| Name: | Aswin Suresh |
| Affiliation: | Indian Institute of Technology Bombay |
| Conference ID : | ASI2024_700 |
| Title : | A Machine-learning Generated Catalog of Long Period Variables from the Palomar Gattini-IR |
| Authors : | Aswin Suresh1, Viraj Karambelkar2, Mansi Kasliwal2 |
| Authors Affiliation: | 1 Department of Physics, IIT Bombay, Powai, Mumbai 400076, India
2 Division of Physics, Mathematics and Astronomy, California Institute of Technology, Pasadena, CA 91125, USA
|
| Mode of Presentation: | Poster |
| Abstract Category : | Stars, Interstellar Medium, and Astrochemistry in Milky Way |
| Abstract : | Stars in the Asymptotic Giant Branch (AGB) phase, dominated by low to intermediate-mass stars in the late stage of evolution, undergo periodic pulsations, leading to the identification of Long Period Variables (LPVs) with periods extending to several hundred days. This phase also encompasses potential core-collapse supernova progenitors, such as massive red supergiants and super-AGB phase stars, exhibiting similar pulsational behaviour. AGB stars gradually shed their mass through stellar winds and mass ejections, enveloping themselves in dust and making them inaccessible to optical time-domain surveys.
Infrared surveys such as the Palomar Gattini-IR (PGIR) can probe these dust-enshrouded regions and uncover populations of AGB stars in the galactic bulge and plane. Gattini is a NIR J-band telescope with a 25 sq. deg. field of view, surveying the northern sky to a depth of 16 AB mag with a cadence of 2 days, rendering it an ideal instrument to study dusty and red sources. Gattini survey operations between September 2018 and April 2021 have produced high cadence lightcurves of more than 60 million stars.
We present a catalog of Long Period Variables from Gattini created using a machine learning approach. Using a comprehensive feature set extracted from Gaussian process interpolated lightcurves, capturing periodicity, variability, phasing and color, we train a gradient-boosted decision tree classifier on resampled, class-balanced data. This allows us to mitigate bias in the classifier and achieves a high precision of 99.9%. We validate the catalog by comparison with LPVs from Gaia and estimate the period-age relation of AGB stars in the near-Infrared regime. |
