hide
Free keywords:
As the number of exoplanets discovered by space missions, such as NASA’s Kepler and TESS, has been increasing rapidly during the past decades, the need for an efficient approach to determining the properties of the exoplanet hosts as well as the exoplanets is urgent. In this work, we developed a neural network approach based on Gaussian mixture models to directly determine stellar properties and corresponding uncertainties for exoplanet host stars. Over 16 million stellar models with different initial parameters generated by the Aarhus STellar Evolution Code were used to train the neural networks. To validate the performance of our neural networks, we compared the neural network predicted stellar parameters with results from other methods, e.g., scaling relations, machine learning, and grid-based modeling, for 47 main-sequence, subgiant, and red-giant stars that have exoplanets orbiting around them. The fractional differences between our neural networks and other methods for masses, radii, and ages are generally small, though systematic deviations for masses and ages are seen during comparisons due to the different input physics and seismic parameters used in the modeling. The average uncertainties predicted by our networks for these stars are of the order of 2.4%, 1.1%, and 17.4% for mass, radius, and age, respectively. Overall, our neural network method is able to determine stellar parameters accurately and precisely in a short amount of time, indicating a promising future for its application in the study of exoplanets and their host stars.
Abstract:
Asteroseismology — techniquesNeural network — starsOscillations — starsFundamental parameters
