajan, N., Wong, O.I. et al. Independent interferometric observable confirms black hole shadow in M87*. Nature 659, 317 (2026), 2026-11-16
- Publisher record (DOI)
The R&D takeaway
For engineering and research leaders dealing with sparse inverse problems, this work illustrates the necessity of decoupling signal verification from calibration assumptions. When sensor coverage is incomplete, tuning algorithmic hyperparameters and instrument calibration in tandem creates severe vulnerabilities to confirmation bias. The technical strategy applied here—restructuring the mathematical framework around invariant, calibration-free observables and stripping out prescriptive structural priors—provides an essential methodology for validating contested observations in any domain reliant on sparse sensor networks.
The R&D Innovate desk