ebook img

A cautionary tale: limitations of a brightness-based spectroscopic approach to chromatic exoplanet radii PDF

0.12 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 A cautionary tale: limitations of a brightness-based spectroscopic approach to chromatic exoplanet radii

Astronomy&Astrophysicsmanuscriptno.Chrom_RM_sim_letter_ref_nobold (cid:13)cESO2017 January6,2017 A cautionary tale: limitations of a brightness-based spectroscopic approach to chromatic exoplanet radii H.M.Cegla1,2,3,C.Lovis2,V.Bourrier2,C.A.Watson1,andA.Wyttenbach2 1 AstrophysicsResearchCentre,SchoolofMathematics&Physics,Queen’sUniversityBelfast,UniversityRoad,BelfastBT71NN, UnitedKingdom 2 ObservatoiredeGenève,UniversitédeGenève,51chemindesMaillettes,1290Versoix,Switzerland 3 SwissNationalScienceFoundationNCCR-PlanetSCHEOPSFellow 7 Received31102016/Accepted23122016 1 0 2 ABSTRACT n Determiningwavelength-dependent exoplanetradiimeasurementsisanexcellentwaytoprobethecompositionofexoplanetatmo- a spheres.Inlightofthis,Borsaetal.(2016)soughttodevelopatechniquetoobtainsuchmeasurementsbycomparingground-based J transmissionspectratotheexpectedbrightnessvariationsduringanexoplanettransit.However,wedemonstratehereinthatthisisnot 5 possibleduetothetransitlightcurvenormalisationnecessarytoremovetheeffectsoftheEarth’satmosphereontheground-based observations.Thisisbecausetherecoverableexoplanetradiusissetbytheplanet-to-starradiusratiowithinthetransitlightcurve;we ] demonstratethisbothanalyticallyandwithsimulatedplanettransits,aswellasthroughareanalysisoftheHD189733bdata. P E Keywords. Methods:dataanalysis–Planetsandsatellites:atmospheres–Planetsandsatellites:fundamentalparameters–Planets andsatellites:HD189733b–Techniques:radialvelocities–Techniques:spectroscopic . h p - o1. Introduction curve to allow one to study the local, occulted stellar profiles r directly by subtracting the in- from out-of-transit observations tTransmission spectroscopy is an essential tool for charac- s (see,e.g.Ceglaetal.2016a).Sincethetransitlightcurveisde- aterising the atmospheres of transiting exoplanets (see e.g. pendentontheplanet-to-starradiusratio,itsetstheplanetradius [Charbonneauetal. 2002; Pontetal. 2013; Madhusudhanetal. thatisrecoverablewhenexaminingthebrightnessratiobetween 2014; Singetal. 2016, and references therein). Snellen (2004) 1 thelocalandout-of-transitprofiles(meaningonecannotrecover demonstratedthat narrowbandexoplanetfeatures(e.g. sodium) v newradiusvariationsfollowingB16). 1could be probed by analysing the shape of the stellar absorp- In this study,we first breakdown the physicalimplications 5tion lines as a planet occults its host star, that is, by studying of the B16 technique, and then simulate the planet transit of 2the chromaticRossiter-McLaughlin(RM) effect;more recently HD189733b to illustrate the impact of the technique’s short- 1DiGloriaetal. (2015) have shown that this effect can also be comingsonthemeasuredplanetradius;wealsoreapplytheB16 0used to probe broadband signatures (e.g. Rayleigh scattering). techniqueto HARPSdata,with a morerigorouserrorpropaga- .Recently, Borsaetal. (2016, hereafter B16) presented a new 1 tion. In Sect. 2, we demonstrate how the choice of the transit 0techniqueusing(aversionof)line-profiletomography,withthe lightcurvenormalisationsetstherecoverableplanetradius.We 7intent of studying chromatic changes in planetary radii. How- present the simulated planet transits, and our reanalysis of the 1ever,we demonstratein thisLetterthatthe techniquein B16 is HARPSdatainSect.3,andshowhowanunderestimationofthe :unfortunatelyflawed. v errorscanleadtospuriousclaimsofplanetaryradiusvariations. i In principle, the application of line-profile tomography is Finally,wesummariseourconclusionsinSect.4. Xwellmotivatedforexoplanetatmospherecharacterisationifcor- 2. Limitationsofabrightness-basedapproachto rrectlyimplemented.Thistechniqueisolatesthestarlightbehind a chromaticexoplanetradii the planet during transit (CollierCameronetal. 2010), and the ratio of the integrated flux within the local profile (behind the In B16, the authors attempted to study passband-dependent planet) to the out-of-transitprofile is equalto the brightnessof planet radius variations by averaging together the cross- theoccultedstarlight.Inturn,theratioofthe occultedstarlight correlationfunctions(CCFs)forsubsetsofHARPSspectralor- is dependent on the planet-to-star radius ratio. As such, if one ders. All of these CCFs, regardless of passband, were contin- had space-based spectra then the planet radius could be recov- uumnormalisedandthenfurtherscaledusingaMandel&Agol ered fromthe spectra alone, anddoingso in variouspassbands (2002)transitlightcurvebasedonthesystemparametersdeter- wouldcharacterisetheplanetaryradiuswavelengthdependency. minedinthefullHARPSpassband,exceptforpassbanddepen- For ground-basedspectra this is notpossible due to, for exam- dentlimbdarkeningcoefficients. ple, transparency variations of the Earth’s atmosphere. To re- For each passband,theycreatedmaster out-of-transitCCFs move these effects, ground-based spectra must first be contin- byaveragingtogetheralltheindividualout-of-transitCCFsina uum normalised; they can then be multiplied by a transit light givennight,andthensubtractingtheout-fromthein-transitdata toobtainCCFsforthestarlightoccultedbytheplanet.Theau- Sendoffprintrequeststo:H.M.Cegla,e-mail:[email protected] thorsthenfittedGaussianfunctionstotherespectiveCCFs,pre- Articlenumber,page1of5 A&Aproofs:manuscriptno.Chrom_RM_sim_letter_ref_nobold sumablytoactasaproxyfortheintegratedfluxwithin,andar- (Borsa,PrivateComm.),suchthat guedthattheratiobetweentheareas(determinedfromtheGaus- CCFB16 =−CCFC16+β , (6) sianfits)ofthelocaltotheout-of-transitCCFscouldserveasa loc loc lc measure of the brightnessof the missing starlight, β, occulted whereCCFC16 is thelocalCCF obtainedfollowingC16.Since duringthein-transitobservations. loc thecontinuumisa freeparameterintheGaussianfits,theratio Theauthorsthencomparedthisempiricalβtotheexpected oftheareasisstillequaltothatinEq.4. brightness ratio based on an approximation to the solution for Assuch,thetechniqueimplementedbyB16representsacir- integratingthestarlightbehindtheplanetgiventheparticularge- cularargument.Moreover,ifonehadthebroadbandphotometry ometryofthesystem(andassumingaparticularfunctionforthe necessaryforthecorrectspectralnormalisation,thentheplanet limbdarkening).Sincethissolutionwasdependentontheplanet radiicouldbedetermineddirectlyfromthelightcurvesalone. radius, the authors argued that they were able to disentangle a We stress that a solely brightness-based approach to trans- planetradiusmeasurementforeachpassband. missionspectroscopyusesonlytheequivalentwidth,andthere- We demonstrate in Sect. 2.1 that the technique outlined by foreprecludesanyretrievalofinformationonexoplanetradii.It B16canonlyrecovertheplanetaryradiussetbythetransitlight istheinclusionofthespectraldimensionthatisnecessarytode- curveusedintheinitialnormalisation.However,weemphasise termineR (i.e.utilisingtheDopplerinformation,asisdonein p that such a limitation is set due to a flux-based approach and CollierCameronetal.2010;DiGloriaetal.2015). would not be present in an RV-based approach,such as that in 3. Systematiceffectsonchromaticradius DiGloriaetal.(2015)orthatintraditionallineprofiletomogra- phy(i.e.followingtheCollierCameronetal.2010formulation). measurements To try to understand how B16 obtained results mimicking 2.1.Transitlightcurvenormalisation Rayleighscattering,wesimulatedthetransitofHD189733band B16startedbydeterminingtheareaoftheirout-of-transitmaster applied their technique. To demonstrate that we could recover CCFs, Aout, for each passband. Since the out-of-transit master ourmodelinputs,wealsopresentresultswhereinweappliedthe CCFs (CCFout) were continuum normalised their area is equal normalisation from C16 with the correct planet radii for each totheirequivalentwidth,EWout,thatis, passband;indoingsowediscoveredtheapproximationsforβin B16underestimatedtheplanetradii,andthuswealsoexploreda Aout = EWout. (1) numericalapproachforcalculatingthisbrightnessratiothatwas moreaccuratethantheapproximationusedinB16. TheauthorsalsomeasuredtheareasofthelocalCCFsbehindthe planet,A .Sincethein-transitCCFs(CCF )arenormalisedby 3.1.Analyticalbrightnessratioapproximation loc in thetransitlightcurve,theareaofthelocalCCF(CCF )is loc Unfortunately,integrating the limb darkened brightness under- A =(1− f )EW =β EW , (2) neath a planet lying off stellar disc centre is not straightfor- loc lc loc lc loc ward and even approximate analytical expressions are quite where f is the flux from the light curve, EW is the equiva- complex(especially if considering ingress and egress regions). lc loc lent width of the local CCF, and β is the fraction of starlight For this reason, B16 used the β formalism presented in lc occultedbytheplanetundertheassumptionsofthetransitlight CollierCameronetal.(2010),whichwasbasedontheanalytical curveusedinthenormalisation. approximationsofOhtaetal.(2005)fortheRMeffect.Therein Inanidealcase,wherethelocalstellarphotosphericprofiles thebrightnessratioforafullyin-transitplanet(i.e.noingressor can be represented by constant Gaussian functions (assumed egressregionswereconsidered)wasdefined,underthestandard bothaboveandinB16),thentheonlydifferencebetweenthelo- linearlimbdarkeninglaw,as: calanddisc-integratedout-of-transitCCFsisthebroadeningby stellarrotationpresentinthedisc-integratedobservations.Since R 21−u +u ∗µ therotationalbroadeningpreservestheequivalentwidth,then β≈ p 1 1 , (7) R ! 1−u /3 ⋆ 1 EW =EW = EW , (3) out loc in whereR isthestellarradius,R istheplanetradius,u isthelin- ⋆ p 1 earlimbdarkeningcoefficient,andµisthecentre-to-limbplanet wheretheEW andtheEW arenotonlyequaltoeachother out loc butarealsoequaltothein-transitequivalentwidth,EW (since position.Wenotethatµ=cos(θ)= 1−x2 −y2,whereθisthe in p p itissimplyasummationoflocalprofilesofequalEW),andthe centre-to-limbangle, and x ,y is tqhe centre of the planet (see p p ratiooftheareasbecomes: CollierCameronetal.2010;Ceglaetal.2016a,fordetails).We alsonotethatOhtaetal.(2005)statethattheaccuracyofthisap- A loc =β . (4) proximationdiminisheswithincreasingR /R ,arguingthatthe A lc p ⋆ out additionalterms in the analytical solution contribute to ∼1% if R /R is≤0.1anduptofewpercentifR /R ∼0.3.Giventhat Hence, under the above assumptions one can only recover p ⋆ p ⋆ R /R isonlypredictedtovaryafewprecentinwavelengthfor the planetradiusinjected into the modeltransitlight curve,re- p ⋆ particularatmosphericcharacteristics,suchasRayleighscatter- gardlessofwhichpassbandisstudied.Thisisbecausethetran- ing, this approximationmay inject systematic errors that could sit light curve normalisation effectively sets the area of the lo- be misinterpreted as having a physical origin (even if the light cal profile. We note, that the transit light curve normalisation curvenormalisationisdonecorrectly). fortheabovefollowsCeglaetal.(2016a,hereafterC16),where CCF =CCF − f ∗CCF ;whereas,thetransitlightcurve loc out lc in 3.2.Numericalbrightnessratioapproximation normalisationinB16was ToinvestigatetheimpactoftheaccuracyofEq.7,wedecidedto CCFB16 =CCF f −CCF +(1− f ) (5) calculateβnumerically.Forthisapproach,thein-transitstarlight loc in lc out lc Articlenumber,page2of5 Ceglaetal.:Limitationsofabrightness-basedapproachtochromaticexoplanetradii Simulated Data HARPS Data 1 in 4 pts Oversampled Borsa 0.158 0.158 d e er v co 0.156 R*0.156 R/R Re*p R/p 0.160 Nightly (1 in 4 pts) 0.158 0.154 0.154 0.156 0.154 CB1166 TTeecchhnniiqquuee 00..115502 AJSuuelgp ‘0 ‘‘00776 Simulated R p 0.152 0.152 400 500 600 700 400 450 500 550 600 650 400 500 600 700 Wavelength (nm) Wavelength (nm) Fig. 1. Recovered average planet radius, R , from the simulated data Fig. 2. Recovered average planet radius, R , for each passband as a p p foreachpassbandasafunctionofwavelength(plottedusingthemid- functionofwavelengthfortheobservedHARPSdata.ResultsfromB16 dlewavelengthinthepassband).Filledcirclesrepresentwhenthesim- areshowningreen(wherethewavelengtherrorbarsrepresentthepass- ulated R was constant, and hollow circles represent when the simu- band wavelength region), and those from our reanalysis are shown in p latedR varied;inbothcasestheseareshowninredwiththeerrorbars red when using the oversampled CCF and in black when using only p reported byB16 for comparison purposes only. Resultsfromthe B16 everyoneinfourpointsintheCCF.Subplot:thenightlyrecoveredR p procedureareshowninblack,whilethosefromthenumericalapproxi- whenusingonlyeveryoneinfourpointsintheCCF(nightindicatedby mationhereinandtheC16formulationareinblue.Thelinesrepresent colour).ThehorizontaldashedlinesshowthemeanR recoveredbythe p linearfitstothesimulatedR shownforviewingease. B16methodonthesimulateddata(i.e.thesolidblackpointsinFig.1). p blocked by the planet is still defined as β = Floc/F⋆, with F⋆ 3.3.Simulatedstar-planetsystem andF definedasthefluxesofthetotalstellardiscandthestel- loc lardiscundertheplanet,respectively.Notetheobservedbright- WeusedthesimulatedstellargridofCeglaetal.(2015,2016b) ness of the un-occulted star can be analytically determined by and injected into each grid cell a Gaussian profile with a full- integratingagivenlimbdarkeninglawovertheprojectedstellar width half-maximum(FWHM) of 5 km s−1 (note this width is disc;foralinearlimbdarkeningthisis similartotheexpectedvalueforthestellarphotosphere).Inthe simulatedstarwedidnotconsideranyastrophysicaleffects(i.e. granulationor starspots etc.) other than rigid body stellar rota- tion,whichwas set to the valueobtainedbyC16, 3.25kms−1. F =R2 2π 1I(µ)µdµdφ=πR2 1− u1 , (8) Wealsoassumedanedge-on(i⋆ =90◦)alignedorbit. ⋆ ⋆Z Z ⋆ 3 ! The transit was sampled in 21 equal steps in phase from 0 0 −0.02−0.02,centredaboutmid-transit,withanadditionalsam- whereφistheazimuthalangle.Aspreviouslystated,calculating ple at phase = 0.03 to serve as a completely out-of-transitref- the flux behind the planet analytically is not trivial. Hence, we erence. We simulated a transit for each of the seven passbands calculatedF numericallybyconstructingasquarestellargrid (from400−700nm)usedinB16,andappliedalinearlimbdark- loc with a width of 2R centred about the planet position (x ,y ), ening using the coefficients (for each passband) these authors p p p withnequalstepsintheverticalandhorizontaldirection.Con- provided.Foreachofthesevenpassbandtransitswe injecteda tributionsfromstepsthatdidnotliebeneaththeplanetand/oron planetwithaconstantradiusequaltothevalueassumedbyB16 the stellar disc were excluded.Thus, we approximatedthe flux for the whole HARPS passband (Rp = 0.1581R⋆, hereafterre- behindtheplanetas ferred to as the broadband Rp), but varied the limb darkening accordingly. 2R 2 Forthissetoftransits,wetestedtheimpactofthetransitlight F ≈ I p , (9) curvenormalisation.Inthefirstcase,wefollowedtheprocedure loc xy X n ! in B16, and in the second case we normalised the data follow- ing C16 and used the numerical approximation in Sect. 3.2 to whereIxyisthelimbdarkenedintensityatagivenpositioninthe estimateRp.Examiningthefirstcaseallowedustoexamineany aforementionedgridand(2R /n)2isthecorrespondingarea. errorsintroducedusingtheβapproximationand/ortheB16nor- p OuraimwastotrytorecoverR asafunctionofwavelength, malisation.Ontheotherhand,thesecondcaseofferedatestcase p wherein the injected R was only used to construct the correct toensurewecouldrecoverthemodelinputs. p lightcurves(actingasifwehadsimultaneousmulti-colourpho- For a secondtest, we repeatedthe above,butvariedthe ra- tometry).Hence,whentryingtorecoverR ,westartedwiththe diusofthesimulatedplanet;forthisweselectedR equaltothe p p broadbandplanetradiusandthen allowedit to varyby upto ± valuesreportedbyB16foreachpassband.Againwetestedtwo 0.005R instepsof0.0001R .Therecoveredplanetradiusthen cases: first followingthe B16 procedure(where the lightcurve ⋆ ⋆ corresponded to the planet radii that minimised the difference limb darkeningvaries in each passband, but the light curve ra- betweenβandA /A . dius remains fixed at the broadbandR ), and the second using loc out p Articlenumber,page3of5 A&Aproofs:manuscriptno.Chrom_RM_sim_letter_ref_nobold theC16normalisation(wenotetheassumedlightcurvehasthe Table1.Bestfitstoobserveddata correctR here)andthenumericalapproximationinSect.3.2. p Data Function BIC χ2 Function BIC χ2 r r 3.3.1. Obtainingtheplanetradiusandsystematicerrors 1in4pts Flat 13.9 2.0 Linear 11.7 1.6 Oversamp. Flat 65.4 10.6 Linear 29.1 5.0 WeexaminedtherecoveredR asafunctionofstellardiscposi- p B16 Flat 13.1 1.9 Linear 5.2 0.3 tion,andfoundonlyaslightdependanceondiscpositionwhen following B16. However, if one point each at the ingress and planetradii(inblack)areconsistentwithaflatline(within1−3σ egressregionswereincludedthenthedependenceondiscposi- of the mean R recovered in the simulated data, i.e. the solid p tion was strong, and including such data would systematically blackpointsinFig.1),asexpectedfromSect.2.Webelievethe decrease the recovered Rp (as the β formulationis not valid in reason B16 report a trend with wavelength, and we do not, is theseregions). largelyduetodifferencesinourerroranalysisandGaussianfit- Moreover,wefoundthat,regardlessoftheB16orC16nor- tingtechniques. malisation, using the β approximation always underestimated InB16,therecoveredR foreachstellardiscpositionandall p the limb darkeningbehind the planet and thereforealso under- threetransitswereaveragedtogethertoprovideoneR foreach p estimated the true planet radius. This is because the analytical passband, and the reported errors came from the rms of these approximationassumesthelimbdarkeningbehindtheplanetis individualplanetradii(i.e.thestandarddeviationdividedbythe constant,andequaltothevaluebehindthecentreoftheplanet.In squarerootofthetotalnumberinthepassband).Inouranalysis, reality,thestellarphotospherebehindtheplanetexhibitsarange wereporttheweightedmeanforeachpassband,withtheweights oflimbdarkening.ThisiswhythenumericalmodelinSect.3.2 being the inverse square of the error for each individualplanet isnecessarytorecovertheR injectedintothesimulateddata. p radii(wheretheerrorwascalculatedbypropagatingtheerrorson The Rp reportedin B16 comesfromaveragingtogetherthe theCCFareasasreportedfromtheGaussianfitsfollowingC16, planet radii recovered across the stellar disc. If the limb dark- andassumingnegligibleerroronthelimbdarkeningandstellar ening effects are sufficiently removed(and the stellar profile is discpositions).Theerrorontheweightedmeanthenwassimply constant), then this providesa goodmeans to boost the signal- thesquarerootoftheinversesumoftheweightssquared.Ifthe to-noise in the reported Rp. In Fig. 1, we present the average errors on individual Rp were all exactly equal to the standard recoveredRp asafunctionofwavelengthfromthesimulations, deviation,thenthetwoapproacheswouldyieldthesameresult. forbothtests(whenR wasconstantandwhenitvaried).Asex- p In addition to this slight difference in error analysis, B16 pectedfromSect.2.1,theB16procedurealwaysresultsinnearly also applied their Gaussian fits to the oversampled CCF grid thesameR ,regardlessofwhetherthetrueR variedornot. p p provided by the HARPS pipeline (Borsa, Private Comm.). We For ournumericalapproachand the C16 normalisation,we cautionagainstsuchanapproach,astheoversamplingwilllead demonstrate accurate recovery of Rp (regardless of whether or toasignificantunderestimationoftheerrors.Hence,wealsofit notweincludeingressandegressdata),butonlyifthelightcurve Gaussianstodatacomposedofeveryoneinfourpointsfromthe normalisationisdonewiththecorrectRpforeachpassband(us- originalCCFs (to compensate for the originalsampling rate of ing the broadband Rp for all passbands meant only the broad- 0.25kms−1forameanpixelwidthof0.82kms−1);theseresults bandRp wasrecovered).Hence,regardlessofthenormalisation areshowninblackinFig.2. (i.e.B16orC16)orthebrightnessformulation(i.e.βorournu- TotestthesignificanceofatrendinR withwavelength,we merical approximation),we could only retrieve the parameters p fitted the data with both a flat line and a linear regression, and injectedintothesystemviathetransitlightcurvenormalisation, calculatedthereducedchi-squared,χ2, andtheBayesian Infor- asexpectedfromSect.2. r mation Criterion (BIC); the results are shown in Table 1. We notethatevenifawavelength-dependentR isfound,itdoesnot 3.4.ReanalysisoftheHARPSdata p confirm the B16 technique is valid, as we have already shown OurapplicationoftheB16procedureonthesimulateddatacan- it is not mathematically possible to retrieve radius variations. not explain the wavelength-dependent planet radii reported in Rather, it would serve as a red flag that we do not fully char- B16. To further investigate this aspect, and to ensure we have acterise the interplayof the variouscomplexitiespresentin the appliedtheB16methodcorrectly,wehavereanalysedthesame observations.Inparticular,stellaractivitycanaltertheobserved threetransitsofHD189733bfollowingtheirtechnique,butus- stellar line shapes and their equivalent widths – which in turn ing the Levenberg-Marquardt least-squares minimisation from couldleadtospuriousradiusvariationsfollowingSect.2.Since MPFIT (Markwardt 2009, and references therein) rather than HD189733isaknownactivestarthisislikelyscenario;andin IDL’s GAUSSFIT1. The results are plotted in Fig. 2, alongside agreementwithFig.5fromB16,whereinthesinglenightanal- thosefromB162.Wedemonstratewecanreproduce(redpoints ysis with the most apparent slope, July 2007, is also the most inFig.2)resultsin1-2σagreementwithB16(ingreen);hence, magneticallyactive(Ceglaetal.2016a).Moreover(andasnoted we are confident we have applied their technique properly (in byB16),McCulloughetal.(2014)havearguedthattheapparent boththe simulatedandobserveddata).However,we arguethat wavelengthdependencyintheirindependentobservationsofthis with the correct treatment of the uncertainties the recovered system are best explained by un-occulted starspots rather than theplanetatmosphere. 1 MPFIT did not produce significantly different results compared to When using the oversampled CCFs, both our analysis and IDL’sGAUSSFIT,butitdidallowustopropagateourerrorsmorethor- B16’sindicate a slight improvementin fit for the modelwith a oughly(seeC16fordetails). wavelength-dependent slope. However, we find a much worse 2 WenotethatweusedthesametransitparametersasB16,butthere fit to the data than that found with the B16 results. The high is a slight difference in the template mask used to obtain the CCFs. χ2 from our reanalysis indicates an underestimation of the un- B16 used the archival data available from the ESO website, where 2 r nights used theG2 mask and 1night used the K5mask, whereasour certaintyinthedata,asonewouldexpectwhenusingtheover- dataalwaysusedtheK5mask.However,thisdifferenceisunlikelyto sampledCCFs. We cannotexplainthe verylow χ2r forthe B16 impacttheanalysissinceeachnighthaditsownmasterCCF . wavelength-dependentfit, which indicatesthe modelis overfit- out Articlenumber,page4of5 Ceglaetal.:Limitationsofabrightness-basedapproachtochromaticexoplanetradii ting the data. We note these tests were only performed on our Charbonneau, D.,Brown,T.M.,Noyes,R.W.,&Gilliland, R.L.2002,ApJ, reanalysisoftheoversampleddataforcomparisonwithB16;for 568,377 ourconclusionsonthebest-fit,wereferthereadertotheanalysis Collier Cameron, A.,Bruce, V. A., Miller, G. R. M., Triaud, A.H. M. J., & Queloz,D.2010,MNRAS,403,151 ontheCCFssampledeveryoneinfourpoints. DiGloria,E.,Snellen,I.A.G.,&Albrecht,S.2015,A&A,580,A84 Fortheproperlysampleddataset,wefoundonlyamarginal Madhusudhan, N.,Knutson,H.,Fortney, J.J.,&Barman, T.2014,Protostars improvement in the fit for the wavelength-dependent model, andPlanetsVI,739 anddonotdeemthisimprovementtobestatisticallysignificant Mandel,K.&Agol,E.2002,ApJL,580,L171 (see Table 1). Moreover,the best-fitflat model(R = 0.1569± Markwardt,C.B.2009,inAstronomicalSocietyofthePacificConferenceSe- p ries,Vol.411,AstronomicalDataAnalysisSoftwareandSystemsXVIII,ed. 0.0003)lieswithin3σofthemeanR predictedbythesimula- p D.A.Bohlender,D.Durand,&P.Dowler,251 tions.Theslightimprovementfortheslopedmodelisalsoheav- McCullough, P.R.,Crouzet, N.,Deming,D.,&Madhusudhan, N.2014,ApJ, ilyinfluencedbyonlyacoupledatapointsfromasingletransit, 791,55 inAugust2007,asshowninthesubplotofFig.2.Ifthebest-fit Ohta,Y.,Taruya,A.,&Suto,Y.2005,ApJ,622,1118 Pont,F.,Sing,D.K.,Gibson,N.P.,etal.2013,MNRAS,432,2917 modelis robust, itshould withstandremovingthe Augusttran- Sing,D.K.,Fortney,J.J.,Nikolov,N.,etal.2016,Nature,529,59 sit; however, doing so means the data is then best-fit by a flat Snellen,I.A.G.2004,MNRAS,353,L1 line(χ2 =1.6,BIC=11.1andχ2 =1.7,BIC=12.6forflatand r r sloped line, respectively).Consequently,we believe B16 likely reportawavelength-dependenttrendinR duetoinsufficienter- p ror analysis, and that its agreement with the literature may be purelycoincidental. 4. Conclusions Weoutlineourconclusionsonapoint-by-pointbasisbelow. – Thetechniquepresentedin B16 usingthe ratioof the areas ofthelocal(starlightbehindtheplanet)totheout-of-transit CCFcannotbeusedtodetermineR ,asR mustbeknown p p aprioriforthetransitlightcurvenormalisationrequiredfor ground-based spectra. This is shown both analytically and usingasimulatedstar-planetsystem. – TheanalyticalβapproximationusedinB16alsointroduces (slight) systematic trends with planet position due to inad- equately accounting for limb darkening and fractional area occultationeffects,andunderestimatesthevalueofR . p – WepostulatethattheR variationsreportedinB16arelikely p duetounderestimatederrors(largelyoriginatingfromuseof oversampled CCFs), as our reanalysis of the HD189733b transitsfurtherdemonstratesthattheonlyR recoverableis p thatinjectedintothetransitlightcurvenormalisation. – Chromatic RM measurements from ground-based spec- tra are not possible without taking the Doppler informa- tion into account. Hence, for future measurements, we ad- vise readers to either follow the works of Snellen (2004); DiGloriaetal. (2015) or to apply the line-profile tomogra- phyof CollierCameronetal. (2010) directly on each spec- tralpassband. Acknowledgements. We thank the referees, I. A. G. Snellen and S. Albrecht, fortheircarefulreadingandconstructivecomments,whichimprovedtheclarity ofthemanuscript.WealsothankF.Borsaforusefuldiscussions.Additionally, HMCthanksE.deMooijforsuggestingweusesimulatedstarstotestthework herein. HMCandCAWgratefully acknowledge supportfromtheLeverhulme Trust(grantRPG-249).HMC,VB,CLandAWacknowledgethefinancialsup- portoftheNationalCentreforCompetenceinResearch“PlanetS”supportedby theSwissNationalScienceFoundation(SNSF).CAWalsoacknowledges sup- portfromSTFCgrantST/L000709/1,andAWacknowledgesadditionalfinancial supportdirectlyfromtheSNSF.ThisresearchhasmadeuseofNASA’sAstro- physicsDataSystemBibliographicServices. References Borsa,F.,Rainer,M.,&Poretti,E.2016,A&A,590,A84 Cegla,H.M.,Lovis,C.,Bourrier,V.,etal.2016a,A&A,588,A127 Cegla,H.M.,Oshagh,M.,Watson,C.A.,etal.2016b,ApJ,819,67 Cegla,H.M.,Watson,C.A.,Shelyag,S.,&Mathioudakis,M.2015,inCam- bridgeWorkshoponCoolStars,StellarSystems,andtheSun,Vol.18,18th CambridgeWorkshoponCoolStars,StellarSystems,andtheSun,ed.G.T. vanBelle&H.C.Harris,567–574 Articlenumber,page5of5

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.