DIAmante TESS AutoRegressive Planet Search (DTARPS): Overview and Methodology

The AutoRegressive Planet Search (ARPS) project develops a statistical methodology for transiting planet detection. It starts with ARIMA, the econometric modeling procedure for stochastic time series that fits both trends and autocorrelation. A novel periodogram, the Transit Comb Filter (TCF), is applied to the ARIMA residuals. TCF demonstrably has better statistical properties and higher sensitivity to smaller planets than the Box-Least Squares periodogram. A decision tree classifier, trained towards injected planets and away from eclipsing binaries, is applied to ARPS features. This procedure was applied to 150K stars in the Kepler 4-year, resulting in the identification of 97 (sub)-Earth candidate exoplanets.

For the TESS Year 1 survey (DTARPS-S), the analysis begins with 0.9M pre-processed light curves from the DIAmante project. 7,743 stars pass the classifier threshold with excellent True Positive recall and False Positive rejection, and excellent completeness in the planetary Radius-Period diagram. The classifier-selected sample is then subjected to a suite of conservative vetting procedures, giving a high-purity ARPS candidate catalog of 642 stars at high-Galactic latitude and 310 stars near the Galactic Plane.