TESS is poised to increase the number of stellar rotation period estimates by an order of magnitude, but the mission’s systematics have complicated period searches. While several efforts attempt to solve this problem by removing systematics, standard methods of data reduction have shown limited success. Here I present a method to predict rotation periods from TESS full-frame image light curves using deep learning. This method relies on a training set of simulated light curves convolved with TESS galaxy light curves to emulate the instrumental noise and systematics observed in stellar signals. The simulations include surface differential rotation, spot evolution, and activity level to make the light curves as realistic as possible. Our approach allows the network to learn the difference between rotation signals and TESS systematics. With the added ability to predict uncertainty in the period, we can determine what regions of parameter space the predictions are most credible, producing a reliable set of rotation periods. I present the first set of rotation periods obtained with this method and explore TESS’s insights to stellar structure and evolution through the lens of rotation.

Poster, thumbnail image, and explanatory video for the poster to be presented at TESS Science Conference II. View arXiv preprint at https://arxiv.org/abs/2104.14566. Find butterpy Python package at https://github.com/zclaytor/butterpy.

{"references": ["https://arxiv.org/abs/2104.14566"]}