New Rotation Periods from the Kepler Bonus Background Light Curves

TL;DR

This study uses a convolutional neural network to analyze Kepler background light curves, identifying over 9,800 new stellar rotation periods and validating deep learning as an effective method for period detection in astrophysical data.

Contribution

The paper introduces a deep learning approach to determine stellar rotation periods from de-blended Kepler background light curves, expanding the dataset with new period measurements.

Findings

Identified 9,811 new rotation periods for Kepler sources.

Validated deep learning method against literature periods with excellent agreement.

Found that up to 63% of background light curves are still blended with foreground sources.

Abstract

The Kepler field hosts the best studied sample of field star rotation periods. However, due to Kepler's large 4" pixels, many of its light curves are at high risk of contamination from background sources. The new Kepler Bonus Background light curves are de-blended using a PSF algorithm, providing light curves of over 400,000 new background sources in addition to over 200,000 re-analyzed Kepler prime targets. These light curves provide the opportunity to search for new rotation periods. Here we apply a convolutional neural network trained on synthetic spot-modulated light curves to regress rotation periods from the Kepler Bonus light curves. We obtained periods for 32,159 total sources, 19,650 of which had previously been measured and 9,811 of which are new periods for both Kepler prime and background sources. Our method also detected 608 pulsation frequencies from asteroseismic oscillations in red giants. We validate our Kepler prime periods against literature values and present the full period sample. We find excellent agreement with previously-known literature periods, validating deep learning as a viable class of period determination methods. Comparing the periods and light curves of foreground-background pairs, we find that as many as 63% of periodic background light curves are still blended with the foreground, highlighting limitations of the de-blending technique.