Source-linked AI summary
Exploring Long-period Architectures: Four New Planet Candidates from Kepler with Periods >342 days
Matthew T. Hansen, Jason A. Dittmann
TL;DR
Long-period planets are underdetected because Kepler provides fewer transits for these systems. The paper applies a CNN-based single-transit pipeline combining photometry and spacecraft diagnostics, finding four new candidates with periods or allowed minimum periods of hundreds of days. Follow-up is constrained by the faintness, long periods, and lengthy transit durations of these systems.
Problem
Kepler’s detection pipeline and transit method favor shorter periods, leaving a dearth of long-period detections and an incomplete view of exoplanet system architectures.
Method
The study applies an adjusted single-transit CNN pipeline using Kepler photometry and onboard spacecraft diagnostics to systems with precise transit timings and suitable periods, followed by manual vetting.
Results
Four new long-period planetary candidates are identified: two have periods of 777.78 and 505.495 days, while two single-transit candidates have shortest allowed periods of 342 and 544 days.
Takeaways & Limitations
The four candidates expand the growing population of long-period transiting planets, but the candidates alone cannot reproduce the inner systems’ observed TTV signals.
Takeaways & Limitations
Ground-based follow-up is effectively ruled out because the candidates have rare, more-than-10 hr transits around faint stars, while RV follow-up would require extensive data and is infeasible for these systems.
Abstract
from arXiv · showhide
The Kepler detection pipeline, as well as the transit method, has a bias towards shorter periods, leaving a dearth of detections at longer orbital periods. This relative lack of detections has left an incomplete picture of the architectures of exoplanet systems within the long-period regime. We have built a single transit detection pipeline, utilizing a classification convolutional neural network and the onboard spacecraft diagnostics of the Kepler spacecraft, to detect long-period planets. We apply our pipeline to all currently known planetary systems in the Kepler field hosting at least one planet with an orbital period longer than 6 days. We manually vet all new signals from our pipeline, and identify four new planetary candidates, all of which are in systems where the inner planets exhibit transit timing variations (TTVs). Two of these candidates, Kepler 1752.02 and Kepler 199.03, cause two transit events that are consistent with periods of $777.78^{+0.01}_{-0.02}$ and $505.495^{+0.004}_{-0.004}$ days, and radii of $3.55^{+0.15}_{-0.15}$ and $2.74^{+0.05}_{-0.05}$ $R_{\oplus}$, respectively. Our remaining two candidates, Kepler 1897.02 and Kepler 1811.02, are single transit candidates with radii $4.81^{+0.20}_{-0.19}$ and $3.25^{+0.28}_{-0.30}$ $R_{\oplus}$, respectively. The shortest orbital periods for these candidates, consistent with the Kepler dataset (gaps and coverage), are 342 days for Kepler 1897.02 and 544 days for Kepler 1811.02. The new planetary candidates, on their own, are incapable of reproducing the observed TTV signals in the inner system. Although difficult to schedule, follow-up observations are needed to further constrain the new candidates and potentially discover the planets causing the perturbations.
1. Introduction
Kepler’s transit surveys are less sensitive to long-period planets because fewer transits are available, leaving a persistent shortage of detections. This work extends a single-transit pipeline using photometry and spacecraft diagnostics to identify additional long-period candidates.
- Motivation: Kepler’s nearly 4 yr baseline enables searches from subday to long orbital periods, but long-period planets provide fewer transits for phase-folding and signal-to-noise enhancement.At long periods, detected planets therefore tend to have larger planet-to-star radius ratios and higher transit signal-to-noise.
- Prior approaches: Visual inspection can preserve long-duration transit signals that automated detrending may remove when transit durations exceed the detrending frequency threshold.
- Prior approaches: Earlier automated searches for single transits imposed a detection threshold more than 25 times the background noise, biasing candidate selection toward larger planets.
- Motivation: Long-period planet detections remain scarce despite Kepler’s extensive photometric coverage.
- Approach: The pipeline combines Kepler photometry with onboard spacecraft diagnostics and avoids dependence on phase-folding to extend detection toward longer periods.
- Study objective: The study applies the adjusted pipeline to systems with precise transit timings and periods longer than twice the pipeline input, identifying four candidates and constraining their physical and orbital parameters.
2. Data
The study constructs standardized transit and no-transit segments from Kepler archive data and trains neural networks on flux with or without spacecraft engineering attributes. Both models achieve similar classification performance and support manual vetting of detected signals.
- Dataset: The training data come from confirmed and candidate planets in the NASA Exoplanet Archive, using the study’s prior strategy for separating training, validation, and testing sets.
- Training set: 128-cadence segments centered on individual transits form the in-transit examples, while windows overlapping flagged transits or large gaps are excluded from the no-transit set.
- Training set: 239,879, 32,062, and 28,827 no-transit segments comprise the training, validation, and testing sets, respectively, creating a 9X class imbalance relative to in-transit segments.
- Preprocessing: The pipeline standardizes flux and engineering attributes separately, expressing each engineering measurement as a median-centered distance in units of median absolute deviation.
- Network construction: The networks classify whether a lightcurve segment contains a transit, using CNN and fully connected blocks with the final sigmoid output representing classification confidence.
- Evaluation: The flux-only model uses a 0.5195 threshold and has AUC 0.81, while the two networks show no significant performance difference.
3. Single Transit Detection Pipeline
The pipeline applies sliding-window neural-network classifications to Kepler lightcurves, converts fractional classifications into threshold crossing events, estimates periods from repeated events, and filters candidates through systematic, visual, and quality checks.
- Sliding-window classification: A 128-long-cadence sliding window scans each stellar lightcurve because transit locations are unknown in advance.The network classifies each window as containing a transit or no transit.
- Fractional classification: A fractional classification rate records how often each time stamp lies in windows classified as containing a transit, accounting for data gaps.This metric labels lightcurve sections despite uneven window coverage near gaps.
- TCE identification: Peaks above the 50% classification threshold and separated by at least 64 long-cadence points are labeled threshold crossing events.The separation requirement prevents double-labeling the same transit, and the KOI 1527 test recovered all seven visible transits.
- Period estimation: Repeated threshold crossing events are converted into boxcar pulses and analyzed with a Lomb–Scargle periodogram, while single events receive no period.Known-planet transits are removed when they occur within 1.3 days of cataloged transit times.
- TCE rejection: Candidate rejection removes poorly detrended stars, common systematic time bins, visually inconsistent signals, and TCEs surrounded by abundant nonzero quality flags.Manual vetting examined raw and detrended lightcurves within a 6-day window; 17 KICs passed visual inspection, and four KICs passed the listed tests.
4. Discussion of New Candidates
The study fits transit models to four new long-period candidates identified in Kepler systems, using multi-transit timing or data-gap constraints to estimate their periods. The candidates do not reproduce the inner planets’ observed TTV signals under the modeled assumptions.
- Model Fitting: The analysis simultaneously fits known planets and new candidates with batman transit models and MCMC-based parameter estimation.The MCMC uses Kepler flux uncertainties in the likelihood and fits stellar density rather than stellar radius and mass.
- Kepler 1752: 777.7 days is the inferred period for Kepler 1752.02 from two observed transits.The candidate has a best-fit radius of 3.55 R⊕ using a stellar radius of 0.788 R☉.
- TTV Constraints: The new candidates cannot reproduce the observed inner-system TTV signals under the circular-orbit modeling assumptions.For Kepler 1752.02, the circular candidate orbit cannot cause Kepler 1752b’s perturbations; analogous failures are reported for Kepler 1897.02 and Kepler 1811.02.
- Kepler 199: 505 days is the adopted orbital period for Kepler 199.03 because a large data gap permits this solution between two detected events.Its best-fit radius is 2.74 R⊕ using a stellar radius of 0.927 R☉.
- Kepler 1897: 342.136 ± 0.001 days is the shortest period allowed for single-transit candidate Kepler 1897.02 given Kepler’s data gaps.The circular-orbit transit fit gives a best-fit radius of 4.81 R⊕, while the candidate is too distant to reproduce Kepler 1897b’s TTVs under zero eccentricity.
- Kepler 1811: Kepler 1811.02 is a single-transit candidate with a 145.9-day circular-orbit lower bound and a 544-day period used for its MCMC fit.Its best-fit radius is 3.25 R⊕, and the modeled candidate cannot cause the observed Kepler 1811b timing perturbations when eccentricities are zero.
5. Discussion
The paper presents four long-period planet candidates and tests whether undetected in-between planets could explain the inner systems’ TTVs. The candidates expand the long-period population, but their configurations and follow-up prospects remain constrained.
- New candidates: Two candidates have periods of 777.78 and 505.495 days, while the two single-transit candidates have shortest allowed periods of 342 and 544 days.Kepler 1752.02 and Kepler 199.03 are the two-transit candidates; Kepler 1897.02 and Kepler 1811.02 are single-transit candidates.
- New candidates: 284 of approximately 4700 Kepler transit candidates have periods greater than 300 days, and only 71 exceed 500 days.The four candidates add to the slowly growing population of long-period transiting planets.
- In-between perturbers: Kepler 1752b and Kepler 1897b can be fit with coplanar, transiting, BLS-undetected in-between perturbers, but multiple stable configurations remain.For Kepler 1752b, 2:1 and 3:2 perturber period ratios produce nearly identical TTV signals.
- Future missions: The candidates’ TTV interpretations remain degenerate, and follow-up is needed despite substantial observational constraints.The candidates’ long periods and transit durations hinder ground-based transit observations, while their low TSM or difficult RV signals limit characterization.
- In-between perturbers: Kepler 199 and Kepler 1811 cannot be fit with coplanar, transiting perturbers small enough to evade BLS detection.A dynamically stable nontransiting configuration was found only for Kepler 1897, leaving other pathways for Kepler 199 outside the paper’s scope.
- System architecture: Kepler 1752.02 and Kepler 199.03 are smaller than their inner companions yet have roughly an order-of-magnitude longer periods.All planets in these systems are Neptune to sub-Neptune sized and lie beyond the ice line.
6. Conclusion
The paper presents a CNN-based single-transit pipeline using Kepler flux and spacecraft engineering data, identifying four long-period planetary candidates. Their periods and system dynamics remain insufficiently constrained, requiring difficult long-term follow-up.
- Pipeline and search: The pipeline combines 128-cadence flux segments with ancillary spacecraft engineering files to classify transit-containing segments.It also compares this flux-engineering network with a flux-only network.
- Pipeline and search: The flux-engineering and flux-only networks show only a marginal difference in classification accuracy.
- Candidate properties: 777.78 days is the period consistent with Kepler 1752.02, while Kepler 199.03 permits a 505-day 2:1 alias because of a data gap.Kepler 199.03 has two transits separated by 1010 days.
- Candidate properties: Kepler 1752.02 and Kepler 199.03 have radii of 3.55 R⊕ and 2.74 R⊕, respectively.The Kepler 199.03 estimate uses a stellar radius of 0.927 R⊙.
- Candidate properties: Kepler 1897.02 has a radius of 4.81 R⊕, while the supplied stellar-radius context for the candidates includes 1.110 R⊙.
- Constraints and follow-up: The systems cannot yet be dynamically modeled to match their observed TTV signals, and follow-up is needed to constrain periods, eccentricities, and masses.The authors specifically mention RV and ground-based photometry, while acknowledging the difficulty of confirming long-period planets.
Data Availability
The paper states that its Kepler data are publicly available through MAST and acknowledges the computational resources and software supporting the work.
- All Kepler data used in the paper are available through MAST.
- The authors acknowledge University of Florida Research Computing and list the software used in the research.
ORCID iDs
The paper provides ORCID identifiers for Matthew T. Hansen and Jason A. Dittmann.
- ORCID identifiers are provided for Matthew T. Hansen and Jason A. Dittmann.