Metastable Helium Reveals an Extended Atmosphere for the Gas Giant HAT-P-18b
Kimberly Paragas, Shreyas Vissapragada, Heather A. Knutson, Antonija Oklop?i?, Yayaati Chachan, Michael Greklek-McKeon, Fei Dai, Samaporn Tinyanont, Gautam Vasisht
DDraft version February 18, 2021
Typeset using L A TEX twocolumn style in AASTeX63
Metastable Helium Reveals an Extended Atmosphere for the Gas Giant HAT-P-18b
Kimberly Paragas, Shreyas Vissapragada, Heather A. Knutson, Antonija Oklopˇci´c, Yayaati Chachan, Michael Greklek-McKeon, Fei Dai, Samaporn Tinyanont, and Gautam Vasisht Astronomy Department, Wesleyan University, 96 Foss Hill, Middletown, CT 06459, USA Division of Geological and Planetary Sciences, California Institute of Technology, 1200 East California Blvd, Pasadena, CA 91125, USA Anton Pannekoek Institute of Astronomy, University of Amsterdam, Science Park 904, 1098 XH Amsterdam, Netherlands Department of Astronomy and Astrophysics, University of California, Santa Cruz, CA 95064, USA Jet Propulsion Laboratory, California Institute of Technology, 4800 Oak Grove Dr, Pasadena, CA 91109, USA
ABSTRACTThe metastable helium line at 1083 nm can be used to probe the extended upper atmospheres ofclose-in exoplanets and thus provide insight into their atmospheric mass loss, which is likely to besignificant in sculpting their population. We used an ultranarrowband filter centered on this lineto observe two transits of the low-density gas giant HAT-P-18b, using the 200” Hale Telescope atPalomar Observatory, and report the detection of its extended upper atmosphere. We constrain theexcess absorption to be 0 . ± .
12% in our 0.635 nm bandpass, exceeding the transit depth from the
Transiting Exoplanet Survey Satellite ( TESS ) by 3 . σ . If we fit this signal with a 1D Parker windmodel, we find that it corresponds to an atmospheric mass loss rate between 8 . +2 . − . × − M J /Gyr and2 . +0 . − . × − M J /Gyr for thermosphere temperatures ranging from 4000 K to 13000 K, respectively.With a J magnitude of 10.8, this is the faintest system for which such a measurement has been madeto date, demonstrating the effectiveness of this approach for surveying mass loss on a diverse sampleof close-in gas giant planets. Keywords: techniques: photometric – planets and satellites: atmospheres – planets and satellites:individual (HAT-P-18b) INTRODUCTIONClose-in exoplanets are exposed to high-energy radia-tion from their host stars, which can lead to atmosphericmass loss. This atmospheric escape appears to shape theobserved short-period exoplanet population (e.g. Lopez& Fortney 2013; Owen & Wu 2013; Fulton et al. 2017),but there are relatively few published measurements ofpresent-day mass loss rates for close-in planets. Prior to2018, most studies of atmospheric escape used the hy-drogen Lyman- α line at UV wavelengths, the H α line atoptical wavelengths, and metal lines at UV and opticalwavelengths (e.g., Vidal-Madjar et al. 2003, 2004; Jensenet al. 2012; Yan & Henning 2018; Cauley et al. 2019). Inrecent years, new theoretical and observational work onthe helium (He) 1083 nm line have shown that it can alsobe used for atmospheric mass loss studies. For planetswith a sufficient population of metastable helium atoms Corresponding author: Kimberly [email protected] in their (potentially escaping) upper atmospheres, the1083 nm line is optically thick at low pressures (highaltitudes), increasing their measured transit depths inthis line by a factor of a few (Oklopˇci´c & Hirata 2018).K-type stars are favorable targets for metastable heliumobservations, as they emit relatively low amounts of mid-UV flux (which depopulates the metastable state), whileemitting relatively high levels of EUV flux (which pop-ulates the metastable state via ground-state ionizationand subsequent recombination; Oklopˇci´c 2019).Excess He I absorption was first detected in the atmo-sphere of the sub-Saturn WASP-107b using the
HubbleSpace Telescope ( HST ; Spake et al. 2018). Since then,this line has been used to detect extended atmospheresin six other planets (5 gas giants and 1 sub-Neptune,with masses ranging from 0 . M J to 1 . M J ) usingboth space and ground based facilities (Allart et al. 2018;Mansfield et al. 2018; Nortmann et al. 2018; Salz et al.2018; Alonso-Floriano et al. 2019; Ninan et al. 2020;Palle et al. 2020). Understanding how the metastablehelium signal scales with stellar EUV flux, planetary a r X i v : . [ a s t r o - ph . E P ] F e b Paragas et al. gravity, and stellocentric distance on a sample of large,well-characterized planets is a prerequisite for He I ob-servations of smaller planets, where measurements andinterpretation are inherently more difficult (Kasper et al.2020). Additionally, detections of outflowing gas giantatmospheres in the He I line provide useful constraintson crucial radiative and collisional processes in theoret-ical atmospheric mass loss models (e.g. Salz et al. 2016;Oklopˇci´c & Hirata 2018).Using the He 1083 nm line, we study the extended at-mosphere of HAT-P-18b (Hartman et al. 2011), which isa Jupiter-sized (0 . ± . R J ), Saturn-mass (0 . ± . M J ), T eq = 841 ±
15 K planet (Esposito et al.2014) orbiting a K2-type star with J = 10 . Spitzer
Space Telescope, and the resulting bright-ness temperature for the planet suggests efficient day-night circulation and/or a nonzero albedo (Wallack etal. 2019).In this work, we characterize the atmospheric massloss of HAT-P-18b for the first time. We observed twotransits of HAT-P-18b with the Hale 200” Telescope atPalomar Observatory using an ultranarrowband filtercentered on the helium 1083 nm line (Vissapragada etal. 2020a) with the Wide-field InfraRed Camera (WIRC;Wilson et al. 2003). Additionally, we use TESS datafrom Sectors 25 and 26 to improve the ephemeris forthis planet and determine its broadband optical transitdepth as a comparison for our helium measurement. InSection 2 we describe the WIRC and TESS observations,and in Section 3 we jointly model the light curves fromboth instruments. We describe our results in Section 4,and offer some concluding thoughts in Section 5. OBSERVATIONS2.1.
WIRC Observations
We observed transits of HAT-P-18b on UT June 62020 (hereafter night 1) and July 8 2020 (night 2). Weused an ultranarrowband filter centered on the helium1083.3 nm line with a FWHM of 0.635 nm (Vissapra-gada et al. 2020a). For these observations we typicallyuse a custom beam-shaping diffuser, which produces astable 3 (cid:48)(cid:48) diameter top hat point spread function (Ste- fansson et al. 2017), to mitigate time-correlated system-atics (Vissapragada et al. 2020b). However, on night 1the weather conditions were poor and we elected to defo-cus the telescope to 1 . (cid:48)(cid:48) Q (3 /
2) OH emission line at elium for HAT-P-18b M e d i a n N o r m a li ze d F l ux D e t r e nd e d F l ux R e s i du a l s Binsize [points/bin]10 -3 -2 R M S A b s o r p ti on P r oxy (a) M e d i a n N o r m a li ze d F l ux D e t r e nd e d F l ux R e s i du a l s Binsize [points/bin]10 -3 -3 -3 -3 -3 R M S A b s o r p ti on P r oxy (b) Figure 1. (a) Results for WIRC night 1 and night 2 shown in (a) and (b), respectively. The top row shows the median-normalized raw light curves of the target (in black) and its comparison stars (in grey) as well as the water absorption proxydescribed in Section 2.1 (dotted blue line) throughout each respective night. The middle row shows the helium light curve withunbinned data in grey and binned data to a 15 minute cadence in black, with the best-fit joint helium model in red and theTESS model in blue, and the residuals of each fit. The bottom row shows the Allan deviation plot for each dataset. >
80 km; Bernath & Colin 2009) can be absorbed byH O while passing through the lower atmosphere. Wecan therefore track the telluric water variation over the
Paragas et al. D e t r e nd e d F l ux R e s i du a l s Figure 2.
Combined transit light curves and residuals for WIRC (left) and TESS (right), with unbinned data in grey andbinned data to a 15 minute cadence in black. The best fit models for WIRC (red) and TESS (blue) are overplotted with the 1 σ confidence interval denoted by the shaded region. night by dividing the time-varying flux in the water-contaminated OH emission line (as measured by ourscaling factors) by that of the uncontaminated R (3 / R (1 /
2) OH lines at 1075 nm and 1078 nm. Ifthe water variation is significant enough to impact ourphotometry, we can utilize this absorption proxy as adecorrelation parameter in our transit fits.We performed aperture photometry on the target starand six comparison stars (the same ones on each night)using the package photutils (Bradley et al. 2016). Wetested different aperture sizes in one pixel steps from3 to 13 pixels in radius. We removed 3 σ outliers fromthe data using a moving median filter. This process isdetailed in Vissapragada et al. (2020a). Our optimalaperture sizes for our night 1 and night 2 observationswere 7 pixels (1 . (cid:48)(cid:48)
75) and 11 pixels (2 . (cid:48)(cid:48)
75) in radius, re-spectively, with the difference arising from our use of aslight defocus on the first night and a diffuser on thesecond night. The raw light curves of the target and thecomparison stars are shown in Figure 1 for both nights.2.2.
TESS Observations
We used the 2-minute cadence TESS observations ofHAT-P-18b obtained during Sectors 25 and 26. TESSobserved the target for 51.5 days starting on May 142020 and June 9 2020 for Sectors 25 and 26, respectively,covering 8 transits in total. We downloaded the Pre-search Data Conditioning Simple Aperture Photometry(PDCSAP) light curve from the Mikulski Archive forSpace Telescopes (MAST) using the lightkurve pack-age (Lightkurve Collaboration et al. 2018). With the transits masked, we removed low-frequency variabilityfrom the data using the Savitzky-Golay filter from scipy (Jones et al. 2001) and rejected 5 σ outliers using a mov-ing median filter. However, we noticed that even afterthe filter was applied, there were still strong uncorrectedsystematics that biased the transit depths of the first twotransits, so we omitted them from our combined fit. Al-though these transits may be recoverable with differentdetrending methods, our constraint on the TESS transitdepth from the remaining six transits were sufficientlyprecise for comparison to the WIRC light curves (i.e.,the uncertainty in the comparison is dominated by theuncertainty on the WIRC transit depth). LIGHT CURVE MODELINGWe simultaneously fit both nights of WIRC data alongwith the corrected TESS photometry using exoplanet (Foreman-Mackey et al. 2020). For each WIRC lightcurve, we fit an instrumental noise model consisting of alinear baseline along with a linear combination of com-parison star light curves, with the weights of the com-parison stars left as free parameters in the fit. Thisis an update from our previous modeling methodology(Vissapragada et al. 2020a), where we used an ordinaryleast-squares method to quickly determine comparisonstar weights at each likelihood evaluation, and it is en-abled by the rapid No U-Turn Sampler (NUTS) samplerthat exoplanet makes available for high-dimensionallight-curve fitting. We also tried including two addi-tional parameters in our instrumental noise model foreach night: the water absorption proxy (as described in elium for HAT-P-18b <
10. Our final systematics modelcontained 15 parameters: two parameters for each of thelinear baselines, four comparison stars for each dataset,the distance from median centroid for each dataset, andthe absorption proxy for the first night.We fit a transit model simultaneously with the sys-tematics. We have three fit parameters that are com-mon to all datasets: the predicted mid-transit time T ,the period P , and the impact parameter b . Initially,we allowed each night of WIRC data to have its owntransit depth in the joint fit. The two transit depths(2 . +0 . − . % for night 1 and 2 . ± .
14% for night 2) werewithin 1 σ of each other, indicating that the magnitude ofhelium absorption appears consistent between these twoepochs. We therefore fit a single transit depth for bothhelium light curves. We fit for the limb darkening coef-ficients [ u , u ] (He) and [ u , u ] (TESS) and transit depths( R p /R (cid:63) ) and ( R p /R (cid:63) ) in each bandpass, bothsampled uniformly (see Table 1). For each WIRC lightcurve, we fit for a jitter parameter describing the excessnoise in addition to the photon noise log( σ extra ). ForTESS, we noticed that the error bars that came withthe PDCSAP fluxes were not an accurate representa-tion of the photon noise and in fact overestimated theobserved scatter in the data. We therefore include ascaling factor k for the TESS error bars.We use the NUTS in PyMC3 (Salvatier et al. 2016) tosample the posterior distributions for our model param-eters. We ran four chains, tuning each for 1500 steps(the “burn-in” period) and then taking 1000 draws ineach chain, achieving good convergence with a Gelman-Rubin (Gelman & Rubin 1992) statistic of < .
006 forall parameters. The priors and posteriors for the phys-ical parameters in our model are given in Table 1 forthe joint fit and the detrended light curve, residuals,and Allan deviation plot for each night of WIRC dataare displayed in Figure 1. The final combined helium and TESS light curves are displayed in Figure 2, andthe posterior distributions for the model parameters arevisualized in Figure 4.
Figure 3.
Atmospheric mass loss model for HAT-P-18b.Each point is a different mass loss model corresponding tospecific T and ˙ M values, and the shading indicates the com-patibility between the model and our observed excess absorp-tion (with the lighter regions indicating the most concordantmodels). 4. RESULTS & DISCUSSIONWe measure a transit depth of 2 . +0 . − . % in the he-lium line. The corner plot for our fit parameters is shownin Figure 4. Our measurement can be compared to theTESS transit depth measurement of 1 . +0 . − . %. Ourmeasured transit depth in the helium bandpass exceedsthat in the TESS bandpass by 0 . ± .
12% (3 . σ ). TheTESS bandpass is between 600 nm to 1000 nm, makingit reasonable to use as a comparison for our measure-ment. Previous studies on our target have shown varia-tion within this range to be limited to variations on theorder of the scale height (Kirk et al. 2017). The differ-ence between the two transit depths exceeds, by an orderof magnitude, the expected change in transit depth fora one (lower-atmospheric) scale-height change in planetradius (0.03%) for this target. Thus, the observed ex-cess absorption cannot be explained by broad absorptionfeatures – for instance, by water – in the lower atmo-sphere. The helium line is near an opacity minimum ofwater anyways so this explanation is disfavored a pri-ori. We conclude that the observed absorption indeedarises from metastable helium in HAT-P-18b’s extendedatmosphere.We use the model described in Oklopˇci´c & Hirata(2018) to convert our measured excess absorption intoa joint constraint on HAT-P-18b’s mass loss rate ˙ M Paragas et al.
Table 1.
Priors and posteriors for joint fit to Palomar/WIRC and TESS dataParameter Prior Posterior Units( R p /R (cid:63) ) (He) U (1 ,
25) 2 . ± .
12 %( R p /R (cid:63) ) (TESS) U (1 ,
25) 1 . +0 . − . % P N (5 . , . . ± . T U (2038 . , .
0) 2038 . ± . TDB b N (0 . , . . +0 . − . – u (He) Kipping (2013) 0 . +0 . − . – u (He) Kipping (2013) 0 . ± .
39 – u (TESS) Kipping (2013) 0 . ± .
16 – u (TESS) Kipping (2013) 0 . +0 . − . – log ( σ extra ) (night 1) U ( − , − − . +0 . − . – log ( σ extra ) (night 2) U ( − , − − . +0 . − . – k (TESS) U (0 . , .
5) 0 . +0 . − . –absorption proxy (night 1) N (0 . , . − . +0 . − . – Note —BTJD
TDB = BJD - 2457000. Note that we omitted the stellar parameters and all ofthe detrending weights except for the absorption proxy for night 1. and upper atmospheric temperature T . This model cal-culates the velocity and density profiles of a 90%/10%H/He 1D Parker wind as a function of ˙ M and T , andthen calculates the level populations for helium given aUV stellar spectrum. We use the MUSCLES UV spec-trum of (cid:15) Eridani (France et al. 2016; Loyd et al. 2016;Youngblood et al. 2016), which is another K2 type star,as a stand-in for the unknown UV spectrum of HAT-P-18. Accounting for the stellar radius and semi-majoraxis of HAT-P-18b, the EUV irradiance of the planetwas 8 W/m integrated between 5.5 ˚A and 911 ˚A. Theresults are shown in Figure 3. HAT-P-18b’s mass lossrate is likely between 8 . +2 . − . × − and 2 . +0 . − . × − M J /Gyr for thermosphere temperatures between 4000and 13000 K, respectively. Using the EUV irradianceabove along with an efficiency parameter ε = 0 .
1, wecan also calculate an energy-limited mass loss rate forHAT-P-18b (e.g. Murray-Clay et al. 2009):˙ M = επR F XUV GM p ≈ × g / s . (1)This estimate agrees well with the inferred mass lossrates in Figure 3 (with uncertainties of a factor of fewaccommodated by similar uncertainties on the efficiencyparameter), suggesting that our observationally-derivedconstraints are energetically feasible.Because (cid:15) Eridani is a relatively young, active star,with log( R HK ) = − .
51 compared to log( R HK ) = − . R HK ) = − .
99) K2.5V star. By trying proxy stars with activity levels on either side of HAT-P-18b’s activ-ity, we can get a sense for the uncertainty on the resultbased on our choice of EUV proxy. With HD 40307as a proxy, we found a best-fit log( ˙ M ) = 9 . +0 . − . at4000 K, and 11 . +0 . − . at 13000 K, nearly identical toour findings for (cid:15) Eri. This is because the mid-UV toEUV flux ratios between the two stars are quite similar(Oklopˇci´c 2019). Although (cid:15)
Eri is a much stronger X-ray emitter, the cross section to X-ray photoionizationof helium is very small compared to the cross sectionin the EUV near the 504 ˚A threshold, so the contribu-tion to the flux-averaged photoionization cross sectionis negligible (Oklopˇci´c & Hirata 2018).We note that there is a strong degeneracy betweenthe mass loss rate and thermosphere temperature dueto the complex dependence of the outflow velocity anddensity on the temperature and the mass loss rate. Thisdegeneracy could be partially resolved with a preciseline shape measurement, but we do not resolve the lineshape in these observations. Due to the faintness ofHAT-P-18, spectrographs on all but the largest tele-scopes may have difficulty resolving the line shape pre-cisely enough to break the degeneracy. Additionally, ourhelium light curve is symmetric across our best-fit mid-transit time. However we cannot exclude the possibilityof an extended egress, as our combined light curve lacksthe precision required to significantly detect a trailinghelium tail for such a faint target. CONCLUSIONS elium for HAT-P-18b ( R p / R ) ( TE SS ) P T b u ( H e ) u ( H e ) u ( TE SS ) u ( TE SS ) l og ( σ e x t r a )( n i gh t ) l og ( σ e x t r a )( n i gh t ) k ( TE SS ) . . ( R p /R ) (He) a b s o r p ti onp r oxy .
80 1 . ( R p /R ) (TESS) . . P .
825 2038 . T . . b . . u (He) . . u (He) . . u (TESS) . . u (TESS) .
25 2 . log( σ extra )(night 1) . . log( σ extra )(night 2) .
85 0 . k (TESS) . . absorptionproxy Figure 4.
Corner plot displaying the posterior probability distributions for the joint model for HAT-P-18b. Note transitdepth ( R p / R ∗ ) values are in %, period P is in days, and predicted mid-transit time T is in BTJD TDB . We omit all of thedecorrelation parameters except for the absorption proxy.
In this work, we use an ultra-narrowband helium fil-ter centered on the 1083 nm line to observe two transitsof HAT-P-18b. We detect 0 . ± .
12% excess heliumabsorption in the planet’s upper atmosphere. This de-tection corresponds to an atmospheric mass loss ratebetween 8 . +2 . − . × − and 2 . +0 . − . × − M J /Gyr,which means HAT-P-18b is losing less than 2% of itsmass per Gyr. This is typical for close-in gas giants, with other helium outflow detections having mass lossrates less than 5% per Gyr (Allart et al. 2018; Mans-field et al. 2018; Spake et al. 2018; Alonso-Floriano etal. 2019).Of the handful of planets with detected helium out-flows, WASP-107b is the most comparable to HAT-P-18b with a similar radius of 0.94 R J , mass of 0.12 M J , separation of 0.55 au, and equilibrium tempera- Paragas et al. ture of 770 K (Anderson et al. 2017). If we assumethat HAT-P-18b has a He line shape similar to thatof WASP-107b, we can invert our excess helium tran-sit depth to obtain an estimate of the underlying pre-dicted line depth of 4 . ± . . ± .
24% depth measured by CARMENESand Keck/NIRSPEC for WASP-107b (Allart et al. 2019;Kirk et al. 2020). The difference may be due to thesmaller gravitational potential of WASP-107b (the massof WASP-107b has recently been suggested to be evenlower by Piaulet et al. 2020), or differences in the EUVspectra of the two stars (WASP-107 is a K6 star whileHAT-P-18 is a K2). Detailed comparative modeling ofthese two planets may make clear the primary controlon the metastable helium signal.This is the faintest system ( J = 10 .
8) with detectedhelium absorption thus far, establishing the effectivenessof our technique for observing such targets with a mid-sized telescope. For reference, the next faintest systemwith detected excess helium absorption is WASP-107bwith a J magnitude of 9.4. Of the 11 planets identified in Kirk et al. (2020) as promising targets for observa-tions of helium outflows, many are challenging targetswith J > .
5. Our photometric technique allows usto begin surveying planets around such faint stars, ex-panding the sample of planets with measured metastablehelium absorption. Further population-level studies ofextended atmospheres in the He 1083 line will greatlyimprove our ability to calibrate the mass loss modelsused to elucidate the long-term evolution of the close-inexoplanet population.ACKNOWLEDGMENTSWe are thankful to the Palomar staff, especially Ka-jse Peffer, Paul Nied, Joel Pearman, Carolyn Heffner,and Kevin Rykoski for their support. KP acknowledgessupport from the Summer Undergraduate Research Fel-lowship (SURF) at the California Institute of Technol-ogy and the NASA CT Space Grant. SV is supportedby an NSF Graduate Research Fellowship and the Paul& Daisy Soros Fellowship for New Americans. HAK ac-knowledges support from NSF CAREER grant 1555095.REFERENCES
Allart, R., Bourrier, V., Lovis, C., et al. 2018, Science, 362,1384Allart, 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.2019, A&A, 629, A110Anderson, D. R., Collier Cameron, A., Delrez, L., et al.2017, A&A, 604, A110Bernath, P. F., & Colin, R. 2009, Journal of MolecularSpectroscopy, 257, 20Bradley, L., Sipocz, B., Robitaille, T., et al. 2016,Astrophysics Source Code Library, ascl:1609.Cauley, W. P., Shkolnik, E. L., Ilyin, I., et al. 2019, AJ,157, 69Cutri, R. M., Skrutskie, M., F., van Dyk, S., et al. 2003,VizerR Online Data Catalog, 246Esposito, M., Covino, E., Harutyunyan, A., et al. 2014,A&A, 564, L13Foreman-Mackey, D., Luger, R., Czekala, I., et al. 2020,exoplanet, v0.3.2, ZenodoFrance, K., Loyd, R. O. P., Youngblood, A., et al. 2016,ApJ, 820, 89Fulton, B. J., Petigura, E. A., Howard, A. W., et al. 2017.AJ, 154, 109 Gelman, A., & Rubin, D. B. 1992, Statistical Science, 7,457. doi:10.1214/ss/1177011136Hartman J. D., Bakos, G. ´A., Sato, B., et al. 2011, ApJ,726, 52Jensen, A. G., Redfield, S., Endl, M. et al. 2012, ApJ, 751,86Jones, E., et al. 2001,
SciPy : Open source scientific toolsfor
Python elium for HAT-P-18b9