Data Driven Modeling of Red Giant Background Oscillation Parameters

Red giants are stars in the late stages of stellar evolution. Their internal structures and properties can be determined through asteroseismology, the study of stellar oscillations. Asteroseismic scaling relations, involving quantities such as the frequency of maximum power (νₘₐₓ) and the large frequency separation (Δν), are commonly used to estimate mass, age, and metallicity, but do not predict oscillation amplitudes or granulation power.

Using the model of Peralta et al. (2018) alongside the 16,000 pre-analysed red giants from Yu et al. (2018), we trained a neural network that predicts the corresponding Lomb–Scargle (LS) periodogram of a red giant, given its mass, age, temperature, and metallicity. Our initial implementation was an approximation that neglected inter-parameter correlations. To correct this, we performed Markov Chain Monte Carlo (MCMC) fitting to quantify the covariance matrix and uncertainty relations between model parameters. These results will inform an improved neural network architecture that properly considers parameter dependencies.

Later on, this model will allow us to generate synthetic integrated power spectra for unresolved globular clusters, providing a new method to infer their physical parameters through the collective contribution of their red giants.