ROSE: a neural network characterises the oscillations of 56,000 red giants observed by TESS, at a rate of 8 ms per star

Trained solely on synthetic spectra, the pipeline processed 274,721 red giant candidates and identified the ‘red clump’ without ever having been provided with evolutionary labels.

The Kepler and TESS space telescopes have recorded the oscillations of hundreds of thousands of red giants, and PLATO and the Roman telescope will add to this data. A team including Nipun Ghanghas and Rafael A. García presents ROSE, a modular machine learning framework designed to automate the asteroseismology of these stars. Its first two modules estimate two key quantities: the frequency of the maximum oscillation power and the large frequency separation, which provide information about the mass and radius of the star.

Uncertainties rather than figures

Each module is trained on simulated spectra and returns a probability distribution, with a reliability index, instead of a single value. On the giants of the TESS continuous observation area, the dispersion remains 4.4% and 1.3% depending on the measured quantity.

Applied to 274,721 candidates observed over at least ten sectors, the pipeline provided reliable measurements for 56,224 stars, in 8 milliseconds each. Notably, the helium burning phase at the core ("red clump") appears on its own as a distinct overdensity, while no module has been informed of the evolutionary stage of the stars.