orbitlight

orbitlight finds new planets, companions and planetary atmospheres in public archives: high-resolution spectra and radial velocities from ESO, the CADC, Keck, the HARPS-RVBank, DACE, the published HIRES, APF and Lick tables, CARMENES's data release and the SOPHIE, ELODIE, TNG and NEID archives, checked against TESS and Gaia. It runs on a GPU. The new planets are in planets.csv, and every signal with its decision is in planets/signals.csv.

Each new planet's contribution column gives what orbitlight's evidence adds. A discovery is a signal orbitlight found in archived velocities or TESS light curves. A mass confirmation is a TESS candidate whose mass archived velocities measure at its transits. A statistical validation is a TESS, Kepler or K2 candidate validated from its light curves; its candidate column names the TOI, community TOI, KOI or K2 candidate. The discovered_by column credits who first found each body: CharlesCNorton for orbitlight's discoveries, for a TESS candidate the TESS pipeline (SPOC or QLP) or the community member who submitted it, and for a Kepler or K2 candidate that mission. In signals.csv, a catalogued planet or companion that orbitlight's own data recover is marked as an independent detection, mass measurement or validation.

Files

File	Contents
planets.csv	New planets: orbits, masses, radii, temperatures, habitable-zone place, inferred character and contribution
discoveries.csv	The planets and companions orbitlight itself found, those of planets.csv and planets/companions.csv whose contribution is discovery
planets/	companions.csv, the new stellar companions; signals.csv, every signal with its decision, evidence and contribution (for the search of TESS's stars, those that pass its vetting); validation.jsonl, each candidate's statistical validation; followup.jsonl, followup.npz and figures.pdf, the records behind each decision with their posterior draws and each system's figures; stars.csv, ledger.csv, eccentric.csv and calibration.json, the census's star models, prediction tests, eccentric-orbit candidates and calibration; dr4_predictions.csv, what each new body of the velocities should show in Gaia DR4's epoch astrometry, written before its release; likely_planets.csv, TESS candidates and transits that pass every test at looser limits of TRICERATOPS's probabilities, listed apart from planets.csv; single_transits.csv, lone dips in the census stars' light curves, each with the orbit its star's velocities best give it and how far that orbit stands above the same search at other times, a list of candidates with no period measured
solar/, giants/, transits/, companions/	Each other mode's results, with the archive queries behind its searches; companions/ also holds GJ 367's line depths, the template of the companion and binary searches

A .jsonl file holds one record per line, and a record's arrays are in the .npz file of the same name under its arrays key. python orbitlight.py publish writes these files from the data roots, each row naming its root.

Modes

planets finds planets and companions in radial velocities and TESS light curves and decides each signal. solar measures the Sun's reflex motion and reads the solar system's atmospheres. giants searches giant planets' spectra for atoms, molecules, and reflected and emitted light. transits tests rocky planets' transits for He I, H-alpha and Na D. companions, binaries and eclipses find and weigh stellar companions. publish writes this repository's files and scans a folder or a clone for what a public file must not hold. pipeline keeps every data root current on one machine by itself: searched, weighed, tried against its controls, decided and published, within a budget of processes. selftest checks the file's methods on synthetic data, with nothing read from any archive. Each mode's --help gives its method.

Decisions

A signal is known if the NASA Exoplanet Archive or the Extrasolar Planets Encyclopaedia (exoplanet.eu) lists a planet of its star within 5% of its period, within 25% for a companion in the Encyclopaedia, if SIMBAD catalogues it, if the TFOPWG lists it as a known or confirmed planet, or if the published HIRES surveys tabulate it at its star (Butler et al. 2017; Rosenthal et al. 2021), within 5% of its period, a companion within 25% or within a factor of two of one they read as a star; a signal they read as activity or an artifact is rejected; an eclipsing binary is also known where a variable-star or eclipsing-binary catalogue, a TOI or a paper SIMBAD links to the star already holds it. A TESS candidate is known from the Encyclopaedia only where it lists the candidate as confirmed, since its candidates include the TOIs themselves, and a body it lists only as retracted is rejected. The catalogued column of signals.csv names each entry. Any other signal is new if it passes its tests and rejected with a reason if it does not. A transit from the search that passes every test before the validation stays a candidate where TRICERATOPS does not validate it or cannot weigh it. A TESS candidate or transit that passes every test with a probability below 10% that its dip is no planet's and a nearby-false-positive probability below 0.3%, short of a validation's limits, is a likely planet, likely in signals.csv and listed in planets/likely_planets.csv.

Signal	New when
Velocity signal	It holds in two independent reductions of the same spectra, or in another spectrograph's velocities of the star, at a false-alarm probability below 1e-3 with both halves of the data agreeing, it is not at a period the California Legacy Survey judged false, no activity indicator has power at its period or its own highest peak there, or where one does, the signal still stands once the star's activity is modelled as a quasi-periodic Gaussian process fitted to that indicator (a false-alarm level below 1e-3 against the process alone, the indicator itself showing nothing at the period, and the period more than 10% from the rotation period, its half and its third), such a signal being a likely planet until another spectrograph's spectra hold it, removing it costs more than 10 in BIC, it is not an alias, TESS's PDCSAP flux shows no variation at P, P/2 or 2P, it lies more than 10% from the star's rotation period and its half, the system stays stable, and another spectrograph's archived spectra, where they cover enough nights, do not fail its orbit
Companion	Its cross-correlation velocities follow the bank's at r ≥ 0.9, and two cycles at a false-alarm probability below 1e-3 or significant Hipparcos-Gaia astrometry fix its period. An orbit resting on astrometry must agree with Gaia's astrometric excess noise, and its true mass is given where it agrees closely
TESS candidate with velocities	Its ephemeris holds to a tenth of a period, its semi-amplitude at the transit phase is positive at 5 sigma, and its free phase agrees within 3 sigma
TESS, Kepler or K2 candidate, validated	TRICERATOPS gives a probability below 1.5% that the dip is no planet's, a planet on an unresolved companion or on a star behind the target counting as a planet (Morton et al. 2016; on_target holds the share on the target, whose radius it is given, and fpp TRICERATOPS's own figure for a planet on the target alone), or below 15% for a body under eight Earth radii on a star that holds another live candidate or planet at another period, and a nearby-false-positive probability below 0.1% (means of ten runs, without the stars the TIC flags as artifacts or duplicates, each nearby star's share weighed by how well the transit's difference images place the dip's source at it against at the star, and without a star eclipsing the target where twice the scatter of Gaia's epoch velocities lies below the reflex of the least star at that period), its radius is below two Jupiter radii, it passes the planet tests below, SPOC's data validation is clean, and its own flattened fold shows its dip at 7.1 sigma for a TOI, at 15 sigma for a Kepler candidate and at 10 sigma for any other, a community TOI also reaching 2.5 times the highest the same search finds in the light curve turned upside down
Transit from orbitlight's search	A blind box search of the star's TESS light curves, its boxes as short as half an hour about a star under 0.6 solar radii, finds it with the star's registered TOIs set aside, it passes the signal and planet tests below, and it is validated as a TESS candidate is at a signal-to-noise of 2.5 times the highest the same search finds in the light curve turned upside down, or weighed with archived velocities as a TESS candidate with velocities is
Eclipsing binary from orbitlight's search	The same search finds eclipses of different odd and even depths or a secondary eclipse, it passes the signal tests below, its eclipses or its size rule out a planet, and under 0.1% of TRICERATOPS's probability lies on a nearby or background star
Orbit from Gaia DR4's epochs	The star's epoch astrometry shows it at a false-alarm probability below 1e-3 against its own residuals shuffled, ESA's orbit tables do not hold it, and the star's velocities hold the amplitude and phase it predicts

The signal tests ask whether a signal from the search is real: three transits or more with none dominating, at least 85% of the signal-to-noise without the strongest of them that transits alike would keep, depths that agree from transit to transit, a signal-to-noise of 7.1 or more on the transit's timescale and of 1.25 times the highest the same search finds in the light curve turned upside down, a period that is neither TESS's orbit nor another signal's, and a dip the folded light shows at 7.1 sigma.

The planet tests ask whether a dip is a planet's on this star: odd and even transits agree against the scatter of the transits themselves and are not half their mean depth apart at 5 sigma, there is no secondary eclipse, the folded light is a transit's box rather than the star's own variation, no catalogued variable star within six arcminutes varies at its period, half or double, the star is not catalogued as a binary at it, no brighter star within two arcminutes shows the signal, the centroids do not place its source 5 sigma off the star, Gaia's RUWE is below 1.4, Gaia holds no non-single-star solution of the star and its epoch velocities do not move (a goodness of fit above 4 at a p-value of 0.01 or less), the star does not spin at 50 km/s or more with the period at its spin's limit, and its orbit and duration fit the star.

A planet from the velocities is tested in the spectra of a spectrograph its velocities did not come from, ESPRESSO's for HARPS's and HARPS's for the others, measured line by line: each orbit's amplitude and phase are fitted freely and set against what its posterior predicts. Its second_test is held at 3 sigma at the predicted phase, failed where the amplitude falls short of the prediction by more than 3 sigma and 30% or a detection lies 3 sigma off that phase, which rejects it, and open otherwise. Each new body's predicted orbit is then tested against observations made after its discovery, and a failed prediction rejects it. A planet's letter follows those of the planets the archive and the Encyclopaedia already list for its star.

Each orbit from the velocities is also held against the star's Gaia DR4 epoch astrometry. K fixes the star's orbit on the sky at each inclination, and the likelihood over inclinations and nodes is set against no orbit and against the same orbit at other periods. Its dr4_test is held where the epochs show that orbit, which gives its true mass, refuted where they leave no room for the smallest orbit K allows, which rejects it, and open otherwise, with dr4_mass_mjup the mass the epochs allow.

planets validate --benchmark scores the validation on TOIs whose fate the TFOPWG has settled, and with --source koi or --source k2 on the Kepler and K2 candidates whose fate the NASA Exoplanet Archive lists, planets calibrate --wobble sets the excess-noise limit on Gaia DR3's published orbits and accelerations, and planets/calibration.json holds their counts. A Kepler or K2 candidate enters the decisions once its mission's benchmark holds 90% of the settled candidates drawn weighed and, of what it passes at even odds, at least 90% are planets. planets validate --control tries the rules on confirmed planets and false positives, and calibration.json holds how often the TFOPWG has found TOIs false with and without another live candidate on their stars, which sets the higher limit. planets validate --null searches a random sample of TESS's stars with their light curves upside down, where no star's dip survives, weighs what passes the search's vetting as a candidate is, and calibration.json holds what the vetting and the decisions let through there against the same stars' real signals. planets validate --inject lays one planet's transits into each light curve of another random sample and searches them, and calibration.json holds what the search finds and its vetting passes by signal-to-noise, period, number of transits and length. planets validate --clear reads each candidate's difference images, and calibration.json holds what they let through among the TOIs the TFOPWG has settled. planets validate --gaia reads Gaia's marks of a double star, and calibration.json holds what they do among the same TOIs. planets validate --scramble moves each star's nights to one another's dates, which keeps every velocity's noise and breaks whatever repeats in time, and calibration.json holds what the census and its vet let through there against the real velocities. planets controls reads from the records, with nothing weighed, what the decision does to each mission's settled candidates and to the open ones, by test, by band of signal-to-noise, radius, period and magnitude, and at each limit of the fold's dip, the noise ratio and the two probabilities, with what the search's own rules keep of laid-in transits, upside-down light curves and the same stars' real signals, and calibration.json holds it. planets controls --indicators measures how often an activity indicator holds power at a known planet's period. planets strict tries the strict-periodicity test of Hara et al. (2022) and planets sparse RELAX (Li and Stoica 1996) and the l1 periodogram (Hara et al. 2017) against the census on the archive's planets; no decision reads them. planets recovery tries a range of periods before the search is carried over it, and planets window measures what the detrending costs a transit of each length.

Running

pip install -r requirements.txt
export ORBITLIGHT_DATA=/path/to/data
python orbitlight.py planets bank && python orbitlight.py planets dace
python orbitlight.py planets census      # then tois, validate, eccentric, followup and ledger
python orbitlight.py publish OUT bank=/path/to/data

python orbitlight.py selftest checks the methods on synthetic data. python orbitlight.py pipeline start then keeps every data root under the folder of the file or ORBITLIGHT_HOME current for as long as it runs, pipeline status says what runs, pipeline pause and pipeline resume hold and release the work between stars, and pipeline decide decides and publishes once on the CPU. python orbitlight.py publish scan OUT, with --git every revision of a clone, reads a folder for home paths, tokens, private addresses and what ORBITLIGHT_SCAN_NAMES adds, through PDFs and arrays, and publish check OUT PUBLIC sets a publication against the public files.

Stars beyond the HARPS-RVBank are searched in data roots of their own, set up by planets rvcat, planets tables, planets carmenes, planets sophie, planets harpsn, planets neid, planets elodie, planets targets and planets measure, and planets peers gives each root the others' velocities of its stars, a spectrograph a root already holds of a star left out so that no spectra count twice. The papers are in seti-tools (https://huggingface.co/phanerozoic/seti-tools), under papers/.

Acknowledgements

Based on data obtained from the ESO Science Archive Facility; from the Canadian Astronomy Data Centre, operated by the National Research Council of Canada with the support of the Canadian Space Agency, and the Canada-France-Hawaii Telescope; from the Keck Observatory Archive, operated by the W. M. Keck Observatory and NExScI under contract with NASA; from the NASA Exoplanet Archive and the Exoplanet Follow-up Observing Program, operated by Caltech under contract with NASA; from the Extrasolar Planets Encyclopaedia (exoplanet.eu); from SIMBAD and VizieR, operated at CDS, Strasbourg; from the HARPS-RVBank (Trifonov et al. 2020); from the published velocities of the California Legacy Survey (Rosenthal et al. 2021), of HIRES (Butler et al. 2017, as Tal-Or et al. 2019 corrected them) and of the Lick planet search (Fischer et al. 2014); from CARMENES's first data release (Ribas et al. 2023), of the instrument at the Centro Astronómico Hispano en Andalucía at Calar Alto; from the SOPHIE and ELODIE archives of the Observatoire de Haute-Provence; from the Telescopio Nazionale Galileo, operated on La Palma by the Fundación Galileo Galilei of INAF, through the archive of INAF's Italian Center for Astronomical Archives; from the NEID spectrograph, built by Penn State University and operated at the WIYN Observatory by NOIRLab under the NN-EXPLORE partnership of NASA and NSF, through the NEID Archive of the NASA Exoplanet Science Institute; and from The Data & Analysis Center for Exoplanets (DACE), a facility of the University of Geneva and a platform of the Swiss NCCR PlanetS. It includes data from the TESS mission, obtained from MAST at the Space Telescope Science Institute and funded by NASA's Explorer Program, and from ESA's Gaia mission, processed by DPAC and read from ESA's archive and the Gaia archive at the Astronomisches Rechen-Institut, Heidelberg, and Hipparcos. The line lists are Kurucz's and petitRADTRANS's (Mollière et al. 2019).

Licence

Code is under the MIT License and everything else under CC BY 4.0; see LICENSE.