The focus of my research is M dwarfs. These low-mass low-luminosity stars are the most abundant stellar type in our Galaxy and in the recent years they gained increasing popularity among
exoplanet searches. A precise determination of stellar parameters is therefore crucial to constrain planetary physical parameters. However, due to their faintness and low temperatures, M
dwarfs pose a great challenge for modern parameter determination techniques. Regardless of considerable improvements over the last decades, e.g., in stellar atmosphere modeling, different methods
still arrive at significantly different results for the very same stars (see left plot and Passegger et al. 2022).
During my work with CARMENES, I developed an algorithm to fit state-of-the-art PHOENIX model spectra to high-resolution M-dwarf spectra to derive fundamental stellar parameters, such as effective temperature, surface gravity, and metallicity. Special care has to be taken on which spectral lines and regions to use for fitting, since
M dwarfs can show fairly high magnetic activity, influencing spectral line shapes and, therefore, the derived stellar parameters. Additionally, it has been shown that for hotter stars (and recently also for cool stars) the neglect of non-local thermodynamic equilibrium (NLTE) in modelling stellar atmospheres can result in significant differences in the calculated spectra and derived stellar parameters. Therefore, I'm also investigating the effects of NLTE on the determination of M-dwarf stellar parameters.
Artificial Intelligence and Machine Learning have become widely used in recent years in many aspects of our every-day lives, such as virtual assistants, image recognition, fraud detection, medical image analysis, photo descriptions, to name a few. Applications in astrophysics range from galaxy classification over characterization of variable stars to determination of fundamental stellar parameters. The right figure presents the basic structure of a deep learning neural network we developed in collaboration with computer scientists (see Passegger et al. 2020 and Bello-García, Passegger et al. 2023). We show that deep learning and deep transfer learning can help to significantly improve the precision and accuracy of M-dwarf stellar parameters compared to other techniques. This work also illustrates that there is still a lot of work to be done in improving stellar atmosphere models, as they are still unable to perfectly resemble real stars and their spectra.
Transit and radial-velocity observations are the most successful techniques to search for extrasolar planets, detecting over 3800 and 1000 exoplanets to date, respectively (see The Extrasolar Planets Encyclopaedia). As mentioned before, a precise stellar parameter determination is key for properly characterizing orbiting planets. During my work in the CARMENES consortium, I derived stellar parameters for several planet-hosting M dwarfs and also collaborated in characterizing some of the planets detected by CARMENES. An example is the Saturn-mass planet around TYC 2187-512-1 (Quirrenbach, Passegger et al. 2022).
During my work with ESPRESSO, I further constrained the parameters of the three-planet system GJ 9827 (Passegger et al. 2024). To do this, I used archival data from the spectrographs HARPS, HARPS-N, and HIRES, and recent data from ESPRESSO. I combined these with photometric transit measurements from the satellites K2 and TESS and used a Gaussian Process procedure to model stellar activity and the planetary orbits and physical parameters.
During my work, I conducted several observing runs on small and large telescopes around the world.