Ultracool Dwarf Atmospheres Reveal Complex Molecular Signatures, Challenging Characterisation

Low-mass stars and substellar objects, though faint and difficult to characterise, are incredibly common throughout our galaxy and crucial for understanding its structure and the potential for Earth-like planets. Pedro Mas Buitrago, Enrique Solano Márquez, and Ana González Marcos, all from Universidad Complutense de Madrid, investigate these ‘ultracool dwarfs’ by combining the power of large astronomical databases, known as the Virtual Observatory, with modern machine learning techniques. This research addresses a long-standing challenge in astronomy, namely accurately identifying and analysing these objects due to their complex atmospheres and faintness. By developing flexible, automated methodologies, the team aims to unlock new insights into the boundary between stars and planets and prepare for the wealth of data expected from next-generation surveys like Euclid and LSST, ultimately advancing our understanding of the Milky Way’s stellar population.

The team meticulously selected targets from the J-PLUS survey, a large-scale photometric study, and conducted follow-up spectroscopic observations to confirm their nature and characterise their properties. These observations provide detailed information about the stars’ composition, temperature, and velocity, allowing for a comprehensive analysis of this stellar population. The research involved careful data processing and quality control to ensure the reliability of the results. The team compiled a detailed catalogue of observed stars, including information about their brightness, colour, and spectral features, serving as a valuable resource for the astronomical community. This catalogue enables further studies of these faint and elusive objects, and the research involved comparing observed data with existing catalogues and theoretical models to validate findings and identify discrepancies, refining our understanding of stellar evolution and the properties of low-mass stars.

Machine Learning Reveals Ultracool Dwarf Properties

This research focuses on low-mass stars and ultracool dwarfs, crucial for understanding the structure of the Milky Way and the search for Earth-like planets. Despite comprising 75% of stars within 10 parsecs of the Sun, their faintness and complex atmospheres have historically made detailed characterisation difficult. This work addresses these challenges through innovative data analysis techniques and the application of machine learning, leveraging large datasets and the Virtual Observatory to facilitate data access and interoperability. Researchers developed methods to identify and characterise ultracool dwarf candidates using multi-filter photometry, enriching the known census of these faint objects.

Furthermore, the team pioneered the use of deep transfer learning, a powerful machine learning technique, to determine atmospheric stellar parameters of M dwarfs from high-resolution spectra, demonstrating significant improvements in accuracy and efficiency compared to traditional methods. Importantly, this technique was successfully adapted to analyse low-resolution spectroscopic data, extending its applicability to the ultracool dwarf domain. The work details the development and application of several techniques, including machine learning algorithms for identifying these objects and analysing their characteristics, as well as methods for detecting flaring events in M dwarfs. Furthermore, the research explores the use of autoencoders and deep transfer learning to improve the estimation of stellar parameters, demonstrating the potential of these advanced techniques for analysing complex astronomical data. The application of these methods to datasets such as those from CARMENES and SpeX has enabled a more detailed characterisation of both M dwarfs and ultracool dwarfs, contributing to a better understanding of their properties and prevalence in the galaxy. The results demonstrate the effectiveness of combining machine learning with established astronomical data sources to address challenges in stellar characterisation and identify potentially interesting objects for further study. Future work includes expanding the application of these methods to larger datasets and exploring new machine learning architectures to further enhance the characterisation of low-mass stars and ultracool dwarfs, ultimately contributing to a more complete census of these objects in our galaxy.