Improving Gyrochronology: a new benchmark data set and age inference model

Gyrochronology uses a star’s rotation period and location on the main sequence (MS) to predict stellar age. It is particularly useful for low mass main sequence stars, the regime in which other stellar dating methods (e.g. isochrone fitting) tend to struggle. Gyrochronology has gained popularity in recent years due to the increasing availability of photometric data, but analytical models have struggled to coherently summarize the uncertainty and intrinsic variance in the photometric time series data. Thus, we have developed a generalized machine learning-based Bayesian inference framework that captures the uncertainty and variance through a probabilistic solution. Our framework implements a normalizing flow -- a neural network-based model that optimizes the transformation of parameter distributions -- to predict rotation period distributions for stars based on their age and colour. We have successfully trained and tested the model on data from eight open clusters with promising results, indicating that a data-driven approach to gyrochronology could improve upon existing models. We have now expanded and standardized our data sample to 30 open clusters; in this talk we will present the results of cluster age recovery testing using this new model.