Identification of high-energy astrophysical point sources via hierarchical Bayesian nonparametric clustering

Andrea Sottosanti, Mauro Bernardi, Alessandra Rosalba Brazzale, Alex Geringer-Sameth, David C. Stenning, Roberto Trotta, David A. van DykarXiv 2021 · DOI · arXiv

Abstract

The light we receive from distant astrophysical objects carries information about their origins and the physical mechanisms that power them. The study of these signals, however, is complicated by the fact that observations are often a mixture of the light emitted by multiple localized sources situated in a spatially-varying background. A general algorithm to achieve robust and accurate source identification in this case remains an open question in astrophysics. This paper focuses on high-energy light (such as X-rays and gamma-rays), for which observatories can detect individual photons (quanta of light), measuring their incoming direction, arrival time, and energy. Our proposed Bayesian methodology uses both the spatial and energy information to identify point sources, that is, separate them from the spatially-varying background, to estimate their number, and to compute the posterior probabilities that each photon originated from each identified source. This is accomplished via a Dirichlet process mixture while the background is simultaneously reconstructed via a flexible Bayesian nonparametric model based on B-splines. Our proposed method is validated with a suite of simulation studies and illustrated with an application to a complex region of the sky observed by the Fermi Gamma-ray Space Telescope.

Introduction

In high-energy astronomy, telescopes like the Fermi Gamma-ray Space Telescope do not capture traditional images. Instead, they record an "event list" of individual photons, each characterized by its sky coordinates, arrival time, and energy. The primary challenge for researchers is "source extraction": distinguishing which photons originated from a specific point source (such as a pulsar or a distant galaxy) and which are part of the diffuse, spatially-varying background.

A simulated high-energy sky region showing point sources embedded in a complex, spatially-varying background.

Traditional methods often struggle with unknown source counts, complex backgrounds that vary due to interstellar dust, and faint signals that are easily lost in noise. Many existing tools rely on rigid physical templates for the background; if these templates are slightly inaccurate, they can either obscure real sources or generate spurious "ghost" sources.

Methodology: A Bayesian Nonparametric Approach

To address these challenges, we propose a fully data-driven, Bayesian nonparametric framework. This approach avoids the biases of fixed templates by modeling both the sources and the background simultaneously.

Modeling Sources and Background

  • The Sources (Dirichlet Process Mixture): Rather than fixing the number of sources a priori, we utilize a Dirichlet Process (DP) prior. This allows the model to infer the number of clusters (sources) directly from the data density. Each source is modeled according to the instrument's Point Spread Function (PSF).
  • The Background (B-splines + DP): To accommodate non-uniform backgrounds, we employ a flexible mixture of B-splines. This allows the background model to adapt to complex spatial structures without requiring a pre-existing physical map of the galaxy.

Joint Spatial-Spectral Modeling

While spatial coordinates are the primary data, incorporating photon energy significantly enhances detection capabilities. Astrophysical sources typically follow a power-law energy distribution. By modeling this spectral signature (using a Pareto distribution), we can better distinguish faint sources from the background. Crucially, the instrument's PSF is energy-dependent, typically becoming narrower at higher energies, which provides additional discriminatory power.

Photon distribution across energy bins, demonstrating the tightening of the Point Spread Function (PSF) at higher energies.

Results and Validation

The methodology was validated through extensive simulations and an application to real data from the Fermi telescope.

Simulation Studies

In simulations of 10° × 10° sky regions, the Joint Spatial-Spectral Model significantly outperformed spatial-only versions. In scenarios where sources were too faint to be identified by position alone, their distinct energy spectra allowed the model to recover 12 out of 15 sources, compared to only 10 for the spatial-only model.

Application to Antlia 2

We applied the model to a 9.4-year dataset of the region surrounding the Antlia 2 dwarf galaxy. This region is particularly challenging due to its proximity to the Galactic plane and high background gradients. Our model successfully recovered 9 known sources from the official 4FGL catalogue and identified 4 additional potential point source candidates. This suggests that standard physical background models may sometimes subsume real, faint point sources into the background noise.

Conclusion

By combining Dirichlet Process mixtures with flexible B-spline modeling, this framework provides a robust, data-driven solution for source identification in high-energy astrophysics. The inclusion of spectral information and the ability to provide full posterior probabilities for photon origin represent significant steps forward in uncertainty quantification for the field. Future work will focus on extending the model to handle diverse spectral indices across different sources and accounting for energy-dependent instrumental sensitivity.

Summary updated

Abstract

The light we receive from distant astrophysical objects carries information about their origins and the physical mechanisms that power them. The study of these signals, however, is complicated by the fact that observations are often a mixture of the light emitted by multiple localized sources situated in a spatially-varying background. A general algorithm to achieve robust and accurate source identification in this case remains an open question in astrophysics. This paper focuses on high-energy light (such as X-rays and gamma-rays), for which observatories can detect individual photons (quanta of light), measuring their incoming direction, arrival time, and energy. Our proposed Bayesian methodology uses both the spatial and energy information to identify point sources, that is, separate them from the spatially-varying background, to estimate their number, and to compute the posterior probabilities that each photon originated from each identified source. This is accomplished via a Dirichlet process mixture while the background is simultaneously reconstructed via a flexible Bayesian nonparametric model based on B-splines. Our proposed method is validated with a suite of simulation studies and illustrated with an application to a complex region of the sky observed by the \emph{Fermi} Gamma-ray Space Telescope.
BibTeX
@misc{sottosanti2021identification,
  author = {Andrea Sottosanti and Mauro Bernardi and Alessandra Rosalba Brazzale and Alex Geringer-Sameth and David C. Stenning and Roberto Trotta and David A. van Dyk},
  title = {{Identification of high-energy astrophysical point sources via hierarchical Bayesian nonparametric clustering}},
  year = {2021},
  eprint = {2104.11492},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2104.11492},
}
Figure