A Search for FeH in Hot-Jupiter Atmospheres with High-Dispersion Spectroscopy
Aurora Kesseli, I.A.G. Snellen, F.J. Alonso-Floriano, P. Molliere, D.B. Serindag
DDraft version September 11, 2020
Typeset using L A TEX twocolumn style in AASTeX62
A Search for FeH in Hot-Jupiter Atmospheres with High-Dispersion Spectroscopy
Aurora Y. Kesseli, I.A.G. Snellen, F.J. Alonso-Floriano, P. Molli`ere, and D.B. Serindag Leiden Observatory, Leiden University, Postbus 9513, 2300 RA, Leiden, The Netherlands Max-Planck-Institut f¨ur Astronomie, K¨onigstuhl 17, 69117 Heidelberg, Germany
Submitted to AAS JournalsABSTRACTMost of the molecules detected thus far in exoplanet atmospheres, such as water and CO, are presentfor a large range of pressures and temperatures. In contrast, metal hydrides exist in much more specificregimes of parameter space, and so can be used as probes of atmospheric conditions. Iron hydride (FeH)is a dominant source of opacity in low-mass stars and brown dwarfs, and evidence for its existence inexoplanets has recently been observed at low resolution. We performed a systematic search of archivalCARMENES near-infrared data for signatures of FeH during transits of 12 exoplanets. These planetsspan a large range of equilibrium temperatures (600 (cid:46) T eq (cid:46) (cid:46) log g (cid:46) ∼
3) in two planets, WASP-33b and MASCARA-2b. Previousmodeling of exoplanet atmospheres indicate that the highest volume mixing ratios (VMRs) of 10 − to 10 − are expected for temperatures between 1800 and 3000K and log g (cid:38)
3. The two planets forwhich we find low-confidence signals are in the regime where strong FeH absorption is expected. Weperformed injection and recovery tests for each planet and determined that FeH would be detected inevery planet for VMRs ≥ − , and could be detected in some planets for VMRs as low as 10 − . .Additional observations are necessary to conclusively detect FeH and assess its role in the temperaturestructures of hot Jupiter atmospheres. INTRODUCTIONHot Jupiters and ultra-hot Jupiters are gas-giant ex-oplanets that orbit extremely close to their host stars( (cid:46) . (cid:46) T eq (cid:46) (cid:46) T eq (cid:46) Corresponding author: Aurora Y. [email protected] formation (Lecavelier Des Etangs et al. 2008; Pont et al.2013; Sing et al. 2015; Stevenson 2016; Helling 2019),and global atmospheric dynamics (Snellen et al. 2010;Kataria et al. 2016).Transit transmission spectroscopy of hot and ultra-hot Jupiters has allowed for the detection of many ele-ments and molecules in their atmospheres. The majorityof these elements and molecules are present for a widerange of pressures and temperatures, and have thus beendetected in both hot and ultra-hot Jupiters, includingCO (Snellen et al. 2010; Brogi et al. 2012; Sheppard et al.2017), Na (Charbonneau et al. 2002; Snellen et al. 2008;Wyttenbach et al. 2015), H (Vidal-Madjar et al. 2003;Yan & Henning 2018; Casasayas-Barris et al. 2019), andHe (Spake et al. 2018; Nortmann et al. 2018; Allart et al.2019; Alonso-Floriano et al. 2019a). While H O beginsto dissociate in ultra-hot Jupiters, it is ubiquitous in ex-oplanets with temperatures cooler than 2000K (Deminget al. 2013; Birkby et al. 2013; Sing et al. 2016; Sheppardet al. 2017). a r X i v : . [ a s t r o - ph . E P ] S e p Kesseli et al.
Metal oxides (TiO, VO, etc.) and metal hydrides(FeH, MgH, TiH, CaH, etc.) exist within more specifictemperature and pressure ranges and therefore couldprove useful as probes of atmospheric conditions (e.g.,Lodders 1999). However, they have proven difficult toconclusively detect even though they are the definingand dominant opacity features in M and L dwarf spectra(Kirkpatrick et al. 1999). Since hot Jupiters have simi-lar equilibrium temperatures as M and L dwarfs, thesemolecules might be expected in their atmospheres. Nu-groho et al. (2017) measured TiO in the atmosphere ofWASP-33b at high resolution, but Herman et al. (2020)was not able to reproduce these results. VO has beenmeasured at low-resolution in the atmosphere of WASP-121b (Evans et al. 2018), but has been difficult to con-firm at high-resolution due to inaccurate line lists (Mer-ritt et al. 2020). There have also been non-detections ofTiO and VO absorption in hot Jupiters with equilibriumtemperatures where TiO would be expected, leading tosuggestions that TiO and VO are trapped in solids onthe much cooler night sides of these exoplanets (Spiegelet al. 2009; Sheppard et al. 2017).Three studies have reported tentative detections ofFeH in four different transiting exoplanets, WASP-62b(Skaf et al. 2020), WASP-79b (Sotzen et al. 2020; Skafet al. 2020), WASP-121b (Evans et al. 2016), andWASP-127b (Skaf et al. 2020). FeH has also been ob-served in young directly imaged exoplanets, such as De-lorme 1 (AB)b (Eriksson et al. 2020). Furthermore,MacDonald & Madhusudhan (2019) published poten-tial evidence of other metal hydrides (TiH, CrH, andScH) in HAT-P-26b. However, all of these studies reliedon low-resolution spectra, where distinguishing specieswith overlapping opacities and differentiating them fromcontinuum opacity can be challenging. In addition, thereported signal to noise ratios (SNRs) of the potentialsignals were all less than five.Even though metal oxides and metal hydrides havesignificantly lower volume mixing ratios (VMRs) thanmore common molecules and elements, such as CO andH O, these exotic species can have large effects on ex-oplanet atmospheres and detecting them can provideimportant constraints to atmospheric models. Opacityfrom TiO and VO are often suggested to be the pri-mary cause of temperature inversions in hot Jupiters(e.g., Hubeny et al. 2003; Fortney et al. 2008). With thedebate over the prevalence of TiO and VO, Lothringeret al. (2018) found that opacity from a combination ofH − , metals, and metal hydrides could produce the re-quired opacity to cause temperature inversions in hotJupiters without the need for TiO and VO. In addi-tion to being a potential contributor to temperature in- version, FeH also traces weather, cloud formation, andcloud dispersal in L and T dwarfs (Burgasser et al.2008).In this paper, we present results of a systematic searchfor FeH in the atmospheres of 12 hot gas-giant plan-ets using high dispersion transmission spectroscopy ofarchival CARMENES data. The targets cover a range ofsurface gravities, equilibrium temperatures, and masses,to explore where in parameter space opacity from FeHis important. In Section 2 we present the data and dis-cuss the reduction process. In Section 3 we introducethe atmospheric models that we used, and in Section4 we explain how we used these models to retrieve thepotential exoplanetary signals. Next, we present the re-sults in Section 5 and discuss the expected VMRs of theexoplanets in our study in Section 6. Finally, in Section7 we summarize our findings. DATA AND REDUCTIONCARMENES is a high resolution echelle spectrograph(Quirrenbach et al. 2018), which is capable of character-izing atmospheres of transiting exoplanets (e.g., Alonso-Floriano et al. 2019b). CARMENES is installed on the3.5 meter telescope at the Calar Alto Observatory andcontains a near-infrared channel (NIR; R ∼ ∼ . < λ < . µm . We downloaded all of the transits of hot gas-giant planets that were publicly available on the CalarAlto Public Archives (CAHA Archive ) through Novem-ber of 2019, and exclude those transits that were severelyaffected by weather conditions like clouds or high hu-midity. Table 1 summarizes the observations of eachexoplanet transit.The exoplanetary systems cover a large range of pa-rameter space in terms of planet mass (0 . ≤ M p ≤ .
38 M J ), radius (0 . ≤ R p ≤ .
89 R J ), surface grav-ity (2 . ≤ log g ≤ . ≤ T eq ≤ http://caha.sdc.cab.inta-csic.es/calto/jsp/searchform.jsp eH in Exoplanetary Atmospheres Table 1.
Transit observations of hot gas giants with CARMENESExoplanet Observation Number of Phase Exposure Time * Avg. SNRDate Spectra Coverage per Spec. (s) per Spec.KELT-9b 2017/08/07 43 -0 . − .
100 306 92WASP-33b 2017/01/05 94 -0 . − .
077 118 56MASCARA-2b 2017/08/23 70 -0 . − .
034 198 91HAT-P-57b 2018/07/08 30 -0 . − .
049 606 51WASP-76b 2018/10/03 44 -0 . − .
079 498 86HAT-P-32Ab 2018/09/01 23 -0 . − .
066 898 44HD 209458b 2018/09/06 91 -0 . − .
037 198 100HD 189733b 2017/09/07 46 -0 . − .
036 198 174WASP-69b 2017/08/22 35 -0 . − .
028 398 86WASP-69b 2017/09/22 31 -0 . − .
020 398 75WASP-107b 2018/02/24 22 -0 . − .
018 956 44HAT-P-11b 2017/08/12 63 -0 . − .
044 406 107HAT-P-11b 2017/09/25 32 -0 . − .
018 456 97HAT-P-11b 2018/07/25 28 -0 . − .
019 498 127GJ 436b 2017/02/02 38 -0 . − .
025 278 85GJ 436b 2017/02/17 36 -0 . − .
021 278 108GJ 436b 2018/04/09 25 -0 . − .
023 278 50GJ 436b 2018/04/16 31 -0 . − .
024 278 115 * All exposure times were chosen so that the change in radial velocity of the planet in a single exposure was smaller than theCARMENES NIR pixel size pipeline does not perform any telluric correction, andany telluric absorption or sky emission that is presentin the spectra is removed later during our analysis pro-cess. The region where the main FeH bandhead exhibitsa peak opacity ( ∼ µ m) is directly between waterbands so there is minimal telluric contamination to be-gin with. As telluric correction is a major hurtle in ana-lyzing ground-based exoplanet observations in the NIR,the position of the FeH bandhead allows us to efficientlyanalyze transits from many planets.In order to uncover the small planetary transit sig-nal, we had to further analyze the data and remove thestellar absorption features and any remnant noise. Wefollowed a similar set of steps as Alonso-Floriano et al.(2019b) and S´anchez-L´opez et al. (2019), except for ourtreatment of the spectral orders. In previous studies,the orders were handled separately until the very end,when the final 1D cross correlation functions were com-bined. This can decrease the significance of the signalif orders with less strong absorption are simply addedto orders of strong absorption, or can lead to a falselyinflated signal if the orders are weighted before they arecombined. Instead, we combined the orders into a sin-gle spectrum from the start (see Section 3 to see whichorders are included). This simplified and reduced the computational time of the process, allowing us to effi-ciently analyze many transits, and did not significantlychange the final cross correlation function. To combinethe orders, we first normalized each order by fitting athird degree polynomial to the continuum and then in-terpolated each order onto a wavelength grid that wassampled uniformly in log wavelength space in 0.2 km s − increments. This spacing is highly oversampled as theCARMENES resolution is about 3 km s − , but resultedin virtually no information being lost in the interpola-tion process and does not cause any problems duringthe later analysis steps. Any regions that overlappedbetween the orders were combined together using an av-erage that was weighted by the uncertainty in each pixel.An example of the single order-combined spectrum isshown in the top panel of Figure 1.Figure 1 outlines the analysis steps for HD 189733as an example of our process. To begin, we removedany 5-sigma outliers from cosmic rays or bad pixels thatwere missed by the automatic reduction pipeline (secondpanel of Figure 1). Next, we removed the stellar absorp-tion features. Since in every case the star’s radial veloc-ity changes by less than the width of a CARMENESpixel over the course of our observations, while that ofthe planet is rapidly increasing at up to 10 km s − per Kesseli et al.
Table 2.
Hot exoplanet system parameters: stellar spectral type, stellar effective temp., stellarradius, system velocity, planet mass, planet radius, planet equilibrium temp., planet log g , orbitalperiod, reference mid-transit time, semi-major axis, semi-amplitude of the planet’s radial velocity,orbital inclination, planet rotation velocity assuming tidal locking KELT-9 WASP-33 MASCARA-2Stellar SpT B9.5-A0 A5 A2SpT Ref. Gaudi et al. 2017 Lehmann et al. 2015 Talens et al. 2018 T eff (star; K) 9600 ±
400 7308 ±
71 8980 +90 − T eff Ref. Borsa et al. 2019 Stassun et al. 2017 Talens et al. 2018 R ∗ ( R Sun ) 2 . ± .
058 1.55 ± ± R ∗ Ref. Borsa et al. 2019 Stassun et al. 2017 Talens et al. 2018 v sys (km s − ) -19.819 ± ± ± v sys Ref. Borsa et al. 2019 Johnson et al. 2015 Talens et al. 2018 M p ( M J ) 2.88 ± ± < M p Ref. Borsa et al. 2019 Chakrabarty & Sengupta 2019 Lund et al. 2017 R p ( R J ) 1.936 ± ± +0 . − . R p Ref. Borsa et al. 2019 Stassun et al. 2017 Lund et al. 2017 T eq (K) 4050 ±
180 2781.70 ± ± T eq Ref. Gaudi et al. 2017 Chakrabarty & Sengupta 2019 Talens et al. 2018log g † < P orb (d) 1.4811235 1.21986983 3.4741070 P Ref. Gaudi et al. 2017 Stassun et al. (2017) Lund et al. 2017 T (d) 2457095.68572 2452984.82964 2457909.5906 T Ref. Gaudi et al. 2017 Turner et al. 2016 Talens et al. 2018 a (au) 0.03462 +0 . − . ± +0 . − . a Ref. Gaudi et al. 2017 Turner et al. 2016 Lund et al. 2017 K p (km s − ) †
269 231 170 i (deg) 86.79 ± i Ref. Gaudi et al. 2017 Chakrabarty & Sengupta 2019 Lund et al. 2017 v rot (km s − ) † T eff (star; K) 6330 ±
124 6250 ±
100 6001 ± T eff Ref. Stassun et al. 2017 West et al. 2016 Wang et al. 2019 R ∗ ( R Sun ) 1.538 +0 . − . ± +0 . − . R ∗ Ref. Gaia Collaboration et al. 2018 West et al. 2016 Gaia Collaboration et al. 2018 v sys (km s − ) -9.62 ± ± ± v sys Ref. Gaia Collaboration et al. 2018 Soubiran et al. 2018 Soubiran et al. 2018 M p ( M J ) 1.41 ± ± +0 . − . M p Ref. Stassun et al. 2017 West et al. 2016 Wang et al. 2019R p ( R J ) 1.74 ± +0 . − . ± R p Ref. Stassun et al. 2017 West et al. 2016 Bonomo et al. 2017 T eq (K) 2200 ±
76 2160 ±
40 1835.7 +6 . − . T eq Ref. Hartman et al. 2015 West et al. 2016 Wang et al. 2019log g † P orb (d) 2.4653 1.809886 2.1500082 P Ref. Stassun et al. 2017 West et al. 2016 Wang et al. 2019 T (d) 2455113.48127 2456107.85507 2455867.402743 T Ref. Hartman et al. 2015 West et al. 2016 Wang et al. 2019 a (au) 0.0406 ± ± +0 . − . a Ref. Hartman et al. 2015 West et al. 2016 Bonomo et al. 2017 K p (km s − ) †
180 198 172 i (deg) 88.26 ± +1 . − . ± i Ref. Hartman et al. 2015 West et al. 2016 Stassun et al. 2017 v rot ((km s − ) † eH in Exoplanetary Atmospheres Table 2. cont.
HD 209458 HD 189733 WASP-69Stellar SpT G0 K0-2 K5SpT Ref. del Burgo & Allende Prieto 2016 Salz et al. 2015 Anderson et al. 2014 T eff (star; K) 6091 ±
10 5052 ±
16 4700 ± T eff Ref. Stassun et al. 2017 Stassun et al. 2017 Stassun et al. 2017 R ∗ ( R Sun ) 1.19 ± ± +0 . − . R ∗ Ref. Stassun et al. 2017 Stassun et al. 2017 Gaia Collaboration et al. 2018 v sys (km s − ) -14.743 ± ± ± v sys Ref. Soubiran et al. 2018 Soubiran et al. 2018 Gaia Collaboration et al. 2018 M p ( M J ) 0.73 ± ± ± M p Ref. Stassun et al. 2017 Bonomo et al. 2017 Casasayas-Barris et al. 2017 R p ( R J ) 1.39 ± ± ± R p Ref. Stassun et al. 2017 Stassun et al. 2017 Casasayas-Barris et al. 2017 T eq (K) 1450 1200 963 ± T eq Ref. Sing et al. 2016 Sing et al. 2016 Anderson et al. 2014log g † P orb (d) 3.52474859 2.21857567 3.868140 ± P Ref. Stassun et al. 2017 Stassun et al. 2017 Stassun et al. 2017 T (d) 2452826.629283 2454279.436714 2455748.83422 T Ref. Bonomo et al. 2017 Agol et al. 2010 Bonomo et al. 2017 a (au) 0.04707 +0 . − . +0 . − . +0 . − . a Ref. Bonomo et al. 2017 Bonomo et al. 2017 Bonomo et al. 2017 K p (km s − ) †
145 153 127 i (deg) 86.71 ± pm ± i Ref. Stassun et al. 2017 Bonomo et al. 2017 Anderson et al. 2014 v rot ((km s − ) † T eff (star; K) 4430 ±
120 4708 ±
84 3479 ± T eff Ref. Anderson et al. 2017 Stassun et al. 2017 Bourrier et al. 2018 R ∗ ( R Sun ) 0.66 ± ± ± R ∗ Ref. Anderson et al. 2017 Yee et al. 2018 Bourrier et al. 2018 v sys (km s − ) 13.74 ± ± ± v sys Ref. Gaia Collaboration et al. 2018 Soubiran et al. 2018 Soubiran et al. 2018 M p ( M J ) 0.12 ± ± ± M p Ref. Anderson et al. 2017 Stassun et al. 2017 Turner et al. 2016 R p ( R J ) 0.94 ± ± ± R p Ref. Anderson et al. 2017 Yee et al. 2018 Turner et al. 2016 T eq (K) 770 ±
60 878 ±
15 686 ± T eq Ref. Anderson et al. 2017 Bakos et al. 2010 Turner et al. 2016log g † P orb (d) 5.721490 ± ± ± P Ref. Anderson et al. 2017 Stassun et al. 2017 Bourrier et al. 2018 T (d) 2456514.4106 2454957.812464 2454865.084034 T Ref. Anderson et al. 2017 Sanchis-Ojeda & Winn 2011 Bourrier et al. 2018 a (au) 0.055 ± +0 . − . ± a Ref. Anderson et al. 2017 Yee et al. 2018 Turner et al. 2016 K p (km s − ) †
105 117 128 i (deg) 89.7 ± ± ± i Ref. Anderson et al. 2017 Stassun et al. 2017 Turner et al. 2016 v rot ((km s − ) † † calculated using the listed parameters Kesseli et al. N o r m F l u x -0.0120.0050.018-0.0120.0050.018 P h a s e Å )-0.0120.0050.018 Figure 1.
Example of the analysis steps of the HD 189733 system. The top plot shows one spectrum that has been order-combined and flattened and contains the full wavelength range that we used in our analysis. Next, we show all 46 of thesigma-clipped and normalized spectra in grayscale. The dark vertical lines are the stellar absorption features mixed with a fewtelluric absorption lines. The third panel shows the data after the stellar lines have been removed and another high-pass filterhas been applied. The dominant variations left are due to telluric lines mostly at the edges of the spectra. The bottom panelshows the final spectra after 9
SYSREM iterations have been applied. The spectra appear uniform and lack any significant trendsin time. hour, we created a time averaged spectrum and then di-vided each time series spectrum by this average. Thisprocess removed any signal that was constant in wave-length over the observation times, but left any signalthat was not constant in time (i.e. any planet absorp-tion). Before this step could be completed, however,we corrected for the change in the stellar line positionsover the course of the observations due to the motionof the Earth (barycentric velocity correction). To calcu-late the barycentric velocity of each exposure, we usedthe barycorr online application to convert the mid ex-posure MJD to a barycentric velocity correction with aprecision of 3 m s − (Wright & Eastman 2014). We next applied a high-pass filter with a width of 1000pixels (200 km s − ) and again performed a 5-sigma clip-ping in case any overall shape differences or outliers re-mained in the spectra. Finally, we divided by the stan-dard deviation of each pixel in time, which effectivelydown-weighted the wavelength regions with large stan-dard deviations due to tellurics, bad pixels or cosmicrays.At this point the dominant remaining features weredue to telluric contamination (see third panel in Figure1). To remove these features, the strong lines are oftenmasked and any residuals are removed with the SYSREM algorithm. We experimented with masking and
SYSREM ,but chose to only use
SYSREM in the following results as eH in Exoplanetary Atmospheres
SYSREM iteratively performs principle com-ponent analysis, allowing for unequal uncertainties ateach wavelength point, to remove systematic trends inphotometric or spectroscopic data due to trends in tem-perature, airmass, and more (Tamuz et al. 2005; Mazehet al. 2007). Numerous studies have thoroughly testedand validated the use of
SYSREM for removing telluric sig-nals in high resolution spectroscopy (e.g., Birkby et al.2017; Nugroho et al. 2017; Cabot et al. 2019). Even withthe extensive testing,
SYSREM cannot be applied blindlyto the spectra, as it can remove the planetary signal iftoo many iterations are applied. To find the ideal num-ber of
SYSREM iterations, we tried a range of values andobserved how the SNR evolved over each iteration. Sec-tion 4.2 shows the results of these tests on the injectedsignal. ATMOSPHERIC TRANSMISSION MODEL
Figure 2.
The cross correlation signal between the Exo-Mol line list and Teegarden’s star for each CARMENES NIRspectral order. The main FeH bandhead begins at 0.99 µ m.Water bands surround the FeH bandhead and are the causeof the decrease in SNR around 0.9, 1.15, and 1.3 µ m. A sec-ond peak in SNR occurs around 1.25 µ m, but it is about athird as strong, and there is significantly more telluric con-tamination in this wavelength region than around the pri-mary peak. We created a model transmission spectrum of FeHfor each exoplanet using the petitRADTRANS package(Molli`ere et al. 2019) to test whether FeH can be ob- served in the atmospheres of these planets. petitRAD-TRANS is a radiative transfer code, designed specifi-cally for spectral characterization of exoplanetary atmo-spheres. The code takes as input a temperature-pressureprofile, the planetary radius, the surface gravity, therelative abundances of the requested species, and themean molecular weight of the atmosphere, and producesa transmission or emission spectrum at low or high spec-tral resolution. The relative abundances of the speciesare required to be in units of mass fractions, and notVMRs, so we multiply by the molecular weight overthe mean molecular weight to convert to mass fractions,where applicable.A pre-computed opacity line list of FeH is availablewith the code. The opacity line list of FeH is sourcedfrom the ExoMol library and uses the empirically deter-mined FeH lines from Wende et al. (2010). We testedthis line list to ensure its accuracy by comparing the listto a high SNR CARMENES spectrum of Teegarden’sstar, an M dwarf with a similar effective temperature(2700 K) to many of the exoplanets in our study. Toaccurately compare the two, we used petitRADTRANSto make a model of FeH using the parameters of Tee-garden’s star (shown in the right panel of Figure 3).We cross correlated each spectral order separately withthe ExoMol line list to determine in which orders FeHwas detectable and at what level. Figure 2 shows thatFeH is strongly detected in the six orders around theprimary bandhead. There is another peak in the SNRfunction around 1.25 µ m, but this region is heavily con-taminated by telluric water and oxygen lines and wefind when these orders are added to our injection andretrieval tests (see Section 4.2), the SNR of our retrievedsignal is actually decreased. We therefore use only the6 orders that span wavelengths 0.98 − µ m.With the validated line list, we created transmissionspectra of FeH using the high-resolution mode (R= 10 )of petitRADTRANS (see Figure 3). The planetary radiiand surface gravities that were used as input are alllisted in Table 2. When available, we used publishedpressure-temperature (PT) profiles (see Table 3). Ifno PT profile was available, we assumed an isothermalprofile at the planet’s equilibrium temperature. Trans-mission spectra are not very sensitive to small changesin the PT profile and so this should be sufficient forour purpose. To determine the mean molecular weight(MMW) of each planet, we implemented a simple equi-librium chemistry code to minimize Gibbs free energy.The code is described in Appendix A2 of Molli`ere et al.(2017), and takes the PT profile, the C/O and Fe/H ra-tios (both assumed to be solar) as input, and returns the Kesseli et al.
Table 3.
Planet atmospheric model parametersExoplanet PT profile H − VMR (0.01 bars) Cloud BaseKELT 9b 4000 K . × − none, , 0.01WASP 33b Haynes et al. 2015 1 . × − none, , 0.01MASCARA 2b 2250 K 3 . × − none, , 0.01HAT-P-57b 2200 K 2 . × − none, , 0.01WASP 76b 2150 K 1 . × − none, , 0.01HAT-P-32Ab Nikolov et al. 2018 1 . × − none, , 0.01, 0.001, 0.0001 bar HD 209458b Brogi et al. 2017 2 . × − none, , 0.01 barHD 189733b 1200 K 3 . × − none, , 0.01 barWASP 69b 1000 K 3 . × − none, , 0.01 barWASP 107b 1000 K 3 . × − none, , 0.01 barHAT-P-11b 1000 K 3 . × − none, , 0.01 barGJ 436b 1000 K 3 . × − none, , 0.01 bar When no published PT profile was available we assumed an isothermal profile at the equilibriumtemperature Bolded values are used in the fiducial models and for most of the analysis Based on the literature, it is unclear if HAT-P-32 has a high cloud deck or not (Damiano et al. 2017;Nortmann et al. 2016), so we explored a larger area of parameter space
MMW and other elemental or molecular abundances ateach pressure in the atmosphere.We also included continuum opacity from H − H col-lisions, H − He collision, and H − opacity in all of theatmospheres. Many recent works have emphasized theimportance of including H − opacity in hot exoplanet at-mospheres (e.g., Freedman et al. 2014; Lothringer et al.2018; Arcangeli et al. 2018). Freedman et al. (2014)showed that H − was the dominant continuum opacitysource around one micron for an atmosphere of 2600 K.The abundances of H − , free electrons, and H are in-cluded in the equilibrium chemical modeling. In somemodels we also experimented with including a cloudbase, below which the atmosphere cannot be probed.Table 3 show the continuum opacity sources exploredin the models, and Figure 3 shows the effect of thesechanges to the continuum opacity on the resulting modelatmospheres.Finally, we tested a range of different VMRs for FeH,from 10 − to 10 − . We did not utilize any chemicalmodeling to calculate the VMRs of FeH at different alti-tudes, but instead kept the VMR of FeH constant. Thischoice was motivated in part to facilitate comparisonswith previous studies, which all used constant VMRs ofFeH. It was also motivated by discrepancies between ob-servations of brown dwarfs and chemical models; manylow-temperature brown dwarfs show evidence for FeHeven though the models predict that FeH should have condensed out of the atmosphere (see Section 6 for moredetails). Even though a constant VMR is less realisticthan a VMR that changes with altitude, the resultingcross correlation functions are not significantly affectedbecause transmission spectra are not very sensitive tothese changes.To make the models more realistic, we took into con-sideration the rotation of the planet in a method similarto Brogi et al. (2016), assuming tidally locked planets.The majority of the planets have rotation velocities thatare smaller than the resolution element of CARMENES(see v rot in Table 2), but for the fastest rotating planetsthis effect is significant. We also reduced the resolutionof the models to match the CARMENES spectrograph(R ∼ SIGNAL RETRIEVAL4.1.
Cross Correlation
To search for the exoplanetary signal in the data, eachspectrum was cross correlated with the model for a widerange of radial velocities, spanning −
250 to +250 kms − . Since we already interpolated the spectra onto auniform grid in radial velocity space, no other steps wererequired to prepare the spectra for cross correlation. We eH in Exoplanetary Atmospheres , cloud base1.571.601.631.661.69 VMR 10 , H0.9 1.0 1.1 1.2 1.3 1.4 1.5 1.6Wavelength ( m)1.571.601.631.661.69 VMR 10 , H P l a n e t R a d i u s ( R J ) N o r m a li z e d F l u x Telluric Absorption (+Offset)petitRADTRANS FeHObserved M7 star
Figure 3.
Left: Example of the atmospheric models from petitRADTRANS for WASP-33b for different continuum opacitysources and FeH VMRs. All of the models include opacity from H − H and H − He collisions, while in the bottom three plotsother sources of continuum opacity are also included. In the second plot we have included a cloud base at 0.1 bar, below whichthe atmosphere cannot be probed. In the bottom two plots, the continuum is set by opacity from H − (with a VMR of 10 − ).As expected, by adding more continuum opacity sources, the line strengths are decreased. The bottom plot shows that bydecreasing the VMR of FeH, the line strengths also decrease. Right: Enlarged version of the second FeH model showing about5% of the wavelength coverage we used, as well as a non-telluric corrected CARMENES spectrum of Teegarden’s star and thetelluric absorption model for this spectrum from molecfit (Smette et al. 2015). Some water lines are present but they are muchweaker and sparser than the FeH lines. The model and observations are in very good agreement and there are thousands of FeHlines densely spaced throughout this wavelength region. normalized the cross correlation functions according toTonry & Davis (1979). After the cross correlation, wewere left with a grid containing a different cross corre-lation function for each time-series spectrum.The planet’s velocity at the time of each spectrum canbe calculated with the following equation v p ( t, K p ) = v sys + K p sin 2 πφ ( t ) (1)where v sys is the systemic radial velocity of the star, K p is the semi-amplitude of the planet’s radial velocity, and φ ( t ) is the orbital phase at the time of the observation.The values of K p , v sys , the time of transit T and orbitalperiod P for each planet are given in Table 2.Using the calculated v p values, we determined if therewas a positive correlation between the model and dataat the planet’s expected velocity. The top panel of Fig-ure 4 shows an example of the cross correlation matrixalong with the calculated planet velocities. If the trans-mission spectrum shows significant FeH absorption, pos-itive correlation should be present along the planet’s ve-locity path. In Figure 4, a signal was injected for clar-ity. To determine the strength of the signal and the resulting SNR of the possible detection, the cross cor-relation functions were each shifted to the planet’s restframe (third panel). We then added together all of theindividual cross correlation functions, weighted by thetransit depth at each phase (bottom panel). We im-plemented the PyTransit software package (Parviainen2015) to model the transit light curve and determine thetransit depth at the observed phases of each exoplanet,using a quadratic limb darkening model originally laidout in (Mandel & Agol 2002). All the parameters usedas input in the model are given in Table 2 except for thelimb darkening coefficients, which are from Claret et al.(2012, 2013).4.2. Injection and Recovery Tests
We performed a series of injection and recovery testsboth to determine the optimal number of
SYSREM iter-ations, and to determine the VMR of FeH that wouldbe detectable in each planet’s atmosphere. We will leavethe discussion of the FeH VMRs for Section 6. To deter-mine the optimal number of
SYSREM iterations for eachsystem, we injected a signal at the expected strength0
Kesseli et al.
60 40 20 0 20 40 60Radial Velocity (km s )0.020.010.000.010.020.03 P h a s e CC v a l u e
60 40 20 0 20 40 60Radial Velocity (km s )0.020.010.000.010.020.03 P h a s e CC v a l u e
60 40 20 0 20 40 60Radial Velocity (km s )0.020.000.02 P h a s e CC v a l u e
150 100 50 0 50 100 150Radial Velocity (km s )0.050.000.050.100.15 CC V a l u e Figure 4.
Example cross correlation functions of HD189733b with a model atmosphere. In all of the plots a plan-etary signal has been injected for demonstration purposes.The top panel shows all 46 cross correlation functions in therest frame of the host star, without any
SYSREM iterations.The red dashed lines indicate when the transit begins andends, while the black dashed line indicates the planet’s ex-pected radial velocity. The signal appears strongest at zerophase due to the somewhat v-shaped transit of HD 189733b.The second panel (as well as the subsequent two) is the sameas the top plot, except the optimal number of
SYSREM itera-tions were applied during the analysis. The third panel showsthe same as the second, except shifted into the planet’s restframe (the signal is now vertical and centered at 0 km s − ).In the bottom panel all of the individual cross correlationfunctions have been weighted by the transit shape and thenadded together to create a single 1D cross correlation func-tion. It is with this function that we can measure the SNRby computing the peak signal and the standard deviation ofthe surrounding noise. This injected signal has a SNR of6.6, and so it would be considered a statistically significantdetection. S N R Figure 5.
Example plot demonstrating how the SNRchanged as a function of
SYSREM iteration. Again, we usedHD 189733 as our example and have injected a signal witha VMR of 10 − . Due to the region being relatively devoidof telluric lines, we recovered the signal with a SNR of 6.0without any SYSREM iterations. However, by using
SYSREM toremove the small telluric contamination, we were able to re-cover the signal at a higher SNR until nine iterations where
SYSREM began to remove the planet’s signal. for a VMR that would be recoverable. We injected thesignal at a different v sys value so as to not be biased byany real signal from the planet. The signal was injectedat each phase with a strength specified by the PyTran-sit transit light curve, discussed in Section 4.1. We thenperformed the full analysis as outlined in previous sec-tions, changing only the number of SYSREM iterations.For each iteration we recorded the resulting SNR. Fig-ure 5 shows an example of how the SNR changed witheach
SYSREM iteration. This analysis was performed sep-arately for each night so that the optimal number of
SYSREM iterations changed on a night-by-night basis. Forthe remaining analysis we used the number of
SYSREM iterations that maximized the SNR. RESULTSWe performed all of the analysis steps on each of the12 exoplanet systems and searched for peaks in the crosscorrelation function along the planet’s expected velocity.We did not find a significant detection of FeH in any ofthe exoplanets in our survey. Figure 6 shows all of theone dimensional cross correlation functions in the restframe of the exoplanet. Two of the planets, WASP-33band MASCARA-2b, show peaks near the planets’ ex-pected velocities at a SNR ∼
3. The peak of WASP-33boccurs at − − , while that of MASCARA-2b oc- eH in Exoplanetary Atmospheres
100 0 1000.10.00.1 CC V a l u e KELT-9
100 0 1000.20.00.2
WASP-33
100 0 1000.20.00.2
MASCARA-2
100 0 1000.10.00.1 CC V a l u e HAT-P-57
100 0 1000.10.00.1
WASP-76
100 0 1000.10.00.1
HAT-P-32
100 0 1000.20.00.2 CC V a l u e HD 209458
100 0 1000.10.00.1
HD 189733
100 0 1000.10.00.1
WASP-69
100 0 100Radial Velocity (km/s)0.050.000.05 CC V a l u e WASP-107
100 0 100Radial Velocity (km/s)0.20.00.2
HAT-P-11
100 0 100Radial Velocity (km/s)0.10.00.1
GJ 436
Figure 6.
One dimensional cross correlation functions between the FeH model atmosphere for each planet and the time seriesspectra. The spectra have all been shifted to the rest frame of the planet using the K p values in Table 2, and then addedtogether over time in the same way as for Figure 4. If FeH is present in a detectable amount in the planet’s atmosphere apeak near 0 km s − would be visible. The dark gray shaded region represents one standard deviation of the cross correlationfunction excluding a 30 km s − region around zero, while the light gray region represents two standard deviations. WASP-33band MASCARA-2b are the only planets with a peak above the two standard deviation near the expected location of 0 km s − . curs at − . − . Although these signals are not sta-tistically significant, a slight offset in the system radialvelocity is often seen in hot Jupiters, and is attributedto winds in the exoplanet’s atmosphere ( v wind ; Snellenet al. 2010; Ehrenreich et al. 2020).By observing a larger sample of exoplanets, the proba-bility of observing a 3-sigma peak due to random Gaus-sian noise in the expected velocity range increases. Weperformed a simple calculation to determine the prob-ability of observing two of these 3-sigma peaks at thecorrect velocity. We assumed that any peak with windvelocities between 0 and -10 km s − would be acceptable,which gives a window that is about 3 times the spectralresolution of CARMENES. The probability of measur- ing a random positive 3-sigma peak is 0.15%. There-fore, with 12 exoplanets observing a 3-sigma noise peakwithin the expected velocity range has a probability of12 × × .
15% = 5 . σ level, which,although interesting, we do not consider statistically sig-nificant enough. This 3-sigma level is a simple order ofmagnitude estimate and could be altered due to effectsof correlated noise from tellurics or stellar residuals.We tested the validity of these signals further and ex-plored whether they could be enhanced by small varia-tions to the model or slightly different K p values. Wefound that for WASP-33b, the model that gave thelargest SNR had a FeH VMR of 10 − . For MASCARA-2 Kesseli et al.
100 75 50 25 0 25 50 75 100Radial Velocity (km s )0.060.040.020.000.020.040.06 P h a s e CC v a l u e
150 100 50 0 50 100 150Radial Velocity (km s )210123 S N R Figure 7.
Cross correlation matrix (top) and one dimen-sional cross correlation function (bottom) of WASP-33b af-ter the 15
SYSREM iterations, using the optimal model and K p value of 248 km s − . The one dimensional correlationfunction has a peak at -5 km s − with a SNR of 2.98, whichalthough interesting we do not consider to be statisticallysignificant considering the size of our exoplanet sample. − maximized the SNR. However, inboth cases the change in SNR for volume mixing ratiodifferences of one order of magnitude (e.g., VMR of 10 − versus 10 − ) is roughly ∼ − to 10 − are reason-able for the VMR of FeH. Figures 7 and 8 show thetwo dimensional and one dimensional cross correlationfunctions for both of these planets.We searched in a large region of K p and v wind pa-rameter space to see if the assumed values representedthe maximum signal, and to investigate any other strongfeatures (see Figure 9). For WASP-33b, we find that theSNR ∼ K p valuehas a maximum at 248 +22 − km s − . For MASCARA-2bwe found that the maximum K p occurred at 158 +41 − kms − . These uncertainties represent the K p values wherethe SNR decreases by one for the peak near the planet’sexpected velocity. We note that these are not one sigmaerror bars as a decrease in one of SNR does not neces-sarily directly correspond to a decrease of one in sigma.Neither parameter search (Figure 9) convincinglyshows a detection of FeH as both plots reveal otherstrong positive correlation peaks that can certainly notbe associated with the planet. WASP-33b is a knownDelta Scuti pulsator, and while we do not see obviousresiduals from the pulsations in our analysis, some of thenoisy areas in the 2D cross correlation functions could
60 40 20 0 20 40 60Radial Velocity (km s )0.030.020.010.000.010.020.03 P h a s e CC v a l u e
150 100 50 0 50 100 150Radial Velocity (km s )202 S N R Figure 8.
Same as Figure 7 but for MASCARA-2b after 8
SYSREM iterations at a K p value of 158 km s − . The rowsof white space are due to a gap in observing coverage, andno data exists on the archive directly before the start ofthe transit. The one dimensional correlation function has apeak at − . − with a SNR of 3.02, which althoughinteresting we do not consider to be statistically significantconsidering the size of our exoplanet sample. be due to residuals from the removal of stellar lines thatwere affected by the pulsations (see area between 25 and75 km s − in Figure 7). A previous analysis of the sameWASP-33 data by Yan et al. (2019) found that signalsfrom Ca could be successfully recovered from the dataafter applying a high-pass filter similar to the one ap-plied in our analysis, but they noted that a particularlystrong negative correlation signal that still remained intheir data could be due to the pulsations. Alternatively,the spurious signals in Figure 9 for WASP-33 could sim-ply be due to the low SNR of the data. Other peaks inthe plot of MASCARA-2b seem to be caused by telluriccontamination, stellar residuals, or some combination ofthe two, as the SNR peak extends down to 0 km s − .Therefore, even though the 1D cross correlation func-tions show SNRs ∼
3, we do not claim a statisticallysignificant FeH detection.Some previous studies have questioned the statisticalsignificance of solely using the SNR metric to judge thequality of the signal (e.g., Brogi et al. 2013). We there-fore also tested the use of the Welch T-test to comparethe signal within and outside the expected exoplanet’strail, as has been done in many previous works (e.g.,Brogi et al. 2013; Alonso-Floriano et al. 2019b). Wefind that the Welch T-test gives a very similar statis-tical significance for the potential signal in WASP-33b, eH in Exoplanetary Atmospheres
100 75 50 25 0 25 50 75 100 V wind (km s )50050100150200250300 K p ( k m s ) S N R
100 50 0 50 100 V wind (km s )50050100150200250 K p ( k m s ) S N R Figure 9.
Parameter space search for cross correlation peaks for a range of K p and v wind values in WASP-33b (left) andMASCARA-2b (right). The dotted black line shows the expected K p value for a v wind of 0 km s − . For WASP-33b, the peakin SNR near the expected value is not the strongest and is one of several peaks of similar SNR. The spurious strong peaksseem to be due to correlated noise in the cross correlation functions, which could be a signature of stellar pulsations. ForMASCARA-2b, other major SNR peaks reside near a K p and v wind of 0 km s − , leading us to believe they are associated withtelluric contamination or stellar residuals. MASCARA-2b had the highest level of telluric contamination of all the spectra weanalyzed, and we found that even with a large number of SYSREM iterations some telluric signal remained visible. The 0 km s − peak seems to extend up toward the potential exoplanet signal peak, which further weakens the case for a true signal. but that it estimates an extremely high significance ofapproximately 5 σ for the signal in MASCARA-2b. TheWelch T-test has been previously shown to often lead toover-inflated confidence estimates (Cabot et al. 2019).Furthermore, the test does not take into account anycorrelated noise, and so we are skeptical of these re-sults and instead prefer to report the significance of theMASCARA-2b detection with a SNR of 3. DISCUSSIONWe searched for signals of FeH in 12 exoplanets span-ning a large range of equilibrium temperatures and log g , but found no conclusive detections of FeH. In two ofthe exoplanets, however, we did see SNR ∼ − − g of 4.0 andobservations from Martin et al. (2017). The spectralindices in Martin et al. (2017) were originally reportedin spectral types, which we converted into temperatureswith the relation from Filippazzo et al. (2015). Metalhydrides, such as FeH, have long been known to be grav-ity dependent, and Martin et al. (2017) used the J-bandFeH spectral index along with other atomic lines to di-vide a large sample of brown dwarfs into surface grav-ity bins. The FeH spectral indices from the low-gravitybrown dwarfs match those from the models very well.Martin et al. (2017) also measured that the FeH spec-tral indices from the higher gravity objects were larger,4 Kesseli et al.
500 1000 1500 2000 2500 3000 3500 4000Temperature (K)2.502.753.003.253.503.754.00 l o g ( g ) SNR < 3SNR 3500 1000 1500 2000 2500 3000 3500 4000Temperature (K)1.01.2 F e H Sp e c t r a l I n d e x BT Settl log (g)=4Martin+2017 low-gMartin+2017 field0.951.001.051.101.15 B T S e tt l F e H I n d e x Figure 10. Top:
Expected FeH feature strength over the parameter space explored in this paper. Each symbol represents aplanet in our study, where the temperatures are the equilibrium temperatures of the planet. The purple circles are planets wherewe do not detect any FeH, while the stars indicate the two ∼ − σ signals (WASP-33b and MASCARA-2b). The blue colormapshows the expected FeH feature strength (darker blue is stronger FeH absorption), calculated by measuring its spectral indexfrom a grid of BT-Settl models with a given effective temperature and surface gravity (Allard et al. 2012). Although we do notconsider the two ∼ − σ signals statistically significant, it is interesting that they lie in the part of parameter space with strongexpected FeH absorption. Bottom:
Comparison between the FeH spectral index calculated from the models with a log g of 4.0and FeH spectral indices of brown dwarfs with similar temperatures from Martin et al. (2017). The red dot-dashed line is thetrend for low gravity brown dwarfs (3.5 (cid:38) log g (cid:38) g (cid:38) g values ispresent in both the models (blue scale in top panel) and observations. helping to validate the trend seen in the top panel ofFigure 10.The other trend that is visible in both panels of Fig-ure 10 is that FeH exhibits the strongest absorption be-tween about 1800 and 3000 K. Below temperatures ofabout 1800 K iron starts to condense out of the atmo-sphere and the abundance quickly drops off (Visscheret al. 2010). At higher temperatures, FeH dissociatesand again the abundance decreases. The observationsof brown dwarfs show that FeH again becomes visible attemperatures of around 1000 K. It is not exactly known why FeH becomes visible again, but Burgasser et al.(2002) suggested that it could be evidence of cloud dis-ruption, which allows deeper layers of the atmosphereto be probed. This suggests that in future studies, FeHcould be a tool for uncovering weather and cloud disper-sal in planetary atmospheres at low temperatures.The two objects in our sample for which we detect aCCF peak near the expected K p and V sys have equi-librium temperatures and log g values correspondingto where FeH is expected to produce a strong signal.WASP-33b and MASCARA-2b have high enough log eH in Exoplanetary Atmospheres K E L T - W A S P - M A S C - H A T - P - W A S P - H A T - P - H D H D W A S P - W A S P - H A T - P - G J L o g V o l u m e M i x i n g R a t i o WASP-62b (Skaf+2020)WASP-79b (Sotzen+2020)WASP-79b (Skaf+2020)WASP-127b (Skaf+2020)Excluded By Our Study
Figure 11.
Results of our injection and recovery tests foreach planet. The black points represent the FeH VMR thatwe are able to recover with an SNR > g values that FeH is still quite abundant, but not toohigh that the scale height is so small that the atmo-sphere cannot be effectively probed with transmissionspectroscopy. Since the highest FeH abundance is ex-pected for high gravity objects, day-side spectroscopymay more amenable to FeH searches due to fact thatthe signal strength does not decrease for smaller scaleheights as it does in transmission spectroscopy. Fur-thermore, emission spectroscopy can probe deeper in theatmosphere, where FeH is thought to be more abundant(Visscher et al. 2010).Visscher et al. (2010) studied the chemical behaviorof iron-bearing gases in giant planets, brown dwarfs andlow-mass stars to derive abundances as a function oftemperature, pressure, and metallicity. They found thatFeH is the second most abundant iron-bearing gas aftermonatomic Fe at temperatures above about 1500 K. Forthese temperatures, and pressures between 0 and 10 − bars (the region of the atmosphere probed by transmis-sion spectroscopy of most molecules), the FeH abun-dance was found to be between 10 − and 10 − .We used injection and recovery tests to determinewhat VMR of FeH we could recover for each planet.We used a SNR of 5 as our detection threshold. Fig- ure 11 shows the FeH VMR limits for each planet forthe range of different model atmospheres shown in Ta-ble 3. For all the planets we could recover abundancesdown to 10 − and for some as low as 10 − . . When weinject an FeH signal with a VMR of 10 − into the spec-tra of MASCARA-2b and a VMR of 10 − into WASP-33b, we recover both signals with a SNR of about 3.Although we do not treat the observed signals as sta-tistically significant, their amplitudes are in the rangefor the expected VMRs. In addition, it indicates thatobtaining a statistically significant detection of FeH inhot Jupiters requires transmission spectra at a signalto noise of about a factor of two better than what arecurrently available, but well within the capabilities ofcurrent facilities and instruments.Two studies have recently published potential FeHdetections in three different exoplanets, WASP-79b,WASP-127b, and WASP-62b (Sotzen et al. 2020; Skafet al. 2020). These are all hot Jupiters with equilibriumtemperatures between 1300 and 1700 K and log g valuesless than 2.9, which means that they have both lowersurface gravities and lower temperatures than the plan-ets for which we obtained potential signals. Sotzen et al.(2020) found a best fit FeH VMR of ∼ − in WASP-79b, while Skaf et al. (2020) retrieved VMRs of 10 − . ,10 − . , and 10 − . from WASP-79b, WASP-127b andWASP-62b, respectively. These retrieved FeH VMRsare between three and five orders of magnitude moreabundant than Visscher et al. (2010) predicted. If FeHexists at this level in any of the planets in our samplewe would have detected it (see Figure 11). While noneof these planets are in our sample, the orders of mag-nitude discrepancy between the potential FeH VMRs isworrying and could point to over-estimation of molecu-lar opacities in low-resolution data due to degeneracieswith clouds, hazes, or H − continuum opacity.Using both low- and high-resolution observations leadsto a more complete and accurate picture of the role ofmetal hydrides and oxides in hot-Jupiter atmospheres.This combination of low- and high-resolution studies hasproved vital for uncovering whether TiO and VO arepresent in WASP-121b (Evans et al. 2016, 2018; Merrittet al. 2020), and highlights the importance of obtainingboth types of data.While all of these methods for estimating the VMRof FeH in exoplanetary atmospheres can give us a gen-eral idea of how the chemical abundances change withtemperature and surface gravity, it is important to notethat they may not give a full picture, and importantphysics still may be missing. Atmospheric models andlow-gravity brown dwarfs both address the issue thatplanets have lower log g values than stars and field6 Kesseli et al. brown dwarfs, but neither have the intense insolationfrom the host star, which can cause ultra-hot Jupitersto host temperature inversions (Pino et al. 2020) andpotentially change the predicted VMRs of FeH. This in-tense insolation also causes large day-to-night temper-ature constrasts, and iron could be rained out on thenight side of hot Jupiters, as seen in WASP-76b (Ehren-reich et al. 2020), which would have unknown affectson the abundance of FeH on the day side and termina-tors. Additionally, the chemical models describing thebehavior of iron in Visscher et al. (2010) assume ther-mochemical equilibrium, which is not always an accu-rate assumption, especially in the upper parts of the at-mosphere that are probed by transmission spectroscopy(Molaverdikhani et al. 2019). Because of these issues,measuring FeH in a variety of exoplanetary atmosphereswill be needed to fully understand how these differenceaffect its abundance. CONCLUSIONSWe searched for FeH in archival near-infrared CAR-MENES spectra of 12 exoplanets spanning a wide rangein temperature and surface gravity. The FeH main band-head is located at 0.99 µ m, which is ideally situated be-tween two water bands, making it relatively free fromtelluric contamination. Since removing tellurics is themost challenging part of ground-based high-dispersiontransmission spectroscopy, the location of the bandheadis ideal to efficiently and accurately search for FeH in alarge sample of planets. To search for the exoplanet’sFeH signal we cross correlated the data with a range ofexoplanet atmospheric models, created using petitRAD-TRANS.We did not find any statistically significant FeH sig-nals in any of the transmission spectra. Two of theplanets, WASP-33b and MASCARA-2b, showed pos-itive correlation near the expected K p and v sys withSNRs of about 3. Even though these peaks seemedpromising in the 1D cross correlation functions, therewere several other peaks with similar or greater signifi-cance when we searched a wider K p and v sys range, andso we do not claim a detection.To put these results into context, we explored whatthe expected VMR of FeH would be for each planet,and where in parameter space FeH would contribute themost opacity. We conclude that opacity from FeH ismost likely important for planets with temperatures be-tween 1700 and 3000 K, and relatively high log g val-ues. However, at lower temperatures FeH could still beimportant if clouds are somehow dispersed. While the signals of WASP-33b and MASCARA-2b are not statis-tically significant, it is interesting that these two planetsreside in the part of parameter space where the expectedFeH opacity is strong.By performing injection and recovery tests we wereable to rule out FeH existing in any of these exoplan-ets’ atmospheres with a VMR greater than 10 − and forsome, as low as VMRs of 10 − . . If WASP-33b containedFeH with a VMR of 10 − and MASCARA-2b a VMR of10 − , our injection and recovery tests indicate that wewould recover a SNR of about 3, similar to what we ex-tract from the data. Chemical modeling of iron in plan-etary atmospheres suggests that the FeH VMR is mostlikely between 10 − and 10 − . Recent results from HSTtransmission spectra retrieve much higher FeH VMRs(between 10 − and 10 − ), which is at odds with ourresults and those from chemical modeling, highlightingthe importance of high-resolution data.We conclude that FeH could potentially exist in the at-mospheres of WASP-33b and MASCARA-2b at a VMRof 10 − − − , but that higher quality data or moretransits is required to reject or confirm the planetarynature of these signals. Measurements of FeH in hot-Jupiter atmospheres is therefore well within the observ-ing limits of future ground-based high-dispersion spec-troscopy studies.A.K., I.S., and D.S. acknowledge funding from theEuropean Research Council (ERC) under the EuropeanUnion’s Horizon 2020 research and innovation programunder grant agreement No 694513. P.M. acknowledgessupport from the European Research Council underthe European Union’s Horizon 2020 research and in-novation program under grant agreement No. 832428.We would like to thank Alex Cridland and AlejandroSanchez-Lopez for useful discussions during the prepa-ration of the manuscript. We would also like to thankthe CARMENES consortium for making their data pub-licly available. This research has made use of the NASAExoplanet Archive, which is operated by the Califor-nia Institute of Technology, under contract with the Na-tional Aeronautics and Space Administration under theExoplanet Exploration Program. Facilities:
CAO:3.5m (CARMENES)
Software: astropy (Astropy Collaboration et al.2018), matplotlib (Hunter 2007), numpy (Van Der Waltet al. 2011)REFERENCES eH in Exoplanetary Atmospheres Agol, E., Cowan, N. B., Knutson, H. A., et al. 2010, ApJ,721, 1861, doi: 10.1088/0004-637X/721/2/1861Allard, F., Homeier, D., & Freytag, B. 2012, Royal Societyof London Philosophical Transactions Series A, 370,2765, doi: 10.1098/rsta.2011.0269Allart, R., Bourrier, V., Lovis, C., et al. 2019, A&A, 623,A58, doi: 10.1051/0004-6361/201834917Alonso-Floriano, F. J., Snellen, I. A. G., Czesla, S., et al.2019a, A&A, 629, A110,doi: 10.1051/0004-6361/201935979Alonso-Floriano, F. J., S´anchez-L´opez, A., Snellen, I. A. G.,et al. 2019b, A&A, 621, A74,doi: 10.1051/0004-6361/201834339Anderson, D. R., Collier Cameron, A., Delrez, L., et al.2014, MNRAS, 445, 1114, doi: 10.1093/mnras/stu1737—. 2017, A&A, 604, A110,doi: 10.1051/0004-6361/201730439Arcangeli, J., D´esert, J.-M., Line, M. R., et al. 2018, ApJL,855, L30, doi: 10.3847/2041-8213/aab272Astropy Collaboration, Price-Whelan, A. M., Sip˝ocz, B. M.,et al. 2018, AJ, 156, 123, doi: 10.3847/1538-3881/aabc4fBakos, G. ´A., Torres, G., P´al, A., et al. 2010, ApJ, 710,1724, doi: 10.1088/0004-637X/710/2/1724Baraffe, I., Homeier, D., Allard, F., & Chabrier, G. 2015,A&A, 577, A42, doi: 10.1051/0004-6361/201425481Birkby, J. L., de Kok, R. J., Brogi, M., et al. 2013,MNRAS, 436, L35, doi: 10.1093/mnrasl/slt107Birkby, J. L., de Kok, R. J., Brogi, M., Schwarz, H., &Snellen, I. A. G. 2017, AJ, 153, 138,doi: 10.3847/1538-3881/aa5c87Bonomo, A. S., Desidera, S., Benatti, S., et al. 2017, A&A,602, A107, doi: 10.1051/0004-6361/201629882Borsa, F., Rainer, M., Bonomo, A. S., et al. 2019, A&A,631, A34, doi: 10.1051/0004-6361/201935718Bourrier, V., Lovis, C., Beust, H., et al. 2018, Nature, 553,477, doi: 10.1038/nature24677Brogi, M., de Kok, R. J., Albrecht, S., et al. 2016, ApJ,817, 106, doi: 10.3847/0004-637X/817/2/106Brogi, M., Line, M., Bean, J., D´esert, J. M., & Schwarz, H.2017, ApJL, 839, L2, doi: 10.3847/2041-8213/aa6933Brogi, M., Snellen, I. A. G., de Kok, R. J., et al. 2012,Nature, 486, 502, doi: 10.1038/nature11161Brogi, M., Snellen, I. A. G., de Kok, R. J., et al. 2013, ApJ,767, 27, doi: 10.1088/0004-637X/767/1/27Burgasser, A. J., Looper, D. L., Kirkpatrick, J. D., Cruz,K. L., & Swift, B. J. 2008, ApJ, 674, 451,doi: 10.1086/524726Burgasser, A. J., Marley, M. S., Ackerman, A. S., et al.2002, ApJL, 571, L151, doi: 10.1086/341343 Butler, R. P., Vogt, S. S., Marcy, G. W., et al. 2004, ApJ,617, 580, doi: 10.1086/425173Caballero, J. A., Gu`ardia, J., L´opez del Fresno, M., et al.2016, 9910, 99100E, doi: 10.1117/12.2233574Cabot, S. H. C., Madhusudhan, N., Hawker, G. A., &Gandhi, S. 2019, MNRAS, 482, 4422,doi: 10.1093/mnras/sty2994Casasayas-Barris, N., Palle, E., Nowak, G., et al. 2017,A&A, 608, A135, doi: 10.1051/0004-6361/201731956Casasayas-Barris, N., Pall´e, E., Yan, F., et al. 2019, A&A,628, A9, doi: 10.1051/0004-6361/201935623Chakrabarty, A., & Sengupta, S. 2019, AJ, 158, 39,doi: 10.3847/1538-3881/ab24ddCharbonneau, D., Brown, T. M., Noyes, R. W., & Gilliland,R. L. 2002, ApJ, 568, 377, doi: 10.1086/338770Claret, A., Hauschildt, P. H., & Witte, S. 2012, A&A, 546,A14, doi: 10.1051/0004-6361/201219849—. 2013, A&A, 552, A16,doi: 10.1051/0004-6361/201220942Cridland, A. J., van Dishoeck, E. F., Alessi, M., & Pudritz,R. E. 2019, A&A, 632, A63,doi: 10.1051/0004-6361/201936105Damiano, M., Morello, G., Tsiaras, A., Zingales, T., &Tinetti, G. 2017, AJ, 154, 39,doi: 10.3847/1538-3881/aa738bdel Burgo, C., & Allende Prieto, C. 2016, MNRAS, 463,1400, doi: 10.1093/mnras/stw2005Deming, D., Wilkins, A., McCullough, P., et al. 2013, ApJ,774, 95, doi: 10.1088/0004-637X/774/2/95Ehrenreich, D., Lovis, C., Allart, R., et al. 2020, Nature,580, 597, doi: 10.1038/s41586-020-2107-1Eriksson, S., Asensio Torres, R., Janson, M., et al. 2020,arXiv e-prints, arXiv:2005.11725.https://arxiv.org/abs/2005.11725Evans, T. M., Sing, D. K., Wakeford, H. R., et al. 2016,ApJL, 822, L4, doi: 10.3847/2041-8205/822/1/L4Evans, T. M., Sing, D. K., Goyal, J. M., et al. 2018, AJ,156, 283, doi: 10.3847/1538-3881/aaebffFilippazzo, J. C., Rice, E. L., Faherty, J., et al. 2015, ApJ,810, 158, doi: 10.1088/0004-637X/810/2/158Fortney, J. J., Lodders, K., Marley, M. S., & Freedman,R. S. 2008, ApJ, 678, 1419, doi: 10.1086/528370Freedman, R. S., Lustig-Yaeger, J., Fortney, J. J., et al.2014, ApJS, 214, 25, doi: 10.1088/0067-0049/214/2/25Gaia Collaboration, Brown, A. G. A., Vallenari, A., et al.2018, A&A, 616, A1, doi: 10.1051/0004-6361/201833051Gaudi, B. S., Stassun, K. G., Collins, K. A., et al. 2017,Nature, 546, 514, doi: 10.1038/nature22392Hartman, J. D., Bakos, G. ´A., Torres, G., et al. 2011, ApJ,742, 59, doi: 10.1088/0004-637X/742/1/59 Kesseli et al.
Hartman, J. D., Bakos, G. ´A., Buchhave, L. A., et al. 2015,AJ, 150, 197, doi: 10.1088/0004-6256/150/6/197Haynes, K., Mandell, A. M., Madhusudhan, N., Deming,D., & Knutson, H. 2015, ApJ, 806, 146,doi: 10.1088/0004-637X/806/2/146Helling, C. 2019, Annual Review of Earth and PlanetarySciences, 47, 583,doi: 10.1146/annurev-earth-053018-060401Herman, M. K., de Mooij, E. J. W., Jayawardhana, R., &Brogi, M. 2020, arXiv e-prints, arXiv:2006.10743.https://arxiv.org/abs/2006.10743Hubeny, I., Burrows, A., & Sudarsky, D. 2003, ApJ, 594,1011, doi: 10.1086/377080Hunter, J. D. 2007, Computing in Science Engineering, 9,90Johnson, M. C., Cochran, W. D., Collier Cameron, A., &Bayliss, D. 2015, ApJL, 810, L23,doi: 10.1088/2041-8205/810/2/L23Kataria, T., Sing, D. K., Lewis, N. K., et al. 2016, ApJ,821, 9, doi: 10.3847/0004-637X/821/1/9Kirkpatrick, J. D., Reid, I. N., Liebert, J., et al. 1999, ApJ,519, 802, doi: 10.1086/307414Lecavelier Des Etangs, A., Pont, F., Vidal-Madjar, A., &Sing, D. 2008, A&A, 481, L83,doi: 10.1051/0004-6361:200809388Lehmann, H., Guenther, E., Sebastian, D., et al. 2015,A&A, 578, L4, doi: 10.1051/0004-6361/201526176Lodders, K. 1999, ApJ, 519, 793, doi: 10.1086/307387Lothringer, J. D., & Barman, T. 2019, ApJ, 876, 69,doi: 10.3847/1538-4357/ab1485Lothringer, J. D., Barman, T., & Koskinen, T. 2018, ApJ,866, 27, doi: 10.3847/1538-4357/aadd9eLund, M. B., Rodriguez, J. E., Zhou, G., et al. 2017, AJ,154, 194, doi: 10.3847/1538-3881/aa8f95MacDonald, R. J., & Madhusudhan, N. 2019, MNRAS, 486,1292, doi: 10.1093/mnras/stz789Mandel, K., & Agol, E. 2002, ApJL, 580, L171,doi: 10.1086/345520Martin, E. C., Mace, G. N., McLean, I. S., et al. 2017, ApJ,838, 73, doi: 10.3847/1538-4357/aa6338Mazeh, T., Tamuz, O., & Zucker, S. 2007, AstronomicalSociety of the Pacific Conference Series, Vol. 366, TheSys-Rem Detrending Algorithm: Implementation andTesting, ed. C. Afonso, D. Weldrake, & T. Henning, 119Merritt, S. R., Gibson, N. P., Nugroho, S. K., et al. 2020,arXiv e-prints, arXiv:2002.02795.https://arxiv.org/abs/2002.02795Molaverdikhani, K., Henning, T., & Molli`ere, P. 2019, ApJ,883, 194, doi: 10.3847/1538-4357/ab3e30 Molli`ere, P., van Boekel, R., Bouwman, J., et al. 2017,A&A, 600, A10, doi: 10.1051/0004-6361/201629800Molli`ere, P., Wardenier, J. P., van Boekel, R., et al. 2019,A&A, 627, A67, doi: 10.1051/0004-6361/201935470Mordasini, C., van Boekel, R., Molli`ere, P., Henning, T., &Benneke, B. 2016, ApJ, 832, 41,doi: 10.3847/0004-637X/832/1/41Nikolov, N., Sing, D. K., Goyal, J., et al. 2018, MNRAS,474, 1705, doi: 10.1093/mnras/stx2865Nortmann, L., Pall´e, E., Murgas, F., et al. 2016, A&A, 594,A65, doi: 10.1051/0004-6361/201527323Nortmann, L., Pall´e, E., Salz, M., et al. 2018, Science, 362,1388, doi: 10.1126/science.aat5348Nugroho, S. K., Kawahara, H., Masuda, K., et al. 2017, AJ,154, 221, doi: 10.3847/1538-3881/aa9433¨Oberg, K. I., Murray-Clay, R., & Bergin, E. A. 2011, ApJL,743, L16, doi: 10.1088/2041-8205/743/1/L16Owen, J. E. 2019, Annual Review of Earth and PlanetarySciences, 47, 67,doi: 10.1146/annurev-earth-053018-060246Parviainen, H. 2015, MNRAS, 450, 3233,doi: 10.1093/mnras/stv894Pino, L., D´esert, J.-M., Brogi, M., et al. 2020, ApJL, 894,L27, doi: 10.3847/2041-8213/ab8c44Pont, F., Sing, D. K., Gibson, N. P., et al. 2013, MNRAS,432, 2917, doi: 10.1093/mnras/stt651Quirrenbach, A., Amado, P. J., Ribas, I., et al. 2018, 10702,107020W, doi: 10.1117/12.2313689Salz, M., Schneider, P. C., Czesla, S., & Schmitt,J. H. M. M. 2015, A&A, 576, A42,doi: 10.1051/0004-6361/201425243S´anchez-L´opez, A., Alonso-Floriano, F. J., L´opez-Puertas,M., et al. 2019, A&A, 630, A53,doi: 10.1051/0004-6361/201936084Sanchis-Ojeda, R., & Winn, J. N. 2011, ApJ, 743, 61,doi: 10.1088/0004-637X/743/1/61Sheppard, K. B., Mandell, A. M., Tamburo, P., et al. 2017,ApJL, 850, L32, doi: 10.3847/2041-8213/aa9ae9Sing, D. K., Wakeford, H. R., Showman, A. P., et al. 2015,MNRAS, 446, 2428, doi: 10.1093/mnras/stu2279Sing, D. K., Fortney, J. J., Nikolov, N., et al. 2016, Nature,529, 59, doi: 10.1038/nature16068Skaf, N., Fabienne Bieger, M., Edwards, B., et al. 2020,arXiv e-prints, arXiv:2005.09615.https://arxiv.org/abs/2005.09615Smette, A., Sana, H., Noll, S., et al. 2015, A&A, 576, A77,doi: 10.1051/0004-6361/201423932Snellen, I. A. G., Albrecht, S., de Mooij, E. J. W., & LePoole, R. S. 2008, A&A, 487, 357,doi: 10.1051/0004-6361:200809762 eH in Exoplanetary Atmospheres19