Modeling Blazar Broadband Emission with a Convolutional Neural Network. I. Synchrotron Self-Compton Model

Abstract

Modeling the multiwavelength spectral energy distributions (SEDs) of blazars provides key insights into the underlying physical processes responsible for the emission. While SED modeling with self-consistent models is computationally demanding, it is essential for a comprehensive understanding of these astrophysical objects. We introduce a novel, efficient method for modeling the SEDs of blazars by the mean of a convolutional neural network (CNN). In this paper, we trained the CNN on a leptonic model that incorporates synchrotron and inverse Compton emissions, as well as self-consistent electron cooling and pair creation–annihilation processes. The CNN is capable of reproducing the radiative signatures of blazars with high accuracy. This approach significantly reduces the computational time, thereby enabling real-time fitting to multiwavelength data sets. As a demonstration, we used the trained CNN with MultiNest to fit the broadband SEDs of Mrk 421 and 1ES 1959+650, successfully obtaining their parameter posterior distributions. This novel framework for fitting the SEDs of blazars will be further extended to incorporate more sophisticated models based on external Compton and hadronic scenarios, allowing for multimessenger constraints in the analysis. The models will be made publicly available via a web interface at the Markarian Multiwavelength Data Center to facilitate self-consistent modeling of multimessenger data from blazar observations.

Modeling Blazar Broadband Emission with Convolutional Neural Networks. II. External Compton Model

“…In this paper, we further develop the analysis technique presented in Bégué et al (2024) and consider the modeling of broadband SEDs of FSRQs. These SEDs are typically characterized by a pronounced Compton dominance, suggesting that, alongside synchrotron photons, an external photon field significantly contributes to the formation of the HE and VHE components.…”

“…As the emission region is assumed to be oriented at a small angle relative to the observer's frame, the emission is amplified by the relativistic Doppler factor, denoted as δ, which we assume to be such that δ = Γ. The electrons are injected into the emitting region as a power law with an exponential cutoff above the minimum Lorentz factor min g , which is the same distribution function we used for the SSC model (Bégué et al 2024). This injection function is given by: where m e denotes the electron mass.…”

“…Shortly after, a similar study using a different network approach was presented by Tzavellas et al (2024). Bégué et al (2024) is the first publication of the project titled "Modeling blazar broadband emission with convolutional neural networks." The primary goal of the project is to transform complex and computationally demanding leptohadronic radiative models into efficient neural network frameworks.…”

“…To address the significant computational challenges posed by traditional radiative models, we introduced an innovative approach for the self-consistent modeling of blazar SEDs (Bégué et al 2024). We employed a Convolutional Neural Network (CNN) trained on a leptonic SSC model that accounts for processes such as electron cooling and pair creationannihilation.…”

“…The simplicity of use and quick computation time also enable us to share the stand-in model directly with the blazar community through the Markarian Multiwavelength Data Center (MMDC). 7 The current paper seeks to build upon the work we initiated in Bégué et al (2024) by developing a CNN dedicated to an EIC model. Any EIC model presents additional challenges due to its extra parameters, which are essential for characterizing the external photon fields.…”