Detection of Ionized Calcium in the Atmosphere of the Ultra-hot Jupiter WASP-76b

In recent years, high-resolution transmission and emission spectroscopy with ground-based observatories has revealed numerous atomic (e.g., Snellen et al. 2008; Seidel et al. 2019; Turner et al. 2020) and molecular (e.g., Snellen et al. 2010; Hawker et al. 2018) species in exoplanet atmospheres. Such observations not only provide us with insight into their chemical compositions, but also allow us to investigate winds (e.g., Snellen et al. 2010; Seidel et al. 2021), planetary rotation patterns (e.g., Snellen et al. 2014; Brogi et al. 2016), and atmospheric escape (e.g., Allart et al. 2018; Nortmann et al. 2018), shedding light on the underlying physical processes in these alien worlds. Following numerous detections for individual exoplanets, it will soon be possible to compare trends across populations and move into the field of comparative exoplanetology at high resolution (as has been done for low-resolution spectra; e.g., Sing et al. 2016).

Several studies have established the potential of GRACES (the Gemini Remote Access to CFHT ESPaDOnS Spectrograph; Chene et al. 2014) at the Gemini North Telescope in characterizing atmospheres of planets ranging from super-Earths (e.g., Esteves et al. 2017) to gas giants (e.g., Deibert et al. 2019). The Gemini Large and Long Program GN-2020B-LP106: "Exploring the Diversity of Exoplanet Atmospheres at High Spectral Resolution" (Exoplanets with Gemini Spectroscopy or ExoGemS for short; PI: Jake Turner) aims to harness this potential by carrying out the first systematic, high-resolution, comparative survey of transiting exoplanets with GRACES. With planned observations of more than 40 exoplanets, the survey will probe atmospheres over a range of planetary masses and temperatures, illuminating the role of these parameters in regulating atmospheres and yielding insights into planetary formation and evolution.

Serving as a benchmark target for ExoGemS, WASP-76b (West et al. 2016) is a transiting ultra-hot Jupiter (i.e., with T_eq ≳ 2200 K; Arcangeli et al. 2018; Bell & Cowan 2018; Parmentier et al. 2018) that orbits its bright (V = 9.52; Høg et al. 2000) F7 V host star in ∼1.8 days (Ehrenreich et al. 2020). With a mass of ∼0.89 M_J (recently refined with radial velocity measurements; Ehrenreich et al. 2020) and an inflated radius of ∼1.85 R_J (Ehrenreich et al. 2020), the planet is ideally suited for atmospheric characterization and has received considerable attention.

Several atomic and ionic species have been detected recently in its atmosphere. Using the High-Accuracy Radial-velocity Planet Searcher (HARPS; Mayor et al. 2003), Seidel et al. (2019) reported a detection of the Na i doublet near 589 nm. Data from the ESPRESSO (Echelle Spectrograph for Rocky Exoplanets and Stable Spectroscopic Observations; Pepe et al. 2021) instrument have provided additional insights (Ehrenreich et al. 2020; Tabernero et al. 2021; Seidel et al. 2021). Using two transits observed with ESPRESSO, Ehrenreich et al. (2020) reported asymmetric absorption due to neutral iron; their result was confirmed by Kesseli & Snellen (2021) using archival HARPS data of four transits, and later investigated through 3D Monte Carlo radiative transfer modeling by Wardenier et al. (2021). Tabernero et al. (2021) carried out a detailed analysis of the spectra presented in Ehrenreich et al. (2020), searching for absorption due to a range of atomic and molecular species. They confirmed the sodium detection of Seidel et al. (2019) and the iron detection of Ehrenreich et al. (2020), and reported detections of Li I, Mg i, Ca ii, Mn i, and K i. They did not detect, but provided upper limits for, Ti i, Cr i, Ni i, TiO, VO, and ZrO.

In this Letter, we present, for the first time in WASP-76b's atmosphere, the detection of the ionized calcium (Ca ii) triplet at ∼850 nm, using transit observations with GRACES. ¹⁰ We also confirm the Na i doublet (Seidel et al. 2019; Tabernero et al. 2021), tentatively recover Li i and K i signals (Tabernero et al. 2021), and report a tentative nondetection of Hα. This work represents the first results from the ExoGemS survey, and the fifth detection of the Ca ii triplet in ultra-hot Jupiters.

In Section 2, we describe our high-resolution optical spectra obtained using GRACES. Our data reduction process, including the removal of telluric and stellar absorption features and corrections for the center-to-limb variation (CLV) and the Rossiter–McLaughlin (RM) effect, is presented in Section 3. Our analysis and modeling routines are described in Section 4, while the results and discussion follow in Section 5, with a conclusion in Section 6. The Appendices present additional details.

We observed one transit of WASP-76b with GRACES, which uses a fiber optic feed to combine the large collecting area of the Gemini North Telescope with the high resolving power of the ESPaDOnS (Echelle SpectroPolarimetric Device for the Observation of Stars; Manset Donati 2003) spectrograph at the Canada–France–Hawaii Telescope (CFHT). GRACES spectra cover the optical wavelengths from 400 to 1050 nm at a resolving power of R ∼ 66,000. In total, 169 spectra (120 in-transit and 49 out-of-transit) were obtained over a period of ∼5.16 hr on 2020 October 11, including observations before, during, and after transit. A full summary of the observing conditions can be found in Appendix A.

The raw data frames were initially reduced with OPERA, the Open source Pipeline for ESPaDOnS Reduction and Analysis (Martioli et al. 2012). OPERA performs an optimal extraction, a bias subtraction, flat-fielding, a blaze correction, a continuum normalization, a wavelength calibration using arc lamps, and a correction to the wavelength solution using telluric lines. The result is a 1D calibrated spectrum for each exposure.

Next, we removed contaminating cosmic rays from the data through the use of a median absolute deviation filtering algorithm with a threshold of 5 times the median absolute deviation. Following Allart et al. (2017), we then flux-scaled each spectrum by dividing out a reference spectrum (in this case chosen to be the first spectrum of the night) and fitting a fourth-order polynomial to a binned version of this ratio. We note that the choice of order does not significantly impact our results. This polynomial fit was then evaluated at the wavelength range of the unbinned spectra and divided out of each individual spectrum in order to renormalize the data to the same continuum flux level. This process allows us to correct for flux level variations between individual exposures. The results of this data reduction process are presented in Appendix B.

3.1. Removal of Telluric and Stellar Features with SysRem

3.2. Correction for the Rossiter–McLaughlin Effect and Center-to-limb Variation

Finally, we corrected for the Rossiter–McLaughlin effect and stellar center-to-limb variation (CLV) using the method outlined in Turner et al. (2020). We note that the star's $v\sin i$ value is low (1.48 ± 0.28 km s⁻¹; Ehrenreich et al. 2020), so the magnitude of these effects (as calculated in Tabernero et al. 2021) is not expected to be highly significant relative to the strength of our detections; nevertheless, we model and correct for them in order to ensure that they do not affect the reported signals.

To model the RM and CLV effects, angle-dependent spectra were generated using the Spectroscopy Made Easy (SME) package (Piskunov & Valenti 2017) for 25 angles and the wavelength ranges matching those of the orders of interest for the stellar parameters taken from Ehrenreich et al. (2020; see Table 1). The star was modeled as a solid rotating body. The code uses a grid-based approach, with the stellar radius set to 510 pixels. The limb-angle-dependent spectra are interpolated onto the stellar grid, taking the rotational velocity and limb-angle into account. The planet was modeled as a black disk. To ensure that the models were treated in a similar way to the data, they were continuum-normalized and subsequently divided by the out-of-transit model spectrum before being used to correct the data.

Table 1. Orbital and Physical Parameters of the WASP-76 System Used in This Analysis

Parameter	Symbol (Unit)	Value	Reference
Stellar radius	R* (R⊙)	1.756 ± 0.071	Ehrenreich et al. (2020)
Stellar mass	M* (M⊙)	1.458 ± 0.02	Ehrenreich et al. (2020)
Magnitude	V (mag)	9.52 ± 0.03	Høg et al. (2000)
System scale	a/R*	${4.08}_{-0.06}^{+0.02}$	Ehrenreich et al. (2020)
Orbital period	P (days)	${1.80988198}_{-0.00000056}^{+0.00000064}$	Ehrenreich et al. (2020)
Transit duration	T14 (minute)	230	Ehrenreich et al. (2020)
Epoch of midtransit	Tc (BJD)	${2458080.626165}_{-0.000367}^{+0.000418}$	Ehrenreich et al. (2020)
Radius ratio	Rp/R*	${0.10852}_{-0.00072}^{+0.00096}$	Ehrenreich et al. (2020)
Planetary radius	Rp (RJ)	${1.854}_{-0.076}^{+0.077}$	Tabernero et al. (2021)
Planetary mass	Mp (MJ)	${0.894}_{-0.013}^{+0.014}$	Ehrenreich et al. (2020)
Inclination	i (degrees)	${89.623}_{-0.034}^{+0.005}$	Ehrenreich et al. (2020)
Systemic velocity	γsys (km s−1)	−1.0733 ± 0.0002	West et al. (2016)
Stellar radial velocity semi-amplitude	K* (m/s)	${116.02}_{-1.35}^{+1.29}$	Ehrenreich et al. (2020)
Planetary radial velocity semi-amplitude	Kp (km s−1)	196.52 ± 0.94	Ehrenreich et al. (2020)
Quadratic limb darkening coefficient	u1	0.393	Ehrenreich et al. (2020)
Quadratic limb darkening coefficient	u2	0.219	Ehrenreich et al. (2020)
Projected equatorial rotational velocity	$v\sin i$ (km s−1)	1.48 ± 0.28	Ehrenreich et al. (2020)

Download table as:  ASCIITypeset image

4.1. Extraction of Transmission Spectra

After reducing the data as described in Section 3, we followed the methods of, e.g., Wyttenbach et al. (2015) and Turner et al. (2020) to extract the planetary signal in the form of a transmission spectrum in the planet's rest frame. The adopted values for the systemic velocity γ_sys and stellar radial velocity semi-amplitude K_* used to shift to the stellar rest frame are presented in Table 1, while the barycentric Earth radial velocity values were calculated with the barycorr applet (Wright & Eastman 2014) using Julian date values provided by OPERA.

For the Keplerian planetary radial velocity semi-amplitude K_p , we tested two approaches: first, we used the value derived in Ehrenreich et al. (2020; see Table 1); following this, we determined a value directly from our data for each detected line species (see Appendix C.1, with K_p values presented in Table 2). We then determined an average K_p value for each species. In cases where the species was not detected, or where K_p could not be reliably extracted from the data, we only created the transmission spectrum with the Ehrenreich et al. (2020) value.

Table 2. Summary of Detected Atmospheric Line Absorption Parameters

Line	λ (nm)	Depth (%)	Vcenter (km s−1)	FWHM (km s−1)	Reff (Rp )	Extracted Kp (km s−1)
Na i D2	588.995	0.270 ± 0.062	1.043 ± 1.647	8.277 ± 2.180	1.11 ± 0.09	⋯
		(0.191 ± 0.048)	(1.258 ± 2.278)	(14.529 ± 4.198)	(1.08 ± 0.08)
Na i D1	589.592	0.354 ± 0.045	−3.426 ± 0.963	15.520 ± 2.268	1.14 ${}_{-0.08}^{+0.09}$	${185.6}_{-27.7}^{+12.2}$
		(0.348 ± 0.046)	(−3.015 ± 1.003)	(15.367 ± 2.362)	(1.14 ${}_{-0.08}^{+0.09}$)
Ca ii	849.802	0.295 ± 0.034	−1.081 ± 0.856	15.357 ± 2.016	1.12 ± 0.08	⋯
		(0.306 ± 0.033)	(−1.828 ± 0.815)	(15.156 ± 1.919)	(1.12 ± 0.08)
Ca ii	854.209	0.574 ± 0.041	0.701 ± 0.697	20.000 ± 1.640	1.22 ± 0.09	${184.3}_{-11.3}^{+11.8}$
		(0.599 ± 0.039)	(1.055 ± 0.625)	(19.542 ± 1.471)	(1.22 ± 0.09)
Ca ii	866.214	0.454 ± 0.024	0.858 ± 0.445	17.532 ± 1.048	1.18 ± 0.08	${183.5}_{-16.0}^{+8.9}$
		(0.445 ± 0.026)	(0.836 ± 0.518)	(18.098 ± 1.221)	(1.18 ± 0.08)

Note. For all lines, the first row represents the results obtained with the K_p value fit in this work (see Appendix C.1), while the second row (in brackets) represents the results obtained with the K_p value from Ehrenreich et al. (2020). Column 1: species detected. Column 2: expected location of species in air wavelengths. Column 3: fitted absorption depth of the line, assuming a Gaussian profile. Column 4: fitted offset from expected line location, assuming a Gaussian profile. In some cases, this could be attributed to atmospheric winds. Column 5: FWHM of the line (derived using a fitted value of σ) assuming a Gaussian profile. Column 6: effective radius at the line center, calculated using Equation (1). Column 7: planetary radial velocity semi-amplitude extracted in this work. For lines without entries, we were unable to accurately extract a K_p value.

Download table as:  ASCIITypeset image

Once the spectra were in the planet's rest frame, we summed over the in-transit frames to create the transmission spectrum. In general, we used all in-transit frames to create the transmission spectrum. However, we note that in the case of the sodium doublet, low signal-to-noise ratio (S/N) remnants of the stellar spectrum were present, which may affect the extraction of the transmission spectrum (see Figure 7 in Appendix C.2). We thus follow, e.g., Seidel et al. (2020) in fitting the stellar sodium lines (prior to the corrections described in Section 3.1), calculating the full width at half-maximum (FWHM) of each line, and masking out all points which fall within the FWHM of the lines before extracting the transmission spectrum. Across both lines, these masked values span 19 pixels.

The transmission spectra were generated separately for each relevant order, and are shown in Figure 1 for the Ca ii triplet and Figure 2 for the Na i sodium doublet. Our tentative detections of Li i and K i and tentative nondetection of Hα are presented in Appendix D.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

As in Turner et al. (2020), we fit each atomic line with a Gaussian that describes the line center, absorption depth, and Gaussian standard deviation σ. We note that while most lines are fit individually, we fit the two lines of the Na i doublet together. The fits were obtained with the curve_fit function from the scipy library, with bounds applied to exclude nonphysical values. From these parameters we determined an associated Doppler shift of the fitted line center from the expected location of the line (which in some cases can be attributed to atmospheric winds) and the FWHM for each line (FWHM = $2\sqrt{2\mathrm{ln}2}\sigma$). Following Yan et al. (2019), we then used the fitted line depth h and the photometric transit depth δ (∼1.2%; see Table 1) to calculate the effective radius R_eff at each line center via:

$$\begin{eqnarray}&&\displaystyle \frac{{R}_{\mathrm{eff}}^{2}}{{R}_{p}^{2}}=\displaystyle \frac{\delta +h}{\delta }.\end{eqnarray} \tag{ 1 }$$

The results are presented in Table 2. We note that our results are generally consistent within the uncertainties regardless of whether we used our measured K_p values or the value derived by Ehrenreich et al. (2020). The exception is the depth of the Na i D2 line (see Table 2), which is slightly lower when the Ehrenreich et al. (2020) value is used. This discrepancy could be due to noise in the data, as the D2 line is located near the edge of the spectral order, where the efficiency of the detector drops steeply and any variations in the blaze profile may impact the line shape. We also acknowledge that atmospheric dynamics may impact the recovery of K_p (e.g., see recent results and discussions from Nugroho et al. 2020 and Wardenier et al. 2021), which is why we have confirmed that our results are consistent regardless of the value used.

4.2. Model of WASP-76b's Atmosphere

Table 2 presents the measured and derived line parameters for the atmospheric species detected through our analysis. Additional details on our tentative results are presented in Appendix D.

5.1. Comparison with Previous Results

5.2. Comparison with Synthetic LTE and NLTE Transmission Spectra

We analyzed transit observations of WASP-76b taken with GRACES at the Gemini North Telescope as part of the Large and Long Program GN-2020B-LP-106: "Exploring the Diversity of Exoplanet Atmospheres at High Spectral Resolution" (ExoGemS), resulting in detections of Na i and Ca ii. The latter represents the first detection of the ionized calcium triplet at ∼850 nm in the atmosphere of WASP-76b, and the fifth such detection in ultra-hot Jupiters.

In a forthcoming paper, we will analyze the full spectral range covered by these observations and search for absorption due to a suite of atomic, ionic, and molecular species modeled at high resolution (following Hood et al. 2020), including the species recently detected by Ehrenreich et al. (2020) and Tabernero et al. (2021).

We thank Miranda Herman and Tomás Cassanelli for the helpful discussions. We also thank Matteo Brogi, Drake Deming, Miranda Herman, and David Lafrenière for their contributions as Co-Investigators of the ExoGemS survey. Finally, we thank the scientific editor Fred Rasio and the anonymous referee for their swift and constructive feedback on this work.

E.K.D. is supported by a Vanier Canada Graduate Scholarship—NSERC. J.J.F. and C.E.H. acknowledge the support of NASA Exoplanets Research Program grant 80NSSC19K0293. R.A. is a Trottier Postdoctoral Fellow and acknowledges support from the Trottier Family Foundation. This work was supported in part through a grant from FRQNT.

This work was based on observations obtained through the Gemini Remote Access to CFHT ESPaDOnS Spectrograph (GRACES). ESPaDOnS is located at the Canada–France–Hawaii Telescope (CFHT), which is operated by the National Research Council of Canada, the Institut National des Sciences de l'Univers of the Centre National de la Recherche Scientifique of France, and the University of Hawai'i. ESPaDOnS is a collaborative project funded by France (CNRS, MENESR, OMP, LATT), Canada (NSERC), CFHT and ESA. ESPaDOnS was remotely controlled from the international Gemini Observatory, a program of NSF's NOIRLab, which is managed by the Association of Universities for Research in Astronomy (AURA) under a cooperative agreement with the National Science Foundation on behalf of the Gemini partnership: the National Science Foundation (United States), National Research Council (Canada), Agencia Nacional de Investigación y Desarrollo (Chile), Ministerio de Ciencia, Tecnología e Innovación (Argentina), Ministério da Ciência, Tecnologia, Inovações e Comunicações (Brazil), and Korea Astronomy and Space Science Institute (Republic of Korea).

This work was enabled by observations made from the Gemini North telescope, located within the Maunakea Science Reserve and adjacent to the summit of Maunakea. We are grateful for the privilege of observing the universe from a place that is unique in both its astronomical quality and its cultural significance.

This research also made use of the NIST Atomic Spectra Database funded [in part] by NISTs Standard Reference Data Program (SRDP) and by NISTs Systems Integration for Manufacturing Applications (SIMA) Program.

A subset of the figures in this work were generated with colorblind safe color schemes provided by ColorBrewer 2.0 (Harrower & Brewer 2003). The original ColorBrewer (v1.0) was funded by the NSF Digital Government program during 2001–02, and was designed at the GeoVISTA Center at Penn State (National Science Foundation grant No. 9983451, 9983459, 9983461). The design and rebuilding of this new version (v2.0) was donated by Axis Maps LLC, winter 2009 and updated in 2013.

Facilities: Gemini:Gillett (GRACES) - , CFHT (ESPaDOnS). -

Software: astropy (Price-Whelan et al. 2018), barycorr (Wright & Eastman 2014) batman (Kreidberg 2015), Cloudy (Ferland et al. 2017), ColorBrewer 2.0 (Harrower & Brewer 2003), IPython (Perez & Granger 2007), Matplotlib (Hunter 2007), Numpy (Harris et al. 2020), OPERA (Martioli et al. 2012), SciPy (Virtanen et al. 2020), SME (Piskunov & Valenti 2017).

We observed one transit of WASP-76b with GRACES at the Gemini North telescope. The spectra cover the optical wavelengths from 400 to 1050 nm, and the resolution of the observations ranged from R ∼ 58,000 to ∼64,000.

In total, we collected 169 spectra that included 120 spectra in-transit and 49 spectra out-of-transit. The observations spanned a period of ∼5.16 hr. We note that approximately 10% of the transit, covering orbital phases of roughly 0.005 to 0.012, was lost due to technical issues at the observatory.

Each spectrum was recorded with an exposure time of 60 seconds, resulting in average signal-to-noise ratios (S/N) per spectral element varying between ∼73.8 and ∼111.1 for the orders relevant to this work. The change in S/N throughout the night for each relevant order is shown in the left panel of Figure 3. The airmass varied between 1.046 and 1.627 over the course of the observations (see the right panel of Figure 3).

Zoom In Zoom Out Reset image size

The individual steps of the data reduction process described in Section 3, including the correction for telluric and stellar absorption using the SysRem algorithm (Section 3.1), are presented in Figure 4 for the two orders containing the Ca ii infrared triplet and Figure 5 for the three other orders considered in this analysis.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

C.1. 2D Phase-folded Maps

Figure 6 shows the results of phase-folding the transmission spectra in order to determine the radial velocity semi-amplitude for each line. After shifting the data to the stellar rest frame, each spectrum was phase-folded to potential planetary radial velocity semi-amplitudes ranging from 0 km s⁻¹ to 300 km s⁻¹, with a step size of 0.1 km s⁻¹, to create the K_p -V_center plots. Note that while the planetary radial velocity semi-amplitude is known to a far greater precision than this from previous studies, we sampled a wide range of radial velocities in order to investigate any potential peaks due to noise at other locations in the data (e.g., a correlation with residual stellar absorption might be expected at a K_p value of 0 km s⁻¹ in the K_p -V_center map). The peaks of the phase-folded K_p -V_center plots were then used to determine the planetary radial velocity semi-amplitudes, which are presented in Table 2. Errors were estimated by determining the point by which this peak velocity value had dropped off by 1σ. These errors are in general large enough that individual lines in a species are consistent with one another, as well as with values reported in previous works (e.g., K_p = 196.52 ± 0.94 km s⁻¹ in Ehrenreich et al. 2020; though we note that our values are in general lower). In each plot, a V_center value of 0 km s⁻¹ corresponds to the expected location of each line, because the data have already been shifted to the stellar rest frame.

Zoom In Zoom Out Reset image size

We note that there were several additional peaks of comparable strengths to the planetary signal in the K_p -V_center map created for the first line of the Ca ii infrared triplet (i.e., the line at ∼849.8; see the top left panel of Figure 6). However, given the proximity of this line to the edge of the spectral order, as well as the fact that this line is the weakest of the three in the triplet, we assume that this is caused by noise present in the data and adopt the peak K_p value consistent with values determined for the other two lines (see Table 2). We also note that the peak K_p value of the Na i D2 line is significantly lower than expected (though there is also an absorption feature present at the expected location); we likewise attribute this to noise at the edge of the spectral order and adopt the measured peak value of the Na i D1 line for the sodium transmission spectrum. In the case of Hα, Li i, and K i, we are unable to recover a peak value from the K_p -V_center maps. We thus only use the Ehrenreich et al. (2020) value to create these transmission spectra.

Finally, we note that the broad signals present in these plots are comparable to that seen in Figure B.1 of Seidel et al. (2021).

C.2. Spectra in Planetary Rest Frame

Figure 7 shows the spectra in the planetary rest frame, after the data reduction steps outlined in Section 3 and using the K_p values measured in Appendix C.1 (or in cases where K_p could not be reliably measured, using the value from Ehrenreich et al. 2020; see Table 1). In theory, a strong planetary signal could be visible by eye in the form of a vertical line at or near the expected absorption line location. In practice, while such lines are marginally visible in the case of the two strongest Ca ii lines, summing over in-transit frames and creating transmission spectra (as described in Section 4) increases the signal strength to the point that the planetary absorption can be seen (e.g., Figures 1 and 2).

Zoom In Zoom Out Reset image size

The transmission spectra generated for the order containing Hα and Li i are presented in Figure 8, while the transmission spectra generated for the two lines of the K i doublet are presented in Figure 9. The process used to generate the spectra is the same as that outlined in Section 4.1.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

In the case of Hα, we note that while there is absorption present in the spectrum at the expected line location, its depth is comparable to the overall noise levels in the spectrum (e.g., note the dip in the spectrum at ∼656.4 nm) and thus could be due to noise rather than any potential planetary signal. This result is largely consistent with those presented in Tabernero et al. (2021). In particular, the upper limit inferred for the second Hα transit analyzed by Tabernero et al. (2021) is at the noise level of our Hα transmission spectrum, while the absorption depth measured for their first transit (0.48% ± 0.12%) is at/above the noise level of our spectrum. This could indicate (as was inferred in Tabernero et al. 2021) that Hα may be variable in the atmosphere of WASP-76b, but further data are needed to confirm this scenario.

We note as well that a low S/N stellar Hα remnant is present in the data (see Figure 7 in Appendix C.2). The broad nature of the stellar Hα line means that this low S/N remnant overlaps with the expected location of planetary absorption across essentially all in-transit frames; we thus do not mask these pixels as we did for the sodium doublet (see Section 4.1), but note that the contamination from this low S/N remnant is likely affecting our ability to confidently extract a transmission spectrum.

In the cases of Li i and K i, the noise levels are slightly lower than the dips present at the expected line locations (note that Li i is in the same spectral order as Hα but is located closer to the center of the order; see Figure 5). We are able to fit Gaussians to each line profile and derive associated parameters as described in Section 4.1; these are presented in Table 3. However, we note that in all cases there are various dips/peaks in the spectra that approach the depths of the fitted line profiles, which could call the validity of these results into question. We thus label these as tentative detections warranting further investigation, and provide additional insights through the bootstrapping analysis described in Appendix E.

Despite only being tentative, the measured line profiles for Li i and K i presented in Figures 8 and 9 are largely consistent with those in Tabernero et al. (2021). In particular, our detected line depths are comparable to those of Tabernero et al. (2021). Our measured blueshift for the K i line at ∼769.9 nm (which could be consistent with atmospheric winds) is significantly different from that reported in Tabernero et al. (2021); however, we note that the values reported by Tabernero et al. (2021) for each of their transits are also significantly different from each other (−10.1 ± 2.7 km s⁻¹ and −2.2 ± 2.1 km s⁻¹). Our measured blueshift of the K i line at ∼766.5 nm is consistent with that of the first transit analyzed in Tabernero et al. (2021); however, we note that this line falls within a dense forest of telluric O₂ lines and thus may have been affected by the telluric absorption correction (in particular, a pair of strong telluric O₂ lines are located directly bluewards of the K i line, perhaps contributing to the absorption depth). We note as well that Tabernero et al. (2021) did not investigate the presence of this line. While the K i line at ∼766.5 nm does appear to be a somewhat clearer detection, the fact that we are unable to confidently recover the line at ∼769.9 nm indicates that systematic effects may be impacting our results, and we thus classify K i as a tentative detection warranting further investigation. The large difference in velocity shifts between the two lines also calls the validity of these tentative results into question, as they would be expected to be consistent with one another. We note that the telluric oxygen lines in this region could significantly affect the line profile of the lines, resulting in spurious shifts. Finally, the large redshift that we measure for the Li i is unlikely to be physical, and could further suggest that the result is indeed spurious, or at the least, affected by noise in the data.

To further assess the contribution of systematic errors to our results, we carried out a bootstrapping or empirical Monte Carlo (EMC) analysis (e.g., Redfield et al. 2008; Casasayas-Barris et al. 2019; Seidel et al. 2019; Allart et al. 2020, among others). For each line, we created three different scenarios: an out–out scenario where we randomly divided all out-of-transit frames into "virtual" in-transit and out-of-transit frames and repeated our analysis; an in–in scenario where we did the same but with in-transit frames; and an in–out scenario, where an increasing number of in-transit frames (up to half of the sample) were randomly removed from the sample and used to create a virtual in-transit data set that was compared with the nominal out-of-transit sample. For the first two scenarios, the distribution of derived absorption depths is expected to be centered at zero, while for the latter we expect to see a distribution centered at negative values for detected species and a distribution centered at zero for nondetections. Note that the widths of the distributions may vary due to the different number of spectra used in each case; this does not affect the analysis. In all cases, we ran 10,000 trials of each scenario.

Absorption depths were calculated via the method described in Redfield et al. (2008), Seidel et al. (2019), and Allart et al. (2020), among others. The average flux in a narrow (0.75 Å) window around the line of interest was compared to the average flux in blue and red passbands outside the area of interest. This is different from the process used to derive absorption line parameters in our main analysis (i.e., fitting a Gaussian), but allowed us to more readily account for cases where absorption depths of 0% were detected but the spectra themselves were noisy (note that a Gaussian fitting algorithm may attempt to fit on a noise spike, and checking each of the 10,000 trials by eye for proper fits was not feasible). We note that in the cases of the Na i D2 and K i lines, we centered the narrow flux windows around the detected locations of the lines accounting for Doppler shifts.

The results for all lines investigated in this work are presented in Figure 10. As can be seen in the figure, the out–out and in–in scenarios are centered around zero, while the in–out scenarios for detected line species are centered around negative absorption depths.

Zoom In Zoom Out Reset image size

In the case of Hα, the in–out scenario is slightly offset from zero, which does indicate a potential detection. However, we note that the out–out and in–in scenarios are also slightly offset from zero; while the widths of these distributions (i.e., the errors) are large enough to place the distributions within error of zero absorption, this could potentially indicate that residual stellar Hα absorption is impacting our results. We thus do not classify this or the transmission spectrum presented in Figure 8 as a detection.

In the case of Li i, the in–out scenario shows a considerable negative offset, while the in–in and out–out scenarios do not. While this would indicate a detection, the transmission spectrum presented in Figure 8 contains a number of dips/peaks adjacent to the absorption line profile that suggest that noise may be impacting our results. As in Appendix D, we thus classify this as a tentative detection warranting further investigation.

Finally, in the case of K i, the in–out scenario of the line at ∼769.9 nm is only slightly offset from zero while the line at ∼766.5 nm is more consistent with a negative absorption depth. While the transmission spectra presented in Figure 9 do indicate a potential detection, the results of the EMC analysis are inconclusive and suggest that these absorption features may in fact be caused by noise or other systematic effects in the data. We thus classify K i as a tentative detection as well, but caution that systematic effects may be impacting our results.