Extraction of Physical Parameters of RRab Variables using Neural Network based Interpolator

TL;DR

This paper introduces a neural network approach to directly infer physical parameters of RRab variable stars from TESS light curves, enabling efficient analysis of stellar properties and relations.

Contribution

It presents a novel neural network framework trained on synthetic models to accurately derive stellar parameters from light curves, extending the method to real TESS data.

Findings

Neural network accurately recovers input parameters from synthetic data.

Derived an empirical period-luminosity-metallicity relation for RRab stars.

Demonstrated the method's applicability to real TESS observations.

Abstract

Determining the physical parameters of pulsating variable stars such as RR Lyrae is essential for understanding their internal structure, pulsation mechanisms, and evolutionary state. In this study, we present a machine learning framework that uses feedforward artificial neural networks (ANNs) to infer stellar parameters-mass ($M$), luminosity (log($L/L_\odot$)), effective temperature (log($T_{\rm eff}$)), and metallicity ($Z$)-directly from Transiting Exoplanet Survey Satellite (TESS) light curves. The network is trained on a synthetic grid of RRab light curves generated from hydrodynamical pulsation models spanning a broad range of physical parameters. We validate the model using synthetic self-inversion tests and demonstrate that the ANN accurately recovers the input parameters with minimal bias. We then apply the trained model to RRab stars observed by the TESS. The observed light curves are phase-folded, corrected for extinction, and passed through the ANN to derive physical parameters. Based on these results, we construct an empirical period-luminosity-metallicity (PLZ) relation: log($L/L_\odot$) = (1.458 $\pm$ 0.028) log($P$/days) + (-0.068 $\pm$ 0.007) [Fe/H] + (2.040 $\pm$ 0.007). This work shows that ANN-based light-curve inversion offers an alternative method for extracting stellar parameters from single-band photometry. The approach can be extended to other classes of pulsators such as Cepheids and Miras.