Gaussian Process Modelling of Asteroseismic Data

The measured properties of stellar oscillations can provide powerful constraints on the internal structure and composition of stars. To begin this process, oscillation frequencies must be extracted from the observational data, typically time series of the star’s brightness or radial velocity. In this paper, a probabilistic model is introduced for inferring the frequencies and amplitudes of stellar oscillation modes from data, assuming that there is some periodic character to the oscillations, but that they may not be exactly sinusoidal. Effectively we fit damped oscillations to the time series, and hence the mode lifetime is also recovered. While this approach is computationally demanding for large time series (> 1500 points), it should at least allow improved analysis of observations of solar-like oscillations in subgiant and red giant stars, as well as sparse observations of semiregular stars, where the number of points in the time series is often low. The method is demonstrated on simulated data and then applied to radial velocity measurements of the red giant star xi Hydrae, yielding a mode lifetime between 0.41 and 2.65 days with 95% posterior probability. The large frequency separation between modes is ambiguous, however we argue that the most plausible value is 6.3 microHz, based on the radial velocity data and the star’s position in the HR diagram.

💡 Research Summary

The paper introduces a Bayesian framework based on Gaussian Processes (GP) to infer the frequencies, amplitudes, and lifetimes of stellar oscillation modes from time‑series observations of brightness or radial velocity. Traditional techniques such as Fourier transforms or Lomb‑Scargle periodograms assume strictly sinusoidal, equally spaced modes and therefore struggle with the non‑sinusoidal, damped nature of solar‑like oscillations, especially in subgiants, red giants, and semiregular variables where data are sparse.
The authors model each mode as a damped harmonic oscillator and embed this physical description directly into the GP covariance kernel:

(k(t_i,t_j)=\sigma^{2}\exp!\big