≡ Menu

Exoplanets: Weeding Out False Positives

The success of the Kepler mission in sifting through a field of more than 150,000 stars to locate transiting planets is undeniable, and the number of planets thus far discovered has been used to estimate how often planets occur around stars like the Sun. Now comes a paper to remind us that statistical analysis based on Kepler results assumes that most of the planet candidates are real and not false positives. Alexandre Santerne, a graduate student at the University of Aix-Marseille, has worked with a team of researchers to study the false positive rate for giant planets orbiting close to their star. 35 percent of these Kepler candidates may be impostors.

The problem is that eclipsing binaries can mimic planetary transits, which is why scientists perform follow-up radial velocity studies or use transit timing variations (TTV) to confirm the existence of the planet. Another technique is to systematically exclude all possible false positive scenarios to a high level of confidence. Whatever the method, it’s clear that validating Kepler’s candidates — making sure that what looks like a planet really is one — has a key role to play if we’re going to interpret the Kepler results properly and extend them to the larger stellar population.

Image: Both Kepler and CoRoT have detected exoplanets by looking for the drop in brightness they cause when they pass in front of their parent star. But as transit studies continue, scientists are working to filter out false positives. Credit: CNES.

Santerne’s team used the SOPHIE spectrograph at Observatoire de Haute-Provence, looking at a selection of Kepler giant planet candidates for follow-up spectroscopic studies. Their sample of 46 candidates represented about 2 percent of the total list of 2321 candidates as of February 2012, and about 22 percent of the giant planet candidates with significant transit depth found thus far in the Kepler data. Their candidates all showed a transit depth greater than 0.4%, an orbital period less than 25 days and a host star brighter than Kepler magnitude 14.7.

Eleven of the candidates had already been confirmed as planets, and the researchers were able to confirm another nine. Two of the candidates turned out to be transiting brown dwarfs and another eleven were in binary star systems. All of that leaves 13 unconfirmed candidates, and leads the team to conclude that the false-positive rate for giant, close-in planets is 35 percent. It’s an interesting result in light of earlier work by Timothy Morton and John Johnson (Caltech), who calculated an expected false positive probability (FPP) of Kepler planets of 5 percent for most candidates.

Morton reacted to the new work in this article in Science News:

…comparisons between the two studies might not be so simple, Morton says, noting that the two groups calculated different things. Instead of looking at impostor rates in a specific population of planets, Morton determined the probability that any candidate — plucked from the sea of twinkling candidates — was real. He also excluded data from obvious impostors.

“Everything here is sort of a game of probabilities,” Morton says, pointing to the abundance of candidates. “It will be impossible to confirm them all with observations.”

The Santerne paper argues that Morton and Johnson did not consider undiluted eclipsing binaries — binary stars that mimic a close-in gas giant — as a source of false positives in the Kepler data, assuming that detailed analysis of Kepler photometry alone would be enough to weed these out. Santerne’s team disagrees:

…we have found that more than 10% of the followed-up candidates are actually low-mass-ratio binary stars, even excluding the two brown dwarfs reported here. This source of false positives is expected to be less important for smaller-radii candidates. However, as it is clearly shown by the cases of KOI-419 and KOI-698, stellar companions in eccentric orbits and with relatively long periods can produce single-eclipse light curves, even for greater mass ratios. It is difficult to imagine how these candidates can be rejected from photometry alone if grazing transits are to be kept.

In other words, it’s easier to mimic a planetary signature than we realized in the case of close-in giant planets. It will take radial velocity follow-up studies of giant planets on much wider orbits to determine whether the false positive rate is as high with them. And we have a lot to do to learn about the reliability of our smaller planet detections:

Only a small fraction of Kepler small candidates are suited for the radial velocity follow-up. These candidates should be followed in radial velocity to determine the true value of FPP and to fill the mass-radius diagram of Neptune and super-Earth like planets. This FPP value for small size candidates is required to correctly derive and discuss the distribution of transiting planet parameters.

The paper is Santerne et al., “SOPHIE velocimetry of Kepler transit candidates VI. A false positive rate of 35% for Kepler close-in giant candidates,” accepted by Astronomy & Astrophysics (preprint). Thanks to Antonio Tavani for the pointer to this one.


Comments on this entry are closed.

  • Ronald June 12, 2012, 11:35

    May sound a bit more dramatic than it really is: 35 % of close-orbit giant planets means only a very small % of all planets, maybe on the order of 1% or less.

  • Paul Gilster June 12, 2012, 13:17

    Yes, and the authors point this out. They’re talking only about gas giants on close orbits.

  • Paul June 12, 2012, 13:49

    I know that this is very important work, and I would certainly choose to fund it.

    But likely, any star that hosts a “Hot Jupiter” is a place that we would never visit. If the gas giant didn’t evolve in it’s present orbit, it probably ejected any “friendlier” planets into deep space on it’s way in.

    So far, I can think of only 1 planet discovered by Kepler in the habitable zone of a “Sun like” star. And it is Neptune size. A tide locked planet around a cool red dwarf may prove interesting indeed, but not like earth.

    Kepler was specifically designed to find earth sized planets in the habitable zones of G stars. I am wondering why none have been found some 3 years into the Kepler mission.

  • NS June 12, 2012, 23:33

    At present Kepler is the only instrument that can detect smaller planets in distant orbits from their stars. If Kepler data alone isn’t enough to verify (or reject) planetary candidates does that mean they’ll just be “possible” until we build another Kepler (or better)?

    My understanding is that it would take an absolute minimum of 2 years (3 transits) to collect enough data to identify an Earth-equivalent planet, and longer unless we were lucky. I don’t know how much additional time the data processing and verification would take but 3 years (or more) overall doesn’t seem unreasonable.

  • Eniac June 13, 2012, 0:33


    May sound a bit more dramatic than it really is: 35 % of close-orbit giant planets means only a very small % of all planets, maybe on the order of 1% or less.

    Can the false positive rate for other types of planets really be much smaller? I thought these close-in giant ones are the easiest to detect and should therefore constitute a lower rather than upper limit on false positive rates, or not?

  • Ronald June 13, 2012, 4:24

    @Paul: “I am wondering why none (earth sized planets in the habitable zones of G stars) have been found some 3 years into the Kepler mission”.

    Observational bias and time. Even the more pessimistic estimates mention a fraction of about 1-2% of solartype stars with earthlike planets in the HZ.

    I expect a large new batch of fascinating (analyzed) data to be released from Kepler either later this year or maybe early next year.

  • Ronald June 13, 2012, 4:49

    A bit further to my previous comment:

    As mentioned in the recent post her on disruptive planets, indeed we should not expect many (earthlike) planets in a system with a close-orbit giant planet (Hot Jupiter), as per the publication “Kepler constraints on planets near hot Jupiters”, by Steffen et al., 2012., though even systems with ‘Warm Jupiters’ and ‘Hot Neptunes’ may have them.

    And with regard to earth analogues (earthlike planets in the HZ of sunlike stars): according to Catanzarite and Shao, 2011 (Occurrence Rate of Earth Analog Planets Orbiting Sunlike Stars), also using Kepler data: “We estimate that 1% to 3% of stars like the Sun are expected to have Earth analog planets, based on the Kepler data release of Feb 2011”, based on extrapolation. “The accuracy of the extrapolation will improve as more data from the Kepler mission is folded in”.

    I do hope that near-future Kepler data will be a bit more positive, because I find their upper limit for ‘earth analogue’ planet (2* Re) and their maximum limits for HZ (0.75 – 1.8 AU) rather optimistic. This leads to about 3% occurrence rate of earth analogues in the HZ of sunlike stars.

    Using Kasting more conservative HZ of 0.95 – 1.37 AU (which I would prefer, at least for the inner boundary) would lead to about 1% occurrence only.

    On the other hand their lower limit for an earth-analogue (0.8 Re) may be a bit pessimistically on the high side.
    And: “The validity of (the estimate) rests on our assumption that the transit detections are complete in (the earth-analogue) region”. If not, “Our extrapolated estimate of (earth-analogues) would accordingly have been much larger”.

    Near future Kepler results will tell us more. Exciting times.

  • Ronald June 13, 2012, 12:20

    @Eniac: “Can the false positive rate for other types of planets really be much smaller?”

    My understanding of this post was, but correct me if wrong, that the false positives for the giant planets are caused by (small) stellar companions. This seems quite unlikely for the much smaller terrestrial planets. I mean a stellar companion to be mistaken for such a planet.

  • Ronald June 14, 2012, 6:39

    Talking about exoplanet detection, great news from ESO: the European E-ELT, which is going to be the largest optical telescope by far (almost 40 m primary mirror diameter), gets the final go-ahead:

    Although the news release mentions that one of the major goals is “tracking down Earth-like planets around other stars in the “habitable zones” I still wonder to what extent this will be possible with a, any, ground-based telescope and in particular this instrument, because of atmospheric absorption and scattering.

    The Executive Summary mentions: “The E-ELT will have the resolution to obtain the first direct images of such objects (Earth-like planets), and even analyse their atmospheres for the biomarker molecules that might indicate the presence of life.”

    But the link to this page (http://www.eso.org/public/teles-instr/e-elt/e-elt_exo.html) mentions: “This will include not only the discovery of planets down to Earth-like masses through indirect measurements of the wobbling motion of stars perturbed by the planets that orbit them, but also the direct imaging of larger planets and possibly even the characterisation of their atmospheres.”

    Can anybody shed more light on this?

  • philw1776 June 14, 2012, 10:48

    Readers ask why haven’t Earthlike planets been found around G & K stars by Kepler yet?

    Because early on Kepler found that stars were much noisier with random brightness variations than thought. The SNR is too low. The extended Kepler mission starting this fall will give a total of 6+ observations believed to be sufficient to get an Earth signal above the noise and finally detect such planets. Of course confirming such discoveries (see above false positive article) is another matter. In a decision of astounding short sightedness, the US science astronomy decadal survey committee in 2011 decided not to recommend any proposed planetary search programs for the next decade.