ebook img

Uncovering the planets and stellar activity of CoRoT-7 using only radial velocities PDF

1.8 MB·
Save to my drive
Quick download
Download
Most books are stored in the elastic cloud where traffic is expensive. For this reason, we have a limit on daily download.

Preview Uncovering the planets and stellar activity of CoRoT-7 using only radial velocities

Astronomy&Astrophysicsmanuscriptno.AA_2015_27899 (cid:13)cESO2016 February4,2016 Uncovering the planets and stellar activity of CoRoT-7 using only radial velocities(cid:63) J.P.Faria1,2,R.D.Haywood3,B.J.Brewer4,P.Figueira1,M.Oshagh1,5,A.Santerne1,andN.C.Santos1,2 1 InstitutodeAstrofísicaeCiênciasdoEspaço,UniversidadedoPorto,CAUP,RuadasEstrelas,PT4150-762Porto,Portugal e-mail:[email protected] 2 DepartamentodeFísicaeAstronomia,FaculdadedeCiências,UniversidadedoPorto,RuaCampoAlegre,4169-007Porto,Portugal 3 Harvard-SmithsonianCenterforAstrophysics,60GardenStreet,Cambridge,MA02138,USA 4 DepartmentofStatistics,TheUniversityofAuckland,PrivateBag92019,Auckland1142,NewZealand 5 InstitutfürAstrophysik,Georg-August-Universität,Friedrich-Hund-Platz1,37077Göttingen,Germany 6 ReceivedDecember7,2015;acceptedJanuary27,2016 1 0 2 ABSTRACT b Stellaractivitycaninducesignalsintheradialvelocitiesofstars,complicatingthedetectionoforbitinglow-massplanets.Wepresent e amethodtodeterminethenumberofplanetarysignalspresentinradial-velocitydatasetsofactivestars,usingonlyradial-velocity F observations.Insteadofconsideringseparatefitswithdifferentnumberofplanets,weuseabirth-deathMarkovchainMonteCarlo 3 algorithmtoinfertheposteriordistributionforthenumberofplanetsinasinglerun.Inanaturalway,themarginaldistributionsfor theorbitalparametersofallplanetsarealsoinferred.ThismethodisappliedtoHARPSdataofCoRoT-7.Weconfidentlyrecoverthe ] orbitsofbothCoRoT-7bandCoRoT-7calthoughthedatashowevidenceforthepresenceofadditionalsignals. P Keywords. methods:dataanalysis–stars:planetarysystems–stars:individual:CoRoT-7–techniques:radialvelocities E . h p 1. Introduction Simultaneous photometric and RV observations have been - o successfullyusedtoconstrainactivity-inducedRVsignals.How- Imagine we have at our disposal a set of spectroscopic obser- r ever,thisapproachrequireseitherajointmodelforphotometric tvations of an unknown star, and we can obtain precise radial- s and RV variations, which can be statistical (e.g. Rajpaul et al. avelocity (RV) measurements, using the cross-correlation func- 2015) or based on a description of the stellar features inducing [tion (CCF) technique (Baranne et al. 1996; Pepe et al. 2002). the signal (e.g. Lanza et al. 2010), or a conversion from pho- 2 Theso-calledlineprofileindicators(suchasthefullwidthathalf tometric variations to RV variations (e.g. Haywood et al. 2014, maximum of the CCF, its bisector span, and associated quanti- v hereafterH14). ties, e.g. Figueira et al. 2013) and activity indicators, like the 5 Here we present a framework that models activity-induced 9logR(cid:48)HK (Noyes et al. 1984), are also usually available but are signalsascorrelatednoiseintheRVobservationsanddoesnot 4notnecessaryforouranalysis.Andwithoutotherinstrumentsat require simultaneous photometric observations. The method is 7hand,wecannotmeasurethestar’sphotometricvariations,asis basedonthefactthat,fordataofsufficientquality,itispossible 0thecaseformosttargetsinRVsurveys. todistinguishifanoscillatingsignalhastheKeplerianshapethat 1. Armed with some statistical artillery, we aim to answer the isexpectedfromaplanet,orsomeotherapproximately-periodic 0following question: how many orbiting planets can be confi- shape as expected from stellar activity. Below we describe our 6dentlydetectedinthesedata,andwhataretheirorbitalparame- model and apply it to High Accuracy Radial Velocity Planet 1tersandminimummasses? Searcher(HARPS;Mayoretal.2003)observationsofCoRoT-7. : Besides planets, other physical processes can induce varia- v itions in the radial velocities of a star. These include stellar os- Xcillations, granulation, spots and faculae/plages, and long-term 2. ABayesianmodelforRVdata rmagneticactivitycycles(see,e.g.Saar&Donahue1997;Santos aetal.2010;Boisseetal.2011;Dumusqueetal.2011a).Someof In this section we present in detail our model for RV data and these activity-induced signals can be mitigated or averaged out describethetwotypesofsignalsweconsider:planetarysignals byadaptingtheobservationalstrategy(Dumusqueetal.2011b). andactivity-inducednoise. Butsignalscausedbythepresenceofactiveregionsonthestel- larsurfacecanshowperiodicitiesandamplitudessimilartothe 2.1. RVplanetarysignals onesinducedbyrealplanetarysignals,andmaybehardertodis- entangle.Indeed,thesesignalscanevenmimicplanetarysignals The dynamical evolution of a planetary system is governed by (e.g. Figueira et al. 2010; Santos et al. 2014; Robertson et al. thegravitationalinteractionsofitsconstituentbodies.Formost 2015). systemswithmultipleplanetsonecanassume,toagoodapprox- imation,thatthemutualplanetaryperturbationsarenegligibleon (cid:63) Alldataandsoftwarepresentedinthisarticleareavailableonline timescales that are comparable to the duration of observations. athttps://github.com/j-faria/exoBD-CoRoT7. The stellar RV perturbations that are due to a multiple-planet Articlenumber,page1of7 A&Aproofs:manuscriptno.AA_2015_27899 systemcanthenbemodelledasalinearsuperpositionofKeple- rianorbits. µP wP µK Each Keplerian can be described with five RV observables: the semi-amplitude K, the orbital period P, the eccentricity e, anorbitalphaseφ,andthelongitudeoftheline-of-sight,ω.We also consider a systemic velocity, vsys, of the centre of mass of P K e η1 η2 η3 η4 thesystem,whichcorrespondstoanRVzero-pointmeasuredby Np HARPS. φ ω 2.2. Gaussianprocessestomodelstellaractivity Npplanets Even if activity signals cannot be easily described analytically vsys Σ s (a complete description would require knowledge of the active regiondistributionandevolution,temperaturecontrastandstel- larparameters,suchasinclinationandlimbdarkening),wecan makesomeassumptionsabouttheirform.Besidesassumingthe signals will be continuous and smooth, stellar rotation induces vi a periodicity or quasi-periodicity, as active regions evolve and ti vkiep σ cycleinandoutofviewonthestellarsurface. i For the purposes of detecting planets using RV measure- ments, the signals caused by stellar activity can then be seen Ndatapoints ascorrelatedquasi-periodicnoise.Gaussianprocesses(GP)are Fig.1.Representationoftherelationsbetweenparametersandobserva- anincreasinglycommontooltodealwithcorrelatednoise(e.g. tionsinourRVmodel,asaprobabilisticgraphicalmodel.Anarrowbe- Robertsetal.2012).Intheirapplicationtoregressionproblems, tweentwonodesindicatesthedirectionofconditionaldependence.The GPscanbeseenaspriordistributionsoverfunctions,whichwill circlednodesaretheparametersofthemodel,whosejointdistribution be constrained by the data (e.g. Rasmussen & Williams 2006). issampledbytheMarkovchainMonteCarlo(MCMC)algorithm.The Forourpurposes,weusetheGPtomodelthestochasticcompo- double circled node v represents the observed RVs. The filled nodes i nentofoursignal–thatis,thestellarnoise. represent deterministic variables: if these variables have parent nodes AGPisdefinedbyitsmeanfunction,thedeterministiccom- (vkep and Σ), they are given by a deterministic function of those par- i ponentofthesignal,anditscovariancefunction,whichdefines ents;iftheydonothaveparents(t andσ),theyareassumedasbeing i i the overall behaviour of the functions under the GP distribu- givenandthusfixed.Thevariablesinsideboxesarerepeatedanumber tion.Whenthecovariancefunctionisevaluatedattheobserved oftimes,asshowninthebottomleftcornerofeachbox. times,thecovariancematrixisobtained.Fromthemanypossible choicesforacovariancefunction,thequasi-periodickernelisthe times(foreachofthe N planets).Anobserveddatasetiscom- mostwidelyusedintheexoplanetliterature(e.g.H14,Grunblatt p posed of N radial-velocity observations v, at times t and with et al. 2015; Rajpaul et al. 2015), which results in a covariance i i associated uncertainty estimates σ. The diagram in Fig. 1 is a matrixoftheform i representationofthejointprobabilitydensityfunction(PDF)for Σij =η21exp−(ti2−ηt22j)2 − 2sin2η(cid:16)24π(tηi−3tj)(cid:17)+(cid:16)σ2i +s2(cid:17)δij. (1) apll(cid:16)Θth,e{vvia},ri(cid:8)avbkielpe(cid:9)s(cid:12)(cid:12)(cid:12):{ti},{σi},I(cid:17), This represents some of our expectations for the activity- inducedRVsignals:thecorrelationsdecayonatimescaleofη where we condition on the information I1. In the remainder of 2 daysand[Note1:thesuggested’to’changesthemeaningof the paper we will include {ti} and {σi} in I, because they are the sentence.] have a periodic component with period η days. assumedtobefixed.Toeasethenotation,wealsogrouptheRV The parameter η4 controls the relative importance of th3e peri- values in a proposition D = {vi}. The (cid:8)vkiep(cid:9) are obtained from odicanddecayingcomponents.Theparameterη representsthe evaluatingthesumofKepleriansignalsattheobservedtimes. 1 amplitudeofthecorrelations.Thiscovariancematrixalsotakes ThejointPDFcanbefactoredandrearrangedtogiveBayes’s additional uncorrelated noise into account, added quadratically theorem, tothediagonal,whereσ arethereportedRVuncertaintiesand i sisafreeparameter. p(Θ|I)p(D|Θ,I) p(Θ|D,I)= , (2) p(D|I) 2.3. Thecompletemodel providinganexpressionfortheposteriordistributionforallthe Our complete model for the RV observations is shown in Fig. parameters,conditionedontheobserveddata.Theposteriordis- 1, in the form of a probabilistic graphical model. The diagram tribution contains all the information about the parameters that relatesalltheparameters(seeTable1fortheirdescription)and isavailabletous.Withthisdistributiontohand,wecanthenin- observablesinthemodel,statingtheirinter-dependencies.From ferhowmanyplanetsaresupportedbythedataandtheirorbital thisgraphicalrepresentationwecanbuildanexpressionforthe parametersandmasses. jointprobabilityoftheparametersofinterestandthedata. Let Θ be the vector of all the parameters in the model: (cid:104) (cid:105) 1 Here, I encodes the assumptions considered when setting up the Θ = N ,µ ,w ,µ ,{P,K,e,φ,ω},v ,η ,η ,η ,η ,s .Theno- p P P K sys 1 2 3 4 problem,e.g.theformoftheGPkernelorthefactthatweignoreplanet- tation means that the subset {P,K,e,φ,ω} will be repeated Np planetinteractions,etc. Articlenumber,page2of7 J.P.Fariaetal.:UncoveringtheplanetsandstellaractivityofCoRoT-7usingonlyradialvelocities 2.4. Determiningthenumberofplanets assumedauniformpriorbetween[Note2:Ibelieve’between’ isbettersuitedherethanthesuggested’of’]10and40days. Tocalculatethejointposteriordistribution,followingEq.(2),we Tosamplethejointposteriordistribution,weusedthealgo- need the three terms on the right-hand side: the prior p(Θ|I), rithmproposedbyBrewer(2014).Theparticulardifficultyhere likelihood p(D|Θ,I),andevidence p(D|I). –since N isnotknown–isthatthesamplingalgorithmneeds Forthelikelihood,thechoicethatreflectsmostgenuinelyour to jump bpetween candidate solutions with different numbers of stateofknowledgeisamultivariateGaussiandistribution2,with mean (cid:8)vkep(cid:9) and covariance matrix Σ. The complete covariance planets.Brewer(2014)proposedamethodthatusesbirth-death matrix cian be obtained, deterministically, from the values of MtheardkioffvucsihvaeinnMestoendtesaCmarplloin(gMfCraMmCew)morokv(eBsrteowienrfeertNapl.,2w0i1th1i)n. η ,η ,η ,η ,σ,and s,accordingtoEq.(1).Thelog-likelihood 1 2 3 4 i Thisway,onecanestimatethevalueoftheevidencewhileim- isthengivenby: proving the mixing in complex posteriors (affected by multi- modalityandphasetransitions).Brewer&Donovan(2015)ap- 1 1 N lnp(D|Θ,I)=− rTΣ−1r− lndetΣ− ln2π, (3) pliedthismethodtoRVdataofνOphandGliese581,although 2 2 2 theseauthorsdidnotincorporateGPsintheiranalysis. whereristhevectorgivenbyv −vkep foralldatapoints. i i The priors used for all the parameters are listed in Table 1. 3. ApplicationtoHARPSdata Most of them are the same as those used by Brewer & Dono- van(2015)andwerechosentorepresentuninformativeorvague We apply the method described in the previous section to priorknowledge.Fortheorbitalperiodsandsemi-amplitudeswe HARPS observations of CoRoT-7. The planet CoRoT-7b was assign hierarchical priors that are conditional on the hyperpa- first announced by Léger et al. (2009) and was the first super- rameters,µ ,w ,andµ ,respectively(seeTable1).Thisreflects P P K Earth with a measured radius. Its orbital period is estimated our belief that knowing the parameters of one planet provides fromthetransitsintheCoRoTlightcurveasbeing P = 0.854 a small amount of information about the parameters of another b days (Léger et al. 2009). A second non-transiting planet, with planet. P = 3.69 days, was detected in a follow-up RV campaign c (Queloz et al. 2009) and a more disputed detection of a third Table1.Meaningandpriordistributionfortheparametersinthemodel. planetarysignalwasreportedbyHatzesetal.(2010). Insomecaseswesampleonthelogarithmoftheparameter. Owingtothehighactivitylevelsofthehoststar,thissystem has since generated a wealth of discussion, resulting in differ- hyperparameters ent estimates for the masses of the planets (Lanza et al. 2010; N numberofplanets U(0,10) p Boisse et al. 2011; Ferraz-Mello et al. 2011; Pont et al. 2011; µ medianorbitalperiod log C(5.9,1) P Hatzesetal.2011).SimultaneousobservationsfromCoRoTand w diversityorbitalperiods U(0.1,3) P HARPSwereobtainedin2012tohelpsettletheseissues(Barros µ meansemi-amplitude log C(0,1) K etal.2014;H14).TheseobservationswereanalysedbyH14with planetparameters anRVmodelthatissimilartooursbutconsideringinformation P orbitalperiod log L(logµ ,w ) fromthesimultaneousphotometricobservations. P P K semi-amplitude E(µK) We emphasise that, in the following, we analyse the full e eccentricity B(1,3.1) set of RV observations. In summary, the star was observed φ orbitalphase U(0,2π) with HARPS in 2009 and 2012, with a total of 177 public RV ω longitudeofline-of-sight U(0,2π) measurements.Theaverageerrorbaronthesemeasurementsis 2 ms-1 (whichincludesphotonandinstrumentalnoise)andthe GPandnoiseparameters RVdispersionis10 ms-1,overthecomplete3-yeartimespan. η amplitudeofcovariance LU(0.1,50) 1 A common procedure in (current) RV studies is to fix the η aperiodictimescale LU(1,100) 2 numberofplanetsandsampletheposteriorfortheremainingpa- η correlationperiod U(10,40) 3 rameters(andpossiblycalculatetheevidence)inastep-by-step η periodicscale LU(0.1,10) 4 approach. If we fix N = 2, we recover the orbital parameters s extrawhitenoise log C(0,1) p ofthetwoknownplanets,CoRoT-7bandCoRoT-7c,withinthe vsys systematicvelocity U(minvi,maxvi) errorsreportedinH14andBarrosetal.(2014).Themethodwe presentinthispaperis,however,muchmoregeneralandallows Notes.Symbolmeaning:U(·,·)–Uniformpriorwithlowerandupper forthefullposteriorforN tobeobtainedfromonerun.Wenow limits;C(·,·)–Cauchypriorwithlocationandscale(thesedistributions p proceedwiththisgeneralmethod. were truncated for numerical reasons); L(·,·) – Laplace prior (some- timescalleddoubleexponential)withlocationandscale;E(·)–Expo- We ran our algorithm on the full set of RVs, using the nentialpriorwithmean;B(α,β)–Betapriorwithshapeparameters,α priors in Table 1, and obtained 16248 effective samples from andβ,anapproximationtothefrequencydistributionofeccentricities the joint posterior distribution3. The evidence for our model is proposedbyKipping(2013);LU(·,·)–Log-uniformpriorwithlower log(p(D|I)) = −530.9.Theresultingmarginalposteriordistri- andupperlimits. butionforN isshowninFig.2. p Theposteriordistributionfor N isoneofthemainoutputs p FormostoftheparametersoftheGPcovariancekernel,we ofourmethod.Buttoactuallydecideonwhatisthenumberof assigned log-uniform priors in sensible ranges. For η , the pa- 3 rameterthatcanbeinterpretedasthestellarrotationperiod,we 3 The computer used to run the simulations was equipped with an Intel(cid:13)R CoreTM i5-4460CPUrunningat3.20GHzand4GBofRAM. 2 Fig.1showsthatallweknowaboutthedistributionofthe{v}areits Therunningtimetoobtain50000samples(fromdiffusivenestedsam- i firstandsecondmoments.ThemultivariateGaussianfollowsfromthe pling’stargetmixturedistribution)wasfourdays.Sincethecomputa- principleofmaximumentropywhenthedistributionisconstrainedto tionalcostofthealgorithmisdominatedbytheinversionofthecovari- havingaspecifiedcovariancematrix(e.g.Cover&Thomas2006). ancematrix,itisexpectedtoscaleroughlywithN3. Articlenumber,page3of7 A&Aproofs:manuscriptno.AA_2015_27899 theposteriordistributionissensitivetothepriordistributions,in 3000 s particularastohowsmallwebelieveK mightbe. e mpl 2500 pp((32)) ≈4.49 Thejointposteriordistributionsfortheorbitalperiods,semi- a amplitudes,andeccentricitiesofthesignalsareshowninFig.3, s or 2000 where the samples for all Keplerians were combined. The fig- eri ure shows a 2-dimensional histogram of the posterior samples, ost 1500 wherethecolourmaprepresentsbincountsandissetinaloga- p of 1000 rithmicscale. p(2) er p(1) ≈∞ The two detected planets are seen as overdensity regions at mb 500 Pb = 0.85 days and Pc = 3.69 days. Their amplitudes and ec- Nu centricitiesarewellconstrained.Thereisaclearposteriorpeak 0 0 1 2 3 4 5 6 7 8 9 10 aroundtwodayswithamplitudeandeccentricitymostlyuncon- strained.Theposterioralsoshowsasmallerpeakat9days,the Numberofplanets periodreportedbyHatzesetal.(2010)asapossiblethirdplanet (seealsoTuomietal.2014). Fig.2.PosteriordistributionforthenumberofplanetsN .Thecounts p MarginalposteriordistributionsfortheparametersoftheGP arenumberofposteriorsamplesinmodelswithagivennumberofplan- and the extra white noise parameter s are shown in Fig. 4. The ets. The two ratios of probabilities between models with 1, 2, and 3 planetsishighlighted;notethatp(0)= p(1)=0. posterior for η3 is particularly interesting as it provides a con- straint to CoRoT-7’s rotation period, obtained exclusively from the RVs. Our inferred value for the stellar rotation period of 1] 10 2in2.a3g0r+−e16e0.m1.108endta(ysseiesToabbtalein2e)dwsoitlheleyafrrloiemrethsteimRVatetsimweh-sicehrieusseadndthies -s m CoRoTlightcurve(Légeretal.2009;Lanzaetal.2010;H14). 1 [ de Alsointerestingisthejointbehaviourofη2,η3,andη4.For u higher values of η , the periodic component of the covariance plit 0.1 function loses imp4ortance relative to the decaying component, m mi-a 0.01 ηG2PgsemtsosomthaslltehreaRndVηs3obneacotimmeesscuanlceoonfstηrain≈ed3.Idnayths.isBcuatsew,hthene 2 e S η isoforderunity(meaningtheperiodiccomponentispresent), 4 thedecayingtimescaleishigherandη isconstrainedaround22 3 0.8 days. The values of η in this situation (20-30 days) are closer 2 0.7 tothestellarrotationperiodandarealsoconsistentwiththeav- y 0.6 erage lifetime of active regions measured in the CoRoT 2012 cit 0.5 photometry (20.6 ± 2.5 days, H14). Our results therefore vali- tri 0.4 datetheapproachtakenby,e.g.H14andGrunblattetal.(2015), n ce 0.3 of modelling the activity-induced RV variations with a GP that c E 0.2 hasthecovariancepropertiesofthelightcurve. 0.1 Considering only the posterior samples with Np = 2, Table 0.0 2liststhemedianvaluesofsomeorbitalparametersforthetwo 0.1 1 10 100 1000 planets, and the maximum likelihood RV curves are shown in Period[days] Fig.5.Ourestimatesfortheorbitalparametersareinagreement, withintheuncertainties,withtheonesobtainedbyH14. Fig.3.Jointposteriordistributionforthesemi-amplitudes(toppanel) andtheeccentricities(bottompanel)togetherwiththeorbitalperiodsof 4. DiscussionandConclusions theKepleriansignals. BuildingontheworkofBrewer&Donovan(2015),wehavede- velopedasimplemethodtoestimatethenumberoforbitingplan- planets orbiting CoRoT-7, and answer our initial question, we etsaroundactivestars,usingonlyRVmeasurements.Weapplied needtoclarifywhatwemeanby“confidentlydetected”.Based thismethodtoHARPSobservationsofCoRoT-7andconfidently onascalesuggestedbyJeffreys(1998)(seealsoKass&Raftery detectCoRoT-7bandCoRoT-7c,whilefindingweakerevidence 1995), some authors (e.g. Tuomi 2011; Feroz et al. 2011) have for two additional signals. In this framework, there is no need proposedthat,toclaimadetectionofNpplanets,theprobability tousephotometricobservations,informationfromtransitdetec- of Np should be 150 times greater than the probability of Np − tions,orauxiliaryactivityindicators. 1. This criterion requires ’strong’ (Jeffreys 1998) evidence for The posterior distribution for N shows evidence for the p detecting a planet, thus considering false positives to be much presenceofextrasignals,eveniftheydonotmeetourdetection worse than false negatives. By applying this rule to our results, criteria. We note that the effects of considering a uniform prior wechooseNp =2asthenumberofplanetsconfidentlydetected forNp(thusgivingconsiderablepriorweighttolargenumbersof inourdataset. planets)andahierarchicalpriorforthesemi-amplitudes(which We emphasise that having confidently detected two planets changes how the Occam’s razor penalty is taken into account) isadifferentmatterfromknowingthatthenumberofplanetsis canbeimportantandwillbestudiedinthefuture. actually two. According to the posterior distribution, it is more Ourchoicesforthelikelihood,covariancefunction,andpri- likelythattherearefourplanets.Butourdetectioncriterionde- orsprovidethemodelwitharobustnessagainstpossibleoutliers pends on the ratio of probabilities for consecutive values of N whichissimilartomostanalysesofRVdata.Ifwehadreasonto p (seeFig.2),notontheprobabilityvaluesthemselves.Ofcourse, suspect the presence of outliers, it would be straightforward to Articlenumber,page4of7 J.P.Fariaetal.:UncoveringtheplanetsandstellaractivityofCoRoT-7usingonlyradialvelocities s = 0.88+0.42 0.54 − η = 9.31+1.53 1 1.31 − 16 1 12 η 8 η = 3.28+14.15 2 1.14 − 80 60 2 η 40 20 η = 22.21+9.79 40 3 −6.40 32 η3 24 16 η = 0.93+3.25 10 4 −0.36 8 6 4 η 4 2 0.5 1.0 1.5 2.0 8 12 16 20 40 60 80 16 24 32 40 2 4 6 8 10 s η1 η2 η3 η4 Fig.4.Marginalised1-and2-DposteriordistributionsfortheparametersoftheGPandtheextrawhitenoise.ThesamplesforallvaluesofN p werecombined.Thetitlesaboveeachcolumnshowthemedianoftheposteriorandtheuncertaintiescalculatedfromthe16%and84%quantiles. Thesolidlinesarekerneldensityestimationsofthemarginaldistributions. extendourmodel,e.g.byscalingeacherrorbarwithacommon In the analysis of H14, the authors modelled the out-of- valueorwithindividualvalueswhosecommonpriorisdefined transit photometry using a GP (as an interpolator) and applied hierarchically. the FF’ method of Aigrain et al. (2012) to obtain the RV sig- nalthatisduetoactivity.Bycomparingtheevidenceofmodels containingthisactivitysignalpluszero,one,two,orthreeplan- Steppingbacktoappreciateourresults,wefindthatthereis ets,theyassertedthatthetwo-planetmodelwasthemostproba- alotofinformationcontainedintheRVs,bothaboutplanetary ble.Tomodelallthequasi-periodicsignalsinthedata,H14in- and (arguably more importantly) activity signals, and that this cludedaGP(withthecovariancepropertiesoftheCoRoTlight information can be recovered. The GP provides a flexible and curve)aspartoftheRVmodel.Thiswasjustifiedbecause“the accommodatingmodelforactivity-inducedsignals,allowingus FF’methodislikelytoprovideanincompleterepresentationof to infer the planetary masses and orbital parameters with more activity-induced RV variations” (H14). We also note that these realisticuncertainties. Articlenumber,page5of7 A&Aproofs:manuscriptno.AA_2015_27899 a andcanberecoveredwiththeflexibilityoftheGP,ifwebetter 70 70 accountfortheuncertaintyassociatedwithit. 60 60 We should finally note two important properties of the 50 50 CoRoT-7system,whichmadeitidealforthisanalysis.First,the m/s] 40 40 amplitudesoftheplanetsignalsaremuchhigherthanthemean V[ 30 30 error bar of the HARPS observations, regardless of the stellar R activitycontamination.Second,thetimesamplingoftheobser- 20 20 vationsisalmostidealforthedetectionofshort-periodplanets, 10 10 andisverydifficulttoobtainaspartofatypicalRVsurvey.We 0 0 highlightheretheimportanceoffurthertests,usingotherwell- 4780 4830 4880 5940 5950 5960 studieddatasetsofactivehoststarsaswellassimulateddatasets, BJD-2450000[days] forassessingthelimitsofapplicabilityofourmethod. b RV-GP-planetc c RV-GP-planetb Nevertheless,theimportanceofthisworkisnotonthespe- 20 20 cific application to CoRoT-7, but instead on providing a sim- 15 15 ple and fast method to infer the number of planetary signals in 10 10 the presence of stellar activity. With small modifications, this s] 5 5 m/ methodcanbeusedtosearchforplanetsaroundstarswithdif- 0 0 [ ferentactivitylevels. V -5 -5 R -10 -10 Acknowledgements. The work presented here grew directly from a collabo- -15 -15 ration that started at the Astro Hack Week 2015. We thank Andrew Collier Cameronforinsightfuldiscussions.Weacknowledgetheexcellentopen-source -20 -20 0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1 Python packages made available to the community (in particular DAFT and George).ThisworkwassupportedbyFundaçãoparaaCiênciaeaTecnologia orbitalphase (FCT)throughtheresearchgrantUID/FIS/04434/2013andthegrantPTDC/FIS- AST/1526/2014. JPF acknowledges support from FCT through the grant ref- Fig.5.Panela:RVmeasurementsofCoRoT-7from2009and2012and erenceSFRH/BD/93848/2013.RDHgratefullyacknowledgesagrantfromthe thetwo-planetbest-fitmodel(blackcurve).Notethedifferentscalesin JohnTempletonFoundation.Theopinionsexpressedinthispublicationarethose theabscissaeontheleftandrightpartsoftheplot.Panelsbandcshow of the authors and do not necessarily reflect the views of the John Temple- thephasedRVcurvesaftersubtractingeachplanetsignalandtheGP. ton Foundation. BJB is supported by a Fast Start grant from the Royal So- ciety of New Zealand. PF and NCS acknowledge support by FCT through Investigador FCT contracts (IF/01037/2013 and IF/00169/2012, respectively), Table2.ParameterestimatesfromourworkandfromH14.Weconsider andPOPH/FSE(EC)byFEDERfundingthroughtheprogramProgramaOp- allmodelsthathaveN = 2andshowthemarginalposteriormedians, eracional de Factores de Competitividade - COMPETE. PF further acknowl- p togetherwiththe16%and84%quantiles. edges support from FCT in the form of an exploratory project of reference IF/01037/2013CP1191/CT0001.MOacknowledgesresearchfundingfromthe DeutscheForschungsgemeinschft(DFG,GermanResearchFoundation)-OS units Thiswork H14 508/1-1. AS is supported by the European Union under a Marie Curie Intra- European Fellowship for Career Development with reference FP7-PEOPLE- P (a) days 0.85424+0.00071 0.85359165± 5×10−8 2013-IEF,number627202. b −0.00126 K ms-1 3.97+0.62 3.42±0.66 b −0.55 e 0.045+0.053 0.12±0.07 b −0.027 References m M 5.53+0.86 4.73±0.95 b ⊕ −0.78 Aigrain,S.,Pont,F.,&Zucker,S.2012,Mon.Not.R.Astron.Soc.,419,3147 P days 3.69686+0.00036 3.70±0.02 Baranne,A.,Queloz,D.,Mayor,M.,etal.1996,A&AS,119,373 c −0.00026 Barros,S.C.C.,Almenara,J.M.,Deleuil,M.,etal.2014,Astron.Astrophys., Kc ms-1 5.55+−00..3341 6.01±0.47 569,A74 e 0.026+0.033 0.12±0.06 Boisse,I.,Bouchy,F.,Hébrard,G.,etal.2011,Astron.Astrophys.,528,A4 c −0.017 Brewer,B.J.2014,arXiv:1411.3921 m M 12.62+0.77 13.56±1.08 Brewer,B.J.&Donovan,C.P.2015,Mon.Not.R.Astron.Soc.,448,3206 c ⊕ −0.72 Brewer,B.J.,Pártay,L.B.,&Csányi,G.2011,Stat.Comput.,21,649 η (b) days 22.50+10.56 23.81±0.03 Cover,T.M.&Thomas,J.A.2006,Elementsofinformationtheory,2ndedn. 3 −6.19 (Hoboken,N.J:Wiley-Interscience) Notes.ThefollowingnotesapplytotheestimatesfromH14: Dumusque,X.,Santos,N.C.,Udry,S.,Lovis,C.,&Bonfils,X.2011a,Astron. (a) AGaussianpriorwasused,centredat0.85359165daysandwitha Astrophys.,527,A82 Dumusque,X.,Udry,S.,Lovis,C.,Santos,N.C.,&Monteiro,M.J.P.F.G. standarddeviationof5.6×10−7 days.(b) Theestimatefortherotation 2011b,Astron.Astrophys.,525,A140 periodwasderivedfromtheCoRoTlightcurve. Feroz,F.,Balan,S.T.,&Hobson,M.P.2011,Mon.Not.R.Astron.Soc.,415, 3462 Ferraz-Mello,S.,TadeudosSantos,M.,Beaugé,C.,Michtchenko,T.A.,&Ro- dríguez,A.2011,Astron.Astrophys.,531,A161 authorsuseaverystrongGaussianpriorforP (andthetimeof b Figueira,P.,Marmier,M.,Bonfils,X.,etal.2010,Astron.Astrophys.,513,L8 periastron), which was nevertheless completely justified by the Figueira,P.,Santos,N.C.,Pepe,F.,Lovis,C.,&Nardetto,N.2013,Astron. transitobservations. Astrophys.,557,A93 Grunblatt,S.K.,Howard,A.W.,&Haywood,R.D.2015,Astrophys.J.,808, Ouranalysisrestsonmuchfewerassumptions–onlytheRV 127 measurementswereused,therotationperiodofthestarwasal- Hatzes,A.P.,Dvorak,R.,Wuchterl,G.,etal.2010,Astron.Astrophys.,520, lowedtovary,thepriorfortheorbitalperiodsismuchlessstrin- A93 gent–butwearestillabletorecoverthesamenumberofcon- Hatzes,A.P.,Fridlund,M.,Nachmani,G.,etal.2011,Astrophys.J.,743,75 Haywood,R.D.,CollierCameron,A.,Queloz,D.,etal.2014,Mon.Not.R. fidentlydetectedplanetsandreliableestimatesoftheorbitalpa- Astron.Soc.,443,2517 rametersandthestellarrotationperiod(seeTable2).Thisshows Jeffreys,H.1998,Theoryofprobability,3rdedn.(Oxford:OxfordUniversity that all this information is contained in the RV measurements Press) Articlenumber,page6of7 J.P.Fariaetal.:UncoveringtheplanetsandstellaractivityofCoRoT-7usingonlyradialvelocities Kass,R.E.&Raftery,A.E.1995,J.Am.Stat.Assoc.,90,773 Kipping,D.M.2013,Mon.Not.R.Astron.Soc.Lett.,434,L51 Lanza,A.F.,Bonomo,A.S.,Moutou,C.,etal.2010,Astron.Astrophys.,520, A53 Léger,A.,Rouan,D.,Schneider,J.,etal.2009,Astron.Astrophys.,506,287 Mayor,M.,Pepe,F.,Queloz,D.,etal.2003,TheMessenger,114,20 Noyes,R.W.,Hartmann,L.W.,Baliunas,S.L.,Duncan,D.K.,&Vaughan, A.H.1984,Astrophys.J.,279,763 Pepe,F.,Mayor,M.,Galland,F.,etal.2002,Astron.Astrophys.,388,632 Pont,F.,Aigrain,S.,&Zucker,S.2011,Mon.Not.R.Astron.Soc.,411,1953 Queloz,D.,Bouchy,F.,Moutou,C.,etal.2009,Astron.Astrophys.,506,303 Rajpaul,V.,Aigrain,S.,Osborne,M.A.,Reece,S.,&Roberts,S.2015,Mon. Not.R.Astron.Soc.,452,2269 Rasmussen,C.E.&Williams,C.K.I.2006,Gaussianprocessesformachine learning(Cambridge:MITPress) Roberts,S.,Osborne,M.,Ebden,M.,etal.2012,Philos.Trans.R.Soc.Lond. Math.Phys.Eng.Sci.,371 Robertson,P.,Roy,A.,&Mahadevan,S.2015,Astrophys.J.,805,L22 Saar,S.H.&Donahue,R.A.1997,Astrophys.J.,485,319 Santos,N.C.,GomesdaSilva,J.,Lovis,C.,&Melo,C.2010,Astron.Astro- phys.,511,A54 Santos,N.C.,Mortier,A.,Faria,J.P.,etal.2014,Astron.Astrophys.,566,A35 Tuomi,M.2011,Astron.Astrophys.,528,L5 Tuomi, M., Anglada-Escude, G., Jenkins, J. S., & Jones, H. R. A. 2014, arXiv:1405.2016 Articlenumber,page7of7

See more

The list of books you might like

Most books are stored in the elastic cloud where traffic is expensive. For this reason, we have a limit on daily download.