Arxiv:1203.5537V3 [Astro-Ph.EP] 2 Nov 2012 Odw Ihik2011)
![Arxiv:1203.5537V3 [Astro-Ph.EP] 2 Nov 2012 Odw Ihik2011)](http://data.docslib.org/img/5129605/arxiv-1203-5537v3-astro-ph-ep-2-nov-2012-odw-ihik2011.jpg)
arXiv:1203.5537v3 [astro-ph.EP] 2 Nov 2012 od&Rso20;W ihik2011). Lithwick & 2003; Wu Murray 2006; circu- & Rasio orbit Wu sev- & of (e.g. its Ford one mechanisms as by perturbation planet (see initiated be- eral dissipation, the that orbit tidal for periapse through eccentric spot star-skirting larized parking an a the with Al- onto came 2012), per- al. body been et 2011). have another Naoz Kenyon may by & Jupiter al. Bromley (e.g. turbed hot et typical 2008; formed Alibert the Lin they 1997; ternatively, & which Ward in- Ida migrated from 1980; 2005; smoothly Tremaine disk have & the Goldreich may Jupiters formed through have Hot not ward could 2012), situ. Murray short in & very Hansen on class (e.g. planets ets enigmatic Jupiters: the (P hot created period as relics that known be planets processes may of same they because the interest of 2008; particular of Rasio are & bits 2011). Ford Ida sys- 2008; & planetary Tremaine Nagasawa Juri´c of & architectures (e.g. dynamical the tems the shape of signature that a processes as interpreted been has that eevd21 pi ;acpe 02Jl ;pbihd21 Au 2012 published 9; July L 2012 using accepted typeset 5; Preprint April 2012 Received ayeolnt aehgl ceti ris trend a orbits, eccentric highly have exoplanets Many 1 H HTECNRCEFC N RT O UIESI MEASURI I. JUPITERS HOT PROTO AND EFFECT PHOTOECCENTRIC THE [email protected] et sn actual using tests ih uvso D116bt esr necnrct of eccentricity an measure to demonstrat b To 17156 HD probability. of transit margina curves the naturally light for posteriors accounts parameter of automatically exploration signa Carlo curve Monte light “photoeccentric” long-cadence the identifie of us with readily part curve are a – light Jupiters durations demonstrate transit eccentric transit that we its show We from Here density. solely periods, star’s eccentricity most. of planet’s of range individual follow-up a an velocity spanning radial Jupiters transiting precludes of hundreds ered icvr value discovery c iet ls-nhtJptr,wihaeulkl ohv omdi situ in measu formed velocity have radial to from unlikely come are which primarily lin Jupiters, planets may hot transiting they close-in individual to interesting: particularly line are ice planets Jupiter-sized tric, otecet ffciemaso dniyn h xrml eccentr headings: Subject extremely (2012). the w al. identifying Finally, et of intervals. Socrates means confidence effective their efficient, and most eccentricities exoplanet ing xpae ria cetiiisoe aubecusaottehisto the about clues valuable offer eccentricities orbital Exoplanet < 0dy)obt ht niesalrplan- smaller unlike that, orbits days) 10 1. A INTRODUCTION T E tl mltajv 5/2/11 v. emulateapj style X e Giant CETIIISO NIIULTASTN PLANETS TRANSITING INDIVIDUAL OF ECCENTRICITIES 0 = lntr ytm,tcnqe:photometric techniques: systems, planetary 20Es aionaBuead C291,Psdn,C 911 CA Pasadena, 249-17, MC Boulevard, California East 1200 Kepler Kepler eevd21 pi ;acpe 02Jl ;pbihd21 A 2012 published 9; July 2012 accepted 5; April 2012 Received . AAEolntSineIsiue(ESI,CTMi oe10 Code Mail CIT (NExScI), Institute Science Exoplanet NASA 67 lnt necnrcor- eccentric on planets ± eateto srnm,Clfri nttt fTechnolo of Institute California Astronomy, of Department aa nec aetetcnqepoe ob ibemto fme of method viable a be to proves technique the case each In data. 0 htmtyadloeycntandselrprmtr.AMarko A parameters. stellar loosely-constrained and photometry . 8bsdo 3rda-eoiymaueet.W rsn w addit two present We measurements. radial-velocity 33 on based 08 7 ot isnAeu,Psdn,C 91125 CA Pasadena, Avenue, Wilson South 770 0Gre t S1,Cmrde A02138 MA Cambridge, MS-10, St, Garden 60 avr-mtsna etrfrAstrophysics for Center Harvard-Smithsonian onAhrJohnson Asher John eea .Dawson I. Rebekah ABSTRACT ito a etdb oge l 21)o apeof sample a pre- on (2012) This the al. in et 0.9. binaries Dong of by eclipsing hot excess tested proto was in diction “super-eccentric” eccentricities several with of Jupiters detect process the the should that in sion predicted caught S12 are lifetime, circularization. that star’s tidal Jupiters the tidal hot over have time proto circularize and that to given long full Jupiters any too undergone hot timescales have failed at that circularization, Jupiters HEM Jupiters, tidal hot If observe hot would (HEM). for we migration” responsible eccentricity were “high as cess oigasre aest esrn h niiulec- individual the measuring the to of papers centricities series 100. of a total voting a find eventually bina- to eccentric expect highly and long-period, ries 14 found they search, lmt aitos(. aai ta.21) n h ef- the and 2008), 2011), al. al. et et Jackson Kataria 2007; (e.g and variations 2010) Mardling Hansen climate (e.g 2008; its al. heating understanding et to tidal Jackson key demon- (e.g. also and history is eccentricity tidal planet describe the Jupiter-sized Measuring we a curves. of eccentrici- light paper, transit individual first from measuring ties for this technique In our predicted strate Jupiters hot S12. proto by super-eccentric the out rule ut21 gust ortse l 21)(eefe 1)rfrt hspro- this to refer S12) (hereafter (2012) al. et Socrates sats fpaeayacietr hois eaede- are we theories, architecture planetary of test a As e 0 = 1 . ueo lntsecnrct even — eccentricity planet’s a of ture 71 ytersotingress/egress/total short their by d c rt o uiespeitdby predicted Jupiters hot proto ic, − +0 u h anns ftehs stars host the of faintness the but ,w s he ulse transit published three use we e, h cl”Jptr eodthe beyond Jupiters “cold” the k 0 ie vrtepras nl and angle periapse the over lizes . . yo lntr ytm.Eccen- systems. planetary of ry ru htti ehdi the is method this that argue e 16 09 aeinmto fmeasuring of method Bayesian n ro nweg fishost its of knowledge prior ing ngo gemn ihthe with agreement good in , Kepler 5 S and USA 25, rements. gs 21 ugust odt,ecnrcte of eccentricities date, To . gy, 0-22, Kepler uiest ihrietf or identify either to Jupiters GPHOTOMETRIC NG Kepler ed na incomplete an in field: a discov- has Chain v asur- ional Kepler Mis- 2 Dawson and Johnson fect of the variation in insolation on the habitability planets. (e.g Spiegel et al. 2010; Dressing et al. 2010) of possi- The work of FQV08 was the basis for several recent ble orbiting rocky exomoons detectable by Kepler (e.g. analyses of high-precision light curves from the Kepler Kipping et al. 2009). mission that have revealed information about the eccen- To date, the measurements of eccentricities of individ- tricity distribution of extra-solar planets and the eccen- ual transiting planets have been made through radial ve- tricities of planets in multi-transiting systems. By com- locity follow-up, except when the planet exhibits tran- paring the distribution of observed transit durations to sit timing variations (e.g. Nesvorny et al. 2012). How- the distribution derived from model populations of eccen- ever, a transit light curve is significantly affected by a tric planets, Moorhead et al. (2011) ruled out extreme planet’s eccentricity, particularly if the photometry is eccentricity distributions. They also identified individual of high quality: we refer to the signature of a planet’s planets with transit durations too long to be consistent eccentricity as the “photoeccentric” effect. One aspect with a circular orbit; these planets are either on eccentric is the asymmetry between ingress and egress shapes orbits (transiting near apoapse) or orbit host stars whose (Burke et al. 2007; Kipping 2008). The eccentricity stellar radii are significantly underestimated. also affects the timing, duration, and existence of sec- Kane et al. (2012) used the distribution of transit du- ondary eclipses (Kane & von Braun 2009; Dong et al. rations to determine that the eccentricity distribution of 2012). The most detectable aspect of the photoeccentric Kepler planets matches that of planets detected by the effect in Kepler photometry for long-period, planet-sized RV method and to discover a trend that small planets companions is the transit event’s duration at a given or- have less eccentric orbits. In contrast, Plavchan et al. bital period P , which is the focus of this work. (2012) found that the distribution of eccentricities in- Depending on the orientation of the planet’s argu- ferred from the transit durations is not in agreement with ment of periapse (ω), the planet moves faster or slower the eccentricity distribution of the RV sample; they sug- during its transit than if it were on a circular or- gested that the difference may be due to errors in the bit with the same orbital period (Barnes 2007, Burke stellar parameters. Finally, Kipping et al. (2012) pre- 2008, Ford, Quinn, and Veras 2008, hereafter FQV08; sented a method that they refer to as Multibody Astero- Moorhead et al. 2011). If the transit ingress and egress density Profiling to constrain eccentricities of planets in durations can be constrained, the duration aspect of the systems in which multiple planets transit. They noted photoeccentric effect can be distinguished from the effect that one can also apply the technique to single transit- of the planet’s impact parameter (b), because although ing planets, but discouraged doing so, except for planets b > 0 shortens the full transit duration (T23, during whose host star densities have been tightly constrained which the full disk of the planet is inside the disk of (e.g. by asteroseismology). FQV08 recommend measur- the star, i.e. from second to third contact), it lengthens ing eccentricities photometrically only for planets with the ingress/egress duration. Therefore, with prior knowl- “well-measured stellar properties” but also point out the edge or assumptions of the stellar parameters, combined weak dependence of eccentricity on stellar density. with measurements from the light curve of the planet’s In this work we apply the idea first proposed by FQV08 period and size (RP /R⋆), one can identify highly eccen- to real data and demonstrate that we can measure the tric planets as those moving at speeds inconsistent with eccentricity of an individual transiting planet from its a circular orbit as they pass in front of their stars (see transit light curve. We show that this technique is par- also 3 of Barnes 2007, 3.1 of FQV08). ticularly well-suited for our goal of identifying highly ec- Barnes§ (2007) presented§ the first comprehensive de- centric, giant planets. In 2, we show that even a loose scription of the effects of orbital eccentricity on a transit prior on the stellar density§ allows for a strong constraint light curve, including that a short transit duration cor- on the planet’s orbital eccentricity. In 3, we argue that responds to a minimum eccentricity, contingent on the Markov Chain Monte Carlo (MCMC) exploration§ of the measurement of b and of the host star’s density. Burke parameter posteriors naturally marginalizes over the pe- (2008) discussed the effect of orbital eccentricity on tran- riapse angle and automatically accounts for the transit sit detection and on the inferred distribution of plan- probability. We include both a mathematical and practi- etary eccentricities. FQV08 laid out the framework for cal framework for transforming the data and prior infor- using photometry to measure both the distribution of ex- mation into an eccentricity posterior. In 4, we measure oplanet eccentricities and, for high signal-to-noise tran- the eccentricity of HD 17156b from ground-based§ tran- sits of stars with known parameters, the eccentricities of sit light curves alone, finding good agreement with the individual planets. They derived expressions linking the nominal value from RV measurements. We also measure orbital eccentricity to the transit duration and presented the eccentricity of a transit signal injected into both short predicted posterior distributions of eccentricity and ω for and long cadence Kepler data and of Kepler Object of In- a given ratio of: 1) the measured total transit duration terest (KOI) 686.01 from long-cadence, publicly-available (i.e. from first to fourth contact, including ingress and +0.18 Kepler data, finding an eccentricity of e =0.62−0.14. In egress) T14 to 2) the T14 expected for a planet on a circu- 5, we present our program of “distilling” highly eccen- lar orbit with the same b, stellar density ρ⋆, and P . Then tric§ Jupiters from the KOI sample and we conclude ( 6) they showed how the distribution of planetary transit du- with prospects for further applications of the photoec-§ rations reveals the underlying eccentricity distribution. centric effect. FQV08 focused on the possibility of measuring the ec- centricity distribution of terrestrial planets, which has 2. PRECISE ECCENTRICITIES FROM LOOSE implications for habitability. Here we will show that the CONSTRAINTS ON STELLAR DENSITY technique they describe for measuring individual planet To first order, a transiting planet’s eccentricity and its eccentricities is particularly well-suited for Jupiter-sized host star’s density depend degenerately on transit light Photoeccentric Effect 3 curve observables. Kipping et al. (2012) harnessed the power of multiple planets transiting the same host star to break this degeneracy (see also Ragozzine & Holman 2010). Yet, as FQV08 first pointed out, although the transit observables depend on the stellar density, this dependence is weak (the ratio of the planet’s semi-major axis to the 1/3 stellar radius a/R⋆ ρ⋆ ). Thus a loose prior on the stellar density should allow for a strong constraint on the eccentricity. ∝ In the limit of a constant star-planet distance during transit and a non-grazing transit (such that the transit is approximately centered at conjunction), Kipping (2010a) derived the following expression (Kipping 2010a Equations 30 and 31) for T14, the duration from first to fourth contact (i.e. the total transit duration including ingress and egress), and for T23, the duration from first to third contact (i.e. the full transit duration during which the full disk of the planet is inside the disk of the star):
1−e2 2 3/2 (1 + / δ1/2)2 (a/R )2( )2 cos2 i P (1 e ) − − ⋆ 1+e sin ω T = − arcsin 2 (1) 14/23 π (1 + e sin ω)2 q 1−e (a/R⋆) 1+e sin ω sin i
2 where P is the orbital period; e is the eccentricity; ω is the argument of periapse; R⋆ is the stellar radius; δ = (Rp/R⋆) is the fractional transit depth with Rp the planetary radius; a is the semi-major axis; and i is the inclination. By combining T14 and T23, we can rewrite Equation (1) as 2 2 1/2 2 2 π [1 + e sin ω] 2 π [1 + e sin ω] 4δ (1 + e sin ω) sin ( T14) sin ( T23)= (2) P (1 e2)3/2 − P (1 e2)3/2 sin2 i (a/R )2(1 e2)2 − − ⋆ − Using the small angle approximation, which is also used by Kipping (2010a), allows us to group the transit light curve observables on the right-hand side: a 2δ1/4P g(e,ω) sin i = (3) R⋆ π T 2 T 2 14 − 23 where p 1+ e sin ω g(e,ω)= (4) √1 e2 − The g notation is inspired by Kipping (2010a) and Kipping et al. (2012)’s variable Ψ, for which Ψ = g3. Dynamically, g is the ratio of the planet’s velocity during transit (approximated as being constant throughout the transit) to the speed expected of a planet with the same period but e = 0. Note that ω is the angle of the periapse from the sky plane, such that ω = 90◦ corresponds to a transit at periapse and ω = 90◦ to a transit at apoapse. For a given − P and δ, T14 and T23 are shortest (longest) and g largest (smallest) when the planet transits at periapse (apoapse). Moreover, if we approximate sin i = 1, we can rewrite Equation (3) as: a 2δ1/4P g(e,ω)= (5) R⋆ π T 2 T 2 14 − 23 Finally, using Kepler’s third law and assuming that the planetp mass is much less than the stellar mass (Mp M⋆), ≪ the transit observables can be expressed in terms of the stellar density ρ⋆: −3 ρ⋆(e,ω)= g(e,ω) ρcirc (6) where 3 2δ1/4 3P ρcirc = ρ⋆(e =0)= 2 (7) " T 2 T 2 # Gπ 14 − 23 Although Equation 6 was derived under several stated approximatp ions, the relationships among ρ⋆, e, and ω are key to understanding how and to what extent we can constrain a transiting planet’s eccentricity using a full light curve model. Because g(e,ω) is raised to such a large power, a small range of g(e,ω) corresponds to a large range in the ratio ρ⋆/ρcirc, i.e. the ratio of the true stellar density to the density measured from fitting a circular transit light curve model. For instance, the assumed value of ρ⋆ would need to be in error by two orders of magnitude to produce ◦ the same effect as a planet with e = 0.9 and ω = 90 . Thus the ρcirc derived from the transit light curve strongly constrains g, even with a weak prior on ρ , because g ρ1/3. ⋆ ∝ ⋆ 4 Dawson and Johnson
2.1. Constraints on ρcirc from the light curve: common rameters of an eccentric planet transiting near periapse. concerns The ingress and egress durations would be longer than One might worry that long-cadence data, such as the expected, and the inferred maximum impact parameter to avoid a grazing orbit (i.e. 1 R /R ) would be too 30-minute binning of most Kepler light curves, cannot − p ⋆ resolve the ingress and egress times sufficiently to con- large. Both of these effects would caused the planet’s or- strain a/R , or equivalently ρ . In other words, one bit to appear less eccentric (or, equivalently, for ρcirc to ⋆ circ appear smaller; see Kipping & Tinetti 2010 for a formal might worry that a/R⋆ is completely degenerate with b, and hence that the denominator of Equation (5) is un- derivation of the effect of blending on the measurement constrained. This is often the case for small planets. of a/R⋆). Therefore, dilution would not cause us to over- However, Jupiter-sized planets have high signal-to-noise estimate a planet’s eccentricity, if the transit duration is transits and longer ingress and egress durations (due to shorter that circular. Moreover, because ρcirc depends the large size of the planet). See 2.1 of FQV08 for an only weakly on the transit depth (Equation 7), the effect analysis of how the precision of Kepler§ data affects con- of blending on the eccentricity measurement is small. We straints on the total, ingress, and egress durations. quantify this effect through an example in the next sub- Furthermore, even if the ingress is unresolved or poorly section. resolved, it is often impossible for the impact parameter Furthermore, if we were to mistakenly attribute an b to account for the short duration of a highly eccen- apparently overly-long transit caused by blending to a tric, Jupiter-sized planet’s non-grazing transit. The max- planet transiting near apoapse, the resulting false eccen- tricity would be quite small. Imagine that the planet is imum non-grazing impact parameter is 1 RP /R⋆ . 0.9 for a Jupiter around a Sun-like star.− Imagine that on a circular orbit, but that the blend causes us to mea- sure ρ = (1 f)ρ , where 0
* 3. We compute ρ / 1.00 R⋆ 1+ e sin ω circ probng = (1 Rp/R⋆) 2 , (9) ρ a − 1 e − where probng is the probability of a non-grazing 0.10 transit, for each point (e,ω) (Winn 2010, Equation 9). We generate a uniform random number be- tween 0 and 1 and discard the point if the random number is greater than the transit probability. 0.01 4. We calculate the periapse distance a (1 e) for R⋆ − each grid point and drop the point if the planet’s 0 100 200 300 periapse would be inside the star (effectively impos- ω (deg) ing a physically-motivated maximum eccentricity, which is most constraining for small a/R⋆). Fig. 1.— The ratio of the circular density to the nominal stellar 5. We downsample to a subset of grid points that fol- density, ρcirc/ρ⋆, required for a circular model to account for the transit observables of an eccentric planet. The ratio is plotted as lows a normal distribution centered on g, with a a function of the planet’s argument of periapse. The solid (dotted, width of σg/g = 0.1, corresponding to a 30% un- dashed) line corresponds to a planet with an eccentricity of 0.95 certainty in the stellar density. To do this, we cal- (0.9, 0.8). For a large range of periapse angles, one would infer a density much larger than the nominal value if one modeled the culate the probability eccentric planet’s orbit as circular. 1 [g(e,ω) g]2 prob = exp − (10) g σ √2π − 2σ2 g g 6 Dawson and Johnson
and discard the point (e,ω) if a uniform random parameters and their uncertainties. Therefore, we can number is greater than probg. rewrite the equation as: We plot the resulting posterior e vs. ω distributions in Figure 2 for two a/R⋆, one large and one small, and prob(e,ω,ρ⋆,X D) prob(D e,ω,ρ⋆,X)prob(ρ⋆) (12) Rp = 0.1. The banana shape of the posterior results | ∝ | from the correlation between e and ω (i.e. Equation 4). Next we marginalize over X and ρ⋆ to obtain The posteriors reveal that, rather than being inextri- cably entwined with ω, the eccentricities deduced from g prob(e,ω D) prob(D e,ω,ρ ,X)prob(ρ )dXdρ | ∝ | ⋆ ⋆ ⋆ are well constrained. A ρcirc consistent with the nominal Z Z value (g = 1 with ρ⋆ constrained to within 30%) is more (13) likely to correspond to a small e (e.g. the probability the two-dimensional joint posterior distribution for ec- that e < 0.32 is 68.3% for a/R⋆ = 10 and that e < 0.35 centricity and ω. The first term under the integral is the is 68.3% for a/R⋆ = 300), while circular densities incon- likelihood of the data given e, ω, ρ⋆ and X. Thus a uni- sistent with the normal values (g significantly different form prior on both these quantities naturally accounts for the transit probability because prob(D e,ω,ρ ,X) is from unity) have a well-defined minimum e, above which | ⋆ the eccentricity posterior falls off gently. For example, for the transit probability; for certain values of e and ω, the g =2.5 and a/R⋆ = 300, the probability that e> 0.69 is observed transit D is more likely to occur. Combina- 99%. Furthermore, the eccentricity is likely to be close to tions of parameters that produce no transits are poor this minimum eccentricity because the range of possible models, resulting in a low likelihood of the data. Evalu- ation of the likelihood prob(D e,ω,ρ ,X) is part of how ω narrows as e 1. For g = 2.5 and a/R⋆ = 300, the | ⋆ probability that→ 0.69 0.8 0.8 0.6 0.6 ecc 0.4 0.4 a/R* = 300 0.2 0.2 0.0 0.0 0.8 0.8 0.6 0.6 ecc 0.4 0.4 a/R* = 10 0.2 0.2 0.0 0.0 0 100 200 0 100 200 0 100 200 0 100 200 0 100 200 ω (deg) ω (deg) ω (deg) ω (deg) ω (deg) Fig. 2.— Contoured eccentricity vs. ω posteriors from Monte Carlo simulations for representative values of g. The points follow a normal distribution centered at the indicated value of g (columns) with a width of 10%, corresponding to a 30% uncertainty in ρ⋆. We show the posteriors for two values of a/R⋆ (rows). The black (gray, light gray) contours represent the {68.3, 95, 99}% probability density levels (i.e. 68% of the posterior is contained within the black contour). Over-plotted as a black-and-white dotted line are histograms illustrating the eccentricity posterior probability distribution marginalized over ω. of the distributions in Figure 2 and 3 have ba- priate if we wish to avoid assumptions about the nana shapes, which often cause conventional chi- magnitude of the eccentricity. Here we implement squared minimization algorithms to remain stuck the prior through regularization (i.e. as an extra in the region of parameter space where they be- term in the jump probability). gan. In contrast, an MCMC exploration will even- tually fully sample the posterior distribution. (See 4. It automatically accounts for the transit probabil- Chib & Greenberg 1995, for a pedagogical proof of ity, because jumps to regions of parameter space this theorem.) Because of the “banana-shaped” e that do not produce a transit are rejected. To ad- vs. ω posterior for high eccentricities (Figure 2 dress what may be a misconception, we empha- and 3), conventional MCMC algorithms, like TAP, size that it is unnecessary — and actually a double require many iterations to converge and fully ex- penalty — to impose transit probability priors on plore parameter space. In our case, we test for the eccentricity or periapse. convergence by plotting e and ω each as a func- tion of chain link and assess if the exploration 5. It provides uncertainties that are more reliable than appears random. We also check to ensure that the estimates based on a simple covariance matrix the ω posterior is symmetric about ω = 90◦. (as obtained from traditional least-squares mini- Asymmetry indicates that the chains have not yet mization) because there is no assumption that the converged. We note that the variables e cos ω uncertainties are normally distributed. The un- and e sin ω also have a banana-shaped posterior. certainties fully account for complicated parame- When feasible, we recommend implementing an ter posteriors and correlations. Therefore we can affine-invariant code such as emcee that more ef- be confident in the constraints on ρcirc even when ficiently explores banana-shaped posteriors (e.g. the ingress and egress are not well-resolved. Foreman-Mackey et al. 2012). In 3.3.1, we de- scribe how to speed up the fit convergence§ by using We caution that although this Bayesian framework g instead of e as a variable while maintaining a uni- is appropriate for obtaining the posteriors of a single form prior in e and ω. planet, selection effects must be carefully considered when making inferences about a population. 3. It allows us to easily impose priors on certain pa- 3.3.1. rameters, such as the stellar density. If desired, one Using g as a variable for faster convergence can impose a prior on the eccentricity. In 4, we Using g (Equation 4) instead of e as a variable in the perform an additional fit for each dataset using§ a transit fit model avoids the MCMC having to explore a Jeffrey’s3 prior on the eccentricity, which is appro- banana-shaped posterior. The g variable allows for faster convergence and prevents the chain from getting stuck. 3 We use a true Jeffrey’s prior prob(e) ∝ 1/e, which we have In order to preserve a uniform prior in e and ω, we must not normalized because we only consider the ratio of probabilities impose a prior on g by adding an additional term to the when assessing a jump in an MCMC chain. For the fits in Section 4, for which emin is well above 0, this prior is sufficient. However, likelihood function. Following the Appendix if e = 0 is a possibility (i.e. for g near 1), the reader may wish 0 0 from the uncertainty in ρcirc and ρ⋆ and solving Equation (4) for to use a modified Jeffrey’s prior, prob(e) ∝ 1/(e + e ), where e ◦ is the noise level. We recommend estimating an upper limit on g e0 using ω = 90 . 8 Dawson and Johnson σρ/ρ = .01 σρ/ρ = .1 σρ/ρ = .2 σρ/ρ = .5 σρ/ρ = 1 0.8 0.8 0.6 0.6 ecc 0.4 0.4 g = 2.5 0.2 0.2 0.0 0.0 0.8 0.8 0.6 0.6 ecc 0.4 0.4 g = 1 0.2 0.2 0.0 0.0 0 100 200 0 100 200 0 100 200 0 100 200 0 100 200 ω (deg) ω (deg) ω (deg) ω (deg) ω (deg) Contoured eccentricity vs. ω posteriors from Monte Carlo simulations for representative values of g (rows; the points follow a normal Fig. 3.— distribution centered g) and uncertainty in ρ⋆ (columns), all for a/R⋆ = 30. The black (gray, light gray) contours represent the [68.3,95,99]% probability density levels. Over-plotted as a black-and-white dotted line are histograms illustrating the eccentricity posterior probability distribution marginalized over ω. of Burke et al. (2007), the transformation from a uniform prior in e to a prior in g is: ∂e prob(g)dg = prob(e) dg ∂g ∂e sin2 ω sin2 ω 1 + g2 1 + sin2 ω 2g sin ω sin2 ω 1+ g2 prob(g) = prob(e) = − ± − 2 2 2 ∂g sin ω 1+ g2 g2 + sin ω p − p (15) where we have assumed prob(e) = 1 and for which the + corresponds to g > 1 and the to g < 1. . Therefore, we add the following term to the log likelihood: − sin2 ω sin2 ω 1 + g2 1 + sin2 ω 2g sin ω sin2 ω 1+ g2 ∆ = ln − ± − (16) 2 2 2 L " sin ω 1+ g2 g2 + sin ω p # − We demonstrate the use of this variablep in 4. We note that in our light curve fits, we use g only to explore parameter space, transforming the variable to e in§ order compute the Keplerian orbit, with no approximations, for the Mandel & Agol (2002) light curve model. 3.4. Obtaining the eccentricity posterior from the circular-fit posterior The Monte Carlo exploration in 3.1 was meant to give us a handle on what the eccentricity and ω posterior § should look like and how they are affected by uncertainty in ρ⋆. However, one could use a more formal version of this exploration to obtain posteriors of eccentricity and ω directly from the posteriors derived from circular fits to the light curve, an approach that was adopted by Kipping et al. (2012). One could maximize the following likelihood for the parameters ρ⋆, e, and ω: 3 2 2 1 [g(e,ω) ρ⋆ ρcirc] 1 [ρ⋆ ρ⋆,measured] = 2 − −2 + ln probng (17) L −2 σρcirc − 2 σρ⋆,measured The first term in the likelihood function demands agreement with the ρcirc derived from the circular fit to the light curve. If the ρcirc posterior is not normal, one could replace this term with the log of the probability of 3 g(e,ω) ρ⋆ given the ρcirc posterior. Note that g(e,ω) can either be computed from the approximation in Equation (4) or by solving and integrating Kepler’s equation to obtain the mean ratio of the transiting planet’s velocity to its Keplerian velocity over the course of the transit. The second term is the prior on ρ⋆ from the stellar parameters independently measured from spectroscopy (or asteroseismology). The final term is the probability of a non-grazing transit (Equation 9). If one uses the variable g instead e, one should add Equation (16) to the likelihood. We warn that this likelihood function drops constants, so although it can be used to generate parameter posteriors, it should not be used Photoeccentric Effect 9 to compute the Bayesian evidence quantity. parameter (with a prior imposed to maintain a uniform In the next section, we demonstrate that this approach eccentricity prior; Equation (16) to posteriors generated yields the same eccentricity and ω posteriors as directly using: b) a Jeffrey’s prior on the eccentricity, c) e instead fitting for the eccentricity from the light curve. of g as a free parameter (to demonstrate that they are equivalent), and d) the likelihood-maximization method 4. DEMONSTRATION: MEASURING THE ECCENTRICITIES OF TRANSITING JUPITERS described in 3.4, using the posterior of ρcirc from the circular fit. The§ four sets of posteriors closely resemble To demonstrate that the duration aspect of the pho- one another. The computation times were about 1 day toeccentric effect allows for precise and accurate mea- for the circular fit, about 1 day for the eccentric fit us- surements of a transiting planet’s eccentricity from the ing g as a parameter, several days for the eccentric fit light curve alone, we apply the method described in 3 using e as a parameter, and thirty minutes for the like- to several test cases. In 4.1 we measure the eccentricity§ § lihood maximization method of 3.4. Note that the fi- of a transiting planet that has a known eccentricity from nal method requires the best-fitting§ parameters resulting RV measurements. In 4.2 we inject a transit into short § from a circular fit to the light curve, including accurate and long cadence Kepler data and compare the resulting parameter posteriors. We therefore caution against using e and ω posteriors. In 4.3, we measure the eccentricity § the parameters listed in the Kepler public data releases of a Kepler candidate that has only long-cadence data for this purpose because those values are the result of available. a least-squares fit and make the assumption of normally 4.1. HD 17156 b: a planet with a large eccentricity distributed parameter uncertainties. However, if one has measured from RVs already precomputed circular fits using an MCMC algo- rithm that incorporates red noise and limb darkening—as HD 17156b was discovered by the Next 2000 Stars we have done for all of the Jupiter-sized KOIs ( 5)—the (N2K) Doppler survey (Fischer et al. 2005, 2007). final method ( 3.4) is advantageous because of§ the de- Fischer et al. (2007) reported that the planet has a large creased computation§ time. orbital eccentricity of e =0.67 0.08. We identified this planet and the relevant references± using exoplanets.org (Wright et al. 2011). Barbieri et al. (2007) reported several partial transits observed by small-telescope observers throughout the Northern Hemisphere, and Barbieri et al. (2009) and Winn et al. (2009) observed 1.00 full transits using high-precision, ground-based photom- etry. Here we demonstrate that the planet’s eccentricity could have been measured from the transit light curve 0.99 data alone. We simultaneously fit three light curves (Figure 4), one from Barbieri et al. (2009) and two from Winn et al. (2009) using TAP (Gazak et al. 2011), which employs 0.98 an MCMC technique to generate a posterior for each pa- rameter of the Mandel and Agol (2002) transit model. Time-correlated, “red” noise is accounted for using the 0.97 Carter & Winn (2009) wavelet-based likelihood function. To achieve the 2N (where N is an integer) data points re- quired by the wavelet-based likelihood function without 0.96 excessive zero-padding, we trimmed the first Winn et al. normalized detrended flux (2009) light curve from 523 data points to 512 data points by removing the last 11 data points in the time series. 0.95 Initially, we fixed the candidate’s eccentricity at 0 and fit for ρcirc with no prior imposed, to see how much it differs from the well-measured value of ρ⋆. Then we refitted the transit light curves with a normal prior imposed on the 0.94 stellar density, this time allowing the eccentricity to vary. −2 0 2 4 In both cases, we treated the limb darkening coefficients (time − mid transit) [hrs] following the literature: we fixed the coefficients for the Barbieri et al. (2009) light curve and left the coefficients Fig. 4.— Light curves of HD 17156 from Barbieri et al. (2009) free for the Winn et al. (2009) light curves. Following (top) and Winn et al. (2009) (middle, bottom). A set of eccentric Winn et al. (2009), we also included linear extinction free model light curves drawn from the posterior are plotted as solid lines. parameters for the two Winn et al. (2009) light curves. (The published Barbieri et al. 2009 light curve was al- ready pre-corrected for extinction.) Based on the circular fit alone, we would infer Figure 5 shows posterior distributions from a circular g(emin,π/2) = 2.0, corresponding to a minimum eccen- tricity of emin = 0.61. From the eccentric fit, we obtain fit (top row) and an eccentric fit (bottom row) with a +0.16 prior imposed on the stellar density from Gilliland et al. a value of e = 0.71−0.09 using a uniform prior on the +0.16 (2011). In Figure 6, we compare the posteriors generated eccentricity and e = 0.69−0.09 using a Jeffrey’s prior. from a) the eccentric fit to the light curve using g as a Therefore, we could have deduced the eccentricity deter- 10 Dawson and Johnson 1.0 ing planet’s eccentricity to high precision and accuracy. 1.0 0.5 Relative N 0.5 0.0 0.01 0.10 1.00 10.00 0.2 0.4 0.6 0.8 Relative N ρ ρ ρ circ e from circ/ * 0.0 0 100 200 0.5 0.6 0.7 0.8 0.9 1.0 ω (deg) e Fig. 7.— Posterior distributions of e and ω for the HD 17156 transiting system, with three different priors on the stellar density: 0.5 the density measured by Gilliland et al. (2011) (solid); the density measured by Gilliland et al. (2011) with uncertainties enlarged to Relative N σρ⋆ /ρ⋆ = 0.2, (dashed) and the density based on the stellar pa- 0.0 rameters from Winn et al. (2009) (dotted). 0 100 200 0.2 0.4 0.6 0.8 ω (deg) e 4.2. Short vs. long cadence Kepler data Fig. 5.— Posterior distributions of e and ω for the HD 17156 transiting system, with eccentricity fixed at 0 (row 1) and free to Kipping (2010b) explored in detail the effects of long vary (row 2). Row 1: Left: ρ⋆ derived from circular fit. The solid integration times and binning on transit light curve mea- line marks the nominal value. Right: Posterior distribution for eccentricity solving Equation (5) for ω = 0 (solid line), ω = 45◦ surements, with a particular focus on long-cadence Ke- (dashed line), and ω = 90◦ (dotted line). Row 2: Left: Posterior pler data. He demonstrated that by binning a finely- distribution for ω from eccentric fit (i.e. a fit to the light curve sampled model to match the cadence of the data, as TAP in which the eccentricity is a free parameter; solid). Gaussian has implemented, one can fit accurate (though less pre- illustrating posterior from Fischer et al. (2007) RV fit (dotted line). Right: Same for eccentricity posterior. cise than from short cadence data) light curve param- eters. Using short and long cadence Kepler data of a planet with known parameters (TrES-2-b), he validated 1.0 this approach. Here we explore, through a test scenario of an eccentric planet injected into short and long Kepler data, whether 0.5 this approach holds (as one would expect) for fitting an eccentric orbit and what value short-cadence data adds to the constraint on eccentricity. We chose parameters for the planet typical of an eccentric Jupiter and main- 0.0 ◦ sequence host star: P = 60 days, i = 89.5 , Rp/R⋆ = 0 100 200 0.5 0.6 0.7 0.8 0.9 0.1, e = 0.8, ω = 90◦, M = R = 1, and limb darkening ω (deg) e ⋆ ⋆ parameters µ1 = µ2 = 0.3. We considered the situation Fig. 6.— Left: Posterior distribution for ω for a fit to the light in which long cadence data is available for Q0-Q6 but curve using g as a free parameter with a uniform prior on the eccen- short-cadence is available only for one quarter (or may tricity (sold line) and Jeffrey’s prior (dotted line). Posterior distri- bution using e instead of ω as a free parameter (dot-dashed line). be in the future). We retrieved Q0-Q6 data from the Posterior distribution using method described in §3.4 (dashed line). Multimission Archive at the Space Telescope Science In- Right: Same as left, for eccentricity posterior. stitute (MAST) for Kepler target star KIC 2306756, se- lected because it has both long and short cadence data. mined from 33 RV measurements — e = 0.67 0.08 Then we applied the TAP MCMC fitting routine to fit a) (Fischer et al. 2007) — from these three transit± light one short-cadence transit (fixing the period at 60 days) curves alone. that took place in a single segment of short-cadence data The host star has a particularly well-constrained den- and b) all seven long-cadence transits. As in 4.1, we performed one set of fits fixing the or- sity from asteroseismology (Gilliland et al. 2011). We ar- § tificially enlarge the error bars on the stellar density from bit as circular and another set with g and ω as free pa- 1% to 20% and repeat the fitting procedure, obtaining an rameters, imposing a prior on the stellar density corre- +0.14 sponding to a 20% uncertainty in the stellar density and eccentricity of e = 0.70−0.09. We also repeat the fitting procedure with a density derived from the stellar param- a prior on g from a uniform prior in e and ω (Equation 16). In both cases, we allowed the limb darkening to be a eters M⋆ and R⋆ determined by Winn et al. (2009) from isocrone fitting. This “pre-asteroseismology” density has free parameter. We plot the resulting posterior distribu- an uncertainty of 10% and, moreover, is about 5% larger tions of eccentricity and ω in Figure 9. From the circular fits, the constraint on ρcirc is somewhat stronger from than the value measured by Gilliland et al. (2011). We +1.0 obtain an eccentricity of e = 0.70+0.16. In Figure 7 and the short cadence data (26.3−1.6 ρ⊙) than from the long −0.11 +1.0 8, we plot the resulting posterior distributions, which are cadence data (25.9−2.7 ρ⊙), as Kipping (2010b) found. very similar. Therefore, even with uncertainties and sys- From the short cadence data, we measure an eccentricity +0.08 tematics in the stellar density, we can measure a transit- of e =0.85−0.05 with a uniform prior on the eccentricity Photoeccentric Effect 11 1.0 0.8 e 0.6 0.4 0 100 200 0 100 200 0 100 200 ω (deg) Fig. 8.— Eccentricity vs. ω posterior distributions for HD 17156 b based on fits using a prior on the stellar density from Gilliland et al. (2011) (left); Gilliland et al. (2011) with error bars enlarged to 20% (middle); and Winn et al. (2009) (left). +0.07 and e = 0.85−0.05 with a Jeffrey’s prior. From the long posterior distributions from a circular fit (top row) and +0.08 an eccentric fit (bottom row) with a prior imposed on cadence data, we measure an eccentricity of e =0.84−0.05 +0.07 the stellar density. We measure the eccentricity to be with a uniform prior on the eccentricity and e =0.84−0.04 +0.18 with a Jeffrey’s prior. Therefore, the long cadence data e =0.62−0.14. is sufficient to obtain a precise eccentricity measurement. We caution that this candidate has not yet been In this case, the 20% uncertainty in the stellar den- validated; Morton & Johnson (2011) estimate a false- sity dominated over the constraint from the transit light positive probability of 8%. If the candidate is a false positive, its orbit (and other properties, such as its ra- curve on ρcirc; however, for very well-constrained stellar properties, we would expect the greater precision of the dius) is likely to be different from that inferred. However, short cadence data to allow for a tighter constraint on we note that if the candidate is a background binary or the eccentricity (see Figure 3). hierarchical triple and is actually larger than a planet, the inferred eccentricity would actually be higher (i.e. 4.3. KOI 686.01, a moderately eccentric, Jupiter-sized if the candidate is actually larger, it must be moving Kepler candidate through its ingress and egress even faster), unless KOI 686 is not the primary and the primary has a higher den- KOI 686.01 was identified by Borucki et al. (2011) and sity than KOI 686. Another possibility, if the candidate Batalha et al. (2012) as a 11.1 REarth candidate that is false positive, is that the assumption of M M may transits its host star every 52.5135651 days. We retrieved p ⋆ no longer hold and ρ◦ (Equation 6) should be≪ compared the Q0-Q6 data from MAST and detrended the light to ρ + ρ rather than ρ tar to obtain g. How- curve using AutoKep (Gazak et al. 2011). We plot the ⋆ companion s ever, even if ρcompanion ρ⋆, the error in g would be only light curves in Figure 10. 1 3 ∼ We obtained a spectrum of KOI 686 using the HIgh ( 2 ) = 12.5%. Resolution Echelle Spectrometer (HIRES) on the Keck Santerne et al. (2012) recently found a false positive I Telescope (Vogt et al. 1994). The spectrum was ob- rate of 35% for Jupiter-sized candidates, comprised of tained with the red cross-disperser and 0′′. 86 slit using the brown dwarfs, undiluted eclipsing binaries, and diluted standard setup of the California Planet Survey (CPS), eclipsing binaries. In the case of diluted eclipsing bi- naries, the blend effects that we discussed in 2 could but with the iodine cell out of the light path. The ex- § tracted spectrum has a median signal-to-noise ratio of be larger than we considered. However, Morton (2012) 40 at 5500 A˚, and a resolution λ/∆λ 55, 000. To notes that most of the false positives that Santerne et al. estimate the stellar temperature, surface≈ gravity, and (2012) discovered through radial-velocity follow-up al- metallicity, we use the SpecMatch code, which searches ready exhibited V-shapes or faint secondary eclipses in through the CPS’s vast library of stellar spectra for stars their light curves. In the search for highly eccentric with Spectroscopy Made Easy (SME; Valenti & Piskunov Jupiters, we recommend a careful inspection of the tran- 1996; Valenti & Fischer 2005) parameters and finds the sit light curve for false-positive signatures and, when pos- best matches. The final values are the weighted mean sible, a single spectroscopic observation and adaptive- of the 10 best matches. We then interpolate these stel- optics imaging to rule out false-positive scenarios. lar parameters onto the Padova stellar evolution tracks If the planetary nature of this object is confirmed, to obtain a stellar mass and radius. We checked these it will be one of a number of Jupiter-sized planets values using the empirical relationships of Torres et al. with orbital periods of 10-100 days and moderate ec- +0.45 centricities, but the first in the Kepler sample with a (2010). We find ρ = 1.02 ρ⊙ (the other stellar ⋆ −0.29 photometrically-measured eccentricity. Many previously parameters for this KOI and parameters for other KOI known, moderately-eccentric planets have orbits inside will be published as part of another work, Johnson et al. the snow line; their eccentricities are thought to be sig- 2012, in prep). natures of the dynamical process(es) that displaced them We then fit circular and eccentric orbits to the transit from their region of formation. light curve, as described above, binning the model light curves to match the 30-minute cadence of the data. We 5. impose a normal prior on the limb-darkening coefficients A PLAN FOR DISTILLING HIGHLY-ECCENTRIC JUPITERS FROM THE Kepler SAMPLE based on the values from Sing (2010). Figure 11 shows 12 Dawson and Johnson 1.0 0.5 Relative N 0.0 1 10 0 100 200 0.7 0.8 0.9 ρ ω circ (deg) e Fig. 9.— Posterior distributions of e and ω for an injected, artificial transit, with eccentricity fixed at 0 (panel 1) and free to vary (panel 2-3). The sold curves are from a fit to seven light curves from the long-cadence data and the dotted to a single light curve from the short cadence data. Left: ρ⋆ derived from circular fit. The dashed curve represents the nominal value and its uncertainty. Middle: Posterior distribution for ω from eccentric fit (solid line). Right: Eccentricity posterior. To test the HEM hypothesis (S12), we are “distill- transit light curve, as we did in 4.1, for HD 17156b us- ing” highly-eccentric, Jupiter-sized planets — proto hot ing ground-based photometry. Comparing§ this technique Jupiters — from the sample of announced Kepler can- to Kipping et al. (2012)’s MAP, MAP is more model in- didates using the publicly released Kepler light curves dependent – requiring no knowledge at all of the stellar (Borucki et al. 2011; Batalha et al. 2012). To identify density – but our technique is applicable to single tran- planets that must be highly eccentric, we are refitting the siting planets, as Jupiter-sized Kepler candidates tend Kepler light curves of all the Jupiter-sized candidates us- to be (e.g. Latham et al. 2011). We are the process of ing the TAP. Initially, we fix the candidate’s eccentricity fitting the orbits of all Jupiter-sized Kepler candidates, at 0. We identify candidates whose posteriors for ρcirc are which will lead to the following prospects: wildly different than the nominal value ρ⋆ from the KIC. From this subset of objects, we obtain spectra of the host 1. For candidates with host stars too faint for RV stars. We refine the stellar parameters using SpecMatch, follow-up (65% of candidates in Borucki et al. 2011 interpolate them onto the Padova stellar evolution tracks are fainter than Kepler magnitude 14), our tech- to obtain a stellar mass and radius, and check the in- nique will provide an estimate of the planet’s ec- ferred M⋆ and R⋆ using the empirical relationships of centricity. We may also be able to deduce the Torres et al. (2010). We validate the candidate using the presence of companions from transit timing vari- method outlined in Morton (2012). Finally, we refit the ations, thereby allowing us to search for “smoking transit light curves with a prior imposed on the stellar gun” perturbers that may be responsible for the in- density, this time allowing the eccentricity to vary. This ner planet’s orbital configuration. In a companion process will allow to us easily identify the most unam- paper (Dawson et al. 2012, in prep), we present biguous highly-eccentric hot Jupiters. the validation and characterization of a KOI with 6. DISCUSSION a high, photometrically-measured eccentricity and transit timing variations. Measuring a transiting planet’s orbital eccentricity was once solely the province of radial-velocity obser- 2. For candidates bright enough for follow-up RV vations. Short-period planets were discovered by tran- measurements, the eccentricity and ω posteriors sits and followed-up with RVs, which sometimes re- from photometric fits allow us to make just a few vealed a sizable eccentricity (e.g. HAT-P-2b, Bakos et al. optimally timed radial velocity measurements to 2007; CoRoT-10b, Bonomo et al. 2010). Long-period pinpoint the planet’s eccentricity, the mass and planets—which, based on the RV distribution, are more host-star density, instead of needing to devote pre- commonly eccentric—were discovered by radial-velocity cious telescope time to sampling the full orbital pe- measurements and, on lucky occasions, found to tran- riod. The tight constraints on eccentricity from sit (e.g. HD 17156 b, Fischer et al. 2007, the planet photometry alone can be combined with radial- discussed in 4.1, as well as HD 806066 b, Naef et al. velocity measurements to constrain the candidate’s § 2001). But now, from its huge, relatively unbiased tar- orbit—either by fitting both datasets simultane- get sample size of 150,000 stars, Kepler has discovered ously or by using the posteriors from the photom- a number of long-period, transiting candidates. Among etry as priors for fitting a model to the RVs. To these are likely to be a substantial number of eccentric maximize the information gain, the prior on the planets (S12), which have enhanced transit probabilities stellar density should remain in place. This serves (Kane et al. 2012). Moorhead et al. (2011); Kane et al. as an additional motivation for measuring the spec- (2012) and Plavchan et al. (2012) have characterized the troscopic properties of candidate host stars in the eccentricity distributions of these candidates based on Kepler field. Kepler photometry. Kipping et al. (2012) are employing MAP to measure the eccentricities of planets in systems 3. We can also measure the spin-orbit angles of in which multiple planets transits. Here we have demon- the candidates orbiting the brightest stars with strated that it is also possible to constrain an individual Rossiter-McLaughlin measurements. Then we can planet’s eccentricity from a set of high signal-to-noise compare the distribution of spin-orbit angles of transits using a Bayesian formalism that employs rela- those planets we have identified as eccentric with tively loosely-constrained priors on the stellar density. the distribution of those we have constrained to be The technique we have presented can be applied to any most likely circular. Photoeccentric Effect 13 1.0 KOI−686 0.5 1.00 Relative N 0.0 0.1 1.0 10.0 0.2 0.4 0.6 0.8 ρ ρ ρ circ e from circ/ * 0.98 1.0 0.5 0.96 Relative N 0.0 0 100 200 0.2 0.4 0.6 0.8 ω (deg) e Fig. 11.— Posterior distributions for KOI 686.01 with eccentric- 0.94 ity fixed at 0 (row 1) and free to vary (row 2). Row 1: Left: ρ⋆ derived from circular fit. The solid line marks the nominal value. Right: Posterior distribution for eccentricity solving Equation (5) for ω = 0 (solid line), ω = 45◦ (dashed line), and ω = 90◦ (dotted line). Row 2: Left: Posterior distribution for ω from eccentric fit (solid). Posterior distribution using method from §3.4 (dotted). Right: Same as left, for eccentricity posterior. 0.92 The Kepler sample has already revealed a wealth of in- normalized detrended flux formation about the dynamics and architectures of plan- etary systems (e.g. Lissauer et al. 2011; Fabrycky et al. 2012) but primarily for closely-packed systems of low mass, multiple-transiting planets. Measuring the eccen- tricities of individual, Jupiter-sized planets in the Kepler 0.90 will allow us to investigate a different regime: plane- tary systems made up of massive planets that potentially underwent violent, mutual gravitational interactions fol- lowed by tidal interactions with the host star. 0.88 We are thankful for the helpful and positive feedback from the anonymous referee. R.I.D. gratefully acknowl- edges support by the National Science Foundation Grad- uate Research Fellowship under grant DGE-1144152, clear and constant guidance from chapter Winn (2010), and the ministry and fellowship of the Bayesian Book −5 0 5 Club. J.A.J. thanks Avi Loeb and the ITC for hosting (time − mid transit) [hrs] him as part of their visitors program, thereby allowing the authors to work together in close proximity at the Fig. 10.— Light curves of KOI 686. A set of eccentric model light CfA during the completion of this work. We thank Sarah curves drawn from the posterior are plotted as solid lines. The Ballard, Zachory Berta, Joshua Carter, Courtney Dress- second-from-bottom curve is a compilation of all the light curves. The bottom points are the residuals multiplied by 10. ing, Subo Dong, Daniel Fabrycky, Jonathan Irwin, Boaz Katz, David Kipping, Timothy Morton, Norman Mur- ray, Ruth Murray-Clay, Peter Plavchan, Gregory Snyder, Aristotle Socrates, and Joshua Winn for helpful discus- 4. S12 argue that HEM mechanisms for producing sions. Several colleagues provided helpful and inspiring hot Jupiters should also produce a population of comments on a manuscript draft: Joshua Carter (who, in highly eccentric (e > 0.9) proto hot Jupiters and addition to other helpful comments, suggested the pro- predict that we should find 3-5 in the Kepler sam- cedure described in 3.4), Daniel Fabrycky, Eric Ford, ple. Moreover, Kepler’s continuous coverage may David Kipping, and§ Ruth Murray-Clay. Special thanks offer the best prospect for detecting highly eccen- to J. Zachary Gazak for helpful modifications to the TAP tric planets, against which RV surveys are biased code. (Jones et al. 2006; O’Toole et al. 2009). In 5, we This paper includes data collected by the Kepler mis- described our process for distilling highly-eccentric§ sion. Funding for the Kepler mission is provided by the Jupiters from the Kepler sample. NASA Science Mission directorate. Some of the data 14 Dawson and Johnson presented in this paper were obtained from the Multi- data is provided by the NASA Office of Space Science via mission Archive at the Space Telescope Science Institute grant NNX09AF08G and by other grants and contracts. (MAST). STScI is operated by the Association of Uni- This research has made use of the Exoplanet Or- versities for Research in Astronomy, Inc., under NASA bit Database and the Exoplanet Data Explorer at contract NAS5-26555. Support for MAST for non-HST exoplanets.org. REFERENCES Alibert, Y., Mordasini, C., Benz, W., & Winisdoerffer, C. 2005, Kataria, T., Showman, A. P., Lewis, N. K., et al. 2011, A&A, 434, 343 EPSC-DPS Joint Meeting 2011, 573 Bakos, G. A.,´ Kov´acs, G., Torres, G., Fischer, D. A., et al. 2007, Kipping, D. M. 2008, MNRAS, 389, 1383 ApJ, 670, 826 —. 2010a, MNRAS, 407, 301 Barbieri, M., Alonso, R., Desidera, S., Sozzetti, A., et al. 2009, —. 2010b, MNRAS, 408, 1758 A&A, 503, 601 Kipping, D. M., Dunn, W. R., Jasinski, J. M., & Manthri, V. P. Barbieri, M., Alonso, R., Laughlin, G., Almenara, J. M., et al. 2012, MNRAS, 2413 2007, A&A, 476, L13 Kipping, D. M., Fossey, S. J., & Campanella, G. 2009, MNRAS, Barnes, J. W. 2007, PASP, 119, 986 400, 398 Batalha, N. M., Borucki, W. J., Koch, D. G., Bryson, S. T., et al. Kipping, D. M., & Tinetti, G. 2010, MNRAS, 407, 2589 2010, ApJ, 713, L109 Latham, D. W., Rowe, J. F., Quinn, S. N., et al. 2011, ApJ, 732, Batalha, N. M., Rowe, J. F., Bryson, S. T., Barclay, T., et al. L24 2012, arXiv:1202.5852 Lissauer, J. J., Ragozzine, D., Fabrycky, D. C., Steffen, J. H., Bonomo, A. S., Santerne, A., Alonso, R., Gazzano, et al. 2010, Ford, E. B., et al. 2011, ApJS, 197, 8 A&A, 520, A65 Mandel, K., & Agol, E. 2002, ApJ, 580, L171 Borucki, W. J., Koch, D. G., Basri, G., Batalha, N., et al. 2011, Mann, A. W., Gaidos, E., L´epine, S., & Hilton, E. J. 2012, ApJ, ApJ, 736, 19 753, 90 Bowler, B. P., Johnson, J. A., Marcy, G. W., Henry, G. W., et al. Mardling, R. A. 2007, MNRAS, 382, 1768 2010, ApJ, 709, 396 Moorhead, A. V., Ford, E. B., Morehead, R. C., Rowe, J., et al. Bromley, B. C., & Kenyon, S. J. 2011, ApJ, 735, 29 2011, ApJS, 197, 1 Burke, C. J. 2008, ApJ, 679, 1566 Morton, T. D. 2012, arXiv:1206.1568 Burke, C. J., McCullough, P. R., Valenti, J. A., Johns-Krull, et Morton, T. D., & Johnson, J. A. 2011, ApJ, 738, 170 al. 2007, ApJ, 671, 2115 Naef, D., Latham, D. W., Mayor, M., Mazeh, T., et al. 2001, Carter, J. A., & Winn, J. N. 2009, ApJ, 704, 51 A&A, 375, L27 Chib, S., & Greenberg, E. 1995, American Statistician, 49, 327 Nagasawa, M., & Ida, S. 2011, ApJ, 742, 72 Dong, S., Katz, B., & Socrates, A. 2012, arXiv:1201.4399 Naoz, S., Farr, W. M., & Rasio, F. A. 2012, arXiv:1206.3529 Dressing, C. D., Spiegel, D. S., Scharf, C. A., Menou, K., & Nesvorny, D., Kipping, D. M., Buchhave, L. A., et al. 2012, Raymond, S. N. 2010, ApJ, 721, 1295 Science, 336, 1133 Fabrycky, D. C., Lissauer, J. J., Ragozzine, D., Rowe, J. F., et al. O’Toole, S. J., Tinney, C. G., Jones, H. R. A., Butler, R. P., et al. 2012, arXiv:1202.6328 2009, MNRAS, 392, 641 Fischer, D. A., Laughlin, G., Butler, P., Marcy, G., et al. 2005, Plavchan, P., Bilinski, C., & Currie, T. 2012, arXiv:1203.1887 ApJ, 620, 481 Ragozzine, D., & Holman, M. J. 2010, arXiv:1006.3727 Fischer, D. A., Vogt, S. S., Marcy, G. W., Butler, R. P., et al. Santerne, A., D´ıaz, R. F., Moutou, C., et al. 2012, 2007, ApJ, 669, 1336 arXiv:1206.0601 Ford, E. B., & Rasio, F. A. 2006, ApJ, 638, L45 Sing, D. K. 2010, A&A, 510, A21 —. 2008, ApJ, 686, 621 Socrates, A., Katz, B., Dong, S., & Tremaine, S. 2012, ApJ, 750, Foreman-Mackey, D., Hogg, D. W., Lang, D., & Goodman, J. 106 2012, arXiv:1202.3665 Spiegel, D. S., Raymond, S. N., Dressing, C. D., Scharf, C. A., & Gazak, J. Z., Johnson, J. A., Tonry, J., Eastman, J., et al. 2011, Mitchell, J. L. 2010, ApJ, 721, 1308 arXiv:1102.1036 Torres, G., Andersen, J., & Gim´enez, A. 2010, A&A Rev., 18, 67 Gilliland, R. L., McCullough, P. R., Nelan, E. P., Brown, et al. Valenti, J. A., & Fischer, D. A. 2005, ApJS, 159, 141 2011, ApJ, 726, 2 Valenti, J. A., & Piskunov, N. 1996, A&AS, 118, 595 Goldreich, P., & Tremaine, S. 1980, ApJ, 241, 425 Vogt, S. S., Allen, S. L., Bigelow, B. C., Bresee, L., et al. 1994, in Hansen, B. M. S. 2010, ApJ, 723, 285 Society of Photo-Optical Instrumentation Engineers (SPIE) Hansen, B. M. S., & Murray, N. 2012, ApJ, 751, 158 Conference Series, Vol. 2198, Society of Photo-Optical Ida, S., & Lin, D. N. C. 2008, ApJ, 673, 487 Instrumentation Engineers (SPIE) Conference Series, ed. Jackson, B., Greenberg, R., & Barnes, R. 2008a, ApJ, 678, 1396 D. L. Crawford & E. R. Craine, 362 —. 2008b, ApJ, 681, 1631 Ward, W. R. 1997, Icarus , 126, 261 Johnson, J. A., Apps, K., Gazak, J. Z., Crepp, et al. 2011, ApJ, Winn, J. N. 2010, Exoplanet Transits and Occultations, ed. 730, 79 Seager, S., 55–77 Jones, H. R. A., Butler, R. P., Tinney, C. G., Marcy, G. W., et Winn, J. N., Holman, M. J., Henry, G. W., Torres, G., et al. al. 2006, MNRAS, 369, 249 2009, ApJ, 693, 794 Juri´c, M., & Tremaine, S. 2008, ApJ, 686, 603 —. 2011b, PASP, 123, 412 Kane, S. R., Ciardi, D. R., Gelino, D. M., & von Braun, K. 2012, Wu, Y., & Lithwick, Y. 2011, ApJ, 735, 109 arXiv:1203.1631 Wu, Y., & Murray, N. 2003, ApJ, 589, 605 Kane, S. R., & von Braun, K. 2009, PASP, 121, 1096