Exploring TeV candidates of Fermi blazars through machine learning - Logistic Regression (LR) selection and TeV candidate identification

Aug 5, 2026

Upload by Patrick

Supervised machine learning is applied using a Logistic Regression (LR) model to select TeV blazar candidates from multiple Fermi catalogs (4FGL-DR2/4LAC-DR2, 3FHL, 3HSP, 2BIGB). LR learns a feature hyperplane from optimized parameters and outputs a logistic probability that a source is a TeV candidate. Sources with logistics above 80% among non-TeV blazars are treated as high-confidence TeV candidates. The method identifies 40 high-confidence candidates and builds broadband SEDs to test detectability with instruments, finding 7 likely detectable by existing/upcoming IACT observatories and 1 via EAS particle detector arrays.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages24

Reading Time1 h 0 min

AccessFree to download

Abstract

1. Introduction

Blazars and observational properties

SED modeling and emission mechanisms

BL Lac subclasses and classification schemes

arXiv :2303 . 10557v1 [ astro-ph .HE] 19 Mar 2023 DRAFT VERSION MARCH 21, 2023 Typeset using LATEX default style in AASTeX631 Exploring TeV candidates ofFermi blazars through machine learning J. T. ZHU , 1, 2, 3, 4 C. LIN , 1, 3, 4 H. B. XIAO ,5 J. H. FAN , 1, 3, 4 D. BASTIERI , 1, 2 AND G. G. WANG1, 3, 4 1 Center for Astrophysics, Guangzhou University, Guangzhou 510006, China 2 Department of Physics and Astronomy “G. Galilei”, University of Padova, Padova I-35131, Italy 3 Astronomy Science and Technology Research Laboratory of Department of Education of Guangdong Province, Guangzhou 510006, China 4 Key Laboratory for Astronomical Observation and Technology of Guangzhou, Guangzhou 510006, China 5 Shanghai Key Lab for Astrophysics, Shanghai Normal University Shanghai, 200234, China (Received XXX; Revised YYY; Accepted ZZZ) ABSTRACT In this work, we make use of a supervised machine learning algorithm based on Logistic Regression (LR) to select TeV blazar candidates from the 4FGL-DR2 / 4LAC-DR2, 3FHL, 3HSP, and 2BIGB catalogs. LR constructs a hyperplane based on a selection of optimal parameters, named features, and hyper-parameters whose values control the learning process and determine the values of features that a learning algorithm ends up learning, to discriminate TeV blazars from non-TeV blazars. In addition, it gives the probability (or logistic) that a source may be considered as a TeV blazar candidate. Non-TeV blazars with logistics greater than 80% are considered high-conﬁdence TeV candidates. Using this technique, we identify 40 high-conﬁdence TeV candidates from the 4FGL-DR2 / 4LAC-DR2 blazars and we build the feature hyper-plane to distinguish TeV and non-TeV blazars. We also calculate the hyper-planes for the 3FHL, 3HSP, and 2BIGB. Finally, we construct the broadband spectral energy distributions (SED) for the 40 candidates, testing for their detectability with various instruments. We ﬁnd that 7 of them are likely to be detected by existing or upcoming IACT observatories, while 1 could be observed with EAS particle detector arrays. Keywords: AGN; galaxies-active; galaxies-Quasars; galaxies-BL Lacertae objects; data analysis; machine learning 1. INTRODUCTION Blazars, an extreme subclass of active galactic nuclei (AGNs), are known for prominent observation properties, such as high energy 􀀍-ray emissions, rapid and signiﬁcant amplitude variability, high luminosity, high and variable polarization, and superluminal motions, etc. (Wills et al. 1992 ; Urry & Padovani 1995 ; Fan 2002 ; Villata et al. 2006 ; Fan et al. 2014 ; Xiao et al. 2015 ; Gupta et al. 2016 ; Xiao et al. 2019 ; Abdollahi et al. 2020 ; Xiao et al. 2020 ; Fan et al. 2021) . Blazars are historically subdivided into two main categories based on the equivalent width (EW) of the optical emission lines: ﬂat spectrum radio quasars (FSRQs) 􀀗 and BL Lacertae objects (BL Lacs) . In general, FSRQs show an EW greater than 5 A, while BL Lacs illustrate no or weak emis- 􀀗 sion lines, with EW less than 5 A. Meanwhile, the spectral energy distributions (SEDs) of blazars are generally characterized by two well-separated bumps, a low-energy one is in the infrared to soft X-ray energy range due to synchrotron emission, and a high-energy one is in the region between hard X-ray to 􀀍-ray that is associated with an inverse Compton (IC) radiation according to leptonic model (IC; e.g., Sikora et al. 1994a) . The seed photons undergoing IC scattering could be from the same electron population producing the synchrotron bump in the so-called self-Compton (SSC) model (Ghisellini et al. 1985 ; Maraschi et al. 1992 ; Bloom & Marscher 1996), e.g. 1ES 0347-121, 1ES 0229+200 (Costamante et al. 2018 ; Aharonian et al. 2007a,b), or from external regions (External Compton model, EC), e.g., from the accretion disk (Dermer & Schlickeiser 1993), broad line region (Sikora et al. 1994b), and dust torus (Bła˙zejowski et al. 2000) . While the hadronic proces

FAQs

What machine learning approach is used to find TeV blazar candidates?

A supervised Logistic Regression (LR) model constructs a feature hyperplane to discriminate TeV from non-TeV blazars and estimates a logistic probability for each source.

How is the “high-confidence” TeV candidate threshold defined?

Non-TeV blazars with logistic values greater than 80% are considered high-confidence TeV candidates.

How many high-confidence TeV candidates are identified, and how are they evaluated?

The technique identifies 40 high-conﬁdence TeV candidates from the 4FGL-DR2/4LAC-DR2 blazars. Broadband SEDs are then built to test detectability with different instruments.

Aug 5, 2026

Upload by Patrick