Characterization of Evolved Massive Stars Using Machine Learning

Massive stars, namely stars born with more than eight times the mass of the Sun, are powerful cosmic engines that exert a strong impact to their surroundings via their large deposition of matter and radiation energy in form of winds, energetic outbursts, and supernova explosions. Our understanding, however, of the physics that govern their nature and evolution is limited due to their intrinsic rarity and thus, the lack of observational parameters and constraints. Moreover, due to the fact that all stars spend 95% of their lifetime at the core hydrogen burning phase (so called main sequence; see Fig. 1), a great −if not the largest− part of the current theoretical framework has been built on the physics of the early stellar stages. On the other hand, the evolved stages of massive stars are those, which are tightly linked to the most energetic phenomena documented in the literature of stellar astronomy (e.g. the luminous blue variables; Humphreys & Davidson 1994). Based on the observations, these phases are collectively characterized by enhanced mass losses, distinctive pulsations (e.g. Lovekin et al. 2014, van Genderen et al. 2019), and high atmospheric instability (e.g. the yellow hypergiants; de Jager 1998) that powers episodic ejections of material (e.g. Lobel et al. 2003, Smith et al. 2014) and leads to the formation of dust/gas structures surrounding the stars, such as shells, disks, and extended nebulae (e.g. Kourniotis et al. 2018, 2022, Kraus et al. 2023). In Fig. 1, we illustrate the diverse types of massive stars on the temperature-luminosity (Hertzsprung-Russell) diagram of stellar evolution.

The main goal of the project is to recover and characterize evolved, and thus rare, massive stars that are contained in the modern astronomical databases. The PhD candidate will develop machine-learning algorithms for the classification of the stars using three fundamental tools; spectroscopy, time-domain photometry, and spectral energy distributions. Classifiers will be trained on the features of known evolved stars and will be applied on environments that are expected to host massive stars, such as Local Group galaxies with evident star formation at a range of metallicities. In addition, machine-learning algorithms will be developed for the regression analysis of physical parameters such as the temperature and chemical abundances, as well as parameters of the stellar wind, mass loss, and circumstellar dust. The latter methods will be trained with synthetic data built using the state-of-the-art theoretical codes. For the purpose of validating a stellar classification, the student will also have the chance to work with telescope data from international observatories and from the Perek 2m telescope, in Czech Republic, which are and will be obtained by colleagues of our group.

Literature:

[1] de Jager, C. 1998, A&A Rev., 8, 145
[2] Ekström, S. et al. 2012, A&A, 537, A146
[3] Humphreys, R. M. & Davidson, K. 1994, PASP, 106, 1025
[4] Kourniotis, M. et al. 2018, MNRAS, 480, 3706
[5] Kourniotis, M. et al. 2022, MNRAS, 511, 4360
[6] Kraus, M. et al. 2023, Galaxies, 11, 76
[7] Lobel, A. et al. 2003, ApJ, 583, 923
[8] Lovekin, C. C. & Guzik, J. A. 2014, MNRAS, 445, 1766
[9] Smith, N. et al. 2011, MNRAS, 415, 773
[10]van Genderen, A. M. et al. 2019, A&A, 631, A48