Sensitivity of CTAO to axion-like particles from blazars: a machine learning approach

TL;DR

This study evaluates CTAO's capability to constrain axion-like particles (ALPs) via ALP-photon mixing signatures in blazar gamma-ray spectra. It compares a novel machine-learning (ML) classifier approach, implemented with XGBoost, to the standard likelihood-ratio test (LRT) by applying both to simulated CTAO observations of Mrk 501 and PKS 2155$-$304 in baseline and flare states, across a grid of $(m_a,g_{aγ})$ parameters. The ML method learns to distinguish ALP-induced spectral distortions (wiggles and high-energy hardening) from ALP-free spectra using excess-count features per energy bin, yielding 2σ exclusion regions in good agreement with the LRT and demonstrating CTAO's potential to extend existing ALP constraints in a wide mass range. These results support the viability of ML-based approaches for future gamma-ray ALP searches and highlight CTAO’s role in probing fundamental physics through high-energy astrophysical observations, while noting the need for systematic tuning and consideration of modeling uncertainties.

Abstract

Blazars are a class of active galactic nuclei, supermassive black holes located at the centres of distant galaxies characterised by strong emission across the entire electromagnetic spectrum, from radio waves to gamma rays. Their relativistic jets, closely aligned to the line of sight from Earth, are a rich and complex environment, characterised by the presence of strong magnetic fields over parsec-scale lengths. Owing to their cosmological distance from Earth, these sources serve as ideal targets to probe non-standard gamma-ray propagation. In particular, axion-like particles (ALPs) could be detected through their coupling to photons, which enables ALP-photon conversions in external magnetic fields, leading to distinct signatures in the blazars' gamma-ray spectra. In this work, we explore a novel approach to constrain the ALP parameter space using gamma-ray observations, based on the use of machine-learning classifiers. We apply this technique to simulated observations of two bright blazars -- Mrk 501 and PKS 2155$-$304 -- with the Cherenkov Telescope Array Observatory (CTAO), a next-generation gamma-ray facility well suited to probe such features, thanks to its improved energy resolution and point-source sensitivity with respect to present ground-based gamma-ray telescopes. The obtained $2σ$ exclusion regions on the ALP parameter space are consistent with those found by applying a standard likelihood-ratio test, and suggest that the CTAO sensitivity to ALPs could be extended beyond existing constraints over a wide mass range.

Figures (10)

• Figure 1: ALP effects on the baseline and flaring spectral shapes chosen for the sources in this work. For large enough values of $g_{a\gamma}$, both spectral distortions ("wiggles") and a hardening at high energies are visible.
• Figure 2: Example simulated SEDs for two 50-hour observations of the Mrk 501 baseline state. Spectral fits are shown for the ALP-less case and for two different ALP parameter sets. The residuals for both cases are computed as $(f-\phi)/\sigma_f$, where $f$ is the measured flux with uncertainty $\sigma_f$ and $\phi(E)$ is the model prediction (see ref. abdalla2021).
• Figure 3: Example likelihood-ratio TS distribution obtained for the baseline state of Mrk 501.
• Figure 4: XGB metrics for the baseline state of Mrk 501 over the ALP parameter space.
• Figure 5: Example distributions of the $\Pi$ statistic defined in eq. \ref{['eq:ts-ml']}, comparing good and suboptimal performance of a classifier algorithm in different regions of the ALP parameter space.