A multi-wavelength look at the GJ 9827 system -- No evidence of extended atmospheres in GJ 9827 b and d from HST and CARMENES data
Ilaria Carleo, Allison Youngblood, Seth Redfield, Nuria Casasayas Barris, Thomas R. Ayres, Hunter Vannier, Luca Fossati, Enric Palle, John H. Livingston, Antonino F. Lanza, Prajwal Niraula, Julián D. Alvarado-Gómez, Guo Chen, Davide Gandolfi, Eike W. Guenther, Jeffrey L. Linsky, Evangelos Nagel, Norio Narita, Lisa Nortmann, Evgenya L. Shkolnik, Monika Stangret
DD raft version J anuary
19, 2021Typeset using L A TEX twocolumn style in AASTeX63
A multi-wavelength look at the GJ 9827 systemNo evidence of extended atmospheres in GJ 9827 b and d from
HST and CARMENES data I laria C arleo , A llison Y oungblood , S eth R edfield , N uria C asasayas B arris , T homas R. A yres , H unter V annier , L uca F ossati , E nric P alle , J ohn H. L ivingston , A ntonino F. L anza , P rajwal N iraula , J uli ´ an D. A lvarado -G´ omez , G uo C hen , D avide G andolfi , E ike W. G uenther , J effrey L. L insky , E vangelos N agel , N orio N arita , L isa N ortmann , E vgenya L. S hkolnik , and M onika S tangret Astronomy Department and Van Vleck Observatory, Wesleyan University, Middletown, CT 06459, USA INAF – Osservatorio Astronomico di Padova, Vicolo dell’Osservatorio 5, I-35122, Padova, Italy Laboratory for Atmospheric and Space Physics, University of Colorado, Boulder, CO 80303, USA Instituto de Astrof´ısica de Canarias, C / V´ıa L´actea s / n, 38205 La Laguna, Spain Departamento de Astrof´ısica, Universidad de La Laguna, 38206 La Laguna, Spain Center for Astrophysics and Space Astronomy, University of Colorado, Boulder, CO 80309 Space Research Institute, Austrian Academy of Sciences, Schmiedlstrasse 6, A-8041 Graz, Austria Department of Astronomy, University of Tokyo, 7-3-1 Hongo, Bunkyo-ku, Tokyo 113-0033, Japan INAF - Osservatorio Astrofisico di Catania, Via S. Sofia 78, I-95123, Catania, Italy Department of Earth, Atmospheric and Planetary Sciences, Massachusetts Institute of Technology, Cambridge, MA 02139 Leibniz Institute for Astrophysics Potsdam An der Sternwarte 16, 14482 Potsdam, Germany Key Laboratory of Planetary Sciences, Purple Mountain Observatory, Chinese Academy of Sciences, Nanjing 210023, PR China Dipartimento di Fisica, Universit`a degli Studi di Torino, via Pietro Giuria 1, I-10125, Torino, Italy Th¨uringer Landessternwarte Tautenburg, Sternwarte 5, D-07778 Tautenburg, Germany JILA, University of Colorado and NIST, Boulder, CO 80309-0440, USA Hamburger Sternwarte, Gojenbergsweg 112, D-21029 Hamburg, Germany Astrobiology Center, NINS, 2-21-1 Osawa, Mitaka, Tokyo 181-8588, Japan National Astronomical Observatory of Japan, NINS, 2-21-1 Osawa, Mitaka, Tokyo 181-8588, Japan JST, PRESTO, 7-3-1 Hongo, Bunkyo-ku, Tokyo 113-0033, Japan Institut f¨ur Astrophysik, Friedrich-Hund-Platz 1, D-37077 G¨ottingen, Germany School of Earth and Space Exploration Arizona State University, 781 S Terrace Road, Tempe, AZ 85281, USA (Received; Revised; Accepted 11 January 2021)
Submitted to AJABSTRACTGJ 9827 is a bright star hosting a planetary system with three transiting planets. As a multi-planet systemwith planets that sprawl within the boundaries of the radius gap between terrestrial and gaseous planets, GJ 9827is an optimal target to study the evolution of the atmospheres of close-in planets with a common evolutionaryhistory and their dependence from stellar irradiation. Here, we report on the
Hubble Space Telescope (HST) andCARMENES transit observations of GJ 9827 planets b and d. We performed a stellar and interstellar mediumcharacterization from the ultraviolet
HST spectra, obtaining fluxes for Ly α and MgII of F (Ly α ) = (5.42 + . − . ) × − erg cm − s − and F (MgII) = (5.64 ± × − erg cm − s − . We also investigated a possible absorptionsignature in Ly α in the atmosphere of GJ 9827 b during a transit event from HST spectra, as well as H α andHe I signature for the atmosphere of GJ 9827 b and d from CARMENES spectra. We found no evidence ofan extended atmosphere in either of the planets. This result is also supported by our analytical estimations ofmass-loss based on the measured radiation fields for all the three planets of this system, which led to a mass-lossrate of 0.4, 0.3, and 0.1 planetary masses per Gyr, for GJ 9827 b, c, and d respectively. These values indicate that Corresponding author: Ilaria [email protected] a r X i v : . [ a s t r o - ph . E P ] J a n C arleo et al .the planets could have lost their volatiles quickly in their evolution and probably do not retain an atmosphere atthe current stage. Keywords:
Exoplanet astronomy: Exoplanet systems — High resolution spectroscopy — stars: activity — ISM:abundances INTRODUCTIONThe
Kepler mission (Borucki et al. 2010) discovered thatplanets between the size of Earth and Neptune are the mostcommon type of exoplanets in our Galaxy (i.e., Borucki et al.2011; Batalha et al. 2013; Rowe et al. 2015). Defined as plan-ets with radii between 1 and 4 R ⊕ , they do not have any ana-logue in our Solar System. This makes them very captivatingtargets for the study of their formation and evolution history,as well as understanding their compositions, interior struc-tures, and atmospheres. Moreover, terrestrial planets may bethe most attractive targets for the search of biosignatures.One of the most interesting and still unexplained character-istics of the sub-Neptune sized planet population is the gapin the radius distribution around 1.6 R ⊕ found by Fulton et al.(2017). Planet below this radius may be naked rocky cores,while those above this value have retained their atmospheres.A possible explanation for this gap suggests that gas-rich su-per Earths (mainly solid, rocky planets with a radius up to1.5 R ⊕ ) will retain or lose their envelope depending on thelevel of irradiation from their host stars (Lopez et al. 2012;Owen & Wu 2017; Loyd et al. 2020). It might be also pos-sible that the mass-loss can be caused by the luminosity ofthe cooling planetary cores (core-powered mass-loss mech-anism, Gupta & Schlichting 2019). In this context, observ-ing small planets is essential to better understand the role ofphoto-evaporation in their evolution. Moreover, observingmulti-planet systems o ff er an extra benefit, since such sys-tems presumably formed under the same initial conditions(i.e., same age, same flux evolution) and provide a uniqueopportunity to compare the compositions of planets with dif-ferent sizes, as well as atmospheric characteristics at di ff erentincident flux.In this paper, we present the case of GJ 9827, a K6V stardiscovered to host three transiting planets in 1:3:5 commen-surability by Kepler / K2 (Niraula et al. 2017; Rodriguez et al.2018). Teske et al. (2018) showed with archival radial ve-locity (RV) observations from the
Magellan
II Planet FinderSpectrograph (PFS) that planet b has a mass of ∼ ± M ⊕ ,classifying it as one of the densest planets with an iron massfraction of (cid:38) ± M ⊕ . Prieto-Arranz et al. (2018) also calculated the incidentfluxes for the three planets, pointing to a rocky composition for planets b and c, and a gaseous composition for planet d,which could possibly retain an extended atmosphere.Rice et al. (2019) combined data from Niraula et al. (2017),Teske et al. (2018), Prieto-Arranz et al. (2018), and a newHARPS-N RV dataset, more precisely constraining the plan-etary masses for GJ 9827 b and d. A more recent analy-sis by Kosiarek et al. (2020) refined the ephemerides from Spitzer observations and, adding all the RVs from the pre-vious work to their multi-year HIRES RV follow-up, moreprecisely constrained the parameters of this multi-planet sys-tem. The orbital periods, masses and radii of the three plan-ets by (Kosiarek et al. 2020) are reported in Table 1 and arethe ones used in our analyses. It is worth to notice that theradii of the three planets span the above-mentioned radiusgap at ∼ R ⊕ , making this system even more appealing foruncovering how super-Earths form and evolve. Straddlingthis rocky / gaseous planets divide, GJ 9827 is ideal for study-ing the simultaneous evolution of planets at di ff erent orbitaldistances, having the stellar properties, including age, con-trolled.A key aspect for the assessment of long-term habitabilityin a planetary system is the atmospheric mass loss of planetsdue to the high-energy environment and stellar wind of itshost star. Several spectral lines sensitive to extended atmo-spheres have provided unique measurements of mass loss, in-cluding Lyman- α (e.g., Vidal-Madjar et al. 2003; Kulow et al.2014; Ehrenreich et al. 2015), H α (e.g., Jensen et al. 2012;Cauley et al. 2015; Casasayas-Barris et al. 2018), and He I ff e et al.2005). This is especially relevant for planets with masses < M J , i.e., the Neptune and super-Earth regime (Owen &Wu 2013). Even in a less catastrophic case, the atmosphericcomposition of the planet can be highly altered (Lopez et al.2012).The photochemistry of planetary atmospheres hosted bycool stars is controlled by the two strongest UV emissionlines in the stellar spectrum, Lyman- α (Ly α ) and Mg II (Mad-husudhan 2019). They can drive significant water loss (e.g., y α , H α , and H e I in GJ 9827 b and d α photons from the line of sight (Wood et al. 2005;Youngblood et al. 2016). The LISM is a rich and complexcollection of clouds that leads to a unique absorption profile,often with more than one absorbing cloud (Redfield & Lin-sky 2004, 2008). To aid in the reconstruction of the intrinsicstellar Ly α flux, we observed and characterized the Mg II stellar emission, which samples a similar level of the stellarchromosphere as Ly α , and yet is significantly less altered byLISM absorption. Given then high column density for HI(log N ∼
18 cm − ), >
70% of the intrinsic stellar chromo-spheric emission HI line can be absorbed by the LISM, andin some cases it can be >
90% (Wood et al. 2005), whereas thelower abundances and column densities for MgII (log N ∼ − ), mean that <
20% of the intrinsic stellar chromosphericemission MgII line is absorbed (Redfield & Linsky 2002).Fitting the MgII lines provides not only a characterization ofthe intrinsic chromospheric line shape for Ly α , but also a fitto the LISM absorption profile. Although Wood et al. (2005)used the line shape of MgII to inform the line shape of Ly α ,the simultaneous fitting of the Ly α , MgII and ISM absorptionthat we perform in our analysis is novel.We present here HST and CARMENES transit observa-tions aimed at characterizing the atmosphere of GJ 9827 b, inparticular to evaluate its transit signature in di ff erent wave-length domains. The CARMENES analysis also includesdata for GJ 9827 d, the only planet of this system with pre-vious atmospheric characterization: Kasper et al. (2020)investigated the 10,830 Å HeI triplet spectra, finding no ab-sorption feature. In Section 2 we describe the observationsand data reduction for HST and CARMENES data. We alsoused
HST data for characterizing the star, with Ly α , MgII,and XUV fluxes estimation in Section 3. We investigate thepossibility of an atmospheric planetary signal in Section 4and finally present our discussion in Section 5 and conclu-sion in Section 6. OBSERVATIONS AND DATA REDUCTION2.1.
HST data
A four-orbit
HST pointing on GJ 9827 was carried out withthe Space Telescope Imaging Spectrograph (STIS) during the28 August 2018 transit of planet b as part of the Cycle 25program GO-15434 (PI: S. Redfield). The first-order far-ultraviolet (FUV) grating G140M was used, with its 1222 Åsetting, covering the wavelength interval 1194–1250 Å, at aresolution of R ≡ λ/ ∆ λ ∼ . The chromospheric H I α line is formed in the range 1-3 × K. Otherimportant emission lines in this region are the Si
III T ∼ × K) and the N V T ∼ × K), although both of these were expectedto be too faint to be significantly detected in the relativelybrief FUV exposures of this faint, 10th-magnitude mid-K-type star. Nevertheless, the peak signal-to-noise (S / N) per2-pixel resolution element (resel) in the combined spectrumat Ly α was about 35. The 52 (cid:48)(cid:48) × (cid:48)(cid:48) narrow long slit waschosen to minimize geocoronal Ly α contamination from theupper atmosphere of the Earth. After the guide stars wereacquired, a peak-up was performed to center GJ 9827, usingthe 31 (cid:48)(cid:48) × (cid:48)(cid:48) NDC long slit in the visible at low resolutionwith the STIS CCD. The rest of the initial orbit was occupiedby the first G140M exposure, of 1.8 ks. The remaining threeorbits had single G140M exposures, of 2.9 ks each. A sum-mary of the four FUV observations is provided in Table 2. Inrelation to the planetary transit, the first two exposures werepre-ingress, the third was in-transit, while the final was post-egress.Three months later, a single-orbit out-of-transit near-ultraviolet (NUV) spectrum of the Mg II (cid:48)(cid:48) × (cid:48)(cid:48) to achieveoptimum resolution ( R ∼ , II ab-sorptions. The 2713 Å setting also captures an importantFe II multiplet near 2600 Å, although unfortunately the NUVcontinuum of the mid-K star was too weak for the Fe II in-terstellar absorption to be detected (the peak S / N per resel atMg II / spectralimages derived from rectified versions of the original long-slit stigmatic spectrograms. The image x -direction is alongthe dispersion, with 0.053 Å pixel − . The image y -direction is the spatial (cross-dispersion) dimension, with0.03 (cid:48)(cid:48) pixel − . The image pixel flux densities tabulated in theX2D files, and associated photometric errors, are providedper Å and per 0.0293 (cid:48)(cid:48) (the latter is the cross-dispersion an-gular pixel size), so the extracted spectrum (and photometricerror) must be multiplied by that angular factor to yield fluxdensities (erg cm − s − Å − ).The upper panel of Fig. 1 illustrates a co-added version ofthe 2D spatial / spectral image of the four G140M exposures.The vertical extent of the image represents a ±
40 pixel slicein the detector y -direction ( ∼ ± (cid:48)(cid:48) ) centered on the appar-ent stellar Ly α feature. The horizontal extent is 1200 pixelsalong the dispersion. The narrow geocoronal stripe is con-spicuous in the y -direction, bisecting the broader stellar Ly α C arleo et al . Table 1.
Summary of GJ 9827 planets’ parameters by Kosiareket al. (2020) adopted for our analyses.Planet Orbital Period Mass Radius(days) (M ⊕ ) (R ⊕ )b 1.2089765 ± × − ± ± ± × − ± ± ± × − ± ± Table 2.
Summary of
HST / STIS exposures of GJ 9827.Dataset Mode-CENWAVE Slit UT Start Time Exposure Time Peak S / N [ λ ](Å) ( (cid:48)(cid:48) × (cid:48)(cid:48) ) (yy–mm–dd.ddd) (ks) (resel − [Å])ODRL01010 G140M–1222 52 × × × × × feature. The red band outlines a 9-pixel flux extraction region( ∼ (cid:48)(cid:48) ) for the stellar spectrum, while the blue dashed bandshighlight flanking regions where the background was sam-pled. The two background bands are 25 pixels wide, begin-ning at ±
15 pixels from the center. The wide bands increasethe signal-to-noise for the background subtraction. In prac-tice, we eliminated the top three of the background values ateach wavelength, in an e ff ort to mitigate hot pixels.The middle panel of Fig. 1 depicts the extracted 1D spec-trum from the co-added image, zoomed into the Ly α fea-ture. The green tracing is the gross spectrum; blue with greyshadow is the background including the geocoronal H I emis-sion feature; and black is the net flux (gross–background).The wavelength scale was set to place the geocoronal Ly α feature at its laboratory wavelength (1215.670 Å). The thinred dashed curve represents the 10 σ photometric error level(per resel), derived from the spatial / spectral values providedin the original X2D files, smoothed, for display, by a dou-ble pass of a rectangular filter 15 pixels wide. The bottompanel shows the extracted Ly α features for the four exposuresseparately, where the geocoronal feature has been subtractedfrom the stellar profile. There are small di ff erences betweenthe Ly α peaks of the four profiles, and the di ff erence in thewing of the red profile (that corresponds to ODRL01010 ofTable 2, the first observation of the sequence) around 1215 Åis close to 3 σ in significance.We reduced the single NUV exposure directly from theCALSTIS pipeline X1D file, which is a tabulation of ex- tracted flux densities and associated photometric errors for27 of the echelle orders contained in the original E230H-2713 spectral image. We merged the individual orders to-gether, tapering the overlapping zones to preserve the opti-mum S / N. GJ 9827 is a relatively faint star for STIS high-resolution echelle spectroscopy, so the main features visibleare the Mg II α emission forms. The peak S / N at the k line isabout 8, and interstellar absorption is apparent in both emis-sion cores.The observed Ly α feature is the combination of the intrin-sic stellar emission profile and the interstellar medium (ISM)attenuation profile (Wood et al. 2005). The core of the stellaremission line originates in the lower transition region and up-per chromosphere ( T ∼ × K), while the outer wingsform deeper in the stellar chromosphere. The Ly α emissioncore is strongly attenuated by neutral hydrogen (H I ) and deu-terium (D I ) gas over the 29.7 pc sightline to GJ 9827. Thestar’s + − radial velocity (Prieto-Arranz et al. 2018)shifts the stellar emission lines away from much of the ISMattenuation centered near 0 km s − (Redfield & Linsky 2008),giving the observed Ly α feature its asymmetric appearance.The Mg II cores are less a ff ected than Ly α owing to the muchsmaller cosmic abundance of magnesium.2.2. CARMENES data y α , H α , and H e I in GJ 9827 b and d Figure 1.
Top : Co-added 2D spectral images (in pixels) of GJ 9827G140M exposures. The red band represent flux extraction region forthe stellar spectrum; the blue bands represent the region where thebackground, including geocoronal Ly-alpha, was sampled.
Middle :Co-added 1D spectrum zoomed in the Ly α feature. The green curveis the total extracted spectrum; the gray shaded outlined by the bluecurve is the background including the geocoronal emission; andblack curve is the net flux (the total flux minus the background). Thered dashed line is the photometric error. Bottom : Separate 1D spec-tra zoomed in the Ly α feature for the four exposures (ODRL01010red, ODRL01020 green, ODRL01030 blue, ODRL01040 black).The thinner dotted-dashed curves are (smoothed) 3-sigma noise lev-els per resel for each of the spectra. The one for exposure 10 (in red)is higher than the others because the exposure was shorter. We observed one transit of GJ 9827b and one transit ofGJ 9827d with the CARMENES spectrograph (Calar Altohigh-Resolution search for M dwarfs with Exo-earths withNear-infrared and optical Echelle Spectrographs; Quirren-bach et al. 2014, 2018) located at Calar Alto Observatory,in 13 August 2018 and 06 November 2018, respectively.CARMENES simultaneously covers the optical (VIS; 520 to960 nm) and near-infrared range (NIR; 960 to 1710 nm), giv-ing access to two important traces of planetary evaporationprocesses: the near-infrared He I triplet at 10830 Å and thevisible H α line at 6562 .
81 Å. GJ 9827 is su ffi ciently bright(V = = ff erent read-out overheads of the VIS and NIR arms, the central timeof each exposure was coincident in both. Resetting of theAtmospheric Dispersion Corrector took place between twoexposures (i.e. during readout) and approximately every 30minutes. For the planet b observations, the exposure time at the be-ginning was 190 s but was then increased to 198 s, and theaveraged S / N achieved is 44 in the He I order and 30 in theH α order. On the other hand, for the planet d observations,the exposure times were 200 s and 197 s, with a S / N of 47and 35 around He I and H α orders, respectively. Due toa cloud crossing, the exposures taken at 19:22, 19:26, and19:30 UT presented S / N below 30 in the He I order and be-low 20 in the H α order. These exposures are discarded fromthe atmospheric analysis. In addition, for technical reasons,the observations were stopped from 19:59 to 20:16 UT. Wenote strong telluric contamination of the He I region in bothnights. A log table of the CARMENES observations is givenin Table 3.We reduced CARMENES observations with theCARMENES pipeline CARACAL (CARMENES Reduc-tion And CALibration; Caballero et al. 2016), which per-forms bias, flat-relative optimal extraction (Zechmeister et al.2014), cosmic-ray correction, and the wavelength calibration(Bauer et al. 2015). The wavelengths are given in vacuumand the reduced spectra referenced to the terrestrial restframe. STELLAR AND INTERSTELLAR MEDIUMCHARACTERIZATIONUV stellar emission dramatically a ff ects the fate of an exo-planet’s atmosphere, both in the physical and chemical com-position and through possible mass loss, especially whenthe planet closely orbits its host star. EUV and X-ray pho-tons heat an exoplanet’s upper atmospheric layers and drivemass loss, whereas UV photons are primarily responsiblefor photochemistry (Madhusudhan 2019). In particular, theLy α emission line at 1215.67 Å dominates the far ultravio-let (FUV) spectrum of late-type stars and is the main sourcefor the photodissociation of molecules such as water andmethane. However, the Ly α emission line is heavily ab-sorbed by neutral hydrogen in the interstellar medium (ISM)between the star and the Earth and is also contaminated bythe geocoronal emission. This necessitates a reconstructionto recover the intrinsic stellar flux as seen by the planet’satmosphere. In the next sections we describe the Ly α andMg II emission line reconstructions for GJ 9827, the derivedproperties of the line-of-sight ISM, and how GJ 9827’s UVemission compares to other K dwarfs.3.1. Intrinsic Lyman- α and Mg II reconstruction To correct for the ISM attenuation of the Ly α and MgII lines and recover the intrinsic stellar emission lines, weuse the approach described in Youngblood et al. (2016) andGarc´ıa Mu˜noz et al. (2020), modified to jointly fit Ly α andMg II. This allows for stronger constraints on the physical C arleo et al . Table 3.
Summary of CARMENES observations.Observing night 2018–08–13 2018–11–06Planet transit GJ 9827 b GJ 9827 dNumber of exposures VIS 43 40Number of exposures NIR 43 39Exposure time VIS (s) 200-192 200Exposure time NIR (s) 190-198 197Airmass change 1.86-1.28 1.51-1.27-1.31Mean S / N in H α order 30 35Mean S / N in He I order 44 47 parameters common to the two transitions. Using a MarkovChain Monte Carlo method (Foreman-Mackey et al. 2013),we simultaneously fit a model of the Ly α and Mg II intrinsicstellar emission lines and ISM absorption lines to the STISG140M and E230H spectra.For Ly α , we parameterized the broad, intrinsic stellaremission line as a single Voigt profile in emission with aGaussian in absorption for the line’s self-reversal. For thenarrow Mg II emission lines, we used a single Gaussianprofile in emission for each line and a single Gaussian foreach line’s self-reversal. We required the radial velocities forthe two Mg II lines to be identical and also required theirFWHMs to be coupled such that FWHM k = × FWHM h ,because high resolution stellar spectra indicate that the k lineis ∼
5% broader than the h line (Brian E. Wood, private com-munication). We did not require the radial velocity of Ly α tomatch the Mg II lines in case of small wavelength o ff sets be-tween the two gratings.The ISM absorbers (H I, D I, Mg II) are modeled as sin-gle Voigt profiles in absorption (see Sec. 3.2 for justificationfor one component), each parameterized by a radial velocity (cid:51) (o ff set from their corresponding stellar emission line’s ra-dial velocity), Doppler width b , and species column density N . We assume that the ISM absorbers are coming from thesame interstellar gas with the same kinematics, so all threeISM species were required to have the same radial velocityo ff set from their stellar emission line’s radial velocity, andthe Doppler widths were all connected assuming pure ther-mal broadening by b MgII = b HI / √ . b DI = b HI / √ I and D I are dominated by thermal broadening andthus this assumption should be fine, but heavier species likeMg II will also have a significant contribution from turbulentbroadening. Therefore, our value of b MgII will underestimate the true Doppler width of MgII. The D I and H I columndensities were fixed to be N (D I) / N (H I) = × − (Woodet al. 2004), and N (Mg II) was not linked to N (D I) or N (HI). The corresponding intrinsic emission lines and ISM ab-sorption lines were multiplied together and convolved to thecorresponding instrument resolution using the STIS G140Mand E230H line spread functions.Given the lack of an in-transit detection (see Section 4.1),we co-added all four orbits’ spectra to maximize the preci-sion of the reconstruction. Figure 2 shows the best fit modeland intrinsic profiles with 68% and 95% confidence intervals.We find an intrinsic Ly α flux with 68% confidence interval(5.42 + . − . ) × − erg cm − s − . For Mg II, we find a best-fitintegrated flux of the h line of (2.30 ± . × − erg cm − s − , and the integrated flux of the k line is (3.34 ± . × − erg cm − s − (Figure 4). The total flux is F(Mg II h + k) = (5.64 ± . × − erg cm − s − . These fluxes and the fittedISM parameters (described in 3.2) are printed in Table 4.In Figure 3, we compare the line core shapes of the re-constructed Ly α and Mg II lines, with ISM attenuation andinstrument resolution e ff ects removed. The Ly α line is > − broader than Mg II, as expected. Their self-reversalshapes are roughly similar, but there are significant discrep-ancies between the depth, width, and asymmetry of the twospecies. For Ly α , the self-reversal was only allowed to devi-ate ±
10 km s − from line center due to significant degeneracybetween the self-reversal depth and ISM absorption, but iscentered almost exactly at line center. The self-reversal in theMg II lines was allowed to vary more widely, and it is read-ily apparent from the observed spectra that the self-reversalis asymmetric (i.e., not centered exactly at the stellar veloc-ity). The Ly α self-reversal depth also appears larger than MgII’s, however, the uncertainty on Ly α ’s self-reversal parame- y α , H α , and H e I in GJ 9827 b and d
Interstellar Medium Characterization
We used the LISM Kinematic Calculator (Redfield & Lin-sky 2008), which calculates whether or not a cloud of LISMtraverses any given sight line, in addition to the radial andtraverse velocities of the clouds in a given direction. In thecase of GJ 9827, the LISM Kinematic Calculator yields notraversing clouds along the sight line. We note that this islikely due to the boundaries of clouds not being well sampledby limited LISM datasets, and this sight line probably doestraverse at least one LISM cloud. There are 5 clouds pass-ing within 20 ◦ of GJ 9827’s sight line, including the LocalInterstellar Cloud (LIC). The radial velocities of these cloudsrange from –7 to +
10 km s − with a weighted average of –3.1 ± − . Given the limitations of our data (low S / Nin Mg II and low spectral resolution for H I), we fit a singleISM cloud component to our spectra. This results in fittedparameters that are likely akin to an average of the true pa-rameters of the multiple clouds along the sight line. From ourreconstructions described in the previous section, we find forinterstellar H I the following parameters: (cid:51) HI = –3.35 + . − . km s − , log N (H I ) = . + . − . cm − , and b HI = + . − . km s − . For interstellar Mg II , we find log N (Mg II ) = + . − . cm − . Recall that Mg II ’s Doppler b parameter isdefined as b HI / √ . ff set between the ISM radialvelocity and the intrinsic stellar radial velocity was defined tobe the same for the two species. This gives (cid:51) MgII = + . − . km s − and b MgII = + . − . km s − . While we expect b MgII to be underestimated, this is within the reasonable range forthe Doppler width of LISM Mg II absorption (Redfield &Linsky 2004). Note that the absolute uncertainty in the STISMAMA wavelength calibration is 0.5–1 pixels, or 6–12 kms − for G140M at Ly α and 1.5–3.0 km s − for E230H at MgII.The constraint on the Mg II interstellar absorbers is weakbecause of the narrowness of the stellar emission line servingas a backlight for the interstellar Mg II ions to absorb against.The interstellar Mg II atoms are Doppler shifted almost –30km s − from an emission line with FWHM =
24 km s − , andno stellar continuum is detected around the stellar emissionlines.From our simultaneously-fitted H I and Mg II column den-sities, we calculate the ratio N (MgII) / N (HI) = + . − . × − for GJ 9827’s sight line. This value overlaps with the N (MgII) / N (HI) ratio from Linsky (2019) ((3 . + . − . ) × − )at the 68% confidence interval (2.0 × − –2.6 × − ). lism.wesleyan.edu / LISMdynamics.html
To verify our measurement of the Mg II ISM absorption,we also applied the ISM fitting technique described in Red-field & Linsky (2002). We assume the interstellar absorbershave a radial velocity equal to the LIC for this line-of-sight( + − ; Redfield & Linsky 2008), and we assume thewings of the Mg II emission line to be symmetric. Mask-ing the blueward, ISM-a ff ected half of the line, we mirroredthe redward half of the Mg II k profile to create the assumedintrinsic stellar profile, which clearly indicated the presenceof ISM absorption in the blue wing. The low S / N ratio inthe wings of the line prevented a free fit to the column den-sity. However, visual inspection using the mirrored profileled to a minimum column density log N (Mg II) ≈ σ uncertainty range of the pre-vious fitting results shown in Figure 4. Using the log N (MgII) = N (MgII) / N (HI) = × − , which is inagreement with the ratio value from Linsky (2019). The con-clusion from using two di ff erent methods is that LISM ab-sorption is present on the blueward wing, although it doesnot significantly alter the Mg II emission profile. Our as-sumption of a single LISM absorption component and ourchoice of the stellar emission profile (i.e., automated Gaus-sian versus a mirrored profile) do not formally enter into ourerror analysis, and given the comparison to the LISM average N (MgII) / N (HI) ratio indicate that our Mg II LISM columndensity may be slightly underestimated.We also searched for other ISM-a ff ected spectral lines inthe spectral range of STIS / E230H (2576 - 2823 Å), such asthe iron lines at 2586 and 2600 Å, but the S / N is too low andthe spectrum does not present any other spectral features.3.3.
GJ 9827 and other K dwarfs
We compare GJ 9827’s Ly α and Mg II fluxes to otherK dwarfs with measured fluxes for both lines (Fig. 5). Tocompare to data from Wood et al. (2005) and Youngbloodet al. (2016), we convert our fluxes into surface fluxes us-ing the 0.579 ± (cid:12) radius from Kosiarek et al. 2020 andthe 29.69 pc distance from Gaia DR2 to obtain F S (Ly α ) = (2.81 + . − . ) × erg cm − s − and F S (MgII) = (2.92 ± × erg cm − s − . GJ 9827’s rotation period is poorly con-strained, but appears to be between 15-30 days with a mostlikely value of 28.72 days (Rice et al. 2019).Compared to other K dwarfs of similar rotation period ,GJ 9827 has approximately 3.0 times less Mg II surfaceflux, and 2.8 times more Ly α surface flux. Mg II is com-monly used as an estimator for the di ffi cult-to-observe Ly α The comparison K dwarfs with rotation period >
15 days include HD 40307,HD 85512, HD 97658, α Cen B, 61 Cyg A, (cid:15)
Ind, 40 Eri A, 36 Oph A, and σ Gem. More information on all the K dwarfs in Figure 5 can be found inWood et al. (2005) and Youngblood et al. (2016). C arleo et al . Table 4.
Stellar and ISM parameters.Parameter Value Units
Stellar fluxes
Intrinsic Ly α flux F(Ly α ) (5.42 + . − . ) × − erg cm − s − Intrinsic MgII h flux F (MgII h) (2.30 ± × − erg cm − s − Intrinsic MgII k flux F (MgII k) (3.34 ± × − erg cm − s − Intrinsic MgII h + k flux F (MgII h + k) (5.64 ± × − erg cm − s − Surface Ly α flux F S (Ly α ) (2.81 + . − . ) × erg cm − s − Surface MgII flux F S (MgII) (2.92 ± × erg cm − s − ISM absorbers’ parameters
HI radial velocity (cid:51) HI -3.4 + . − . km s − HI Doppler width b HI + . − . km s − HI column density log N (HI) 18.22 + . − . cm − MgII radial velocity (cid:51)
MgII + . − . km s − MgII Doppler width b MgII + . − . km s − MgII column density log N (MgII) 12.11 + . − . cm − N (MgII) / N (HI) 8.0 + . − . × − N ote —The stellar radial velocity is + − (Prieto-Arranz et al. 2018). The v HI and v MgII values are derived from the same free parameter in the fit, a radialvelocity o ff set from the stellar H I and Mg II emission lines centroids. The v HI and v MgII values di ff er because of di ff erences in the absolute wavelengthsolution between the two gratings (G140M and E230H). The b HI and b MgII values are also derived from the same free parameter ( b HI ) under the assumptionof thermal line broadening ( b MgII = b HI / √ . line (Wood et al. 2005), and if we relied on the Mg II obser-vation to estimate GJ 9827’s Ly α emission, we would haveunderpredicted it by almost a factor of 5. Figure 2 showshow this MgII-derived Ly α profile (both intrinsic and ob-served) would appear and how the Ly α spectrum stronglyrules out this flux level at the > σ level. This underestima-tion could have significant consequences for the atmospheresof GJ 9827’s planets, because Ly α has a strong e ff ect on pho-tochemistry and as a proxy for the EUV (Linsky et al. 2014),could have implications for the atmospheric escape from theplanets.We estimate that there is a 0.001% probability thatGJ 9827’s Ly α flux is consistent with the other K dwarfs inFigure 5, and a 16% probability that its Mg II flux is consis-tent with other K dwarfs. To calculate this, we drew 10 ran-dom samples from GJ 9827’s Ly α and Mg II flux posteriordistributions as well as 10 random samples from a normaldistribution describing the red best-fit lines in the middle andright panels of Figure 5, and determined what percentage ofthe posterior samples overlapped with samples from the best-fit lines. The normal distributions describing the best-fit lineshad means equal to zero, and standard deviations equal to thestandard deviations of all data points about the best-fit line, normalized by the best-fit line. For Ly α , the standard de-viation is 0.29 and for Mg II it is 0.46. The samples fromthe Ly α and Mg II posterior distributions were also cast asdi ff erences from the best-fit line and then normalized by thebest-fit line.What is causing the apparently significant discrepancy inthe Mg II - Ly α flux ratio for this star? It is possible thatthis ratio is within the expected scatter of K dwarf UV flux-flux relations, and more UV spectroscopic observations Kdwarfs are needed to quantify that typical scatter. Mg IIand Ly α form in slightly di ff erent regions in the stellar at-mospheres and therefore their emission mechanisms are notexactly coupled, but in practice the scatter is likely dominatedby the non-simultaneity of the Mg II and Ly α observations.For example, the GJ 9827 Mg II and Ly α observations weretaken 100 days apart, or approximately 3.5 stellar rotationperiods apart. Stellar surface inhomogeneities (e.g., activeregions, faculae, plage) as well as evolution of these featuresresponsible for much of the Mg II and Ly α emission couldcause deviations in the expected Mg II - Ly α flux ratio.Here we consider some additional e ff ects, including metal-licity and rotation evolution. GJ 9827 has slightly sub-solarmetallicity ([M / H] = − .
26 to − .
5; Rice et al. 2019), which y α , H α , and H e I in GJ 9827 b and d α aswell as GJ 9827’s potentially anomalously-high Ly α flux. Adetailed investigation into the e ff ect of metallicity on relativeMg II and Ly α line strengths in K dwarfs would be needed todetermine that. Another possible explanation is the observedrotation evolution of Ly α luminosity compared to less opti-cally thick chromospheric lines like C II. Pineda et al. ( un-der review ) showed with a sample of young and field age Mdwarfs that Ly α luminosity declines much more slowly withincreasing stellar rotation period (a proxy for stellar age) thanother far-UV lines like C II. Assuming that Mg II behavesmore similarly to lines like C II rather than Ly α , GJ 9827could be at a point in its rotational evolution when its Mg IIluminosity has decreased significantly but Ly α has not beenimpacted as much by stellar spindown. More UV observa-tions of low-activity K dwarfs are needed to determine if thisincreasing Ly α / Mg II flux ratio with increasing rotation pe-riod is a real e ff ect. Figure 2.
The reconstruction of the Ly α profile is shown in the toptwo panels (the middle panel is a zoomed version of the top panelwith no changes). The STIS spectrum is represented in black witherror bars (the error bars are generally smaller than the black linewidth). The best-fit model and 68% and 95% confidence intervalsis shown as the pink line, dark gray shading, and light gray shad-ing, respectively (the confidence intervals are also generally thinnerthan the width of the pink line). The intrinsic stellar emission linecorresponding to the best-fit model is shown as the dashed blue line,with the 68% and 95% confidence intervals shown as dark-blue andlight-blue shading, respectively. The bottom panel shows the resid-uals (data-model) / (data uncertainty). The dashed gold and greenlines show how the intrinsic and observed (ISM attenuated) spec-tra would respectively appear if the intrinsic fluxes were consistentwith Mg II - Ly α fluxes from the literature (Section 3.3). Figure 3.
The intrinsic stellar profiles for Ly α and Mg II are shownhere, with ISM attenuation and instrumental resolution e ff ects re-moved. Only the Mg II k line is shown for visual clarity, but theh line appears very similar to the k line. The profiles have beennormalized to their peak values, and 68% confidence intervals areshown as the shaded regions. Figure 4.
Reconstructed (orange dashed line) and fitted (attenuated;solid green line) Mg II profiles are shown. The ISM absorption isweak and centered at 2796.38 Å and 2803.56 Å; the e ff ect on theresultant profile is small. In the left panel, the dotted green lineshows the model of the attenuated profile with the 2- σ upper limiton the Mg II column density (log N (Mg II) < / (data uncertainty).4. SEARCHING FOR PLANETARY ATMOSPHERICABSORPTION SIGNAL4.1.
Investigating Lyman- α We investigate the behaviour of GJ 9827’s observed Ly α profiles during the di ff erent phases of the transit. Startingfrom the four spectra obtained in Section 2.1 (one per orbit)we calculated the in-transit absorption depth as 1 − F IN / F OUT ,where F IN is the flux of the Ly α line during the transit and F OUT is the out-of-transit flux. Fig. 6 shows the out-of-transit(black solid line) and in-transit (red solid line) spectra for0 C arleo et al . Figure 5.
GJ 9827’s Ly α and Mg II surface fluxes (corrected for ISM absorption) and rotation period are compared to other F, G, K, and M starsfrom the literature (Wood et al. 2005; Youngblood et al. 2016). The literature sample is shown in colored, open symbols with correspondingbest-fit linear regressions (dashed lines from Wood et al. 2005, solid lines from Youngblood et al. 2016). The example error bars in each panelapply to colored symbols without error bars. In the left panel, the 68%, 95%, and 99.7% confidence intervals from our simultaneous Ly α andMg II reconstruction are shown as black contours. In the right two panels, the black box and whisker symbol shows the median Ly α and MgII surface flux with the 68% and 95% confidence intervals as the box and whiskers, respectively at the assumed stellar rotation period (28.72days; Rice et al. 2019). GJ 9827 b, where the out-of-transit spectrum ( F OUT ) is ob-tained by averaging the three out-of-transit spectra and thesingle in-transit spectrum represents F IN . We find that theobserved Ly α spectra are very similar in and out of the tran-sit, and there is no evident planetary absorption during thetransit. The largest apparent absorption depths occur in thespectral region most strongly contaminated by ISM absorp-tion and geocoronal emission. Figure 6.
Top . Ly α line of GJ 9827 during the transit (red line)and outside (black line). Bottom . In-transit absorption depth, 1 − F IN / F OUT . Furthermore, we compare pre-ingress and post-egressspectra, as in Kulow et al. (2014), to search for a possible at-mospheric comet-like tail. McCann et al. (2019) showed thatthe stellar wind can significantly shape the planetary outflow, creating strong absorption signatures many hours before andafter the optical transit. Fig. 7 shows pre-ingress and post-egress spectra in the top panel, and the di ff erence betweenthese spectra in the bottom panel. No evident di ff erence isfound. Figure 7.
Top . Pre-ingress Ly α spectrum (black line) (black line)and post-egress Ly α spectrum (red line). Bottom . Di ff erence be-tween pre-ingress and post-egress spectra. We integrated the flux in the Ly α blue wing from -250 to-75 km s − in the stellar rest frame (1214.787-1215.496 Å)to obtain the average fluxes of each of the four orbits (units10 − erg cm − s − ): 2.20 ± ± ± ± batman package (Kreid- y α , H α , and H e I in GJ 9827 b and d α relative to the star (R Ly α / R (cid:63) ) was the only freeparameter, and we find 1- σ , 2- σ , and 3- σ upper limits of0.36, 0.48, and 0.57 for R Ly α / R (cid:63) in the blue wing. We re-peated this upper limit calculation for the Ly α red wing ( + +
250 km s − in the stellar rest frame; 1215.841-1216.813Å) and find average fluxes of each of the four orbits (units10 − erg cm − s − ) of 1.10 ± ± ± ± σ , 2- σ , and 3- σ upper limits on R Ly α / R (cid:63) in the red wing are 0.21, 0.27, and 0.32.4.2. Investigating He I and H α Prieto-Arranz et al. (2018) suggested that the GJ 9827planetary system is an excellent laboratory to test atmo-spheric evolution and planetary mass-loss rates. We inves-tigate the presence of evaporation traces in the atmospheresof GJ 9827 b and GJ 9827 d using CARMENES observa-tions. In particular, we use the data obtained with the near-infrared channel to study the He I triplet lines (at 10829 .
09 Å,10830 .
25 Å and 10830 .
34 Å), and the visible channel data tostudy the H α line (at 6562 .
81 Å).As a first step, we correct the observed spectra of telluricabsorption contamination using molecfit (Smette et al.2015 and Kausch et al. 2015), assuming the parameters pre-sented in Nagel et al. (submitted) for the CARMENES in-strumental line spread function model. In particular, theHe I region is contaminated by telluric absorption of watervapor and telluric emission of OH (Nortmann et al. 2018;Salz et al. 2018). The water vapour absorption is correctedwith molecfit and the OH emission lines are masked, foll-wing the methods described in Palle et al. (2020). The tel-luric line removal and masked regions are illustrated for eachplanet / night in the top panels of Figures 8 and 9.After removing the telluric contamination, we can extractthe transmission spectrum in both He I and H α regions us-ing the same approach, presented in di ff erent studies suchas Wyttenbach et al. (2015), Casasayas-Barris et al. (2019),and Chen et al. (2020). CARMENES observations are ref-erenced to the Earth’s rest frame. Thus, we shift the spec-tra to the stellar rest frame considering the barycentric radialvelocity information and the system velocity (31 .
95 km s − ;Prieto-Arranz et al. 2018). After computing the ratio of eachstellar spectrum to the master out-of-transit spectrum (com-bination of all out-of-transit spectra using the simple aver-age) we move the residuals to the planet rest frame (see mid-dle panels of Figures 8 and 9). To do this, we calculatethe planet’s radial velocity using the radial velocity semi-amplitudes K b p = . − and K d p = . − , forGJ 9827 b and GJ 9827 d, respectively. These values are cal-culated assuming the stellar radial velocity semi-amplitude( K (cid:63) ), and the planet and star masses reported by Kosiarek et al. (2020), and using K p = K (cid:63) M (cid:63) / M p (Birkby 2018). Fi-nally, we combine all in-transit residuals in the planet restframe to obtain the transmission spectrum. In the He I region,the masked intervals due to OH contamination are the samefor all spectra in the stellar rest frame, but change when mov-ing the spectra to the planet rest frame. When combining thein-transit residuals to extract the transmission spectrum, weonly include the non-masked pixels in the calculation. Thefinal transmission spectra are presented in the bottom panelsof Figures 8 and 9. We note that di ff erent K (cid:63) and mass val-ues are reported in the literature, which result in di ff erent K p values (see Rice et al. 2019 or Prieto-Arranz et al. 2018, forexample). However, these di ff erent K p values do not havesignificant impact on the derived transmission spectra.It is important to notice, however, that using the param-eters from Kosiarek et al. (2020), the mid-transit time ofGJ 9827 d’s observations di ff ers from that obtained using theparameters from Prieto-Arranz et al. 2018, Rodriguez et al.2018, and Rice et al. 2019. This di ff erence is produced,mainly, by the di ff erences in the derived orbital period value.The orbital period derived by Kosiarek et al. (2020) di ff ersby more than 30 s from the ones presented in the previousstudies, and this di ff erence is propagated along the di ff erentepochs. To be on the safe side, we have repeated the anal-ysis using the parameters from the di ff erent references andthe resulting transmission spectra do not show any signifi-cant feature in any of the cases.In the two-dimensional residual maps presented in Fig-ures 8 and 9, we are not able to visually distinguish ab-sorption features during the transit that could have planetaryorigin, for either of He I or H α . The overall transmissionspectrum does not show significant absorption features ei-ther. The excess absorption measured in the transmissionspectra of GJ 9827 b using a 0 . I and H α lines core is − . ± . − . ± . + . ± . − . ± . ≈ nHR p / R (cid:63) ) of the annular areaof one ( n =
1) atmospheric scale height ( H ) during the transitis around 3 × − for both planets, respectively, assuming anatmosphere dominated by H / He mixture and near solar com-position ( µ = . I signals arecomparable to those created by an annular area of n = / N ratio of the observations, especially in the lines core,we find no evidence for an extended H / He upper atmospherearound GJ 9827 b or GJ 9827 d. This is consistent with thenon-detection He I presented for GJ 9827 d by Kasper et al.(2020).2 C arleo et al . N o r m . F l u x HeIGJ 9827b - 2018-08-13 -2.0-1.00.01.0 t i m e - T m i d [ h ] Wavelength (Å) -2.00.02.0 - ( % ) HeIGJ 9827d - 2018-11-06
Wavelength (Å)
Figure 8.
CARMENES transmission spectroscopy results around the near-infrared He I triplet for GJ 9827 b (left column) and GJ 9827 d (rightcolumn). Top panel: GJ 9827’s spectra in the He I region. In the spectra we mark in red the telluric absorption lines and in blue the OH telluricemission lines region. In black we plot the spectra used to construct the transmission spectra, where the telluric absorption lines are correctedwith Molecfit and the telluric OH emission lines are masked. The vertical / oblique cyan broken lines mark the positions of the triplet lines.Middle panel: 2D map of the residuals after dividing each spectrum by the master out-of-transit spectrum, in the stellar rest frame. The bluehorizontal lines show the first and fourth contacts of the transit. The color scale shows the relative flux (F in / F out -1) in %. The white verticalregions correspond to the masked emission lines and the horizontal white regions to time stamps for which observations were discarded ormissing. The vertical / oblique cyan broken lines mark the expected trace of the triplet lines during transit. The data are binned by 0.003 and0.0008 in orbital phase for GJ 9827 b and GJ 9827 d, respectively. Bottom panel: The transmission spectrum for GJ 9827 b and GJ 9827 d. Thered dashed line marks the zero absorption level. Again, the cyan dashed vertical lines show the position of the He I triplet lines. All wavelengthsare presented in the vacuum. N o r m . F l u x HGJ 9827b - 2018-08-13 -2.0-1.00.01.0 t i m e - T m i d [ h ] Wavelength (Å) -4.0-2.00.02.04.0 - ( % ) HGJ 9827d - 2018-11-06
Wavelength (Å)
Figure 9.
Same as Fig. 8 but around the H α line. In this wavelength region only telluric absorption contamination is observed and no spectralregions are masked. y α , H α , and H e I in GJ 9827 b and d DISCUSSIONThe non-detection of planetary Ly α absorption during thetransit of GJ 9827 b can be the result of several factors. Theparticular radial velocity of the host star is such that theinterstellar medium absorbs the line core and most of theblue wing of the Ly α line, while the red wing is almostintact. This is relevant, because past Ly α transit observa-tions obtained for both hot Jupiters and warm Neptunes haveshown that the planetary atmospheric absorption is strongestin the blue wing and it is typically caused by energetic neu-tral atoms, which are fast stellar wind protons that receivedelectrons from slow planetary hydrogen atoms via charge ex-change and are moving towards us at velocities of the or-der of 100 km s − (e.g., Vidal-Madjar et al. 2003; Kislyakovaet al. 2014; Ehrenreich et al. 2015; Khodachenko et al. 2017;Shaikhislamov et al. 2020). Weak planetary atmospheric ab-sorption in the red wing of the Ly α line, attributed to natu-ral and thermal line broadening, has been observed before,for example in the case of HD209458b (Vidal-Madjar et al.2003), but this absorption extends to a few tens of km s − atmaximum into both line wings (Kislyakova et al. 2014; Kho-dachenko et al. 2017). In the particular case of GJ 9827, theinterstellar medium absorption is too close to the blue wingto enable detecting planetary absorption at low velocities andin the red wing the observed stellar emission flux may be tooweak to allow detecting the planetary absorption, if present,above the noise level.It has been suggested that for low-mass planets Ly α ab-sorption may probe the presence of large amounts of waterin planetary atmospheres; in this case the hydrogen is the re-sult of water dissociation and further dragging of the lighterhydrogen in the upper layers as a result of the stellar high-energy irradiation (e.g., Bourrier et al. 2017). In this case, thehydrogen originating from the atmospheric water vapor maybe detectable, but, as described above, the specific configu-ration of the stellar emission and interstellar medium absorp-tion hamper detecting the planetary absorption feature. Fur-thermore, a too large amount of water would also hamper thedetection of hydrogen at Ly α because of the reduced atmo-spheric scale height due to the high mean molecular weight(Garc´ıa Mu˜noz et al. 2020).It is also possible that absorption is not observed becausethe planetary atmosphere does not present enough hydro-gen to be detectable. In fact, GJ 9827 b has a bulk den-sity of 7.47 + . . g cm − , thus consistent with a primarilyrocky composition (Kosiarek et al. 2020). Such a high av-erage density may exclude the possibility that the planethosts a primary, hydrogen-dominated atmosphere or an at-mosphere holding large quantities of water. This is indeedthe most likely explanation for the lack of planetary Ly α ab-sorption. Given the high average density of the planet, therather old age of the system, and the short orbital separation, we can expect that the planet has lost its primary, hydrogen-dominated envelope through escape in the first few hundredsof Myrs (e.g., Kubyshkina et al. 2018a,b), developing thena secondary (e.g., CO -dominated) atmosphere as a result ofmagma ocean solidification. If this process happened whilethe star was still active, it is even possible that the planet hasalso lost this secondary atmosphere through hydrodynamicescape (Kulikov et al. 2006; Tian 2009; Garc´ıa Mu˜noz et al.2020) leaving behind the bare surface exposed to the actionof the stellar wind, which may have then led to the formationof a mineral exosphere (e.g., Miguel et al. 2011; Vidotto et al.2018), not dissimilar from that of Mercury (e.g., Pfleger et al.2015).The conclusion that GJ 9827 b has most likely lost itsprimary hydrogen-dominated envelope is supported also bycalculations of the expected planetary mass-loss rate. Weemployed the stellar distance and relations of Linsky et al.(2014) to estimate the stellar high-energy emission (X-rayand EUV; hereafter XUV) from the reconstructed Ly α flux,obtaining an XUV flux at 1 AU of 13.01 erg cm − s − . Wefurther inserted the XUV flux, scaled to the distance ofplanet b, and the system parameters given by Kosiarek et al.(2020) in the “hydro-based approximation” presented byKubyshkina et al. (2018c) that enables one to analyticallyderive hydrogen atmospheric mass-loss rates for planets be-low 40 M ⊕ accounting for all e ff ects included in the hydrody-namic modelling (both core-powered mass-loss and photoe-vaporation). For GJ 9827 b, we obtained a mass-loss rate of3.6 × g s − , that is about 1.9 M ⊕ Gyr − (or about 0.4 plan-etary masses per Gyr). Considering that the star is ≈
10 Gyrold (certainly older than 5 Gyr; Rice et al. 2019) and that thestar was more active in the past, it is safe to conclude that theplanet has lost its primary hydrogen-dominated atmosphere.This is further supported by the rather small restricted Jeansescape parameter Λ (Fossati et al. 2017) of about 20.6, whichalone indicates that the planet is subject to intense mass loss,partially driven by the high atmospheric temperature and lowplanetary gravity (i.e., core powered mass loss).We followed the same procedure to estimate the atmo-spheric hydrogen mass-loss rates of GJ 9827 c and GJ 9827 dobtaining 1.0 × g s − and 5.1 × g s − , respectively. Weremark that for all three planets, the mass-loss rates ob-tained from the hydro-based approximation are within a fac-tor of five from those obtained from directly interpolating thegrid of hydrodynamic upper atmosphere models presented byKubyshkina et al. (2018b). These values correspond to about0.5 M ⊕ Gyr − (or about 0.3 planetary masses per Gyr) forGJ 9827 c and about 0.3 M ⊕ Gyr − (or 0.1 planetary massesper Gyr) for GJ 9827 d. These values and the bulk densityof GJ 9827 c (6.1 g cm − ) indicate that it is very unlikely theplanet still holds part of its primary hydrogen-dominated at-4 C arleo et al .mosphere. It is therefore possible that GJ 9827 c has followedan evolutionary path similar to that of GJ 9827 b.For GJ 9827 d, the lower bulk density of 2.51 g cm − sug-gests that the planet may still host part of its primary atmo-sphere. This may be possible given that the planet appearsto have a more sustainable mass-loss rate, suggesting that forthis planet mass loss is primarily driven by atmospheric heat-ing due to absorption of the stellar XUV emission, which is ingeneral weaker than core-powered mass loss. However, thelarge di ff erence between the mass-loss rates we obtained as-suming a pure hydrogen atmosphere and those obtained fromthe constraint given by the non-detection of neutral He in theplanetary transmission spectrum ( < . × g s − ; Kasperet al. 2020) suggest that the planetary atmosphere may notbe hydrogen-dominated. For GJ 9827 d, mass loss is drivenby the stellar high-energy emission and therefore its estimatedirectly depend on it. However a ten times lower stellarXUV emission compared to what is derived from the Ly α flux would bring the planetary mass-loss rate closer to theupper limit given by Kasper et al. (2020). Such lower stellarXUV emission would be close to what is indicated by the MgII h&k resonance lines. In fact, following Linsky et al. (2013,2014), the measured Mg II h&k emission flux would implyan XUV flux at 1 AU of 5.97 erg cm − s − that is almost 3times lower than what predicted from the Ly α emission flux.Furthermore one has to consider the rather large uncertaintieson the employed conversions that make this value consistentwith a ten times lower XUV flux. Nevertheless, even a tentimes lower XUV flux would still be about ten times higherthan what suggested by the non-detection of the He lines.Interestingly, GJ 9827 d has a bulk density similar to that of π Men c, which also has a predicted hydrogen mass-loss rateof the order of 10 g s − (Gandolfi et al. 2018; Garc´ıa Mu˜nozet al. 2020; Shaikhislamov et al. 2020) and for which Ly α transit observations led to a non-detection of the planetaryatmosphere (Garc´ıa Mu˜noz et al. 2020). Transit observationsof GJ 9827 d, and of other similar planets such as π Men c,aiming at characterising their atmospheres would be veryvaluable for understanding the nature of pu ff y super-Earths(Garc´ıa Mu˜noz et al. 2020). CONCLUSIONSIn this paper we presented a search for exospheres aroundtwo planets orbiting GJ 9827, a K6 bright star discovered tohost three super-Earths in 1:3:5 commensurability from theKepler / K2 mission. We observed GJ 9827 b with
HST andGJ 9827 b and d with CARMENES during transit in orderto characterize their atmospheres via the Ly α , H α , and HeI transitions, and we found no evidence of an extended at-mosphere in either of the planets. Theoretical calculationsof the mass-loss rate supported our results, predicting escape rates of 4.3 × g s − , 7.2 × g s − and 3.3 × g s − forGJ 9827 b, c and d, respectively, making them unlikely to stillretain their hydrogen-dominated atmosphere.We also made use of the HST spectra in order to character-ize GJ 9827’s high energy emission, which was used for theabove escape rate calculations, and the ISM absorption alongits sightline. We reconstructed the intrinsic Ly α and MgIIstellar fluxes, necessary because of attenuating H I and MgII interstellar gas between us and the star, finding F (Ly α ) = (5.42 + . − . ) × − erg cm − s − and F (MgII) = (5.64 ± × − erg cm − s − . We report that GJ 9827 is Doppler-shifted +
30 km s − from the velocity frame of the absorbingISM gas, which results in almost negligible attenuation ofthe narrow Mg II lines, but dramatic absorption of the broadLy α line. However, the reconstructed intrinsic Ly α flux is in-consistent with the literature predictions based on its Mg IIemission (Wood et al. 2005; Youngblood et al. 2016). Com-paring GJ 9827 to other K dwarfs as well as M dwarfs, wefound it to have a significantly high Ly α surface flux and asignificantly low Mg II surface flux. This could have im-portant implications on the planetary atmospheres in the sys-tem as Ly α and Mg II, the two brightest emission lines inGJ 9827’s UV spectrum, have a large e ff ect on atmosphericphotochemistry, potentially controlling which are the domi-nant species in the atmosphere. GJ 9827’s Ly α and Mg IIflux discrepancy also highlights the importance of cautionwhen using UV scaling relations for atmospheric escape cal-culations or photochemistry calculations. Not acknowledg-ing the natural variability between individual stars could bedetrimental to our assumptions about the composition or evenpresence of exoplanet atmospheres.As a nearby system of planets transiting a bright star,GJ 9827 is being intensely studied for a variety of reasons.More HST
STIS UV transit observations are planned forGJ 9827 b, that will allow us to confirm our results presentedhere and investigate possible variations in the stellar flux. Wehave also observed transits of all three Super-Earths orbitingGJ 9827 with
Spitzer (Livingston et al in prep.). These, to-gether with our approved Cycle 1 GO CHEOPS transit obser-vations, will further enhance dynamical constraints via tran-sit timing variations that will provide invaluable measure-ments in the infrared to complement our
Hubble observa-tions, as well as facilitate e ffi cient future observations (e.g.,with JWST ) for a system that will be intensely characterizedin the years to come.ACKNOWLEDGMENTSBased on observations made with the NASA / ESA HubbleSpace Telescope, obtained from the data archive at the SpaceTelescope Science Institute. STScI is operated by the Asso-ciation of Universities for Research in Astronomy, Inc. underNASA contract NAS 5-26555. y α , H α , and H e I in GJ 9827 b and d
Allart, R., Bourrier, V., Lovis, C., et al. 2018, Science, 362, 1384,doi: 10.1126 / science.aat5879Bara ff e, I., Chabrier, G., Barman, T. S., et al. 2005, A&A, 436,L47, doi: 10.1051 / / / / / / / // arxiv.org / abs / / science.1185402Borucki, W. J., Koch, D. G., Basri, G., et al. 2011, ApJ, 736, 19,doi: 10.1088 / / / / / / / / / / / / / / / // arxiv.org / abs / / / / nature14501Foreman-Mackey, D., Hogg, D. W., Lang, D., & Goodman, J.2013, PASP, 125, 306, doi: 10.1086 / / / / / / / / / aa80ebGandolfi, D., Barrag´an, O., Livingston, J. H., et al. 2018, A&A,619, L10, doi: 10.1051 / / / / ab61 ff Gupta, A., & Schlichting, H. E. 2019, MNRAS, 487, 24,doi: 10.1093 / mnras / stz1230Harman, C. E., Schwieterman, E. W., Schottelkotte, J. C., &Kasting, J. F. 2015, ApJ, 812, 137,doi: 10.1088 / / / / / / / / // arxiv.org / abs / / / / / aa88adKislyakova, K. G., Holmstr¨om, M., Lammer, H., Odert, P., &Khodachenko, M. L. 2014, Science, 346, 981,doi: 10.1126 / science.1257829Kosiarek, M. R., Berardo, D. A., Crossfield, I. J. M., et al. 2020,arXiv e-prints, arXiv:2009.03398.https: // arxiv.org / abs / // ascl.net / / / / / / / aae586Kulikov, Y. N., Lammer, H., Lichtenegger, H. I. M., et al. 2006,Planet. Space Sci., 54, 1425, doi: 10.1016 / j.pss.2006.04.021Kulow, J. R., France, K., Linsky, J., & Loyd, R. O. P. 2014, ApJ,786, 132, doi: 10.1088 / / / / ff ects on ExoplanetAtmospheres, Vol. 955 (Springer),doi: 10.1007 / / / / / / / / / / / abb36fLopez, E. D., Fortney, J. J., & Miller, N. 2012, ApJ, 761, 59,doi: 10.1088 / / / / / / ab6605 arleo et al . Luger, R., Barnes, R., Lopez, E., et al. 2015, Astrobiology, 15, 57,doi: 10.1089 / ast.2014.1215Madhusudhan, N. 2019, ARA&A, 57, 617,doi: 10.1146 / annurev-astro-081817-051846McCann, J., Murray-Clay, R. A., Kratter, K., & Krumholz, M. R.2019, ApJ, 873, 89, doi: 10.3847 / / ab05b8Miguel, Y., Kaltenegger, L., Fegley, B., & Schaefer, L. 2011,ApJL, 742, L19, doi: 10.1088 / / / / L19Miguel, Y., Kaltenegger, L., Linsky, J. L., & Rugheimer, S. 2015,MNRAS, 446, 345, doi: 10.1093 / mnras / stu2107Murray-Clay, R. A., Chiang, E. I., & Murray, N. 2009, ApJ, 693,23, doi: 10.1088 / / / / / / aa957cNortmann, L., Pall´e, E., Salz, M., et al. 2018, Science, 362, 1388,doi: 10.1126 / science.aat5348Owen, J. E., & Wu, Y. 2013, ApJ, 775, 105,doi: 10.1088 / / / / / / aa890aPalle, E., Nortmann, L., Casasayas-Barris, N., et al. 2020, arXive-prints, arXiv:2004.12812. https: // arxiv.org / abs / / j.pss.2015.04.016Prieto-Arranz, J., Palle, E., Gandolfi, D., et al. 2018, A&A, 618,A116, doi: 10.1051 / / / / / / / / mnras / stz130Rodriguez, J. E., Vanderburg, A., Eastman, J. D., et al. 2018, AJ,155, 72, doi: 10.3847 / / aaa292Rowe, J. F., Coughlin, J. L., Antoci, V., et al. 2015, ApJS, 217, 16,doi: 10.1088 / / / / / / // arxiv.org / abs / / / / s41586-018-0067-5Teske, J. K., Wang, S., Wolfgang, A., et al. 2018, AJ, 155, 148,doi: 10.3847 / / aaab56Tian, F. 2009, ApJ, 703, 905, doi: 10.1088 / / / / / j.epsl.2013.10.024Vidal-Madjar, A., Lecavelier des Etangs, A., D´esert, J. M., et al.2003, Nature, 422, 143, doi: 10.1038 / nature01448Vidotto, A. A., Lichtenegger, H., Fossati, L., et al. 2018, MNRAS,481, 5296, doi: 10.1093 / mnras / sty2130Wood, B. E., Linsky, J. L., H´ebrard, G., et al. 2004, ApJ, 609, 838,doi: 10.1086 / / / / / / / / / //