🔭 astrophysics

A Mathematical Modeling Method of the EHT Synthesized Beam for Radio Synthesis Imaging

This paper proposes LiRBM, a high-precision modeling method based on linear radial basis functions that significantly outperforms existing techniques in accurately representing the Event Horizon Telescope's synthesized beam to facilitate improved black hole image restoration and quantitative analysis.

Original authors: Guoshengyi Wu, Li Zhang

Published 2026-09-11

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Guoshengyi Wu, Li Zhang

Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

To see a black hole, astronomers must look through a telescope that is not a single tube of glass, but a planet-sized network of radio dishes scattered across the globe. This network, known as the Event Horizon Telescope, links telescopes from Hawaii to the South Pole to create a single instrument with the resolving power of a mirror the size of Earth. Because the dishes are so far apart and do not cover every inch of the sky, the telescope does not produce a clean, sharp photograph. Instead, it captures a blurry, noisy image that is heavily distorted by the gaps in its own coverage. This distortion is caused by a specific pattern of light and dark ripples, called the synthesized beam, which acts like a fingerprint of the telescope's limitations. To recover the true shape of a black hole from this muddled data, scientists must understand this fingerprint with extreme precision. If they cannot describe the pattern of the ripples mathematically, they cannot remove the blur, and the image of the black hole remains hidden behind the noise.

For years, researchers have tried to clean up these images using methods that treat the telescope's distortion as a simple, local problem. Some approaches assume the blur looks like a smooth, single hill, while others try to fix the image by subtracting small, repeated patches of the distortion. These methods work well enough for some tasks, but they struggle when the distortion is complex, with sharp peaks and deep valleys that stretch far from the center. In a new study, researchers Guoshengyi Wu and Li Zhang from Guizhou University proposed a different way to map this distortion. They treated the telescope's error pattern not as a simple shape to be guessed, but as a complex landscape that could be rebuilt from scratch using a specific mathematical tool. Their goal was to create a single, continuous equation that could describe every ripple and dip in the telescope's distortion, allowing for a much more accurate reconstruction of the black hole's true image.

The team focused on the data from the first-ever image of the supermassive black hole in the galaxy M87, captured in 2017. They took the raw data representing the telescope's distortion—a grid of pixels showing where the signal was strong and where it was weak—and applied a technique called linear radial basis function modeling. Imagine trying to recreate a complex, bumpy terrain by placing a set of pegs at specific points on a map and then stretching a flexible sheet over them. The researchers placed thousands of these "pegs" across the image, each one acting as a reference point. They then calculated how the distance between any new point on the map and these pegs influenced the height of the sheet. By combining the influence of all these pegs, they could generate a smooth, continuous surface that matched the original messy data with incredible accuracy. Unlike older methods that relied on guessing the shape of the blur or fitting it to a simple curve, this approach let the data itself dictate the shape, using the distance between points to build the model.

The results showed that this new method, which the authors call LiRBM, was far superior to the standard techniques used in the field. When the team compared their model against the traditional methods, the difference was stark. The old methods left behind large errors, with the model failing to capture the fine details of the ripples and the deep valleys of the distortion. In contrast, the new model matched the original data with a level of precision that was more than 88 percent better in terms of error reduction. It also preserved the structure of the image with a similarity score of nearly 0.99, meaning the mathematical model looked almost identical to the real data. Most importantly, the model did not just smooth over the details; it kept the sharp peaks and the specific positions of the ripples intact, which is crucial for knowing exactly how the telescope has distorted the light.

This success matters because the quality of the final black hole image depends entirely on how well scientists can remove the telescope's own signature from the data. If the model of the distortion is even slightly wrong, the process of cleaning the image can introduce new errors or erase real features of the black hole. By providing a highly accurate, continuous mathematical description of the distortion, this new method gives astronomers a better tool to restore the true image. It allows them to simulate how the telescope degrades an image and to test different ways of cleaning it up without guessing. The researchers found that their model worked best when it used a simple, linear relationship between distance and signal strength, avoiding the need for complex, adjustable knobs that often caused older models to become unstable.

The study does not claim to have solved every problem in black hole imaging, but it offers a significant step forward in how the data is handled. The team demonstrated that their method could be applied to the specific data from the 2017 M87 observations, creating a reproducible map of the telescope's behavior. While the model is currently tailored to this specific set of observations, the approach itself is flexible enough to be tested on other black hole images and different telescope configurations. The researchers acknowledge that future work will need to test the method on a wider variety of data to ensure it remains stable under different conditions. For now, however, they have shown that by treating the telescope's distortion as a complex landscape to be mapped rather than a simple shape to be guessed, it is possible to see the universe with much greater clarity. This new way of modeling the blur brings the hidden details of the black hole one step closer to being fully revealed.

Technical Summary: A Mathematical Modeling Method of the EHT Synthesized Beam for Radio Synthesis Imaging

Problem Statement
The Event Horizon Telescope (EHT) utilizes very-long-baseline interferometry (VLBI) to image black holes, a process heavily influenced by the synthesized beam, or Point Spread Function (PSF). Under sparse spatial frequency sampling, the imaging process is convolved with a complex synthesized beam characterized by a central mainlobe and intricate sidelobe structures. Accurate image restoration and error propagation analysis require a continuous functional representation of this PSF. While existing methods such as Högbom CLEAN, Clark CLEAN, and Asp-Clean utilize local approximations (e.g., finite beam patches or single Gaussian mainlobes) primarily for deconvolution and computational acceleration, they often fail to provide a high-precision, continuous mathematical model capable of capturing the full spatial morphology, including complex sidelobe oscillations and polarity. Previous PSF modeling approaches in other fields (e.g., adaptive optics, optical microscopy) have employed blind deconvolution, deep learning, or polynomial fitting, but these often struggle with the specific structural complexity and sparse sampling constraints inherent to EHT data.

Methodology
The paper proposes LiRBM (Linear RBF-based Beam Modeling), a high-precision modeling method based on Linear Radial Basis Functions (RBFs). The core methodology involves:

Data Sampling: Using sampled image pixels from a 128×128 pixel EHT synthesized beam (derived from 2017 M87* observations) as RBF centers. The study utilizes 4,096 center points selected via uniform sampling.
Model Construction: The model constructs a continuous function $f(x,y)$ by performing a weighted superposition of distance responses from these centers. The general form is $f(x,y) = \sum_{i=1}^{N} w_i \phi(r_i)$ , where $r_i$ is the Euclidean distance between a query point and the $i$ -th center, and $w_i$ are weight coefficients solved via a linear system.
Basis Function Selection: The authors evaluate six basis function types: Linear ( $\phi(r)=r$ ), Thin Plate ( $\phi(r)=r^2\ln(r)$ ), Gaussian, Multiquadric, Inverse, and Cubic. A key methodological choice is the use of the Linear basis function, which does not explicitly contain a shape parameter $\epsilon$ . This avoids the numerical stability issues and parameter sensitivity associated with functions like Gaussian or Multiquadric, where $\epsilon$ must be tuned to match the spatial scale. The method leverages the distance relationships between sample points to construct continuous functions, making it particularly suitable for spatial-data interpolation and continuous spatial-response fitting.
Evaluation Protocol: The model is evaluated using Mean Absolute Residual (MAR), Maximum Absolute Residual (MaxAR), Residual Standard Deviation ( $\sigma_e$ ), Root Mean Square Error (RMS), and Structural Similarity Index (SSIM). Evaluations are conducted on the full PSF, the mainlobe region, and the combined mainlobe and first sidelobe region.

Key Contributions

LiRBM Framework: The introduction of a continuous, functional representation of the EHT synthesized beam using Linear Radial Basis Functions, utilizing distance relationships to effectively represent the beam's spatial structure without requiring shape parameter tuning.
Comparative Analysis: A systematic comparison of six RBF basis functions, demonstrating that the Linear basis function outperforms others (including Thin Plate, Gaussian, and Cubic) in terms of error metrics and stability for this specific application.
Benchmarking Against Established Methods: A direct quantitative comparison of LiRBM against the approximation strategies used in Clark CLEAN (local beam patches) and Asp-Clean2016 (single Gaussian mainlobe), recalculating their performance on the same EHT PSF data to isolate modeling accuracy from deconvolution efficiency.
Reproducible Model Output: The provision of a complete set of 4,096 center coordinates and corresponding weight coefficients, allowing for the exact reproduction of the continuous PSF model for subsequent quantitative analysis.

Results

Global Performance: Among the six basis functions tested, the Linear basis function achieved the lowest error metrics: Mean Absolute Residual (0.010511), Maximum Absolute Residual (0.124657), and Residual Standard Deviation (0.015405). Compared to the second-best method (Thin Plate), LiRBM reduced the mean absolute residual by 40.40% and the residual standard deviation by 38.10%.
Comparison with General Methods: When compared to the approximation methods underlying Clark CLEAN and Asp-Clean2016, LiRBM demonstrated superior fidelity. LiRBM achieved an RMS of 0.015405 and an SSIM of 0.992112. In contrast, the Clark CLEAN approximation yielded an RMS of 0.138998 and SSIM of 0.303455, while Asp-Clean2016 yielded an RMS of 0.216112 and SSIM of 0.229640.
Local Structural Preservation:
- Mainlobe: In the mainlobe region, LiRBM achieved an RMS of 0.039695 and SSIM of 0.958808, successfully preserving the peak position, width, and contour.
- Mainlobe + First Sidelobe: In the region including the first sidelobe, the RMS decreased to 0.030138 and SSIM increased to 0.987305. The model successfully maintained the polarity, relative position, and oscillation sequence of the positive mainlobe and negative sidelobe without introducing polarity reversals or misalignments.
Residual Distribution: Residuals were found to be distributed around zero with no concentrated high-error regions, indicating uniform error control across the beam cross-section.

Significance and Claims
The paper claims that LiRBM establishes a higher-precision mathematical model for synthesized beams compared to existing local approximation or Gaussian-fitting methods. By providing a continuous functional representation, the method supports subsequent quantitative analysis, degradation simulation, and astronomical image restoration. The authors emphasize that the Linear basis function's lack of dependence on a tunable shape parameter $\epsilon$ contributes to its stability and ability to match the spatial structure of the EHT synthesized beam effectively. The study concludes that LiRBM can accurately characterize the spatial structure of EHT images, preserving critical morphological features such as the central mainlobe, sidelobe oscillations, and edge attenuation, which are often lost or approximated in standard deconvolution workflows. The authors note that the current model is specific to the experimental samples and parameter conditions used and suggest that future work should involve cross-validation with different beam structures and noise levels to further assess generalization.

More like this