Latent-space Field Tension for Astrophysical Component Detection An application to X-ray imaging

TL;DR

The paper tackles the challenge of disentangling diffuse, point-like, and extended astrophysical emission in multi-frequency X-ray imaging by introducing latent-space field tension as a diagnostic for model misspecification within a fully Bayesian, information-field-theory framework. It develops a multi-component, multi-frequency sky model with principled priors for diffuse, point-like, and extended structures, and performs posterior inference via geometry-aware variational inference (geoVI) to enable scalable, uncertainty-quantified reconstructions that are instrument-agnostic. The authors demonstrate the approach on synthetic tests and apply it to real SRG/eROSITA observations of SN1987A in the Large Magellanic Cloud, achieving sub-pixel localization of point sources, robust separation of extended emission, and detailed spectral-spatial reconstructions, including explicit modeling of extended sources like the Tarantula Nebula and 30 Doradus C. The method’s generality, scalability, and ability to provide uncertainty quantification make it well suited for upcoming multi-wavelength and multi-messenger surveys, with public release of the J-UBIK-based tools anticipated.

Abstract

Modern observatories are designed to deliver increasingly detailed views of astrophysical signals. To fully realize the potential of these observations, principled data-analysis methods are required to effectively separate and reconstruct the underlying astrophysical components from data corrupted by noise and instrumental effects. In this work, we introduce a novel multi-frequency Bayesian model of the sky emission field that leverages latent-space tension as an indicator of model misspecification, enabling automated separation of diffuse, point-like, and extended astrophysical emission components across wavelength bands. Deviations from latent-space prior expectations are used as diagnostics for model misspecification, thus systematically guiding the introduction of new sky components, such as point-like and extended sources. We demonstrate the effectiveness of this method on synthetic multi-frequency imaging data and apply it to observational X-ray data from the eROSITA Early Data Release (EDR) of the SN1987A region in the Large Magellanic Cloud (LMC). Our results highlight the method's capability to reconstruct astrophysical components with high accuracy, achieving sub-pixel localization of point sources, robust separation of extended emission, and detailed uncertainty quantification. The developed methodology offers a general and well-founded framework applicable to a wide variety of astronomical datasets, and is therefore well suited to support the analysis needs of next-generation multi-wavelength and multi-messenger surveys.

Figures (19)

• Figure 1: 1D example of the automatic component modeling method. The top panel displays the background $b$ (in blue) and the total signal $s$ (in orange) which is the superposition of the background $b$ and the foreground line $f$ signal components. The bottom panel shows the signal response (solid green line), obtained by convolving the total signal $s$ with a Gaussian PSF. Additionally, it displays a synthetic data realization of the signal response (in red), along with the $2 \sigma$ shot noise level contours associated with the background (dotted green lines). In both panels the vertical dashed-dotted gray lines represent the ground truth line component $f$ locations. We note that a large fraction of the line signal $f$ is buried beneath the background noise level.
• Figure 2: Background-only reconstruction. The top panel displays the posterior mean of the background-only reconstruction (solid blue line) in comparison to the ground truth background signal response (dashed black line). The bottom panel shows corresponding latent position-space excitations $\xi$ (purple solid line). The detection level for the line component was set to $\sigma_\text{thresh} = 0.6$ in prior units and it is displayed in red (solid horizontal line). The detected line positions for the given threshold $\sigma_\text{thresh}$ are displayed as dashed vertical gray lines in both panels.
• Figure 3: Results of one-dimensional automatic component modeling method. The top panel displays the posterior mean of the background component $b$ (solid blue line) in comparison to the ground truth background signal (dashed black line). The central panel depicts the posterior mean of the line component $f$ (solid orange line) in comparison to the ground truth line signal (dashed black line). The bottom panel shows the posterior mean of the total signal $s$ (solid green line) in comparison to the ground truth signal (dashed black line). The data is also shown in red. Shaded regions denote the $\pm 2\sigma$ posterior contours for each component in the respective panels.
• Figure 4: Setup of the synthetic component separation problem for imaging. The top row displays a realization of the diffuse emission component (left), the point-source component (middle), and the convolved point-source component (right), which has undergone Gaussian PSF convolution and multiplication by a spatially constant exposure of $80000s$. The bottom row presents the corresponding sky emission, obtained as the sum of the diffuse and point-source components (left), the convolved and exposure-multiplied sky emission (middle), and data generated by a Poisson realization of the observed simulated sky (right).
• Figure 5: Interpolation of the point-source field flux onto a regular pixel grid. The left panel shows the interpolation result for two, unit-flux point sources located at $\qty(0.5, 0.5)$ (center of the pixel) and at $\qty(2.75, 2.6)$ on a $4\times4$ pixel grid. The location of the point sources is marked with a red cross. The right panel displays a random realization of $1000$ uniformly distributed point sources on a $128\times128$ pixel grid with flux values uniformly distributed between $0$ and $100$. A white rectangle in the lower-left corner indicates a region of the same size as that shown in the left panel, providing a spatial scale reference.