An Atomic Spectral Survey of WASP-76b: Resolving Chemical Gradients and Asymmetries

Ultrahot Jupiters (UHJs) are an exotic class of giant, highly irradiated exoplanet and have dayside temperatures greater than 2200 K (Parmentier et al. 2018). These planets are likely tidally locked and are characterized by clear atmospheres dominated by atomic and ionic features from dissociated molecules on the daysides, and cooler cloudy atmospheres on the nightsides. The large-scale heights of UHJs make them ideal laboratories to study how tidally locked planets absorb and redistribute the intense radiation from their host stars.

High-resolution transit transmission spectroscopy has proven a powerful tool to probe the atmospheres of UHJs due to its ability to resolve many narrow-line absorbers (Fe, Ti, V, Mg, etc.) that are invisible at low-spectral resolution. This technique has been used to confirm the presence of a vast assortment of atoms and ions thus showing that most molecules and condensates are dissociated on the limbs of the UHJs. Hoeijmakers et al. (2018, 2019) first explored the atomic/ionic composition, up to atomic number 78, of the hottest known exoplanet, KELT-9b (T_eq = 4050 K; Gaudi et al. 2017), and uncovered signals from neutral Fe i, Mg i, and Na i, as well as ionized Sc ii, Cr ii, Y II, Fe ii, and Ti ii. Subsequent studies of WASP-121b (T_eq = 2358 K; Delrez et al. 2016), MASCARA-2b (T_eq = 2260 K; Talens et al. 2018), and WASP-76b (T_eq = 2160 K; West et al. 2016) have extended this work to cooler UHJs. Recent papers have built upon each other to detect more new species and confirm previous detections of atoms and ions in the atmosphere of WASP-121b, and the list of detected species now includes Mg i, Na i, Ca i, Ca ii, Cr i, Fe i, Fe ii, Ni i, V i, H i, Li i, K i, and Sc ii (Ben-Yami et al. 2020; Hoeijmakers et al. 2020a; Borsa et al. 2021; Merritt et al. 2021). In MASCARA-2b, neutral Fe i, Na i, and Mg i, as well as ionized Ca ii, Fe ii, and Cr ii have been identified (Casasayas-Barris et al. 2019; Stangret et al. 2020; Nugroho et al. 2020; Hoeijmakers et al. 2020b). Tabernero et al. (2021) detected absorption from neutral Li i, Na i, Mg i, Fe i, K i, and Mn i, and ionized Ca ii in the atmosphere of WASP-76b. Planets cooler than UHJs exhibit absorption from far fewer metals (Allart et al. 2020; Casasayas-Barris et al. 2021a), seemingly because many elements have condensed into molecules, clouds, and hazes (e.g., Kataria et al. 2016; Gao et al. 2021).

Recently, Ehrenreich et al. (2020) explored the transition region between the hot dissociated dayside and cool cloudy nightside of WASP-76b using the Echelle Spectrograph for Rocky Exoplanets and Stable Spectroscopic Observations (ESPRESSO) on the Very Large Telescope (VLT). They observed asymmetric absorption in the cross-correlation signal from Fe i, where, at the end of the transit, the signal was blueshifted by ∼10 km s⁻¹ from its expected position, while at the beginning of the transit, the signal was at rest. This same asymmetry was confirmed in Kesseli & Snellen (2021) using four transits with the HARPS spectrograph and a different analysis method. As WASP-76b transits in front of its host star, its viewing angle changes by ϕ = 30° (Ehrenreich et al. 2020) and assuming tidal locking, more of the illuminated morningside is visible at the start of the transit, while more of its illuminated eveningside is visible at the end of the transit. Due to the interplay between the planet's rotation and the measured global wind blowing from the hot dayside to the cool nightside (Seidel et al. 2021), the eveningside of the planet is expected to have a larger blueshift than the morningside, but this velocity change would only be able to be resolved if the atmospheric temperature profile or abundance of Fe i were not uniform over the surface of the planet. Recent in-depth global circulation modeling (GCM) by Wardenier et al. (2021) and Savel et al. (2021) confirmed that a hotter evening terminator, a cooler morning terminator, and either condensation of Fe i or clouds obscuring the cooler morningside were necessary to explain the asymmetry.

By detecting more of these asymmetric transit signals, we can directly measure abundances of individual elements across the surface of exoplanets, and hence indirectly infer which cloud species are forming. Lothringer et al. (2020) suggested a similar method to probe condensate formation and rain-out by using low-resolution near-UV spectral indices, as most metals absorb strongly at these wavelengths. Furthermore, the high-spectral resolution technique discussed here is complementary to proposed techniques for resolving the different temperature structures and low-resolution spectra of the morning and evening terminators with the James Webb Space Telescope (Espinoza & Jones 2021). By combining information from both low-resolution and high-resolution techniques, we can extract information on the dynamics and velocity structure, temperatures, and abundances of a range of species in the morning and evening terminators separately.

In this paper, we aim to search for asymmetric transit signals within the full inventory of atoms and ions that are accessible at the visible wavelengths covered by ESPRESSO, using the same data set that was used for the original Fe condensation result (Ehrenreich et al. 2020). In Section 2, we will assess which atoms and ions contribute significant opacity at visible wavelengths and the temperatures seen in UHJs. We will then use these findings to search for all observable species in the ESPRESSO transit observations of WASP-76b. We discuss the data and our pre-processing steps in Section 3. We then explain how we made the models of each species in Section 4 and the subsequent cross-correlation analysis in Section 5. We present the results of the cross-correlation search in Section 6, and discuss which of the detected species exhibit this asymmetric signal. We discuss the implication of and what physical information about the exoplanet atmosphere can be extrapolated in Section 7. Finally, we conclude in Section 8.

Since our goal is to search for asymmetric transit signals in the cross-correlation maps of each element, we began by exploring which elements would produce detectable signals so as to avoid the unnecessary task of performing the full cross-correlation analysis for elements that would not be detectable. Elements with a low observability score should either have no absorption lines in the visible portion of the spectrum or are expected to be so under-abundant that even if they do present features, the signal would be smaller than that of the base noise level or continuum opacity sources. A similar analysis that also aimed to determine which elements and ions would be detectable at high spectral resolution was performed in Hoeijmakers et al. (2019).

To determine an observability score for each element, we used atomic opacities from the Data and Analysis Center for Exoplanets (DACE) Opacity database. ³ These opacities were generated with the open-source HELIOS-K opacity calculator (Grimm & Heng 2015; Grimm et al. 2021) using input from the Kurucz (Kurucz 2018), National Institute of Standards and Technology (NIST; Kramida et al. 2019), and The Vienna Atomic Line Data Base (VALD3; Ryabchikova et al. 2015) spectroscopic databases. Each database contains opacities for atoms and ions at temperatures ranging from 2500 to 6100 K and a single pressure of 10⁻⁸ bars. We chose to use the opacities generated using NIST because it was the only database that contained opacities for all of the atoms and ions (both the Kurucz and VALD3 databases did not contain opacity data for Br, Kr, and I). We downloaded data for two temperatures, 2500 and 4000 K, which spans the expected temperature range of UHJs. Hoeijmakers et al. (2019) found that the average altitude where most of the detected absorption lines formed in the UHJ Kelt-9b was between ∼10⁻⁴ and 10⁻⁶ bars, and that the cores of the strongest absorption lines could be produced at even higher altitudes. While 10⁻⁸ bars is higher in the atmosphere than the expected formation location, it should give a reasonable estimate for each element as pressure broadening is not expected to be too important at these high altitudes and low pressures.

We choose not to implement chemical modeling and instead simply scaled each element by its solar abundance from Asplund et al. (2009), which will give us a useful upper limit on detectability of each species. Chemistry in hot Jupiters is complicated, and correctly modeling these processes would require a much more in-depth atmospheric model including full temperature pressure profiles at a range of longitudes, extensive chemical reactions and condensation information, and assumptions about equilibrium versus nonequilibrium processes, which is beyond the scope of this paper. This type of in-depth chemical study has been performed previously on other UHJs (e.g., Kataria et al. 2016; Gao et al. 2021), and in later sections, we will refer to these results to help interpret any deviations that arise between the data and the predictions presented here. To estimate the abundances, we scaled the opacities by each element's expected mass density given solar abundances from Asplund et al. (2009). Tc, Po, At, and Rn are not included in Asplund et al. (2009) and so we did not include them in our calculations (shown in gray in Figure 1). We used the values derived from the solar photosphere when available, except for Li. We used the Li abundance derived from solar system meteorites, which is also given in Asplund et al. (2009), since photospheric abundance of Li drastically changes with age and the current photospheric abundance does not represent the initial value that would have been available in the disk during planet formation (Thévenin et al. 2017).

Zoom In Zoom Out Reset image size

We also added a continuum opacity, below which each element's opacity would no longer contribute. We used petitRADTRANS (Mollière et al. 2019) to find the expected levels of continuum opacity from H₂ − H₂ and H₂ − He collisions, as well as H⁻ at 3 mbar, which is where the continuum opacity is expected to begin to become optically thick in hot-Jupiter atmospheres (Hoeijmakers et al. 2019). This resulted in a base opacity near 10⁻³ cm² g⁻¹. We did not include Rayleigh scattering as a source of continuum opacity, as Fu et al. (2021) found that Rayleigh scattering was not consistent with the observed slope of the Hubble Space Telescope (HST) transmission spectra of WASP-76b, which may be too hot to host significant aerosols. We experimented with slightly different base opacity levels, but found that this did not significantly change our results as most of the signal comes from the strong absorption lines that are well above the continuum.

We calculated an observability score for each species, x, following similar methods as Mollière & Snellen (2019), and given by:

$$\begin{eqnarray}&&{{Observability}}_{x}={\int }_{3500\,\mathring{\rm A} }^{8000\,\mathring{\rm A} }{ln}({\tau }_{\mathrm{all}})-{\int }_{3500\,\mathring{\rm A} }^{8000\,\mathring{\rm A} }{ln}({\tau }_{\mathrm{all}-{\rm{x}}})\end{eqnarray} \tag{ 1 }$$

where τ_all is the combined scaled opacities of all of the species and the continuum opacity, and τ_all−x is the same combined opacities except without the contribution from the relevant species, x. These scores were scaled so the most observable species was normalized to 1. The wavelength region of 3500 to 8000 Å corresponds to the ESPRESSO wavelength coverage (as well as similar wavelength coverage of other high-resolution spectrographs like HARPS and EXPRES, barring the reddest 1000 Å). Figure 1 shows the results of this test. As expected, we find that many of the elements that are the most observable are those that are readily detected in hot-Jupiter atmospheres, such as Fe, Na, and Ca.

We note that atoms with few, but strong, absorption lines (e.g., H, Li, Na, Mg, Al, and K) are shown to have a lower observability than atoms with many lines (e.g., Fe, Ti, V) because our method sums line strengths over a wide wavelength region. However, by focusing only on the region around these single strong lines, they can be readily observed.

We now turn to the ESPRESSO data set of WASP-76b to search for all of the elements that we have deemed observable and to search for asymmetries in the transits of each. This is the ideal data set for our study as this same data set led to the first detection of an asymmetry in the absorption signal of Fe (Ehrenreich et al. 2020). The combination of ESPRESSO's stability and high spectral resolution (R ∼ 138,000; Pepe et al. 2010) with the collecting area of the European Southern Observatory's VLT (8m) is unprecedented, and this type of study is only possible because of the unique combination.

WASP-76b is a well-studied UHJ that orbits a bright (V = 9.5) F7 star (see Table 1 for the system parameters). Two transits of WASP-76b were observed on 2018 September 2 (night 1) and 2018 October 30 (night 2). Thirty-five spectra were recorded with an exposure time of 600 s on night 1, while 70 spectra were recorded with an exposure time of 300 s on night 2. These spectra cover about 30 minutes before the transit, the full transit duration, and about 30 minutes after the transit. The data were reduced with version 1.3.2 of the ESPRESSO pipeline to produce 1D blaze-corrected, stitched spectra. The given wavelength solutions have already been corrected for the barycentric velocity. More information about the reduction procedure and the data can be found in the original publication where the data were presented (Ehrenreich et al. 2020). This data set has also been used in two subsequent papers, Tabernero et al. (2021) and Seidel et al. (2021). Tabernero et al. (2021) identified individual absorption lines from Li i, Na i, Mg i, Ca ii, Mn i, K i, and Fe i, and also searched for Fe i, Ti i, Cr i, Ni i, TiO, VO, and ZrO using the cross-correlation technique, but only detected the previously seen Fe i signal. Seidel et al. (2021) performed a concentrated analysis of the Na i doublet and found evidence for a uniform day-to-night wind, as well as a vertical wind in the upper atmosphere, which acts to significantly broaden the Na i doublet.

Table 1. WASP-76 System Parameters from Ehrenreich et al. (2020)

Parameter	Value
M* (MSun)	1.458 ± 0.021
R* (RSun)	1.756 ± 0.071
Teff (star; K)	6329 ± 65
vsys (km s−1)	-1.16
$v\sin i$ (km s−1)	1.48 ± 0.28
K* (m s−1)	${116.02}_{-1.35}^{+1.29}$
Planet
Mp (MJ )	${0.894}_{-0.013}^{+0.014}$
Rp (RJ )	${1.854}_{-0.076}^{+0.077}$
Teq (K)	2228 ± 122
Porb (d)	${1.80988198}_{-0.00000056}^{+0.00000064}$
T0 (d)	${2458080.626165}_{-0.000367}^{+0.000418}$
a (au)	0.0330 ± 0.0002
Kp (km s−1)	196.52 ± 0.94
i (deg)	${89.623}_{-0.034}^{+0.005}$

Download table as:  ASCIITypeset image

Using the 1D spectra from the pipeline, we performed a series of steps to ready the spectra for cross correlation.

3.1. Telluric Correction

3.2. Spectral Cleaning

We created a separate transmission model for each observable atom and ion using petitRADTRANS (Mollière et al. 2019). petitRADTRANS calculates transmission spectra of exoplanet atmospheres at either low- or high-spectral resolution, given a pressure-temperature (PT) profile, the exoplanet's radius and surface gravity, the mean molecular weight of the atmosphere, and the abundances of the requested atoms or molecules. The code has successfully been used on many previous high-resolution transit transmission observations (e.g., Kesseli et al. 2020; Kesseli & Snellen 2021; Landman et al. 2021).

We input the planetary radius and surface gravity for WASP-76b calculated from the parameters in Table 1. We chose to use an isothermal PT profile with a temperature of 3000 K. An isothermal PT profile is often used for transmission spectroscopy as this type of spectrum is not very dependent on the underlying PT profile. Indeed, Seidel et al. (2021) found that an isothermal profile fit this ESPRESSO data well and was preferred over the more complex profiles that they tested. A temperature of 3000 K was motivated as an intermediate choice between Landman et al. (2021), who retrieved a best-fit temperature of ${2701}_{-279}^{+448}$, and Seidel et al. (2021), who found a best-fit temperature of 3389 ± 227 K. Figure 2 shows how changing the temperature changes the resulting model spectrum, and in Section 6.1 we test how our results change if we use isothermal PT profiles of 2000 and 4000 K.

Zoom In Zoom Out Reset image size

As has been discussed in many recent papers, the abundance and the pressure below which the atmosphere cannot be probed are degenerate because we are only sensitive to the line strength above the continuum (Welbanks & Madhusudhan 2019; Merritt et al. 2021; Seidel et al. 2021). We therefore chose to use solar abundance as our input for each species, and nominally set a uniform gray cloud deck (P_cloud) at 1 mbar. This gray cloud deck serves to mimic continuum opacity from a variety of sources, including collisionally induced opacity, H⁻ opacity, and clouds. Hoeijmakers et al. (2019) found that at the temperatures seen in UHJs, the atmosphere generally becomes opaque around millibar pressures due to H⁻, which motivated our selection of this gray atmospheric base. For the majority of the species, the neutral atomic abundance is expected to be similar to the solar abundance of that species because a temperature of 3000 K is too cool to ionize most atoms, but too warm for significant condensation. In Section 6.1 we explore how using different pressure cloud decks can alter our results.

petitRADTRANS has pre-computed opacity files for a range of neutral and ionized atoms, as well as many molecules. Not every atom that we wanted to test was available. For those, we utilized the same opacity data from DACE that we used in Section 2. These files can be used to create the necessary input files to run petitRADTRANS. While the database contains opacities calculated for a large range of temperatures, unfortunately, it only has atomic data at a single pressure (10⁻⁸ bar). Therefore, the pressure broadening profiles will not be exactly correct for those species that we added ourselves into petitRADTRANS. We tested the extent to which this simplification affected our results by creating our own input files using DACE opacities for Cr i and Na i, and compared them to the models made with the pre-computed petitRADTRANS files. We found that the vast majority of features had no change, but as expected, the wings of the strongest three absorption features of Cr i and the Na i D doublet had slight deviations of a few percent the height of the feature, but no noticeable changes to the resulting cross-correlation grids. Therefore we do not expect any significant difference in signals between the petitRADTRANS pre-computed opacities with those we computed from the DACE opacities.

Before we used the models in the cross-correlation analysis, we reduced the spectral resolution to match that of the ESPRESSO instrument by convolving each model with a Gaussian kernel with an FWHM of 2.2 km s⁻¹. Since the asymmetric transit signature of Fe i is best modeled as iron condensing out of the atmosphere on the morningside, we do not rotationally broaden the model spectra because we do not expect the atoms to be uniformly distributed across the limbs of the planet and thus create a net broadening effect. Examples of Cr i and Na i models that have been convolved to ESPRESSO resolution and are ready for cross correlation are shown in Figure 2.

To determine which species are present in the atmosphere of WASP-76b, we cross correlated each cleaned spectrum from night 1 and night 2 with the >40 individual models of each neutral atom or ion. Cross correlation has proven extremely successful at uncovering the very small signals present in exoplanet atmospheres (on the level of ∼10−1000 parts per million) because it effectively combines the weak signals from tens to hundreds of individual absorption lines (e.g., Snellen et al. 2010). For each individual neutral atom or ion, we followed similar steps as those laid out in Hoeijmakers et al. (2020a) and Kesseli & Snellen (2021) to perform the cross-correlation analysis, using the following equation:

$$\begin{eqnarray}&&c(\nu ,t)=\sum _{i=0}^{M}-{x}_{i}(t){T}_{i}(\nu ).\end{eqnarray} \tag{ 2 }$$

Here, c(ν, t) is the 2D cross-correlation grid at each radial velocity (ν) and observation time (t). x_i (t) is the observed spectrum at each time, and T_i (ν) is the model of each element, shifted to every radial velocity. The radial velocities range from −300 to +300 km s⁻¹ in 0.5 km s⁻¹ steps. The models are normalized such that ${\sum }_{i=0}^{M}{T}_{i}(\nu )=1$. The negative sign is included so that excess absorption from the exoplanet will appear as a positive residual, as is standard for cross correlation. In the end, the cross-correlation function acts as a weighted sum for each spectrum, and pixels where the absorption is strong in the model (T_i (ν)) are weighted much higher than pixels where there is little or no absorption from the relevant atom, creating an excess signal when the model is aligned with absorption lines in the planet.

The spectra used in the cross correlation still contained the contribution from the host star, so the cross-correlation grids are dominated by any signal from it. To remove this signal, we averaged together all of the cross-correlation functions that did not occur as the planet was transiting, and then divided each cross-correlation function by this average out-of-transit function. The final cross-correlation grid for each night is multiplied by 10⁶ to convert the residual cross correlations into parts per million (ppm). The resulting residual cross-correlation grid for each night of observations using the Cr i model is shown in panels (a) and (b) of Figure 3.

Zoom In Zoom Out Reset image size

We chose to report the cross-correlation values in ppm, as in Tabernero et al. (2021). The residual amplitude values from the cross-correlation functions reported here are simply a weighted average of the transmission depths from the residual spectrum within all of the lines of the relevant atom (Equation (2)). Larger residual amplitudes correspond to deeper lines in the planet's spectrum and larger transit depths. Therefore, the reported cross-correlation amplitudes give a general idea of where in the atmosphere the majority of the signal arises, and larger residual amplitudes mean the signal was produced on average higher in the atmosphere than smaller residual amplitudes. We do note, however, that unlike the residual amplitudes reported for single lines, these residual amplitudes are model dependent as the model (T_i (ν)) acts as the weights and does not directly correlate to the formation location of any individual line or lines. To ensure that all of the trends we measured are model independent, we tested models with different temperatures, abundances, and continuum opacities, which is discussed in Section 7.

We then combined the two nights together into a single 2D cross-correlation grid to maximize the S/N. We interpolated each cross-correlation grid onto a new grid that was uniformly spaced in phases with a step size of 0.004. We chose this step size as it approximately corresponded with the largest phase step in both nights and so none of the data will be extrapolated at a finer spacing than was observed. After both nights were interpolated onto the same grid, they were averaged together. Night 2 had finer temporal spacing, and so to account for this, we weighted the nights by the number of spectra that were taken during the transit on each night. We also weighted the nights by the average S/N of the stellar spectra over the full course of observations. The combined cross-correlation grid for the same example with Cr is shown in panel (c) of Figure 3.

5.1. Removal of the Rossiter–McLaughlin Effect and Other Stellar Residuals

We performed the same cross-correlation analysis for each neutral atom or ion. We detected Li i, Na i, Mg i, Ca ii, V i, Cr i, Mn i, Fe i, Ni i, and Sr ii with an S/N > 5, and we tentatively detected Co i and K i with a peak S/N > 4. We detected H i with an S/N of 5.77, but there is significant noise around the stellar rest frame due to the low S/N of the spectra near the center of the H α feature, and so we only call this a tentative detection. Tabernero et al. (2021) also found tentative evidence for H α absorption in their single-line analysis, which agrees with our tentative detection. Fe i, Mn i, Na i, Ca ii, Li i, K i, and Mg i were previously detected in Tabernero et al. (2021), but V i, Cr i, Ni i, and Sr ii are detected, and Co i is tentatively detected here, for the first time in WASP-76b.

All of the detections and tentative detections are shown in Figure 4. Each row contains three plots with results of the cross-correlation analysis of a different neutral atom or ion. The left panel shows the 2D cross-correlation grid in the rest frame of the planet, the middle panel shows the averaged 1D cross-correlation function for the beginning of the transit (turquoise) and end of the transit (dark purple) separately, and the right panel shows our systematic search for all prominent signals in the 2D cross-correlation grid for a range of K_p and v_wind values.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

To determine which of these detections and tentative detections exhibit an asymmetry, we fit Gaussians to the 1D cross-correlation functions from the beginning and end of the transit separately (middle panel of Figure 4), using the python module lmfit. We record the central radial velocities in Table 2, where RV₁ is the central radial velocity of the Gaussian fit to the average cross-correlation function between phases ϕ = −0.04 and −0.02, and RV₂ is the central radial velocity for the phases ϕ = +0.02 to +0.04. The uncertainties that we record use the same method as Kesseli & Snellen (2021), and are computed by converting the FWHM of the fitted Gaussian to the standard deviation and then dividing by the S/N of the peak. We tested that these values gave a true measurement of uncertainty by recording the central radial velocity measurements for each individual Fe i cross-correlation function in the same phase ranges for the nights individually before they had been interpolated onto the uniform phase grids. We then took the standard deviation of all of these individual measurements and divided by the square root of the number of measurements, as expected for the standard deviation of a sample. We found uncertainties that were consistent within 20% for these two methods. We also tested using the uncertainties returned by lmfit, but found that these were about half the size of the uncertainties we calculated and the ones given by our test using the many measurements of Fe i, and so we did not use these. For all of the species that we detected confidently (S/N > 5), the measured RV offsets between the first and second half of the transit are discrepant by more than 2σ for Ca ii, V i, Cr i, Mn i, Fe i, and Sr ii. The two RVs are discrepant by 1σ for Mg i and Ni i, while the RVs are consistent for Li i and Na i.

Table 2. Parameters of Detected Species

Atom	Peak	RV1	RV2	Amplitude 1	Amplitude 2	Kp	FWHM
	S/N	(km s−1)	(km s−1)	(ppm)	(ppm)	(km s−1)	(km s−1)
H i	5.31	−14 ± 3.51	2.08 ± 1.78	989 ± 141	1709 ± 229	${219}_{-20}^{+40}$	15.25 ± 2.26
Li i	6.03	−5.06 ± 2.07	−5.28 ± 2.89	822 ± 94	1195 ± 156	${235}_{-24}^{+16}$	21.23 ± 3.39
Na i	10.63	−3.80 ± 0.79	−6.08 ± 1.49	1285 ± 135	941 ± 116	${209}_{-20}^{+14}$	17.37 ± 2.01
Mg i	6.94	−3.32 ± 1.29	−7.61 ± 1.25	1049 ± 128	1240 ± 163	${189}_{-30}^{+26}$	15.4 ± 2.02
K i	4.22	−6.97 ± 0.83	−10.83 ± 2.04	1157 ± 166	743 ± 138	${197}_{-12}^{+14}$	9.35 ± 2.0
Ca ii	7.8	8.05 ± 2.3	−2.52 ± 1.53	387 ± 53	653 ± 78	${197}_{-38}^{+16}$	22.5 ± 2.28
V i	8.19	−5.61 ± 0.68	−9.85 ± 0.85	228 ± 29	411 ± 52	${175}_{-10}^{+14}$	10.89 ± 1.41
Cr i	6.82	−6.59 ± 2.17	−12.91 ± 0.68	207 ± 31	482 ± 60	${173}_{-14}^{+12}$	10.03 ± 1.21
Mn i	5.82	−1.67 ± 2.1	−8.75 ± 1.21	305 ± 68	639 ± 83	${179}_{-10}^{+40}$	11.98 ± 1.62
Fe i	14.24	−3.89 ± 0.48	−10.26 ± 0.24	440 ± 47	833 ± 83	${188}_{-16}^{+10}$	9.05 ± 0.5
Co i	4.03	−1.42 ± 2.16	−12.55 ± 0.8	173 ± 53	434 ± 65	${173}_{-16}^{+14}$	8.13 ± 1.5
Ni i	5.01	−2.83 ± 3.02	−8.44 ± 1.31	259 ± 43	429 ± 69	${173}_{-12}^{+16}$	9.21 ± 2.18
Sr ii	5.77	1.55 ± 1.6	−9.44 ± 1.3	2268 ± 327	2319 ± 376	${152}_{-8}^{+20}$	11.32 ± 2.2

Download table as:  ASCIITypeset image

Table 2 also contains the peak amplitude of the Gaussians fit to the 1D cross-correlation functions for the beginning and end of the transit separately. The uncertainties for these measurements are given by taking the standard deviation of the noise outside of the peak and dividing by the square root of the number of pixels within the peak. We have also listed the FWHM values of the Gaussians and the associated errors produced by lmfit. Finally, we list the values of K_p (planet semi-amplitude velocity) that give the largest S/N, as seen in the right panels of Figure 4 for each detection or tentative detection. The uncertainties here are simply the K_p values corresponding to a decrease in S/N of 1 from the peak. For any of the plots that were multi-peaked (i.e., Ca ii), we made sure that the uncertainties extended to include both peaks.

6.1. Dependence on Model Parameters

We plotted the results from Table 2 in Figure 5 to determine if there were any overarching trends that spanned all of the detected species and would help to explain the results. For each relation, we calculate a Pearson correlation coefficient to determine if the two variables are correlated and the strength of the relation. Pearson correlation coefficient values larger than 0.5 or less than −0.5 are considered to show correlation, while values close to 0.0 are uncorrelated. The top-left plot shows the planet semi-amplitude velocity (K_p ) versus the radial velocity difference. It is intuitive that K_p and the radial velocity difference are correlated as they are essentially measuring the same thing; if RV₂ is more blueshifted than RV₁, the highest S/N would result from a shallower slope, and hence smaller K_p . We measured a Pearson correlation coefficient of −0.69, which denotes moderately strong negative correlation. The only point that does not clearly fit the trend (although is still within the uncertainty) is Ca ii, which has a double-peaked structure in both the K_p versus v_wind diagram and in the 1D cross-correlation function for the first half of the transit.

Zoom In Zoom Out Reset image size

The top right panel of Figure 5 shows the relative amplitude difference versus the radial velocity difference. While there does not seem to be a clear trend between these parameters (Pearson coeff. = −0.2), by examining the difference in amplitude, we can gain information about the temperature structure of the planet. All of the transition metals (V i, Cr i, Mn i, Fe i, Ni i, and a tentative detection of Co i) and Ca ii show stronger absorption during the second half of the transit when the eveningside of the planet is more in view. Li i, Mg i, and Sr ii show slightly stronger absorption during the second half of the transit, while Na i and K i actually show stronger absorption during the first half of the transit when the morningside of the planet is more in view. Chemical modeling from Kataria et al. (2016) predicted that Na i and K i would be more abundant on the nightsides than the daysides of UHJs, and so we hypothesize that they are also more abundant on the cooler morningside of the planet than on the hotter eveningside, where Na and K are expected to be mostly ionized. In contrast, most species are expected to show stronger absorption on the eveningside where the temperature is higher, and hence scale height is larger.

In the middle panels of Figure 5, we plot the ionization temperature (for ions) or the condensation temperature (for the neutral atoms) versus the radial velocity difference and the measured K_p . The relation containing the radial velocity difference shows the highest degree of correlation, with a Pearson value of 0.88, which signals strong positive correlation. As expected, we find a similar relation between the ionization or condensation temperature and K_p , and the Pearson coefficient still points toward moderate correlation with a value of −0.511.

Li i and Na i both show very little radial velocity difference between the beginning and end of the transit, and with condensation temperatures lower than 800 K, both atoms are expected to remain in the gas phase even on the cold nightside of the planet. All of the transition metals have mildly blueshifted (∼−3 km s⁻¹) absorption at the beginning of the transit and absorption that is more blueshifted (∼−10 km s⁻¹) at the end of the transit. The radial velocities of the first half and second half are offset by more than 2σ for all except for Ni i, which has consistent radial velocities with the other species, but larger uncertainties due to the lower S/N of the detection. The transition metals also all have comparable condensation temperatures that would result in their removal from the gas phase on the nightside of the planet. The ions exhibit the largest discrepancies between their two radial velocities and also are expected to be highly temperature sensitive. The correlation between the ionization or condensation temperature and the strength of the asymmetry could point toward condensation or chemical gradients causing the asymmetries, as species with more uniform distributions across the planet's surface are expected to show less of an asymmetry (Wardenier et al. 2021).

Finally, in the bottom panels of Figure 5, we plot the FWHM versus the radial velocity difference (left) and K_p (right). The absorption spectra of atoms such as H i, Li i, Na i, and Ca ii, are dominated by a few, very strong lines. The line cores of species that have stronger lines (or are more abundant) become optically thick at higher altitudes. Our measurements of the residual amplitudes of these species are on average larger than the residual amplitudes of the transition metals (thousands instead of hundreds), confirming that the signals originate higher in altitude. The FWHMs of these signals are also on average larger than those of the transition metals, which tend to have FWHMs of ∼10 km s⁻¹. These broad absorption lines were seen in Seidel et al. (2021), and explained as originating from a vertical wind in the upper atmosphere, which significantly broadens the lines. The trend showing K_p versus FWHM has a Pearson coefficient of 0.756, indicating strong correlation. The correlation can be explained by a day-to-night wind in the lower atmosphere that causes an asymmetry but minimal broadening, while the upper atmosphere (or exosphere) is dominated by a vertical wind or outflow and is not highly rotation phase sensitive but significantly broadens the lines.

To disentangle these two physical scenarios and to better interpret these signals, more GCM and chemical modeling is required. As GCMs are generally used to model the lower atmosphere (<1 microbar) and absorption such as that from Ca ii extend significantly higher in the atmosphere (Tabernero et al. 2021), nonhydrostatic GCM models that include atmospheric escape (e.g., Dobbs-Dixon & Agol 2013; Mayne et al. 2014; Mendonça et al. 2016) may be needed to fully explain the signals we detect. As both the temperature relation and the FWHM exhibit outliers, a combination of both condensation and a two-layered atmosphere could also best explain the measurements outlined here.

7.1. Nondetections

We plot all of the observability scores, as defined in Section 2 and shown in Figure 1, as well as our detections and nondetections in Figure 6. For each nondetection, we did not find a peak with an S/N > 4 near the exoplanet's expected velocity, and Figures 7 (neutral atoms) and 8 (ions) show the resulting K_p versus v_wind parameter space search. Most of the species that have high observability scores in Figure 1 are detected. The only species that are not detected or tentatively detected that have an observability score above 0.25 that we do not find are: Ca i, Ti i, Ti ii, Sc i, Sc ii, Al i, and Zr i. Although we do not detect Ca i, we do detect Ca ii, which leads us to conclude that Ca is mostly ionized at the temperatures and pressures probed in the terminator of WASP-76b.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Since we expect to be able to detect both Ti i and ii, as well as Sc i and ii, these nondetections cannot be explained by ionization. Sc i and ii are only on the edge of what we expect to be able to detect, and so it is not overly surprising that we do not find these species. In contrast, Ti i is expected to be the second most observable atom or ion after Fe i. Ti ii has been observed in the hottest UHJ, Kelt-9b (Hoeijmakers et al. 2019), but neither Ti i nor II has been confidently detected in any of the other cooler UHJs, including WASP-121b (Hoeijmakers et al. 2020a; Merritt et al. 2021) and MASCARA-2b (Hoeijmakers et al. 2020b). Furthermore, TiO was also not detected in WASP-121b (Hoeijmakers et al. 2020a; Merritt et al. 2020) or previously in WASP-76b (Tabernero et al. 2021). Ti i, Ti ii, and TiO make up the main gas-phase Ti-bearing species for the range of temperatures (2000–4000 K) expected along the terminator of WASP-76b (Lodders 2002), and as none of them are detected, this supports the hypothesis of Ti being trapped in condensates (e.g., Spiegel et al. 2009; Parmentier et al. 2013).

Since we do not expect to measure ionized Al and Zr, a significant portion of both of these species may be ionized, and therefore the neutral atoms may be less observable than our estimates and explain the nondetections. Alternatively, Al could have condensed into clouds; Al₂O₃ condenses at the highest temperature (∼2000 K) of all of the common condensate species shown in Gao et al. (2021). As we will discuss later when we compare our results to recent GCM modeling results (Section 7.3), Al₂O₃ clouds are invoked in Savel et al. (2021) to explain the Fe i asymmetry. We recover strong detections of Mg and Li even though these species are expected to be of similar or even lower observability, and so our nondetection of the strong Al i doublet at 4000 Å could be evidence supporting the cloud hypothesis put forth by Savel et al. (2021). In addition to Al₂O₃ clouds, the cloud species with the second highest condensation temperature in Gao et al. (2021) is TiO₂, and so condensation could potentially explain the intriguing nondetections of both Ti and Al.

7.2. Comparisons to Previous Observations of WASP-76b

7.3. Comparisons to GCM Modeling Results

We searched for absorption from >30 atoms and ions in the atmosphere of WASP-76b using the cross-correlation technique on two archival ESPRESSO transits. We detected absorption from Li i, Na i, Mg i, Ca ii, V i, Cr i, Mn i, Fe i, Ni i, and Sr ii, and tentatively detect H i, K i, and Co i. The detected absorption from V i, Cr i, Ni i, Sr ii, and Co i are novel for this planet.

For each of the detected or tentatively detected atoms and ions, we analyzed the beginning of the transit separately from the end of the transit to measure two different radial velocities and amplitudes of the planet absorption. We measure stronger absorption during the second half of the transit for all of the transition metals, and stronger absorption during the first half of the transit for Na i and K i. This observation points toward a temperature asymmetry between the morning and eveningside of the planet, as Na i and K i are expected to be less abundant (mostly ionized) on the hotter eveningside of the planet, while the transition metals are more visible on this side of the planet due to the hotter atmosphere, which has a larger scale height.

We find evidence for a trend between the strength of the measured asymmetry (difference between the expected radial velocity at the beginning and end of the transit) and the condensation or ionization temperature. Na i and Li i show consistent radial velocities, while the transition metals show differences of ∼7 km s⁻¹, and the ions show the largest asymmetries. This could be evidence that rain-out or recombination in the cooler morningside of the planet is causing the asymmetries. Alternatively, the differing degrees of asymmetry could be caused by the different species absorbing at different locations in the atmosphere, as we find correlation between the FWHM of the signals and the strength of the asymmetry. The lower atmosphere could be dominated by a day-to-night wind, which would cause narrower absorption lines with varying Doppler shifts, while the upper atmosphere could be dominated by a vertical wind or outflow, which would cause less phase dependence but broader lines.

Finally, we compare all of the detection S/Ns to what we expected to be able to detect and find a few notable absences. Ti i and Ti ii should both be observable, but we find neither, and Tabernero et al. (2021) also did not detect TiO. Furthermore, Al i has a strong doublet at 4000 Å, which is also not detected. These nondetections could point toward condensation into Al₂O₃ and TiO₂ clouds, as these are both expected to condense at temperatures around 2000 K (Gao et al. 2021). More GCM and chemical modeling, especially disequilibrium chemical modeling, would help to interpret these nondetections, as well as the many new time resolved asymmetric Doppler shifts observed in the atmosphere of WASP-76b.

A.K., I.S., N.C., and A.S. acknowledge funding from the European Research Council under the European Union's Horizon 2020 research and innovation program under grant agreement No. 694513. P.M. acknowledges support from the European Research Council under the European Union's Horizon 2020 research and innovation program under grant agreement No. 832428-Origins. The authors would like to thank the ESPRESSO team for sharing their data, without which we could not perform this work. We would also like to thank Arjun Savel for providing extra insight on recent GCM modeling work of WASP-76b. Finally we would like to thank Karan Molaverdikhani for computing many of the atom and ion opacities available in petitRADTRANS.