Into the UV: The Atmosphere of the Hot Jupiter HAT-P-41b Revealed
Nikole K. Lewis, Hannah R. Wakeford, Ryan J. MacDonald, Jayesh M. Goyal, David K. Sing, Joanna Barstow, Diana Powell, Tiffany Kataria, Ishan Mishra, Mark S. Marley, Natasha E. Batalha, Julie I. Moses, Peter Gao, Tom J. Wilson, Katy L. Chubb, Thomas Mikal-Evans, Nikolay Nikolov, Nor Pirzkal, Jessica J. Spake, Kevin B. Stevenson, Jeff Valenti, Xi Zhang
DDraft version October 20, 2020
Typeset using L A TEX twocolumn style in AASTeX63
Into the UV: The Atmosphere of the Hot Jupiter HAT-P-41b Revealed
N. K. Lewis , H. R. Wakeford , R. J. MacDonald , J. M. Goyal , D. K. Sing ,
3, 4
J. Barstow ,
5, 6
D. Powell , T. Kataria , I. Mishra , M. S. Marley , N. E. Batalha , J. I. Moses , P. Gao , T. J. Wilson , K. L. Chubb , T. Mikal-Evans , N. Nikolov ,
15, 16
N. Pirzkal, J. J. Spake, K. B. Stevenson , J. Valenti , and X. Zhang Department of Astronomy and Carl Sagan Institute, Cornell University, 122 Sciences Drive, Ithaca, NY 14853, USA School of Physics, University of Bristol, HH Wills Physics Laboratory, Tyndall Avenue, Bristol BS8 1TL, UK Department of Physics and Astronomy, Johns Hopkins University, Baltimore, MD 21218, USA Department of Earth and Planetary Sciences, Johns Hopkins University, Baltimore, MD 21218, USA School of Physical Sciences, Open University, Walton Hall, Kents Hill, Milton Keynes, MK7 6AA UK Department of Physics and Astronomy, University College London, Gower St, London WC1E 6BT, UK Department of Astronomy and Astrophysics, University of California, Santa Cruz, CA 95064, USA Jet Propulsion Laboratory, California Institute of Technology, 4800 Oak Grove Drive, Pasadena, CA 91109, USA NASA Ames Research Center, MS 245-3, Mountain View, CA 94035, USA Space Science Institute, 4765 Walnut Street, Suite B, Boulder, CO 80301, USA Department of Astronomy, University of California, Berkeley, CA 94720, USA University of Exeter, Physics Building, Stocker Road, Exeter, Devon, Ex4 4QL, UK SRON Netherlands Institute for Space Research, Sorbonnelaan 2, 3584 CA, Utrecht, Netherlands Kavli Institute for Astrophysics and Space Research, Massachusetts Institute of Technology, 77 Massachusetts Avenue, 37-241,Cambridge, MA 02139, USA Space Telescope Science Institute, 3700 San Martin Drive, Baltimore, MD 21218, USA Department of Physics & Astronomy, Johns Hopkins University, Baltimore, MD 21218, USA Space Telescope Science Institute 3700 San Martin Drive Baltimore, MD 21218, USA Division of Geological and Planetary Sciences, California Institute of Technology, Pasadena, CA 91125, USA Johns Hopkins APL, 11100 Johns Hopkins Rd, Laurel, MD 20723, USA Department of Earth and Planetary Sciences, University of California, Santa Cruz, CA 95064, USA (Published October 8, 2020)
Submitted to ApJLABSTRACTFor solar-system objects, ultraviolet spectroscopy has been critical in identifying sources for strato-spheric heating and measuring the abundances of a variety of hydrocarbon and sulfur-bearing species,produced via photochemical mechanisms, as well as oxygen and ozone. To date, less than 20 exoplan-ets have been probed in this critical wavelength range (0.2-0.4 µ m). Here we use data from Hubble’snewly implemented WFC3 UVIS G280 grism to probe the atmosphere of the hot Jupiter HAT-P-41b inthe ultraviolet through optical in combination with observations at infrared wavelengths. We analyzeand interpret HAT-P-41b’s 0.2-5.0 µ m transmission spectrum using a broad range of methodologiesincluding multiple treatments of data systematics as well as comparisons with atmospheric forward,cloud microphysical, and multiple atmospheric retrieval models. Although some analysis and interpre-tation methods favor the presence of clouds or potentially a combination of Na, VO, AlO, and CrH toexplain the ultraviolet through optical portions of HAT-P-41b’s transmission spectrum, we find thatthe presence of a significant H − opacity provides the most robust explanation. We obtain a constraintfor the abundance of H − , log(H − ) = − . ± .
62 in HAT-P-41b’s atmosphere, which is several ordersof magnitude larger than predictions from equilibrium chemistry for a ∼ Corresponding author: Nikole K. [email protected] a r X i v : . [ a s t r o - ph . E P ] O c t Lewis et al. show that a combination of photochemical and collisional processes on hot hydrogen-dominated exo-planets can readily supply the necessary amount of H − and suggest that such processes are at work inHAT-P-41b and many other hot Jupiter atmospheres. Keywords:
Exoplanet atmospheres (487); Observational astronomy (1145); Exoplanet atmosphericcomposition (2021); Spectroscopy (1558) INTRODUCTIONFor more than 50 years now a variety of space-basedobservatories have provided a window into the ultravi-olet (UV) properties of planetary objects (see review inBrosch et al. 2006). The UV provides a unique per-spective on a number of physical processes in planetaryatmospheres, such as photodissociation and photoion-ization, as well as containing unique spectral indicatorsfor a broad range of atoms, ions, and molecules. In thesolar system the energy source for stratospheric heatingis often absorption of solar UV radiation, usually by highaltitude molecules or hazes produced by photochemicalprocesses, ozone in Earth’s stratosphere being the pro-totypical example. Other examples include UV absorp-tion by aerosols in the giant planet stratospheres (e.g.Zhang et al. 2015). We now know that these UV-drivenprocesses and spectral signatures are not just limitedto the solar system. Exoplanets have been observed toalso possess stratospheres (e.g. Evans et al. 2017) anddisplay signatures of a variety of atmospheric aerosols(e.g. Sing et al. 2016) and ionic species (e.g. Sing et al.2019; Hoeijmakers et al. 2019). However, the numberof exoplanet transits observed at UV and near-UV (0.2-0.4 µ m) wavelengths is only about a dozen, which limitsour ability to fully explore the atmospheric processesshaping these worlds. Here we expand the sample ofexoplanets with UV through infrared (IR) observationswith our exploration of HAT-P-41b that leverages datausing a new observing strategy with the Hubble SpaceTelescope .HAT-P-41b is an inflated ‘hot Jupiter’ (0.8 M J ,1.7 R J , 1940 K) discovered orbiting an F-type star byHartman et al. (2012). The quiet nature of the hoststar, highly inflated planetary atmosphere, and short or-bital period make HAT-P-41b an ideal target for spec-troscopic analysis to probe the physics and chemistryat work in this planet’s atmosphere. For this rea-son, HAT-P-41b was targeted for five observations aspart of the Panchromatic Exoplanet Treasury Survey(PanCET GO-14767, PIs: Sing & Lopez-Morales) usingthree modes on Space Telescope Imaging Spectrograph(STIS) (E230M, 2 × G430L, G750L gratings) and theG141 infrared grism on Wide Field Camera 3 (WFC3),that would provide a spectrum from 0.22–1.7 µ m. Tofurther leverage Hubble ’s spectroscopic capabilities, our team also selected HAT-P-41b as the prime target totest the use of WFC3’s UVIS G280 grism (GO-15288,PIs: Sing & Lewis) on exoplanet time series studies.WFC3’s UVIS grism provides continuous coverage in asingle observation from the UV to the optical and hasthe potential to replace the equivalent three modes onSTIS traditionally used. While the WFC3/UVIS grismpresents some challenges for both observational strate-gies and data reduction, it is able to produce high pre-cision spectroscopy from the UV through visible (0.2–0.8 µ m) at higher resolution than the combination ofthe optical STIS modes (see Wakeford et al. 2020 forfull details).Here we present a detailed exploration of the physicsand chemistry that is shaping the transmission spec-trum of HAT-P-41b from 0.2-5.0 µ m. We have leverageddata from the newly commissioned WFC3-UVIS G280mode (0.2-0.8 µ m) and combined with observations ofthe system from WFC3-IR G141 mode (1.1-1.7 µ m)and Spitzer’s
Infrared Array Camera (IRAC) (3.6 and4.5 µ m channels). We present an independent analysisof the PanCET WFC3 G141 data that was previouslypublished as part of a population study by Tsiaras et al.(2018), and combine it with the data published by Wake-ford et al. (2020). In interpreting our transmission spec-trum of HAT-P-41b we employ a broad range of analysistools and techniques including comparisons with atmo-spheric forward model grids, three-dimensional generalcirculation models, aerosol microphysics models, andthree different atmospheric retrieval tools. We applythis bevvy of analysis tools to transmission spectra ofHAT-P-41b that utilize two different reduction methodsfor the WFC3 UVIS G280 data to test the robustness ofour interpretation to data reduction method employed.This work highlights that our interpretation of exoplanetatmospheric characterization observations are served byexploring multiple reductions of the same data and mul-tiple analyses that can provide complementary views ofthe processes shaping exoplanetary atmospheres. OBSERVATIONSObservations of the transiting hot Jupiter HAT-P-41bwere obtained with
Hubble and
Spitzer
Space Telescopesover three different observing programs to construct thetransmission spectrum from 0.2–5 µ m. Details of the re- AT-P-41b’s Atmosphere Revealed µ m (13044, PI: Deming) datacan be found in Wakeford et al. (2020). In that workthey present different analysis methods on the newlyimplemented UVIS grism mode, that result in two inde-pendent but consistent transmission spectra of approxi-mately 60 measurements in 10 nm bins from 0.2–0.8 µ m.The Spitzer measurements at 3.6 and 4.5 µ m are used tofurther constrain the system parameters, a/R ∗ , inclina-tion, and orbital period which are then fixed in the spec-troscopic analysis to prevent arbitrary offsets betweenthe measurements. Here we outline the data analysisperformed on the Hubble WFC3 IR G141 transmissionspectra taken as part of the Hubble
PanCET program(GO-14767, PI: Sing & Lopez-Morales) used to measurethe planet’s near-IR spectrum, which is critical to thefull interpretation of the atmosphere.A single transit of HAT-P-41b was observed on 2016October 16 using
Hubble ’s WFC3 IR G141 grism, 1.1–1.7 µ m. The observations were conducted in spatial scanmode on the 256 subarray with an exposure time of 81seconds and a scan rate of 0.072”/s resulting in a scanlength of approximately 45 pixels on the detector. Thenearby companion is also visible in the scan and over-laps with the target spectrum by approximately 30 pix-els. We therefore use the difference imaging technique(e.g., Kreidberg et al. 2014; Evans et al. 2016) to sam-ple the target scan (NSAMP = 12) and reconstruct thetarget spectrum without the influence of the companionstar. To ensure a robust interpretation of the full plan-etary transmission spectrum we opt not to use the pub-lished G141 transmission spectrum presented by Tsiaraset al. (2018) as Wakeford et al. (2020) provides updatedHAT-P-41b system parameters that should be used con-sistently across reductions of the Hubble
WFC3/UVISG280, WFC3/IR G141 and
Spitzer
IRAC 3.6 and 4.5 µ mobservations used to construct the UV through IR spec-trum considered in this study.We first analyse the broadband lightcurve from 1.1–1.7 µ m prior to dividing the spectrum into multiple spec-troscopic bins. The observation spanned 5 HST orbits,in this analysis we follow standard practices and dis-card the first orbit and the first exposure in each or-bit as they are subject to additional systematics (e.g.,Deming et al. 2013; Sing et al. 2016). We analyze thetransit time-series data using the instrument system-atics marginalization technique outlined by Wakefordet al. (2016), that has been successfully applied to arange of datasets (e.g., Kilpatrick et al. 2018; Sing et al.2016; Wakeford et al. 2017, 2018, 2020). For a con-sistent analysis between all datasets we fix the system parameters determined in Wakeford et al. (2020); or-bital period = 2.69404861 days, inclination = 89.17 o , anda/R ∗ = 5.55. We also apply the same technique used tocalculate the limb-darkening coefficients over the desiredwavelengths; 3D stellar models using a 4-parameter non-linear limb-darkening law. This ensures that there areno offsets between each dataset that would compromisethe interpretation, which would be the case if using pre-viously published analysis of these data from Tsiaraset al. (2018) that employ different assumptions (see Ap-pendix A for more details).We obtained a transit depth precision of 12 ppm onthe broadband lightcurve, with an average precision of40 ppm in 47 nm bins from 1.1–1.7 µ m. We tested arange of binning options along the spectrum and foundthe 47 nm bins to be consistent in the structure of theresultant transmission spectrum while minimizing thescatter of the light curve residuals. Each lightcurve isindependently analysed correcting for systematics usinga grid of 50 sudo-stochastic polynomial models that ac-count for observatory and instrument based systematics(see Wakeford et al. 2016). For each lightcurve we cal-culate the maximum likelihood estimation based on theAkaike information criterion (AIC) for each of the 50 po-tential systematic models. These are then used as an ap-proximation of the evidence for that systematic correc-tion and converted into a normalized weight. The tran-sit depth measured based on the fit of each systematicmodel is then marginalized based on the weight assignedto the model such that the marginalized transit depthis representative of the evidence across all 50 models.This method is applied to each spectroscopic light curvemeasured in each wavelength bin (see spectroscopic lightcurves in the supplementary material). The full trans-mission spectrum we obtain for HAT-P-41b from 0.2–5 µ m, combining the IR measurements with the UVISand Spitzer data, is presented in Figure 1. SPECTRAL ANALYSISOur initial comparison of HAT-P-41b’s transmissionspectrum with theoretical predictions for the planet(Figure 1) highlights that further exploration of thephysical and chemical processes shaping the spectrumis needed. Here we employ a multiple modelling ap-proach to ensure that we fully explore these processesand our interpretation of HAT-P-41b’s spectrum is ro-bust. We first employ one-dimensional (1D) scalableforward models that assume equilibrium chemistry andprovide a range of parameterizations to represent thepresence of aerosols in the atmosphere. We then lever-age predictions from a three-dimensional (3D) generalcirculation model for HAT-P-41b to guide exploration of
Lewis et al. T r a n s i t D e p t h ( % ) Uniform cloud10x scattering1100x scatteringWFC3/UVIS G280 JitterWFC3/UVIS G280 MarginalizedWFC3/IR G141 + Spitzer
Figure 1.
HAT-P-41b’s transmission spectrum from 0.2–5.0 µ m (red/blue/grey points) obtained via observations with Hubble’s
WFC3 UVIS G280 (0.2–0.8 µ m) and IR G141 (1.1–1.7 µ m) grisms and Spitzer’s
IRAC Ch1 (3.6 µ m) and Ch2 (4.5 µ m)photometry. We include two reductions of the WFC3 UVIS G280 data that employ marginalization (blue circles) and jitterdecorrelation (red squares) treatments of systematics. Theoretical transmission spectra from atmospheric models specific toHAT-P-41b from Goyal et al. (2019) are shown for comparison. Atmospheric models with large amounts of scattering due tothe presence of small particles in the atmosphere (light green line) best match the Hubble
WFC3 IR G141 and
Spitzer
Ch1 andCh2 observations, but fail to match the
Hubble
WFC3 UVIS G280 observations where models with low scattering (teal line)and/or a uniform cloud deck (dark purple line) are preferred. This highlights the need for NUV (0.2–0.4 µ m) and optical inadditional to IR (1.0-5.0 µ m) observations to robustly probe exoplanet atmospheres. aerosol formation using a microphysical model. Finally,we conduct a series of atmospheric retrieval analyses toinfer the atmospheric composition of HAT-P-41b. In allour analyses we consider both the marginalization andjitter decorrelation treatments of the systematics for the Hubble
WFC3 UVIS G280 data. Exploration of the sim-ilarities and differences in our inferences from these for-ward and inverse modelling approaches will allow us torobustly identify the physics and chemistry at work inHAT-P-41b’s atmosphere.3.1.
Comparisons with Forward Models
We begin our exploration of HAT-P-41b’s atmosphereby comparing our observed spectra with a grid of 1D for-ward models. The Goyal et al. (2019) library of ‘generic’exoplanet transmission spectra spans a broad rangeof atmospheric compositions, temperature, and aerosolproperties, under the assumption of equilibrium chem-istry. Our fitting procedure follows a similar methodol-ogy to Wakeford et al. (2018), repeated for each WFC3UVIS G280 data reduction. The resulting best fittingmodels for the jitter detrending and marginalization re-ductions are shown in Figure 2 (top panel). Though the forward model grid can fit most of the observations,mismatches occur in several spectral regions, especiallynear 1.5 and 3.6 µ m. As a result, the χ ν values for thisequilibrium model were 2.24 and 2.60 for the jitter andmarginalization cases, respectively.We find slight variations between the preferred for-ward models for each data reduction. The best fittingmodel for the marginalization data reduction estimatesthe atmospheric temperature at the pressures probedvia transmission as 1400 K, with a solar metallicity, so-lar carbon-to-oxygen (C/O) ratio (0.56), and a signif-icant cloud deck. In contrast, the best fitting modelfor the jitter detrending reduction approach prefers acolder atmosphere (900 K), 10 × solar metallicity, subso-lar C/O (0.35), and a significant cloud deck. The fits arebroadly consistent within the limits provided by theirpseudo-probability distributions . However, note thatthe best fitting model in the marginalization case al-lows for the presence of VO, which can be present under All our pseudo-probability distributions are available in the on-line supplementary material.
AT-P-41b’s Atmosphere Revealed Wavelength ( m) T r a n s i t D e p t h ( R p / R ∗ ) × WFC3 G280 WFC3 G141 Spitzer
HAT-P-41b: Chemical Equilibrium
ATMO fit (Marginalization)ATMO fit (Jitter)Marginalization DataJitter Data
Wavelength ( m) T r a n s i t D e p t h ( R p / R ∗ ) × WFC3 G280 WFC3 G141 Spitzer
HAT-P-41b: Self-consistent Clouds
CARMAMarginalization DataJitter Data
Figure 2.
Forward model fits to HAT-P-41b’s transmission spectrum. Top: chemical equilibrium grid fit. Two independentWFC3 UVIS G280 data reduction techniques are shown: marginalization (blue) and jitter decorrelation (red). For each datareduction, the best fitting model from the Goyal et al. (2019) grid is displayed. Both fits favor a cloud deck to explain thecontinuum UVIS data, while the marginalization fit additionally includes VO opacity to fit optical substructure. Neither modelcan fit the two reddest WFC3 IR G141 data points nor the 3.6 µ m Spitzer point. Bottom: self-consistent microphysicalcloud model fit. Vertical cloud distributions are computed using CARMA (Gao et al. 2018), assuming solar metallicity, withtransmission spectra computed as in Powell et al. (2019). The assumed temperature and vertical mixing profiles are perturbedfrom the globally averaged GCM profiles of Figure 3 to identify the model with minimal residuals. The best-fitting model favorsa population of Al O clouds at ∼ − − − bar. The clouds become optically thin at longer wavelengths, resulting in animproved fit to the 3.6 µ m Spitzer point, but struggle to capture the WFC3 G141 H O feature in the optically thick regime.
Lewis et al. chemical equilibrium assumptions at 1400 K at relevantatmospheric pressures, to explain the spectral structureshortwards of 1 µ m (Figure 2, top panel). This illus-trates a degree of sensitivity to the chosen data reduc-tion technique - we further quantify this in section 3.3.Our forward model comparison indicates clouds couldplay a key role in shaping HAT-P-41b’s transmissionspectrum. However, our grid includes clouds via a sim-ple cloud deck pressure decoupled from the atmosphericdynamics and thermochemical structure. We thereforeverified the plausibility of the formation and advectionof clouds in HAT-P-41b’s atmosphere by running a gen-eral circulation model using the SPARC/MITgcm (e.g.Showman et al. 2009; Kataria et al. 2016). We as-sume the same physical parameters of HAT-P-41b usedthroughout this study and a solar metallicity atmo-spheric composition consistent with the preferred 1D at-mospheric models. The resulting globally and spatially-averaged temperature and vertical mixing profiles forthe dayside, nightside, and each terminator are shown inFigure 3. The temperature profiles of the west (morn-ing; green profiles) and east (evening; purple profiles)terminators cross multiple condensation curves indicat-ing that a broad range of cloud species may be presentin HAT-P-41b’s observable atmosphere. The ∼
200 Kdifference between the terminators may drive differingcloud properties on each limb of the planet. With theirplausibility and constraints on vertical mixing estab-lished, we turn to detailed microphysical modelling to in-vestigate the physical nature and composition of cloudsin the atmosphere of HAT-P-41b.3.2.
Predictions for Cloud Formation
We simulate cloud distributions for HAT-P-41b usingthe Community Aerosol and Radiation Model for Atmo-spheres (CARMA). CARMA is a one-dimensional bin-scheme aerosol microphysics model that computes verti-cal and size distributions of aerosol particles. CARMAsolves a discretized continuity equation that accountsfor aerosol nucleation, condensation, evaporation, andtransport (see Gao et al. 2018, and references therein).The specific microphysical setup that we use for mod-eling condensational clouds in this work is described byPowell et al. (2019) and has been shown to reproducetrends in Hot Jupiter cloudiness across a broad rangeof parameter space. In this work we do not vary themicrophysical parameters of the condensate species andwe assume that the volatile species in the atmospherehave a solar abundance. We post-process these resultsto calculate transmission spectra using a modified ver-sion of Exo-Transmit that includes a Mie theory pre- scription for the cloud opacities as described by Powellet al. (2019).To model the observations of HAT-P-41b, we mustchoose input planetary properties. As the atmosphericmetallicity and microphysical parameters are held con-stant, the remaining tunable parameters are the tem-perature profile and the amount of atmospheric verticalmixing. For both parameters, we begin by consideringthe globally averaged 3D GCM temperature and verti-cal mixing profile shown in Figure 3. We then vary bothprofiles by a constant factor, as both parameters are notwell constrained, to derive cloud properties that give riseto simulated spectra that best match the observations.The best-fitting model, shown in Figure 2 (bottompanel), is 400 K hotter than the globally-averaged tem-perature profile from the 3D GCM and has two orders-of-magnitude less global mixing. The increase in tem-perature limits the supersaturation of the condensiblespecies and thus the formation of clouds. This increasein temperature is consistent with temperature variationsseen in the GCM, especially limb-to-limb variations (e.g.Kataria et al. 2016). Reducing the amount of verticalmixing further limits the formation of clouds as well asthe vertical extent in the atmosphere of the cloud parti-cles (see Powell et al. 2019). The level of reduction in thevertical mixing calculated from the GCM is consistentwith studies of GCM tracer transport where the derivedmixing of tracers is commonly two orders of magnitudeless efficient than the transport of atmospheric gases(e.g. Parmentier et al. 2013). Both of these effects giverise to a population of large aluminum (Al O ) cloudsat ∼ − − − bar that dominate the cloud opacity.These clouds are optically thick at short wavelengthsand optically thin at longer wavelengths (Vahidinia et al.2014). The results of this one-dimensional globally av-eraged cloud modeling result in a similar spectral fit tothe first order forward models considered in section 3.1(e.g. the jitter fit in the top panel of Figure 2). How-ever, the inclusion of wavelength-dependant cloud opac-ities improves the fit at 3.6 µ m. Nevertheless, this cloudmodel fails to reproduce the full shape of the H O fea-ture centered at 1.4 µ m. The corresponding χ ν valuesfor this cloud model were 2.67 and 2.85 for the jitterand marginalization cases, respectively. As an addi-tional sanity check, we ran simple Mie theory modelcomparisons (Wakeford & Sing 2015), which requiredlarge ( ∼ µ m) particles that result in gray opac-ities out to 2 µ m and miss key molecular absorptionfeatures. In seeking a model capable of explaining theobservations over the full wavelength range, we turn nowto retrieval analyses. AT-P-41b’s Atmosphere Revealed
600 800 1000 1200 1400 1600 1800 2000 2200Temperature (K)10 -5 -4 -3 -2 -1 P r ess u r e ( b a r s ) M g S i O M g S i O C r M n S N a S Zn S KC l F e A l O C a T i O Global average Dayside Average Nightside AverageEast term averageWest term averageCARMA best fit k zz (cm s -1 )10 -5 -4 -3 -2 -1 P r ess u r e ( b a r s ) Global average Dayside Average Nightside AverageEast term averageWest term averageCARMA best fit
Figure 3.
Average pressure-temperature (PT) profiles (left) and vertical diffusion coefficient ( k zz , right) derived from three-dimensional general circulation model for HAT-P-41b. The average PT profiles intersect with the condensation curves of anumber of potential cloud species (dotted lines in left panel), which indicates that clouds could play a critical role throughoutHAT-P-41b’s atmosphere. The strength of the predicted vertical mixing in HAT-P-41b’s atmosphere (right panel) highlightsthat clouds formed at the bar level or below could be easily advected into the observable portion of the atmosphere near amillibar. Profiles for temperature and k zz that provided the CARMA model best fit to the data are shown for comparison. Atmospheric Retrieval Analyses
Atmospheric retrievals relax many aforementioned as-sumptions, such as chemical equilibrium, opting insteadto parametrize the atmospheric state. Bayesian sam-pling techniques explore millions of potential states,comparing their resultant transmission spectra with ob-servations to derive posterior probability distributionsfor each model parameter. This inverse approach allowsatmospheric properties, such as chemical abundances,temperature profiles, and cloud properties, to be re-trieved directly from the data.Our retrieval philosophy employs two central princi-ples to ensure robust atmospheric inferences. First, weconduct retrievals for each data reduction separately, es-tablishing the sensitivity of derived atmospheric proper-ties to different reduction techniques. Secondly, threedifferent retrieval codes are independently applied toeach data reduction: POSEIDON (MacDonald & Mad-husudhan 2017), NEMESIS (Irwin et al. 2008; Barstowet al. 2017; Krissansen-Totton et al. 2018), and ATMO(Amundsen et al. 2014; Tremblin et al. 2015; Wakefordet al. 2017). We can thereby quantify the impact on ourresults of differing modelling choices (e.g. molecular linelists, temperature profiles, cloud parametrizations).We consider an extensive range of potential atmo-spheric components. Our investigations include opac-ity due to the following chemical species: H , He,H − , Na, K, Li, TiO, VO, AlO, SiO, TiH, CrH, FeH,AlH, CaH, SiH, H O, CH , CO, CO , NH , HCN,NO, H S, SH, PH , and C H . Isothermal (NEME-SIS and ATMO) and non-isothermal temperature struc- tures (POSEIDON, Madhusudhan & Seager 2009) wereconsidered. Three cloud models were examined: (i)an opaque cloud deck with a vertically-uniform haze(ATMO, Wakeford et al. 2017); (ii) a single cloud deck,with variable top and base pressures, and power lawextinction with a variable index (NEMESIS, Barstowet al. 2017); and (iii) patchy clouds and haze aroundthe terminator (POSEIDON, MacDonald & Madhusud-han 2017). Although these cloud prescriptions do notexplicitly include Mie scattering calculations, our modelfitting exercise with CARMA (see Section 3.2) demon-strates that more complex microphysical prescriptionsfor clouds do not fully explain the observed spectrumof HAT-P-41b. The influence of stellar contaminationwas also considered in the retrieval process (NEMESIS,e.g. Pinhas et al. 2018). Iterative expansion of the con-sidered molecules was performed between the three re-trieval codes until a minimal basis set was identified. Allthree codes use nested sampling to explore the parame-ter space, either via MultiNest (NEMESIS and POSEI-DON, Feroz & Hobson 2008; Feroz et al. 2009, 2013;Buchner et al. 2014) or dynesty (ATMO, Speagle 2020).Our retrieved transmission spectra are compared tothe observations of HAT-P-41b in Figure 4. Contrast-ing with the forward models of the previous sections, theretrievals prefer a combination of gas phase optical opac-ity sources instead of clouds. Specifically, at least one of Lewis et al.
Wavelength ( m) T r a n s i t D e p t h ( R p / R ∗ ) × WFC3 G280 WFC3 G141 Spitzer
HAT-P-41b: Jitter Retrieval
POSEIDONNEMESISATMOData
Wavelength ( m) T r a n s i t D e p t h ( R p / R ∗ ) × WFC3 G280 WFC3 G141 Spitzer
HAT-P-41b: Marginalization Retrieval
POSEIDONNEMESISATMOData
Figure 4.
Atmospheric retrievals of HAT-P-41b’s transmission spectrum. Top: jitter decorrelation reduction. Bottom: sys-tematic marginalization reduction. Each panel shows the median retrieved spectrum (solid lines) and 1 σ confidence regions(shading) from three retrieval codes: POSEIDON (purple), NEMESIS (green), and ATMO (blue) (MacDonald & Madhusudhan2017; Barstow et al. 2017; Wakeford et al. 2017). All models are binned to a common spectral resolution ( R = 100) for clarity.The three codes achieve an excellent fit across the full wavelength range, concurring on the presence of at least one significantopacity source in the visible and near-UV (H − , Na, CrH, VO, or AlO - see Appendix C for a spectral decomposition), H O inthe near-infrared, and an absence of clouds in the observable atmosphere. This interpretation holds for both reductions.
AT-P-41b’s Atmosphere Revealed − , AlO, CrH, or VO , in addition to Na, are requiredto explain the WFC3 UVIS G280 observations. By in-voking chemical species with strong near-UV to visibleopacities, but weak infrared opacities, the infrared ob-servations can be well fit by H O alone. In particular,there is a clear preference for the bound-free opacity ofthe hydrogen anion, H − , which provides a smooth con-tinuum across the UVIS range before falling off rapidlyas the ionization threshold ( ∼ µ m) is approached.This continuum has similar spectral characteristics to acloud deck across the UVIS range, potentially explainingthe preference for clouds in our forward models (whichdo not include H − ). The other inferred opacity sourcesare somewhat sensitive to the retrieval code and chosendata reduction, as we demonstrate below. Nevertheless,all three retrieval codes agree on the interpretation of astrong near-UV to visible chemical opacity source, with-out the need for clouds in the observable atmosphere.3.4. The Atmospheric Composition of HAT-P-41b
We detect the presence of H O in HAT-P-41b’s at-mosphere at (cid:38) σ confidence (jitter: 6 . σ ; marginal-ization: 5 . σ ). The visible and near-UV observationsadditionally require at least one other prominent opac-ity source, with candidates identified as H − , CrH, AlO,VO, or Na. Besides these species, our initial retrievals- ranging in complexity from 12 to 37 free parameters- included many parameters left largely unconstrainedby the present observations (e.g. temperature structureand cloud properties). Consequently, the best-fitting χ ν ranged from ∼ O, H − , Na, CrH, AlO, andVO. This best-fitting model attains χ ν = 1.50 and 1.72for the jitter and marginalization cases, respectively.The χ ν values obtained by our best-fitting retrievalmodels demonstrate a greatly improved quality offit compared to the chemical equilibrium and self-consistent cloud models considered in Sections 3.1 and3.2. However, we note that our chi-squared values stillsuggest some level of tension between the data and mod-els. Under frequentist metrics one could still considerrejecting all the models presented here, but this argu-ment assumes that both the data and models perfectly Absorption data for H − , AlO, CrH and VO are taken from John(1988), Patrascu et al. (2015), Bernath (2020), and McKemmishet al. (2016), respectively. Note that NEMESIS does not currently support CrH, whileATMO does not support AlO. capture all noise sources and atmospheric physics. Tac-tics such as error bar inflation (as done in studies suchas Line et al. 2015; Zhang et al. 2018; Col´on et al. 2020)and incorporating models of additional complexity (e.g.see discussion in Gibson 2014; Parviainen 2018) couldbe used to further reduce these chi-squared values, butcould obfuscate our physical interpretation for HAT-P-41b from these data and associated inter-model com-parisons. It is important to note that both H O andUV-optical opacity sources are independently requiredto explain the observations, irrespective of chi-squaredtests, as established by our Bayesian model comparisons.With reference to our best-fitting ‘minimal’ model, weconducted a series of Bayesian model comparisons withPOSEIDON to compute detection significances for eachinferred UV-optical chemical species. The jitter reduc-tion yields moderate evidence for H − (2 . σ ), weak ev-idence for CrH (2 . σ ) and AlO (2 . σ ), and a tentativehint of Na (2 . σ ). The marginalization reduction yieldsmoderate evidence for Na (2 . σ ), weak evidence for H − (2 . σ ) and CrH (2 . σ ), and tentative hints of AlO (1 . σ )and VO (1 . σ ). The specific spectral features givingrise to these inferences are shown in Appendix C (seeFigure 7). The differing significances for Na and VOhighlight the sensitivity of some atmospheric inferencesto specific data reductions. However, our most rigor-ous conclusion holds for both reductions: at least one ofH − , CrH, AlO, or VO is required at > σ (jitter: 5 . σ ;marginalization: 5 . σ ) to explain HAT-P-41b’s trans-mission spectrum.The retrieved abundances for each inferred chemicalspecies are shown in Figure 5. All three codes provideprecise H O abundances. Across both data reductionsand all three retrieval codes, the retrieved H O abun-dance spans log(H O) ≈ -3.4 to -1.6 (with a mean preci-sion of 0.5 dex). The H O abundances from ATMO are (cid:38) σ - see Table 1. These results illustratethat the chosen data reduction technique at visible wave-lengths can alter retrieved H O abundances by (cid:46) − abundance,consistent across all three retrieval codes and both datareductions, with a mean value of log(H − ) = − . ± . O and The abundances for other included chemical species are con-strained only by upper bounds - see the supplementary material. Lewis et al. log ( X H O ) P r o b a b ili t y d e n s i t y ( n o r m a li z e d ) H O Jitter
POSEIDONNEMESISATMO
12 11 10 9 8 7 6 log ( X H − )
12 11 10 9 8 7 60.00.20.40.60.81.012 11 10 9 8 7 60.00.20.40.60.81.0 H −
10 9 8 7 6 5 4 3 log ( X AlO )
10 9 8 7 6 5 4 30.00.20.40.60.81.010 9 8 7 6 5 4 30.00.20.40.60.81.0
AlO
10 9 8 7 6 5 4 3 2 1 log ( X CrH )
10 9 8 7 6 5 4 3 2 10.00.20.40.60.81.010 9 8 7 6 5 4 3 2 10.00.20.40.60.81.0
CrH
10 9 8 7 6 5 4 3 2 log ( X VO )
10 9 8 7 6 5 4 3 20.00.20.40.60.81.010 9 8 7 6 5 4 3 20.00.20.40.60.81.0 VO
10 9 8 7 6 5 4 3 2 1 log ( X Na )
10 9 8 7 6 5 4 3 2 10.00.20.40.60.81.010 9 8 7 6 5 4 3 2 10.00.20.40.60.81.0 Na log ( X H O ) P r o b a b ili t y d e n s i t y ( n o r m a li z e d ) H O Marginalization
POSEIDONNEMESISATMO
12 11 10 9 8 7 6 log ( X H − )
12 11 10 9 8 7 60.00.20.40.60.81.012 11 10 9 8 7 60.00.20.40.60.81.0 H −
10 9 8 7 6 5 4 3 log ( X AlO )
10 9 8 7 6 5 4 30.00.20.40.60.81.010 9 8 7 6 5 4 30.00.20.40.60.81.0
AlO
10 9 8 7 6 5 4 3 2 1 log ( X CrH )
10 9 8 7 6 5 4 3 2 10.00.20.40.60.81.010 9 8 7 6 5 4 3 2 10.00.20.40.60.81.0
CrH
10 9 8 7 6 5 4 3 2 log ( X VO )
10 9 8 7 6 5 4 3 20.00.20.40.60.81.010 9 8 7 6 5 4 3 20.00.20.40.60.81.0 VO
10 9 8 7 6 5 4 3 2 1 log ( X Na )
10 9 8 7 6 5 4 3 2 10.00.20.40.60.81.010 9 8 7 6 5 4 3 2 10.00.20.40.60.81.0 Na Figure 5.
Retrieved atmospheric composition of HAT-P-41b. The histograms show marginalized posterior probability dis-tributions for the volume mixing ratios of each chemical species inferred by at least one retrieval code. The posteriors fromPOSEIDON (purple), NEMESIS (green), and ATMO (blue) are compared. Where a retrieval code does not include a givenspecies, no histogram is shown. The error bars give the median retrieved abundances and ± σ confidence levels. The retrievalsagree on a 5 σ detection of H O and evidence of at least one visible to near-UV absorber at 3 σ confidence. The retrievedabundances from each code, and for each data reduction, are broadly consistent within their respective 1 σ confidence regions. H − abundances are robust to both modelling choices anddata reduction techniques. We note that these abun-dances are with respect to the original ‘full’ retrievals(see Table 1), ensuring that model uncertainty (overlap-ping absorption features, cloud-chemistry degeneracies,etc.) are automatically encoded in the quoted values.However, biases may still arise from neglected modelcomplexity. In particular, retrieved H O and H − abun-dances from 1D retrieval methods can be biased by up to1 dex if compositional gradients arise between the morn-ing and evening terminators (MacDonald et al. 2020).Such a bias would cause a slight overestimation in ourretrieved H O abundances, and underestimate in ourH − abundances (see MacDonald et al. 2020, Figure 3).The precise H O abundances we obtain can be con-verted into estimates of the atmospheric metallicity.By ‘metallicity’, we refer to the atmospheric O/H ra-tio relative to that of its star ([Fe/H] = 0.21, Stassunet al. (2017)). The retrieved molecular abundances aremapped into O/H ratios as in MacDonald & Madhusud-han (2019). We note that our metallicities should onlybe considered accurate to a factor of two , given that the At HAT-P-41b’s temperature, approximately half of the atmo-spheric O is expected to reside in CO (Madhusudhan 2012). current data are insensitive to CO. The metallicities de-rived by POSEIDON and NEMESIS are consistent withthe stellar metallicity of HAT-P-41 for the marginal-ization reduction: 1 . +6 . − . × stellar and 1 . +3 . − . × stellar, respectively. The ATMO retrievals find ∼ × higher metallicities (due to the aforementioned higherH O abundances): 4 . +10 . − . × stellar for the marginal-ization reduction. Comparatively, the jitter reductionfavours slightly super-stellar metallicities: 3 . +5 . − . × stellar (POSEIDON), 1 . +4 . − . × stellar (NEMESIS),and 9 . +8 . − . × stellar (ATMO). Nevertheless, all de-rived metallicities are consistent with the stellar valueto 2 σ . Overall, the metallicities derived by the differentretrieval codes are consistent with each other. We con-clude that the metallicity of HAT-P-41b’s atmosphere isconsistent with being stellar, or slightly super-stellar, inline with expected mass-metallicity trends.3.5. The Case for Disequilibrium Chemistry
Our retrieved abundances for visible to near-UV ab-sorbers require an atmosphere in chemical disequilib-rium. Here we consider disequilibrium mechanisms thatmight enhance the atmospheric abundance of AlO, CrH,and H − in HAT-P-41b’s atmosphere, the species com-mon to both the jitter and marginalization data reduc-tion analyses (see discussion in section 3.4). At the AT-P-41b’s Atmosphere Revealed ( ∼ ±
200 K),the equilibrium abundances of CrH and AlO for a solarmetallicity atmosphere are respectively ∼ O abundance. Even at depth inHAT-P-41b’s atmosphere ( P ∼ T ∼ − , which had the highest statistical sig-nificance (2.6–2.9 σ ) of the UV-optical absorbing specieswe considered in our retrievals. Under equilibrium ther-mochemistry, H − is not expected to exist in appreciable( ≥ − mole fraction) quantities for T (cid:46) − opac-ity by conducting an additional ATMO retrieval withchemical equilibrium enforced. This retrieval raised theatmospheric temperature to (cid:38) T eq and predictions from the GCM (Figure 3), to create H − opacity with an abundance consistent with that foundby our free retrievals (see the supplementary material).We explored the potential for photochemistry to en-hance the abundance of H − in a hot Jupiter with T (cid:46) ∼ − via radiative electron attach-ment (H + e − → H − + γ ), collisional electron attach-ment (H + e − + M → H − + M), and dissociative elec-tron attachment (H + e − → H + H − ). For conditionsrelevant to HAT-P-41b, the latter process likely domi-nates, due to the prevalence of H and the moderatelylarge rate coefficient of order of 10 − cm s − (Janev Our retrieved temperature profiles are available in the supple-mentary material. et al. 2003). The destruction of H − can occur throughcollisional detachment (e.g. H − + H → H + e − orH − + H O → OH − + H ) with rate coefficients for suchreactions on the order of a few × − cm s − (Bruhnset al. 2010; Martinez et al. 2010). From the 0.1 mbaratmospheric density from the global-average GCM re-sults described above and the mixing ratios of e − , H,and H from the model of Lavvas et al. (2014), we canestimate number densities for e − , H, and H near the ∼ ∼ ,2 × , and 6 × cm − respectively. Assum-ing that in steady state the H − production and lossrates balance each other, and considering productionsolely by H dissociative electron attachment and de-struction solely by collisional detachment with atomicH, we estimate a number density for H − on the order of10 cm − . This value corresponds to a H − mixing ratioof 2 × − , consistent with our retrieved value for theabundance of H − and roughly six orders of magnitudelarger than expectations from equilibrium chemistry fora planet like HAT-P-41b. This order-of-magnitude esti-mate highlights that significant enhancement of H − dueto photochemical processes is likely present in HAT-P-41b and many other exoplanet atmospheres, thus shap-ing their UV-optical spectra. The production of H − willbe particularly enhanced for hot planets that receive ahigh extreme UV flux from their host stars and haveNa in the gas phase, which increases electron produc-tion (Lavvas et al. 2014). Such conditions are expectedfor HAT-P-41b (this work; Hartman et al. 2012; Linskyet al. 2014) and other hot Jupiters orbiting F-stars. DISCUSSION AND CONCLUSIONSOur analysis of the transmission spectrum of thehot Jupiter HAT-P-41b from 0.2-5.0 µ m representsone of the most comprehensive explorations of an ex-oplanet atmosphere to date. In particular, our analy-sis includes new high-precision information at UV/NUVwavelengths provided by Hubble’s WFC3 UVIS G280grism. We leveraged multiple reductions of the WFC3UVIS G280 data and multiple spectral analysis tools toobtain a more complete and robust picture of the phys-ical and chemical processes at work in HAT-P-41b’s at-mosphere. We find that: • The presence of a significant cloud deck, composedof aluminum bearing species, provides a plausibleexplanation for the UV, optical and 4.5 µ m por-tions of HAT-P-41b’s transmission spectrum, butis discrepant with observations in the NIR (1.1-1.7 µ m) and at 3.6 µ m. This highlights the needfor broad wavelength coverage from the UV to IR2 Lewis et al. to constrain atmospheric properties, in particularaerosols, in exoplanet atmospheres. • Our use of multiple reductions of the
Hubble
WFC3 UVIS G280 observations and multiple in-terpretation methods shows that in most areas weobtain a consistent picture for HAT-P-41b’s at-mosphere, which proves the robustness of our re-sults. We also highlight that potentially spuriousconclusions can be drawn when relying on singledata reduction and interpretation techniques. Inparticular, the presence of VO in HAT-P-41b’s at-mosphere is more strongly preferred for data wherethe marginalization approach is used to correct forsystematics in the WFC3 UVIS G280 data and inretrievals performed with the ATMO model. • We find evidence for the presence of the hydro-gen anion, H − , in HAT-P-41b and provide pre-cise constraints for its abundance: log(H − ) = − . ± .
62. This represents an abundance forH − several orders of magnitude larger than whatwould be expected via equilibrium chemistry forHAT-P-41b given its equilibrium temperature of ∼ ∼ − .In the future, the James Webb Space Telescope(JWST) will provide the exoplanet community withhigh-precision spectroscopic observations of exoplanetatmospheres spanning 0.6-14 µ m (Beichman et al. 2014).The complexities encountered in the reduction, analy- sis, and interpretation of the HAT-P-41b observationspresented here will also be encountered with JWST ob-servations; as such this study serves both to highlightthe challenges and provides a needed test-bed for fu-ture transiting exoplanet observations. Additionally ashighlighted by this work, observations in the UV/NUVcritically complement atmospheric transmission obser-vations in the optical and infrared, probing the presenceof a range of chemical species and giving insights intoprocesses occurring in upper atmospheres such as strato-spheric heating and photochemistry.ACKNOWLEDGMENTSThis research is based on observations made with theNASA/ESA Hubble Space Telescope obtained from theSpace Telescope Science Institute, which is operated bythe Association of Universities for Research in Astron-omy, Inc., under NASA contract NAS 5–26555. Theseobservations are associated with programs GO-15288and GO-14767. This work is based in part on obser-vations made with the Spitzer Space Telescope, which isoperated by the Jet Propulsion Laboratory, CaliforniaInstitute of Technology under a contract with NASA.These observations are associated with program 13044.We thank Patrick Irwin for the use of NEMESIS, andJake Taylor for assistance with the inclusion of H − opac-ity within the NEMESIS forward model. Software:
IDL Astronomy user’s library (Landsman1995), NumPy (Oliphant 2006–), SciPy (Virtanen et al.2019), MatPlotLib (Caswell et al. 2019), AstroPy (As-tropy Collaboration et al. 2018), Photutils (Bradley et al.2019), MultiNest (Feroz & Hobson 2008; Feroz et al.2009, 2013), PyMultiNest (Buchner et al. 2014), dynesty(Speagle 2020), ATMO (Amundsen et al. 2014; Trem-blin et al. 2015, 2016; Wakeford et al. 2017; Goyal et al.2018), NEMESIS (Irwin et al. 2008; Barstow et al. 2017),POSEIDON (MacDonald & Madhusudhan 2017).
Facilities:
HST(WFC3), Spitzer(IRAC)APPENDIX A. WFC3-IR G141 TRANSMISSION COMPARISONFigure 6 shows the comparison between the reduction presented here and by Tsiaras et al. (2018)[T18] for
Hubble’s
WFC3 G141 grism observations of HAT-P-41b. To more accurately compare the shape of the transmission spectrathe T18 spectrum was shifted down in altitude by 0.011%. This offset is likely caused by differences in the systemparameters used in the light curve fitting stage of the analysis between T18 and this study which leverages updatedparameters for the HAT-P-41b system published in Wakeford et al. (2020). Using common system parameters acrossall datasets considered in this study ensures a consistent analysis across the entire transmission spectrum and therefore
AT-P-41b’s Atmosphere Revealed χ ν = 1.38 with 11 degrees of freedom(DOF), the published spectra from T18 without an offset has χ ν = 2.70 with 25 DOF, and when shifted in altitude tothe model χ ν = 1.53 with 24 DOF. This demonstrates the similarity in the shape of the reduced transmission spectraand highlights the effect of offsets between different analysis techniques. T r a n s i t D e p t h ( % ) POSEIDON model full resolutionPOSEIDON model R=100Tsiaras et al. (2018)Tsiaras et al. (2018) [-0.011% fit offset]This work
Figure 6.
Measured near-IR transmission spectrum of HAT-P-41b from HST/WFC3-IR G141 from this work (black) andthe best fitting POSEIDON model using the jitter data (corresponding to the 6 parameter ‘minimal’ model) (grey & red).Also plotted are the published spectra by Tsiaras et al. (2018)[T18] (dark blue) and T18 shifted in altitude by -0.011% to themodel by minimizing the chi-squared (light blue). This demonstrates the differences and similarities to the previously publishedspectrum and the reduction presented in this paper for use in the full UV-IR transmission spectrum.B.
FULL ATMOSPHERIC RETRIEVAL RESULTS AND MODEL COMPARISONTable 1 summarises the retrieved values for the 8 common parameters our atmospheric retrievals found necessary toexplain HAT-P-41b’s observed transmission spectrum. All three retrieval codes reach good agreement, despite theirvarying complexity, with all obtaining a precise H − abundance constraint and inferring a stellar (or slightly super-stellar) O/H ratio. The Bayesian evidence and reduced chi-squared statistics both prefer a ‘minimal’ model (withonly the 8 parameters in Table 1), which nevertheless yields consistent parameter constraints with the more complexretrieval models. C. SPECTRAL EVIDENCE OF UV-VISIBLE ABSORBERS IN HAT-P-41B’S ATMOSPHEREFigure 7 shows a spectral decomposition of our best-fitting model transmission spectra. This illustrates whichfeatures in the observations are attributed to specific chemical species. Note that these opacity contributions are relative to the H − continuum, which serves to ‘boost’ the transit depth contributions of other absorbing species. Both4 Lewis et al. data reductions produce similar best-fitting models, with the only difference being a slight preference to include VOfor the systematic marginalization reduction.
Table 1.
Atmospheric Retrieval Analysis Summary
Data Reduction Jitter Marginalization
Retrieval POSEIDON NEMESIS ATMO ‘Minimal’ POSEIDON NEMESIS ATMO ‘Minimal’
Parameters T (K) 1148 +194 − +193 − +196 − +248 − +209 − +194 − +272 − +274 − R p , ref ( R J ) 1 . +0 . − . . +0 . − . . +0 . − . . +0 . − . . +0 . − . . +0 . − . . +0 . − . . +0 . − . log( X H O ) − . +0 . − . − . +0 . − . − . +0 . − . − . +0 . − . − . +0 . − . − . +0 . − . − . +0 . − . − . +0 . − . log( X H − ) − . +0 . − . − . +0 . − . − . +0 . − . − . +0 . − . − . +0 . − . − . +0 . − . − . +0 . − . − . +0 . − . log( X AlO ) − . +0 . − . − . +0 . − . — − . +0 . − . − . +0 . − . − . +1 . − . — − . +0 . − . log( X CrH ) − . +0 . − . — − . +2 . − . − . +0 . − . − . +1 . − . — − . +1 . − . − . +1 . − . log( X VO ) − . +1 . − . − . +1 . − . − . +1 . − . − . +1 . − . − . +1 . − . − . +0 . − . − . +1 . − . − . +1 . − . log( X Na ) − . +1 . − . − . +1 . − . − . +2 . − . − . +1 . − . − . +0 . − . − . +0 . − . − . +0 . − . − . +0 . − . Derived Properties
O/H ( × stellar) 3 . +5 . − . . +4 . − . . +8 . − . . +4 . − . . +6 . − . . +3 . − . . +10 . − . . +0 . − . Statistics ln(Evidence) 473 . . . . . . . . χ ν, min .
55 1 .
90 1 .
73 1 .
50 3 .
02 2 .
37 1 .
97 1 . N param
37 17 12 8 37 17 12 8d.o.f. 32 52 57 61 32 52 57 61
Note —All retrievals here have ‘free composition’, without the assumption of chemical equilibrium. The ‘minimal’ model containsonly the 8 free parameters found necessary to fit either data reduction (i.e. those listed in the table). Only parameters withbounded constraints (i.e. both lower and upper bounds) are included - see the online supplementary material for full posteriordistributions. R p , ref is defined at P = 10 bar for NEMESIS and POSEIDON, and 1 mbar for ATMO. The NEMESIS retrievalsuse a different evidence normalizing factor to ATMO and POSEIDON. The stellar O/H is assumed equal to HAT-P-41’s stellar[Fe/H] (0.21, Stassun et al. (2017)). Equilibrium retrievals with similar complexity to the minimal model are omitted, due totheir relatively poor fits (e.g. an ATMO equilibrium retrieval for the marginalization reduction obtained χ ν, min = 2 .
38 for 62degrees of freedom).
REFERENCES
Amundsen, D. S., Baraffe, I., Tremblin, P., et al. 2014,A&A, 564, A59, doi: 10.1051/0004-6361/201323169Astropy Collaboration, Price-Whelan, A. M., Sip˝ocz, B. M.,et al. 2018, AJ, 156, 123, doi: 10.3847/1538-3881/aabc4fBarstow, J. K., Aigrain, S., Irwin, P. G. J., & Sing, D. K.2017, ApJ, 834, 50, doi: 10.3847/1538-4357/834/1/50Beichman, C., Benneke, B., Knutson, H., et al. 2014, PASP,126, 1134, doi: 10.1086/679566Bernath, P. F. 2020, J. Quant. Spectrosc. Radiat. Transf.,240, 106687, doi: 10.1016/j.jqsrt.2019.106687 Bradley, L., Sip˝ocz, B., Robitaille, T., et al. 2019,astropy/photutils: v0.7.2, v0.7.2, Zenodo,doi: 10.5281/zenodo.3568287Brosch, N., Davies, J., & Festou, M. 2006, Astrophysics andSpace Science, 303, 103, doi: 10.1007/s10509-005-9027-2Bruhns, H., Kreckel, H., Miller, K. A., Urbain, X., & Savin,D. W. 2010, Phys. Rev. A, 82, 042708,doi: 10.1103/PhysRevA.82.042708Buchner, J., Georgakakis, A., Nandra, K., et al. 2014,Astronomy and Astrophysics, 564, A125,doi: 10.1051/0004-6361/201322971
AT-P-41b’s Atmosphere Revealed Wavelength ( m) T r a n s i t D e p t h ( R p / R ∗ ) × WFC3 G280 WFC3 G141
Opacity Contributions H AlO Na CrH CrHH − H O Jitter
ModelH OH − AlOCrH VONaH + HeBinned ModelObservations -6-5-4-3-2-101234 S c a l e H e i g h t s Wavelength ( m) T r a n s i t D e p t h ( R p / R ∗ ) × WFC3 G280 WFC3 G141
Opacity Contributions H AlO VO Na VO CrH CrHH − H O Marginalization
ModelH OH − AlOCrH VONaH + HeBinned ModelObservations -6-5-4-3-2-101234 S c a l e H e i g h t s Figure 7.
Opacity contributions to the best-fitting model transmission spectra of HAT-P-41b. Each panel shows the maximumlikelihood spectrum (green shading) from the ‘minimal’ POSEIDON retrievals for each WFC3 G280 data reduction (top: jitterdecorrelation; bottom: systematic marginalization). The UV-visible spectrum is shaped by H − bound-free opacity (black curve)from ∼ µ m. The opacity contributions of other retrieved species (H O, Na, CrH, AlO, and VO) are depicted relativeto the H − continuum (colored curves). H Rayleigh scattering contributes opacity for wavelengths (cid:46) µ m. The best-fittingmodel, binned to the resolution of each set of observations, is overlaid for comparison (gold diamonds). Lewis et al.
AT-P-41b’s Atmosphere Revealed17