The GTC exoplanet transit spectroscopy survey IX. Detection of Haze, Na, K, and Li in the super-Neptune WASP-127b
G. Chen, E. Palle, L. Welbanks, J. Prieto-Arranz, N. Madhusudhan, S. Gandhi, N. Casasayas-Barris, F. Murgas, L. Nortmann, N. Crouzet, H. Parviainen, D. Gandolfi
AAstronomy & Astrophysics manuscript no. ms c (cid:13)
ESO 2018May 31, 2018
The GTC exoplanet transit spectroscopy survey IX.
Detection of haze, Na, K, and Li in the super-Neptune WASP-127b
G. Chen , , , E. Pallé , , L. Welbanks , J. Prieto-Arranz , , N. Madhusudhan , S. Gandhi , N. Casasayas-Barris , , F.Murgas , , L. Nortmann , , N. Crouzet , , H. Parviainen , , and D. Gandolfi Instituto de Astrofísica de Canarias, Vía Láctea s / n, E-38205 La Laguna, Tenerife, Spaine-mail: [email protected] Departamento de Astrofísica, Universidad de La Laguna, Spain Key Laboratory of Planetary Sciences, Purple Mountain Observatory, Chinese Academy of Sciences, Nanjing 210008, China Institute of Astronomy, University of Cambridge, Madingley Road, Cambridge CB3 0HA, UK Dipartimento di Fisica, Universitá di Torino, Via P. Giuria 1, I-10125, Torino, ItalyReceived March 16, 2018; accepted May 18, 2018
ABSTRACT
Exoplanets with relatively clear atmospheres are prime targets for detailed studies of chemical compositions and abundances in theiratmospheres. Alkali metals have long been suggested to exhibit broad wings due to pressure broadening, but most of the alkalidetections only show very narrow absorption cores, probably because of the presence of clouds. We report the strong detection of thepressure-broadened spectral profiles of Na, K, and Li absorption in the atmosphere of the super-Neptune WASP-127b, at 4.1 σ , 5.0 σ ,and 3.4 σ , respectively. We performed a spectral retrieval modeling on the high-quality optical transmission spectrum newly acquiredwith the 10.4 m Gran Telescopio Canarias (GTC), in combination with the re-analyzed optical transmission spectrum obtained withthe 2.5 m Nordic Optical Telescope (NOT). By assuming a patchy cloudy model, we retrieved the abundances of Na, K, and Li, whichare super-solar at 3.7 σ for K and 5.1 σ for Li (and only 1.8 σ for Na). We constrained the presence of haze coverage to be around 52%.We also found a hint of water absorption, but cannot constrain it with the global retrieval owing to larger uncertainties in the probedwavelengths. WASP-127b will be extremely valuable for atmospheric characterization in the era of James Webb Space Telescope. Key words.
Planetary systems – Planets and satellites: individual: WASP-127b – Planets and satellites: atmospheres – Techniques:spectroscopic
1. Introduction
The characterization of exoplanet atmospheres could play a crit-ical role in connecting to planet formation histories (e.g., Öberget al. 2011; Madhusudhan et al. 2014; Mordasini et al. 2016).The inference of atmospheric metallicity and elemental ratiosmust rely on the diagnosis of spectral features originated in plan-etary atmospheres. Transmission spectroscopy has taught us thatclouds and hazes are common in hot Jupiter atmospheres (e.g.,Sing et al. 2016), which mute spectral features and lead to de-generation in the parameter space. Nevertheless, great e ff ort hasbeen made with the WFC3 instrument on the Hubble Space Tele-scope (HST) to search for the water feature within 1.1–1.7 µ m.Based on the water abundance, an atmospheric metallicity en-richment trend has been noticed from massive hot Jupiters towarm Neptunes (e.g., Kreidberg et al. 2014; Wakeford et al.2017; Arcangeli et al. 2018; Nikolov et al. 2018).WASP-127b is one of the rare short-period super-Neptunesin the transition gap from Jupiter mass to Neptune mass (Mazehet al. 2016), and the characterization of its atmosphere couldhelp explain its formation mechanisms. This object has a massof 0 . ± . M Jup and a radius of 1 . ± . R Jup , and or-bits a G5 star every 4.17 days (Lam et al. 2017). Its large at-mospheric scale height H eq / R (cid:63) = kT / ( µ gR (cid:63) ) = . T eq = µ = . − , g p = .
14 m s − , R (cid:63) = . R (cid:12) ), together with its bright host star ( V = . / VO absorption.We report the detection of scattering haze, Na, and K, and ahint of water based on the new data acquired with the 10.4 mGran Telescopio Canarias (GTC) together with the publishedNOT data. This paper is organized as follows. In Sect. 2, wesummarize the observations and data reduction. In Sect. 3, wedetail the light-curve analysis. In Sect. 4, we discuss the atmo-spheric properties inferred from the transmission spectrum. Wepresent additional figures and tables in the Appendices.
2. Observations and data reduction
One transit of WASP-127b was observed on the night of January19, 2018 with the OSIRIS spectrograph (Sánchez et al. 2012) atthe GTC. The CCD chip 1 of OSIRIS was configured with 2 × (cid:48)(cid:48) per pixel) and 200 kHz readout mode to recordthe spectral data, while chip 2 was switched o ff . The spectral datawere collected using the R1000R grism and a customized 40 (cid:48)(cid:48) -wide slit. A reference star (TYC 4916-897-1; r (cid:48) mag = .
0) at aseparation of 40.5 (cid:48)(cid:48) was simultaneously monitored with WASP-127 ( r (cid:48) mag = . Article number, page 1 of 14 a r X i v : . [ a s t r o - ph . E P ] M a y & A proofs: manuscript no. ms
600 700 800 900 1000Wavelength (nm)0.51.01.52.02.53.0 C o un t s ( ) Na I H α K I WASP-127Reference
Fig. 1.
Example stellar spectra of WASP-127 (red) and the reference star(blue) obtained with the R1000R grism of GTC / OSIRIS on the nightof January 19, 2018. The color-shaded areas indicate the divided pass-bands that are used to create the spectroscopic light curves.
The observation lasted from 00:26 UT to 07:16 UT, whilethe morning twilight started at 06:42 UT. Two jumps in starlocations occurred because of guiding problems, and the starswere roughly put back to the original location afterward (seeAppendix A). Three exposure times were tested in the first 36exposures, and fixed to 6 sec for the remaining 777 exposuresuntil the end. The duty cycle is roughly 20% owing to the read-out overheads of ∼
23 sec. The night was dark and mostly clear.The airmass decreased from 2.01 to 1.19 and then rose to 1.89 inthe end. The observation was slightly defocused and the spatialpoint spread function is well Gaussian. The measured full widthat half maximum (FWHM) of the spatial profile varied between1.25 (cid:48)(cid:48) and 3.12 (cid:48)(cid:48) , resulting in a seeing-limited spectral resolutionof ∼
20 Å.The two-dimensional spectral images were calibrated foroverscan, bias, flat field, and sky background following themethod described in Chen et al. (2017a,b). The sky backgroundmodel was constructed in the wavelength space, where no curva-ture of sky lines exists, and was then subtracted from the originalspectral image after being transformed back to the original pixelspace. This process made use of wavelength solutions that werecreated using the line lists from the HeAr, Ne, and Xe arc lamps,which were acquired with the R1000R grism and the 1.3 (cid:48)(cid:48) slit.The one-dimensional spectra were extracted using the optimalextraction algorithm (Horne 1986) that has a fixed aperture di-ameter of 21 pixels ( ∼ (cid:48)(cid:48) ). This diameter value has been opti-mized over a wide range of aperture sizes and results in the leastscatter in the white-color light curve.To create light curves, the UT time stamp was shifted to eachmid-exposure time and converted to the Barycentric Julian Datein the Barycentric Dynamical Time standard (BJD TDB ; Eastmanet al. 2010). The flux of each star at each exposure was integratedover any given wavelength range, where the counts of the twoedge pixels were fractionally added and those of the in-betweenpixels were directly summed. The light curve was recorded asthe flux ratios between the target and reference stars after be-ing normalized by the out-of-transit data points. The white-colorlight curve was integrated over 535–908 nm, but excluding the755–768 nm region to minimize the contamination of the strongtelluric O -A band. The spectroscopic light curves were createdto have a wavelength span of 5 nm at wavelengths shorter than908 nm and 10 nm at longer wavelengths (see Fig. 1).To avoid ruining the systematics training process, the datataken during the big jump, when the first guiding loss occurred,
200 100 0 100 2000.9880.9900.9920.9940.9960.9981.0001.002 R e l a t i v e f l u x
200 100 0 100 2000.9880.9900.9920.9940.9960.9981.0001.002 R e l a t i v e f l u x
200 100 0 100 200Time from mid-transit [min]21012 O - C [ − ] σ =
340 ppm
Fig. 2.
White-color light curve obtained with GTC / OSIRIS. From top tobottom are the raw light curve (i.e., the normalized target-to-referenceflux ratios), the light curve corrected for systematics, and the best-fittingresiduals. The red line shows the best-fitting model. were not used in the subsequent analyses. A few exposures atthe beginning and end of the observations were not used becausethey are either at very high airmass ( ∼
2) or in the late phase ofmorning twilight (see the gray shaded areas in Fig. A.1).
3. Light-curve analysis
The light-curve data were modeled by the analytic transit model(Mandel & Agol 2002) together with the Gaussian process (GP;Rasmussen & Williams 2006) to account for the systematicstrends. The transit model was assumed as the mean functionof GP, and implemented using the Python package batman (Kreidberg 2015), where the quadratic limb darkening law wasadopted. The limb darkening coe ffi cients (LDC) were calculatedusing the Kurucz ATLAS9 stellar atmosphere models with stel-lar e ff ective temperature T e ff = g (cid:63) = .
9, and metallicity [Fe / H] = − .
18 (Espinoza & Jordán 2015).The two LDC were always fitted with Gaussian priors of thewidth σ LDC = . george (Ambikasaran et al. 2015), with the correlated systemat-ics modeled by the covariance matrix in the form of the squaredexponential (SE) kernel, k SE ( x i , x j , θ ) = A exp (cid:34) − N (cid:88) α = (cid:18) x α, i − x α, j L α (cid:19) (cid:35) . (1)The GP input variables x i , j could be time or auxiliary trends suchas position drifts ( x , y ) and seeing variations ( s x , s y ) measured inthe cross-dispersion and spatial directions. Uniform priors wereused for all the GP hyperparameters. When time is used as a Article number, page 2 of 14. Chen et al.: The GTC exoplanet transit spectroscopy survey IX.
GP input, the corresponding scale hyperparameter L t is alwaysforced to be no shorter than the ingress / egress time (0.01749days in this case). The posterior distributions were determinedby the Python package emcee (Foreman-Mackey et al. 2013)with the A ffi ne Invariant Markov chain Monte Carlo (MCMC)Ensemble sampler.We chose to jointly analyze our GTC / OSIRIS light curvestogether with the NOT / ALFOSC light curves published by Palleet al. (2017). For the white-color light curves, the fitted parame-ters were inclination i , scaled semimajor axis a / R (cid:63) , mid-transittime T mid , planet-to-star radius ratio R p / R (cid:63) , LDC u and u , andthe GP hyperparameters. Period P was fixed to the literaturevalue. Eccentricity e was fixed to zero. In this joint modeling,the white-color light curves of the two transits shared the same i and a / R (cid:63) , but were allowed to have di ff erent T mid , R p / R (cid:63) , u ,and u . The NOT / ALFOSC data adopted time and seeing ( s y ,measured as the FWHM of the spatial profile) as the GP inputs,while the GTC / OSIRIS data adopted time, seeing s y , and cross-dispersion drift ∆ x as the GP inputs. Our MCMC process for thewhite-color light-curve joint analysis consisted of 90 walkers,each with two burn-ins of 500 steps and another 3000 steps.The spectroscopic light curves were modeled individuallyfor each transit. We first constructed the empirical common-mode noise model by dividing the white-color light-curve databy the best-fitting analytic transit model. Every spectroscopiclight curve was divided by this common-mode noise. The model-ing of the corrected spectroscopic light curves adopted the sameGP inputs as their white-color light curves. The fitted parame-ters were R p / R (cid:63) , u , u , and the GP hyperparameters. The otherparameters, including i , a / R (cid:63) , and T mid , were fixed to the best-fitting values derived from the white-color light curve of corre-sponding transit. Our MCMC process for the spectroscopic lightcurves consisted of 32 walkers, each with two burn-ins of 500steps and another 2500 steps.To combine the two data sets, a corrective constant o ff set of ∆ R p / R (cid:63) = . / ALFOSCtransmission spectrum. This was determined in the commonwavelength range 525–590 nm between the NOT / ALFOSC andGTC / OSIRIS stellar spectra. The wavelength was limited up to590 nm to avoid any imperfect correction of second-order con-tamination in the NOT / ALFOSC stellar spectra. Light curves ofthis 525–590 nm band were created after the di ff erence in instru-mental response function was corrected. A joint modeling of theNOT / ALFOSC and GTC / OSIRIS 525–590 nm light curves wasperformed in a similar manner to the spectroscopic light curves,except that they shared the same u and u . We obtained R p / R (cid:63) = . ± . / ALFOSC, R p / R (cid:63) = . ± . / OSIRIS, and ∆ R p / R (cid:63) = . ± . ff er-ence. Such an o ff set could come from an instrumental bias withinthe 1 σ uncertainty or variation of stellar flux baseline caused bystellar activity. Given the consistency in the spectral shapes be-tween the NOT / ALFOSC and GTC / OSIRIS transmission spec-tra, it is not likely caused by the stellar activity.
4. Results and discussion
We present the derived parameters from the white-color light-curve joint analysis in Table 1. In addition to having refined thetransit parameters i and a / R (cid:63) , we also have revised the orbitalephemeris by combining our two mid-transit times with that inthe discovery paper. We show the GTC / OSIRIS white-color lightcurve in Fig. 2 and present the spectroscopic light curves andtheir transit depths in Appendix C.
Table 1.
Derived parameters from the GTC / OSIRIS and NOT / ALFOSCjoint analysisParameter GTC / OSIRIS NOT / ALFOSC
Transit parameters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .R p / R (cid:63) ± ± i [ ◦ ] 87.88 ± a / R (cid:63) ± u ± ± u ± ± Mid-transit times . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .T mid [MJD a ] 8138.670144 ± ± Revised ephemeris . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .T [MJD a ] 7248.741276 ± P [days] 4.17807015 ± Notes. ( a ) MJD = BJD
TDB − The GTC / OSIRIS and NOT / ALFOSC transmission spectraagree well with each other in the common wavelength rangeof 520–945 nm (see Fig. C.1), although the latter has a lowerspectral resolution and larger uncertainties. The large scatter ofNOT / ALFOSC data might be attributed to the imperfect correc-tion of the second-order contamination (Stanishev 2007; Palleet al. 2017). Given that the GTC / OSIRIS data have su ffi cientquality at a higher spectral resolution, we chose not to includethe NOT / ALFOSC data at wavelengths longer than 590 nm inthe subsequent analysis and discussion. We verified that includ-ing the extra NOT / ALFOSC data ( λ >
590 nm) does not changeour subsequent results, but introduces more noise.The combined transmission spectrum spans 5.6 atmosphericscale heights ( H / R (cid:63) = . σ level, and the cloud-free 1400 K solar composition fiducial at-mospheric model (Kempton et al. 2017) at 10.8 σ level. How-ever, the spectrum does exhibit four distinct broad spectral fea-tures, which can be well explained by the linearly scaled fiducialmodel individually. At the blue wavelengths ( λ <
580 nm), thecombined transmission spectrum shows a slope m obs = dR p d ln λ = ( − . ± . R (cid:63) . (2)This can be converted to a dimensionless slope S = m obs / H eq = − . ± .
6, which is consistent with hazes composed of sulphide(e.g., MnS and ZnS) with particle sizes less than 0.1 µ m (Pinhas& Madhusudhan 2017). Focusing on the GTC / OSIRIS measure-ments, the transmission spectrum shows the pressure-broadenedwings of Na and K centered at around 589.3 nm and 768.2 nm,respectively. At red wavelengths ( λ >
900 nm), the transmissionspectrum starts to bump and then drop, which is consistent withthe water absorption feature.
To globally retrieve the atmospheric properties at the day-nightterminator region, we performed a spectral retrieval modeling onthe combined transmission spectrum of WASP-127b. We usedan atmospheric retrieval code for transmission spectra adapted
Article number, page 3 of 14 & A proofs: manuscript no. ms
400 500 600 700 800 900 1000Wavelength [nm]0.900.951.001.051.101.151.20 T r a n s i t d e p t h [ % ] Haze Na Li K H O Retrieved model2 σ σ NOT/ALFOSCGTC/OSIRIS N u m b e r o f s c a l e h e i g h t s Fig. 3.
Transmission spectrum of WASP-127b and retrieved models. The blue circles and black squares with error bars are the observed spectrumby NOT / ALFOSC and GTC / OSIRIS, respectively. This spectrum shows an enhanced slope at the blue optical, strong absorption peaks at 589.3 nm,670.8 nm, and 768.2 nm, and another bump at the red optical. These features can be explained by the model spectrum when including opacitiesresulting from haze, Na, Li, K, and H O, respectively. The red curve shows the retrieved median model while the shaded areas show the 1 σ and2 σ confidence regions. The yellow diamonds show the binned version for the retrieved median model. from recent works (Gandhi & Madhusudhan 2018) and usedthe cloud / haze parameterization of MacDonald & Madhusudhan(2017). The haze is included as σ = a σ ( λ/λ ) γ , where γ is thescattering slope, a is the Rayleigh-enhancement factor, and σ isthe H Rayleigh scattering cross section (5 . × − m ) at thereference wavelength λ =
350 nm. The model computes line-by-line radiative transfer in a transmission geometry, assuminga plane parallel planetary atmosphere in hydrostatic equilibriumand local thermodynamic equilibrium. The model also assumesthe reference pressure as a free parameter, which is the pressureat an assumed radius R p = . R Jup (Lam et al. 2017). Consid-ering R p as a free parameter does not significantly change our re-sults. The chemical composition and temperature profile are freeparameters in the model. We performed a set of retrievals con-sidering the molecules, metal oxides and hydrides, and atomicspecies that could be present in hot Jupiter atmospheres: i.e.,H O, CO, CH , NH , CO , TiO, AlO, FeH, TiH, CrH, Na, K, Li,V, and Fe (Madhusudhan et al. 2016). Our atmospheric pressure-temperature (P-T) model consists of six parameters (Madhusud-han & Seager 2009) and we consider models in the range fromclear to cloudy, both with and without scattering hazes and withinhomogeneous cloud coverage. The opacities for the chemicalspecies are adopted from Gandhi & Madhusudhan (2017, 2018).The Bayesian inference and parameter estimation is conductedusing the nested sampling algorithm implemented via the Multi-Nest application (Feroz et al. 2009), as pursued in previous stud-ies (MacDonald & Madhusudhan 2017; Gandhi & Madhusudhan2018).The optical transmission spectrum of WASP-127b providesstrong constraints on its atmospheric composition. We report thedetection of K at a confidence level of 5.0 σ , Na at 4.1 σ , and Li at3.4 σ in the spectrum along with an indication of H O (see TableB.1 for the Bayesian model comparison). Figure 3 shows the best-fit spectrum to the data along with the significance contours.The models without Na, K, or Li fail to explain the peaks inabsorption at ∼ ∼ ∼ ffi cult to ac-curately retrieve chemical abundances in uniformly cloudy at-mospheres (Benneke & Seager 2012; Gri ffi th 2014; Heng &Kitzmann 2017). However, this degeneracy can be broken or re-duced if the atmosphere is cloud-free (Gri ffi th 2014; Heng &Kitzmann 2017) or partially cloudy (MacDonald & Madhusud-han 2017), where H Rayleigh scattering or pressure-broadenedline wings can help constrain the reference pressure level. Ourpartially cloudy model enables us to reduce the impact of thisdegeneracy and account for any remaining correlation in the de-rived uncertainties. In particular, the presence of optical spectrahelps further mitigate the degeneracy, especially when the spec-trum is not flat.Our modeling retrieved volume mixing ratios of log( X Na ) = − . + . − . , log( X K ) = − . + . − . , and log( X Li ) = − . + . − . forNa, K, and Li, respectively. The retrieved Na abundance is tenta-tively super-solar (Asplund et al. 2009) at 1.8 σ , while the abun-dances of K and Li are super-solar at 3.7 σ and 5.1 σ , respectively.The Li abundance log( A Li ) = log( X Li ) +
12 is also significantlyhigher than the super-solar value of the host star (1 . ± . ff e & Chabrier 2010). We haveverified that the retrieved volume mixing ratios remain well con-sistent within 1 σ error bar even if the reference planet radiusis a free parameter instead of assumed as R p = . R Jup (see
Article number, page 4 of 14. Chen et al.: The GTC exoplanet transit spectroscopy survey IX.
Appendix B for retrieval tests). The retrieved haze is ∼ Rayleighscattering, has a coverage of φ = + − %, and a power-law ex-ponent of γ = − . + . − . . The P-T profile is relatively uncon-strained by the data (see Fig. B.1).Our models also considered the presence of water in the at-mosphere of the planet. Although the spectral shape in the wave-length range 833–1018 nm resembles a water absorption owingto low flux and fringing e ff ect, the current error in the data within900-1018 nm is relatively large and does not constrain the abun-dance of water in the global atmospheric retrieval. We find anominal water signature with a relatively weak abundance con-straint of log(H O) = − . + . − . , which can be confirmed withHST near-infrared spectroscopy in the near term, and with JamesWebb Space Telescope (JWST) in the future.
5. Conclusions
We have observed one transit of the super-Neptune WASP-127busing the long-slit mode of GTC / OSIRIS. We revised the tran-sit parameters by jointly analyzing the GTC / OSIRIS white-colorlight curve with the already published NOT / ALFOSC light curveusing Gaussian processes. The resulting transmission spectrafrom the two transits are consistent in spectral shape. With thecombined transmission spectrum, we detected a scattering hazeat the blue wavelengths, the pressure-broadened spectral profilesof Na, K and Li absorption, and found a hint of water absorp-tion at the red wavelengths. We inferred a tentatively super-solarabundance for Na, significantly super-solar abundances for Kand Li, a coverage of ∼
52% for haze, and a weakly constrainedabundance for water based on the spectral retrieval model-ing. Our results showcase that large-aperture ground-based tele-scopes could result in high-quality spectroscopy that is compa-rable to or even better than what HST can do (also see Sedaghatiet al. 2017). Thanks to its rare physical parameters and the con-firmation of relatively clear sky, WASP-127b will become abenchmark for exoplanet atmospheric characterization in the eraof JWST and high-resolution spectroscopy at 30 m class tele-scopes.
Acknowledgements.
This research is based on observations made with the GranTelescopio Canarias (GTC), installed in the Spanish Observatorio del Roquede los Muchachos, operated on the island of La Palma by the Instituto de As-trofísica de Canarias. This work is partly financed by the Spanish Ministry ofEconomics and Competitiveness through grant ESP2013-48391-C4-2-R. G.C.acknowledges the support by the National Natural Science Foundation of China(Grant No. 11503088) and the Natural Science Foundation of Jiangsu Province(Grant No. BK20151051), and the Minor Planet Foundation of the Purple Moun-tain Observatory. This research has made use of Matplotlib (Hunter 2007) andthe VizieR catalog access tool, CDS, Strasbourg, France (Ochsenbein et al.2000). The authors thank the anonymous referee for useful comments on themanuscript.
References
Ambikasaran, S., Foreman-Mackey, D., Greengard, L., Hogg, D. W., & O’Neil,M. 2015, IEEE Transactions on Pattern Analysis and Machine Intelligence,38 [ arXiv:1403.6015 ]Arcangeli, J., Désert, J.-M., Line, M. R., et al. 2018, ApJ, 855, L30Asplund, M., Grevesse, N., Sauval, A. J., & Scott, P. 2009, ARA&A, 47, 481Bara ff e, I. & Chabrier, G. 2010, A&A, 521, A44Benneke, B. & Seager, S. 2012, ApJ, 753, 100Bouvier, J. 2008, A&A, 489, L53Chen, G., Guenther, E. W., Pallé, E., et al. 2017a, A&A, 600, A138Chen, G., Pallé, E., Nortmann, L., et al. 2017b, A&A, 600, L11Eastman, J., Siverd, R., & Gaudi, B. S. 2010, PASP, 122, 935Espinoza, N. & Jordán, A. 2015, MNRAS, 450, 1879Feroz, F., Hobson, M. P., & Bridges, M. 2009, MNRAS, 398, 1601 Foreman-Mackey, D., Hogg, D. W., Lang, D., & Goodman, J. 2013, PASP, 125,306Gandhi, S. & Madhusudhan, N. 2017, MNRAS, 472, 2334Gandhi, S. & Madhusudhan, N. 2018, MNRAS, 474, 271Gri ffi th, C. A. 2014, Philosophical Transactions of the Royal Society of LondonSeries A, 372, 20130086Heng, K. & Kitzmann, D. 2017, MNRAS, 470, 2972Horne, K. 1986, PASP, 98, 609Hunter, J. D. 2007, Computing in Science and Engineering, 9, 90Israelian, G., Delgado Mena, E., Santos, N. C., et al. 2009, Nature, 462, 189Kempton, E. M.-R., Lupu, R., Owusu-Asare, A., Slough, P., & Cale, B. 2017,PASP, 129, 044402Kreidberg, L. 2015, PASP, 127, 1161Kreidberg, L., Bean, J. L., Désert, J.-M., et al. 2014, ApJ, 793, L27Lam, K. W. F., Faedi, F., Brown, D. J. A., et al. 2017, A&A, 599, A3MacDonald, R. J. & Madhusudhan, N. 2017, MNRAS, 469, 1979Madhusudhan, N., Agúndez, M., Moses, J. I., & Hu, Y. 2016, Space Sci. Rev.,205, 285Madhusudhan, N., Amin, M. A., & Kennedy, G. M. 2014, ApJ, 794, L12Madhusudhan, N. & Seager, S. 2009, ApJ, 707, 24Mandel, K. & Agol, E. 2002, ApJ, 580, L171Mazeh, T., Holczer, T., & Faigler, S. 2016, A&A, 589, A75Mordasini, C., van Boekel, R., Mollière, P., Henning, T., & Benneke, B. 2016,ApJ, 832, 41Nikolov, N., Sing, D. K., Fortney, J. J., et al. 2018, Nature, 000, 00Öberg, K. I., Murray-Clay, R., & Bergin, E. A. 2011, ApJ, 743, L16Ochsenbein, F., Bauer, P., & Marcout, J. 2000, A&AS, 143, 23Palle, E., Chen, G., Prieto-Arranz, J., et al. 2017, A&A, 602, L15Pinhas, A. & Madhusudhan, N. 2017, MNRAS, 471, 4355Rasmussen, C. E. & Williams, C. K. I. 2006, Gaussian Processes for MachineLearningSánchez, B., Aguiar-González, M., Barreto, R., et al. 2012, in Society of Photo-Optical Instrumentation Engineers (SPIE) Conference Series, Vol. 8446, So-ciety of Photo-Optical Instrumentation Engineers (SPIE) Conference Series,4Sedaghati, E., Bo ffi n, H. M. J., MacDonald, R. J., et al. 2017, Nature, 549, 238Sing, D. K., Fortney, J. J., Nikolov, N., et al. 2016, Nature, 529, 59Stanishev, V. 2007, Astronomische Nachrichten, 328, 948Wakeford, H. R., Sing, D. K., Kataria, T., et al. 2017, Science, 356, 628 Article number, page 5 of 14 & A proofs: manuscript no. ms
Appendix A: Spectral movements caused byguiding problems
Guiding problems occurred twice during the observation. At01:10 UT, the target jumped to a location ∼
180 pixels away inthe spatial direction. The target was drifting for another 20 pixelsuntil it was put back to a location that is close to the original loca-tion. It lost guiding again at around 06:12 UT with slow driftingfor ∼
10 pixels as well. The inspection of absorption lines in theacquired spectra shows that such a drift and jump also exist in thecross-dispersion direction. The absorption lines jumped ∼
60 pix-els for the first jump and the position was ∼ ∼
25 pixels away. Figure A.1shows the drift of spectra location on the CCD in both cross-dispersion and spatial directions, and the raw flux sequence ofthe two stars.
200 100 0 100 200200150100500 R e l a t i v e d r i f t [ p i x ] T T T T Cross dispersionSpatial direction
200 100 0 100 200Time from mid-transit [min]0.800.850.900.951.00 N o r m a li z e d r a w f l u x ReferenceWASP-127
Fig. A.1.
Spectral movements and the raw flux time series. The toppanel shows the drift of spectra location in the cross-dispersion (blue)and spatial (green) directions. The bottom panel shows the normalizedraw flux of WASP-127 (black) and its reference star (red). The grayshaded areas indicate the discarded exposures that are not included insubsequent light-curve modeling, where the two regions on the edge areimpacted by high airmass, the region on the right edge is also impactedby morning twilight, and the middle region has severely lost guiding.The vertical dashed lines indicate the first ( T ), second ( T ), third ( T ),and fourth ( T ) contact of the transit event. Appendix B: Additional retrieval test on thetransmission spectrum
The retrieval results presented in Sect. 4.2, Table B.1, andFig. B.1 have assumed a reference planet radius of R p = . R Jup , and the pressure at this assumed reference radius isa free parameter. We also tested setting the reference planet ra-dius as a free parameter in the retrieval modeling. This additionaltest resulted in consistent posterior distributions with the originaltest. The resulting posterior distributions are shown in Fig. B.2and Fig. B.3. We present the comparison between these two sce-narios in Table B.2 along with the prior on each parameter.
Table B.1.
Bayesian model comparison detections of atmospheric com-positions at the terminator of WASP-127b
Model Evidence Best-fit Bayes factor Detectionln Z i χ r , min B i of Ref.Reference 619.0 1.4 Ref. Ref.No K 608.1 1.8 54838.7 5.0 σ No Na 612.1 1.6 1008.3 4.1 σ No Li 614.6 1.6 79.6 3.4 σ No H O 617.8 1.5 3.4 2.1 σ Table B.2.
Prior information and best-fitting retrieval parameters
Retrieval 1 Retrieval 2Parameter Prior Fixed R ref Free R ref (adopted)log( X Na ) U ( − , − . − . + . − . − . + . − . log( X K ) U ( − , − . − . + . − . − . + . − . log( X Li ) U ( − , − . − . + . − . − . + . − . log( X H O ) U ( − , − . − . + . − . − . + . − . log( P ref ) [bar] U ( − , − . + . − . − . + . − . log( P cloud ) [bar] U ( − , − . + . − . − . + . − . log( a ) U ( − ,
10) 4 . + . − . . + . − . γ U ( − , − . + . − . − . + . − . φ U (0 ,
1) 0 . + . − . . + . − . R ref [ R Jup ] U (0 . , .
0) 1.37 (fixed) 1 . + . − . -5 -4 -3 -2 -1 P r e ss u r e [ b a r ] Retrieved median fit2 σ σ Fig. B.1.
Retrieved pressure temperature profile. The shaded areas showthe 1 σ and 2 σ confidence regions. Appendix C: Additional figures and tables
Figures C.2–C.4 show the GTC / OSIRIS spectroscopic lightcurves after removing the common-mode systematics and best-fitting light-curve residuals. Table C.1 shows the transit depthsof the NOT / ALFOSC transmission spectrum (Palle et al. 2017),which has been re-analyzed using the method described in this
Article number, page 6 of 14. Chen et al.: The GTC exoplanet transit spectroscopy survey IX.
Fig. B.2.
Marginalized posterior probability densities for the retrieved species and haze parameters with the reference planet radius fixed at R p = . R Jup . paper. Table C.2 shows the transit depths of the newly derivedGTC / OSIRIS transmission spectrum.
Article number, page 7 of 14 & A proofs: manuscript no. ms
Fig. B.3.
Marginalized posterior probability densities for the retrieved species and haze parameters with the reference planet radius as a freeparameter.Article number, page 8 of 14. Chen et al.: The GTC exoplanet transit spectroscopy survey IX.
GTC/OSIRISNOT/ALFOSC ( χ ν = 2 . Flat ( χ ν = 2 . Rayleigh scattering ( χ ν = 3 . Exo-transmit model: 1 × solar, Na+K+H O, clear ( χ ν = 1 . Retrieved model
400 500 600 700 800 900 1000Wavelength [nm]0.0900.0950.1000.1050.110 R p / R N u m b e r o f s c a l e h e i g h t s Fig. C.1.
Transmission spectrum of WASP-127b. The GTC / OSIRIS and NOT / ALFOSC measurements are shown in squares and circles witherror bars, respectively. A flat line is used to represent the optically thick clouds. The green line shows a representative pure Rayleigh scatteringatmosphere. The purple line shows a fiducial atmospheric model with the solar abundance, with Na, K, and H O included, and with an isothermaltemperature of 1400 K (Kempton et al. 2017). The red line shows the retrieved model spectrum from our retrieval modeling. The binned modelsare shown in diamonds. One atmospheric scale height is equivalent to H eq / R (cid:63) = . ff set as a free parameter. Article number, page 9 of 14 & A proofs: manuscript no. ms
200 100 0 100 200Time from mid-transit [min]0.680.690.700.710.720.730.740.750.760.770.780.790.800.810.820.830.840.850.860.870.880.890.900.910.920.930.940.950.960.970.980.991.001.01 R e l a t i v e f l u x + O ff s e t
200 100 0 100 200Time from mid-transit [min]0.680.690.700.710.720.730.740.750.760.770.780.790.800.810.820.830.840.850.860.870.880.890.900.910.920.930.940.950.960.970.980.991.001.01 O - C + O ff s e t σ = σ = σ = σ = σ = σ = σ = σ = σ = σ = σ = σ = σ = σ = σ = σ = σ = σ = σ = σ = σ =
988 ppm σ = σ =
967 ppm σ =
974 ppm σ =
940 ppm σ = σ =
978 ppm σ = σ = σ = σ =
967 ppm σ =
953 ppm
Fig. C.2.
Spectroscopic light curves after removing the common-mode systematics ( left panel ) and corresponding best-fitting residuals ( rightpanel ) of WASP-127b obtained with the R1000R grism of GTC / OSIRIS. The passbands have been indicated in Fig. 1. This shows the passbandsfrom 524.2 nm to 679.2 nm, which are continued in Fig. C.3 and Fig. C.4.Article number, page 10 of 14. Chen et al.: The GTC exoplanet transit spectroscopy survey IX.
200 100 0 100 200Time from mid-transit [min]0.680.690.700.710.720.730.740.750.760.770.780.790.800.810.820.830.840.850.860.870.880.890.900.910.920.930.940.950.960.970.980.991.001.01 R e l a t i v e f l u x + O ff s e t
200 100 0 100 200Time from mid-transit [min]0.680.690.700.710.720.730.740.750.760.770.780.790.800.810.820.830.840.850.860.870.880.890.900.910.920.930.940.950.960.970.980.991.001.01 O - C + O ff s e t σ = σ = σ = σ =
975 ppm σ =
973 ppm σ =
995 ppm σ = σ =
974 ppm σ = σ = σ = σ = σ = σ =
998 ppm σ = σ = σ = σ = σ = σ = σ = σ = σ = σ = σ = σ = σ = σ = σ = σ = σ = σ = Fig. C.3.
Same as Fig. C.2, but for the passbands from 684.2 nm to 850.5 nm. Article number, page 11 of 14 & A proofs: manuscript no. ms
200 100 0 100 200Time from mid-transit [min]0.610.620.630.640.650.660.670.680.690.700.710.720.730.740.750.760.770.780.790.800.810.820.830.840.850.860.870.880.890.900.910.920.930.940.950.960.970.980.991.001.01 R e l a t i v e f l u x + O ff s e t
200 100 0 100 200Time from mid-transit [min]0.610.620.630.640.650.660.670.680.690.700.710.720.730.740.750.760.770.780.790.800.810.820.830.840.850.860.870.880.890.900.910.920.930.940.950.960.970.980.991.001.01 O - C + O ff s e t σ = σ = σ = σ = σ = σ = σ = σ = σ = σ = σ = σ = σ = σ = σ = σ = σ = σ = σ = σ = σ = σ = Fig. C.4.
Same as Fig. C.2, but for the passbands from 855.5 nm to 1013.0 nm.Article number, page 12 of 14. Chen et al.: The GTC exoplanet transit spectroscopy survey IX.
Table C.1.
Transmission spectrum obtained with NOT / ALFOSC R p / R (cid:63) Center Width1 4050 200 0.1128 ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± Notes.
The corrective o ff set ∆ R p / R (cid:63) = . & A proofs: manuscript no. ms
Table C.2.
Transmission spectrum obtained with GTC / OSIRIS R p / R (cid:63) R p / R (cid:63) Center Width Center Width1 5242 50 0.1000 ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ±±