Exoplanet transit search at the detection limit: detection and false alarm vetting pipeline

TL;DR

This work introduces a fast, three-stage pipeline for exoplanet transit searches at Kepler’s detection limit, combining aggressive preprocessing to remove localized defects, per-transit vetting, and a Bayes-factor–based detection statistic to identify and extract multiple candidates from a single scan. Through injection studies and application to Kepler candidates, the authors demonstrate improved completeness at a fixed false-alarm rate and show that many long-period, low-SNR unconfirmed candidates—some Earth-like in the habitable zone—are likely false alarms, potentially affecting estimates of $\eta_{\oplus}$. The approach hinges on refined modeling of systematics, Gaussianization of outliers, adaptive power-spectrum estimation, and a careful treatment of transit duration priors, with explicit handling of transit-timing variations and harmonics. Overall, the pipeline advances routine exoplanet detection near the detection limit and provides a framework for robust vetting that can influence population inferences such as habitable-zone planet occurrence.

Abstract

One of the primary mission goals of the Kepler space telescope was to detect Earth-like terrestrial planets in the habitable zone around Sun-like stars. These planets are at the detection limit, where the Kepler detection and vetting pipeline produced unreliable planet candidates. We present a novel pipeline that improves the removal of localized defects prior to the planet search, improves vetting at the level of individual transits and introduces a Bayes factor test statistic and an algorithm for extracting multiple candidates from a single detection run. We show with injections in the Kepler data that the introduced novelties improve pipeline's completeness at a fixed false alarm rate. We apply the pipeline to the stars with previously identified planet candidates and show that our pipeline successfully recovers the previously confirmed candidates, but flags a considerable portion of unconfirmed candidates as likely false alarms, especially in the long period, low signal-to-noise ratio regime. In particular, several known Earth-like candidates in the habitable zone, such as KOI 8063.01, 8107.01 and 8242.01, are identified as false alarms, which could have a significant impact on the estimates of $η_{\oplus}$, i.e., the occurrence of Earth-like planets in the habitable zone.

Figures (15)

• Figure 1: Light curve around the transits of the exoplanet candidate KOI 7894.01 with a Robovetter disposition score of $84 \%$ and reliability of $89\%$bryson_occurrence_2021. Shaded area is the transit region. The left-most panel is the folded light curve, which seems like a genuine exoplanet signature. The next five panels are individual transits of the candidate. The second and the fourth transits are apparently discontinuities in the light curve, which clearly reveals that KOI 7894.01 is a false alarm. The moral of this example is that the individual transits contain important information that is lost in the folded light curve. The method proposed in this work easily discards this candidate, while the existing methods fail.
• Figure 2: Data is composed of non-periodic and possibly of periodic sources. Periodic sources are mostly of astrophysical origin and are distinguished from the planets by the Bayesian hypothesis testing. Non-periodic sources are mostly noise systematics and are less well understood. They need to be filtered out, which is the main focus of this work.
• Figure 3: Flowchart of the data analysis. The pipeline is composed of three stages: preprocessing (yellow, orange and red), detection (pink) and vetting (purple).
• Figure 4: Detection limit as a function of the period of the planet. Noise-only stellar variability simulations are performed and searched for a planet signature. In each period bin, the best candidate is found and its test statistic is shown. The confidence regions are quartiles over 1024 noise simulations, the solid line is the median. Left: approximation to the Bayes factor is used as a test statistic. The detection limit is roughly independent of the period, so a fixed detection cutoff is close-to-optimal. Right: signal-to-noise ratio is used as a test statistic. The detection limit is significantly higher at larger periods. This is because of the look-elsewhere effect bayer_look-elsewhere_2020robnik_statistical_2022: phase range is larger at larger periods, $t_0 < \phi < t_0 + P$, making the prior volume larger and thus increasing the effective multiplicity of trials. This demonstrates that using a fixed detection cutoff on $SNR$ is suboptimal, as it has to be designed to control the false alarms at high periods, while a lower cutoff could be used at lower periods.
• Figure 5: Transit duration prior is shown. Blue and orange lines are showing the prior for planets with orbit eccentricities 0 and 0.5 respectively. Pale versions of both colors show the corresponding priors if $\tau_K$ was known exactly, i.e. if stellar density had no uncertainty. Stronger lines are shown for 15 percent relative error on $\tau_K$. Uncertainty in the $\tau_K$ broadens the prior. The black line is the prior which is marginalized over the eccentricity, assuming a beta distribution prior for the eccentricity with parameters $\alpha = 0.867$, $\beta = 3.03$ as fitted to the Kepler planets kipping_bayesian_2014.