ACCESS: An Optical Transmission Spectrum of the High-gravity Hot Jupiter HAT-P-23b

We observed five transits of HAT-P-23b between 2016 and 2018 with the 6.5 m Magellan Baade Telescope and Inamori-Magellan Areal Camera and Spectrograph (IMACS, Dressler et al. 2006) as part of ACCESS, a multi-institutional effort to build a comparative library of visual transmission spectra of transiting exoplanets. For these observations, we used the IMACS f/2 camera. This camera has a 274 diameter field of view (FoV), in which we can observe simultaneously the target and several comparison stars, allowing for effective removal of common instrumental and atmospheric systematics. We selected comparison stars less than 0.5 mag brighter and 1 mag fainter than HAT-P-23b and closest in B − V/J − K color space, following Rackham et al. (2017). We show the selected comparison stars in Table 1.

Table 1. Target and Comparison Star Magnitudes and Coordinates from https://vizier.u-strasbg.fr/viz-bin/VizieR?-source=I/322

Star	R.A.(J2000) (h:m:s)	Decl. (J2000) (d:m:s)	B	V	J	K	Da
HAT-P-23	20:24:29.7	16:45:43.8	13.4	12.1	11.1	10.8	0.00
comp4	20:23:29.4	16:49:08.0	13.5	12.5	10.7	10.0	0.50
comp5	20:24:15.1	16:42:08.4	13.8	12.7	11.0	10.4	0.36

Note.

^a $D\equiv \sqrt{{[{\left(B-V\right)}_{c}-{\left(B-V\right)}_{t}]}^{2}+{[{\left(J-K\right)}_{c}-{\left(J-K\right)}_{t}]}^{2}}$, where c and t refer to the given comparison star and HAT-P-23, respectively, in magnitudes.

Download table as:  ASCIITypeset image

For ACCESS we have used several combinations of custom multislit masks, grisms, and blocking filters over time to refine our observations as the survey progresses. For the first transit of HAT-P-23b, in particular, we used 20'' × 12'' wide slits for the science masks to adequately sample the sky background and narrower, 20'' × 05, slits for the calibration masks to perform high-precision wavelength calibration. We used the 150 lines mm⁻¹ grism blazed at 188 (150–18.8) with the WB5600 blocking filter for this setup. We switched to a new field placement and a new science mask with wider 90'' × 10'' slits for the last four transits to better sample the sky and get more uniform wavelength coverage. We used the 300 lines mm⁻¹ grism blazed at 175 (300–17.5) with the GG495 blocking filter for the second and third transits, the 300 lines mm⁻¹ grism blazed at 175 (300–17.5) with the GG455 blocking filter for the fourth transit, and the 300 lines mm⁻¹ grism blazed at 267 (300–26.7) with the GG455 blocking filter for the fifth transit.

On average, these setups provide a resolving power of R ∼ 1200, or approximately 4.7 Å per resolution element, and access a full wavelength coverage of 5050–9840 Å. In practice, the signal-to-noise ratio (S/N) blueward of ≈5400 Å and redward of ≈9300 Å dropped to less than roughly 25% of peak counts. For this reason, we omitted measurements outside of this range for the rest of the study. Finally, we omitted data taken during twilight and at airmasses of Z ≥ 2, where large deviations in the white-light curve became apparent.

We summarize our observing configurations and conditions for each night in Table 2. We also note that the difference in wavelength coverage between nights comes from our choice to use wavelength ranges that maximized the integrated flux for a given night over uniformity in wavelength coverage. This might produce slight variations in the observed wavelength-binned transit depths, which can contribute to spurious slopes in the final transmission spectrum (McGruder et al. 2020). For HAT-P-23b, however, we recover a transmission spectrum with no typical activity-related slope (see Section 7), so this does not appear to be a significant effect in this case.

We observed the first transit (Transit 1) on 2016 June 22, the second (Transit 2) on 2017 June 10, the third (Transit 3) on 2018 June 4, the fourth (Transit 4) on 2018 June 21, and the fifth (Transit 5) on 2018 August 22. We used blocking filters for all nights to reduce contamination from light at higher orders, while also truncating the spectral range to 5300–9200 Å. We made observations in multi-object spectroscopy mode with 2 × 2 binning in fast readout mode (31 s) for all nights to take advantage of the reduced readout noise. Depending on the instrument setup and seeing conditions, we adjusted the individual exposure times between 15 and 100 s to keep the number of counts per pixel between 30,000 and 49,000 counts (ADU; gain = 1 e⁻/ADU on f/2 camera), i.e., within the linearity limit of the CCD (Bixel et al. 2019).

With the calibration mask in place, we took a series of arcs using a HeNeAr lamp before each transit time-series observation. The narrower slit width of the calibration mask increased the spectral resolution of the wavelength calibration and also avoided saturation of the CCD from the arc lamps. We took a sequence of high-S/N flats with a quartz lamp through the science mask to characterize the pixel-to-pixel variations in the CCD. As in our past studies (Rackham et al. 2017; Bixel et al. 2019; Espinoza et al. 2019; McGruder et al. 2020; Weaver et al. 2020), we did not apply a flat-field correction to the science images because it introduces additional noise in the data and does not improve the final results.

During the Transit 5 observation, an incorrect grism (300–26.7) was mistakenly inserted into the instrument in place of the intended 300–17.5 grism used on previous nights. This led to the grism metadata stored in the header files being incorrectly labeled as the 300–17.5 grism instead of 300–26.7, which we did not notice until after the data had been collected and the spectra examined and compared with previous data. The end result is data that are still usable, but slightly less dispersed relative to the other nights (1.25 Å pixel⁻¹ versus 1.341 Å pixel⁻¹), which we account for in the rest of our analysis. We show representative science frames from the CCD array in Figure 10 of the Appendix.

We reduced the raw data using the ACCESS pipeline described in previous ACCESS publications (Rackham et al. 2017; Bixel et al. 2019; Espinoza et al. 2019; McGruder et al. 2020; Weaver et al. 2020). The detailed functions of the pipeline, including standard bias and flat calibration, bad pixel and cosmic-ray correction, sky subtraction, spectrum extraction, and wavelength calibration are described in Jordán et al. (2013) and Rackham et al. (2017). We briefly summarize the data reduction here.

We applied the wavelength solution found with the arc lamps to the first science image spectra and the remaining image spectra were then cross-correlated with the first image. We calibrated the spectra from all stars to the same reference frame by identifying shifts between the Hα absorption line minimum of the median spectra and the air wavelength of Hα, and interpolated the spectra onto a common wavelength grid using b-splines. We aligned all spectra to within 2 Å, which is less than our average resolution element of 4.7 Å (see Section 2.1).

The final results of the pipeline are sets of wavelength-calibrated and extracted spectra for the target and each comparison star for each night as shown in Figure 1. We use those spectra to produce integrated white-light (Figure 11 in the Appendix) and spectrally binned light curves. The series of white-light curves produced in this fashion informed which comparison stars to omit in the rest of the analysis on a per-transit basis, which we describe in Section 3.3.

Zoom In Zoom Out Reset image size

For our main analysis, we used a set of semi-uniform bins with a main width of 200 Å, which we found to be a good compromise between spectral resolution and signal-to-noise considerations. In addition, we interspersed narrower bins centered around the air wavelength locations of Na i-D, K i, and Na i-8200 with widths of 185.80 Å, 54.00 Å, and 160.80 Å, respectively. We made the K i bin relatively narrow to avoid nearby tellurics. To check the results from our analysis based on this wavelength binning scheme, we separately used a set of five bins centered around each of these three species, set by the full width of their respective absorption lines (45, 10, and 40 Å). We share our two wavelength bin schemes in Tables 5 and 6 of the Appendix.

For each transit, we divided the white-light curve of HAT-P-23 by each of the comparison stars common to each night (comp4 and comp5) to select which data points to use for each night (Figure 11). For a given comparison star, we defined fluxes more than 2σ apart from adjacent points in the divided white-light curve to be potential outliers for that night, where 1σ was defined to be the photometric noise for that night. Because this set could contain false positives caused by an actual non-outlying point surrounded by two positive outliers, we then manually went back and removed points of this type from the potential set of outliers. Conversely, we also manually included false negatives in the outlier set. These were points not initially marked as outliers because they were part of an overall feature, for example the large decrease in flux. Finally, we omitted the union of these selected outliers from all nights from the rest our of analysis. With this set of selected data points and comparison stars for each transit epoch, we next performed the detrending of the resulting light curves as described in Section 4.

We used Gaussian process (GP) regression techniques combined with principal component analysis (PCA) to detrend our reduced data, following Espinoza et al. (2019). GPs are a flexible and robust tool for modeling data in the machine learning community (Rasmussen & Williams 2005) that has gained popularity in the exoplanets field (see, e.g., Aigrain et al. 2012; Gibson et al. 2012). Gibson et al. (2012) provides a good overview of this methodology applied to exoplanet transit light curves. Applying this methodology for a collection of N measurements ( f ), such as the flux of a star measured over a time series, the log marginal likelihood of the data can be written as

$$\begin{eqnarray}&&{\rm{log}}{ \mathcal L }({\boldsymbol{r}}| {\boldsymbol{X}},{\boldsymbol{\theta }},{\boldsymbol{\phi }})=-\displaystyle \frac{1}{2}{{\boldsymbol{r}}}^{{\rm{\top }}}{{\rm{\Sigma }}}^{-1}{\boldsymbol{r}}-\displaystyle \frac{1}{2}{\rm{log}}\left|{\rm{\Sigma }}\right|-\frac{N}{2}{\rm{log}}(2\pi ),\end{eqnarray} \tag{ 1 }$$

where r ≡ f − T( t , ϕ ) is the vector of residuals between the data and analytic transit function T; X is the N × K matrix for K additional parameters, where each row is the vector of measurements x _n = (x_n,1, ⋯, x_n,K ) at a given time n; θ are the hyperparameters of the GP; ϕ are the transit model parameters; and Σ is the covariance of the joint probability distribution of the set of observations f . In our analysis, we used six systematics parameters: time, FWHM of the spectra on the CCD, airmass, position of the pixel trace through each spectrum on the chip of the CCD, sky flux, and shift in wavelength space of the trace.

We used the Python package batman (Kreidberg 2015) to generate our analytic transit model. We used the open-source package ld-exosim ¹⁸ (Espinoza & Jordán 2016) to identify the appropriate limb darkening law for this model, which we found to be the logarithmic law. From here, the log posterior distribution $\mathrm{log}{ \mathcal P }({\boldsymbol{\theta }},{\boldsymbol{\phi }}| {\boldsymbol{f}},{\boldsymbol{X}})$ can be determined by placing explicit priors on the maximum covariance hyperparameters and the scale length hyperparameters. From ${ \mathcal P }$, the transit parameters can then be inferred by optimizing with respect to θ and ϕ .

We accomplished the above optimization problem with the Bayesian inference tool, PyMultiNest (Buchner et al. 2014), using 1000 live points and computed the log likelihoods from the GP with the george (Ambikasaran et al. 2015) package. We implemented this detrending scheme by simultaneously fitting the data with a Matérn-3/2 kernel for the GP under the assumption that points closer to each other are more correlated than points farther apart. The PCA methodology follows from Jordán et al. (2013) and Espinoza et al. (2019), where M signals, S_i (t), can be extracted from M comparison stars and linearly reconstructed according to the eigenvalues, λ_i , of each signal. This allows for the optimal extraction of information from each comparison star to inform how the total flux of HAT-P-23 varies over the course of the night.

In past work, we depended on the ratio of the semimajor axis to stellar radius (a/R_s) as an input transit parameter. With Gaia, we are now able to invert the problem by using the newly determined measurements of the stellar density (ρ_s ) from Gaia Data Release 2 (Gaia Collaboration et al. 2018) to place tighter constraints on a/R_s. Applying our standard assumption of a circular Keplerian orbit with period P then yields $a/{R}_{{\rm{s}}}={[({{GP}}^{2})/(3\pi )]}^{1/3}{\rho }_{{\rm{s}}}^{1/3}$, where G is the gravitational constant. In addition, we implemented the technique introduced by Espinoza (2018) for reparameterizing the impact parameter (b) and planet-to-stellar radius ratio (p ≡ R_p/R_s) to further improve our transit parameter sampling. In this technique, these transit parameters are expressed as functions of the random variables r₁ ∼ U(0, 1) and r₂ ∼ U(0, 1), and efficiently sampled from a region bounded by the physical constraints of transiting systems (i.e., 0 < b < 1 + p). This is a natural analog to the uniform sampling method of limb darkening coefficients introduced in Kipping (2013).

Finally, we Bayesian model-averaged the principal components together, which we determined by detrending with one, then two, up to M principal components, to create the final detrended white-light curves and model parameters of interest. We present the resulting white-light curves in Figure 2, the best-fit parameters in Table 3, and associated corner plots in Figure 12 of the Appendix. To create the transmission spectrum for each night (Figure 4), we applied the same detrending methodology to the wavelength-binned light curves (described in Section 3.2), keeping all transit parameters fixed to the fitted white-light curve values except for the transit depth and limb darkening parameters. Because the impact parameter needed to be held fixed for the binned fitting, we sampled directly for it instead of uniformly sampling in (b, p) space. We used a truncated normal distribution for this sampling, given by

$$\begin{eqnarray}&&\xi =\displaystyle \frac{x-\mu }{\sigma },\alpha =\displaystyle \frac{a-\mu }{\sigma },\beta =\displaystyle \frac{b-\mu }{\sigma }\end{eqnarray} \tag{ 2 }$$

$$\begin{eqnarray}&&Z={\rm{\Phi }}(\beta )-{\rm{\Phi }}(\alpha ),\end{eqnarray} \tag{ 3 }$$

where x is the binned planet-to-star radius ratio, μ is the mean fitted white-light curve depth, σ is the corresponding 1σ uncertainty on that depth, a = 0, b = 1, and Φ( · ) is the cumulative distribution function.

Zoom In Zoom Out Reset image size

Table 3. Fitted Model-averaged Transit Parameters from the GP+PCA Detrending Method Shown in Figure 2

Parameter a	Transit 1	Transit 2	Transit 3	Transit 4	Transit 5
δ	${12584.79890}_{-647.93827}^{+633.98979}$	${12716.26467}_{-459.31331}^{+421.08962}$	${13571.18577}_{-650.42694}^{+598.68259}$	${13204.48610}_{-465.37117}^{+472.94430}$	${12831.57340}_{-489.88214}^{+456.65963}$
Rp/Rs	${0.11218}_{-0.00289}^{+0.00283}$	${0.11277}_{-0.00204}^{+0.00187}$	${0.11650}_{-0.00279}^{+0.00257}$	${0.11491}_{-0.00202}^{+0.00206}$	${0.11328}_{-0.00216}^{+0.00202}$
t0	${4852.26537}_{-0.00015}^{+0.00015}$	${4852.26540}_{-0.00014}^{+0.00014}$	${4852.26541}_{-0.00013}^{+0.00014}$	${4852.26545}_{-0.00015}^{+0.00015}$	${4852.26543}_{-0.00014}^{+0.00013}$
P	${1.21289}_{-1.14e-07}^{+1.13e-07}$	${1.21289}_{-8.76e-08}^{+8.45e-08}$	${1.21289}_{-8.78e-08}^{+8.76e-08}$	${1.21289}_{-7.84e-08}^{+7.42e-08}$	${1.21289}_{-7.13e-08}^{+7.00e-08}$
ρs	${1.07230}_{-0.07603}^{+0.08406}$	${0.96894}_{-0.05683}^{+0.06463}$	${1.01504}_{-0.05063}^{+0.05602}$	${1.03666}_{-0.06079}^{+0.06581}$	${1.02626}_{-0.07204}^{+0.07229}$
i	${84.33384}_{-0.80101}^{+0.88921}$	${83.14471}_{-0.61691}^{+0.73406}$	${83.64261}_{-0.51338}^{+0.56957}$	${83.76319}_{-0.64068}^{+0.69346}$	${83.73923}_{-0.74358}^{+0.73109}$
b	${0.43173}_{-0.05892}^{+0.04848}$	${0.50417}_{-0.04412}^{+0.03445}$	${0.47517}_{-0.03537}^{+0.02969}$	${0.46904}_{-0.04320}^{+0.03843}$	${0.46936}_{-0.04503}^{+0.04278}$
a/Rs	${4.36890}_{-0.10580}^{+0.11130}$	${4.22377}_{-0.08424}^{+0.09190}$	${4.28971}_{-0.07254}^{+0.07750}$	${4.31996}_{-0.08614}^{+0.08955}$	${4.30547}_{-0.10320}^{+0.09880}$
u	${0.24859}_{-0.10754}^{+0.10290}$	${0.38699}_{-0.07472}^{+0.07698}$	${0.37032}_{-0.11732}^{+0.10186}$	${0.40126}_{-0.09298}^{+0.08643}$	${0.32686}_{-0.08796}^{+0.07151}$

Notes. We share the associated corner plots in Figure 12 of the Appendix.

^a Parameter definitions: $\delta ={\left({R}_{{\rm{p}}}/{R}_{{\rm{s}}}\right)}^{2}$—transit depth (ppm), R_p/R_s—planet radius/stellar radius, t₀ mid-transit time (JD – 2,450,000), P—period (days), ρ_s —stellar density (g cm^–3), i—inclination (deg), b—impact parameter, a/R_s—semimajor axis/stellar radius, u—linear limb darkening coefficient.

Download table as:  ASCIITypeset image

The uncertainties in the wavelength-binned transit depths from our individual transits (Table 7) range from 208 ppm to 1445 ppm, which is not precise enough to detect the atmosphere of HAT-P-23b. For example, a hydrogen-dominated composition for the atmosphere of HAT-P-23b would produce a signal ΔD of 384 ppm at five scale heights (based on Equation (11) of Miller-Ricci et al. (2009), using the most up-to-date system parameters from the literature shared in Section 1). Therefore, we needed to combine the transmission spectra from the five transits to be sensitive to atmospheric features of the planet.

We offset each transmission spectrum by the difference in its white-light curve depth from the weighted mean white-light curve depth for all nights and then average the offset transmission spectra together, weighted by the wavelength-dependent uncertainty estimated from the wavelength-binned fitting. We combined the asymmetrical uncertainties according to Model 1 of Barlow (2003) and present the final results in Figure 4.

Zoom In Zoom Out Reset image size

To verify our methodology, we also repeated our analysis up to this point using the wavelength binning scheme described in Section 3.2 that is centered around Na i-D, K i, and Na i-8200. The resulting combined transmission spectrum (Figure 5) shows no excess absorption near the species we examined, in agreement with our retrieval analysis discussed in Section 7.

Zoom In Zoom Out Reset image size

We used a generic grid of forward model transmission spectra as detailed in Goyal et al. (2019) to interpret the observations of HAT-P-23b. This grid of models has been generated using the 1D–2D radiative–convective equilibrium model ATMO (Amundsen et al. 2014; Tremblin et al. 2015; Drummond et al. 2016; Goyal et al. 2018). The model transmission spectra in the grid are generated for a Jupiter-like (radius and mass) planet around a Sun-like star. However, this can be scaled to any hot Jupiter exoplanet with H₂–He-dominated atmosphere using a scaling relationship detailed in Goyal et al. (2019). Each model in the grid assumes chemical equilibrium abundances and isothermal pressure–temperature (P–T) profiles. The model spectra include H₂–H₂, H₂–He collision-induced absorption and opacities due to H₂O, CO₂, CO, CH₄, NH₃, Na, K, Li, Rb, Cs, TiO, VO, FeH, PH₃, H₂S, HCN, SO₂, and C₂H₂. The entire grid spans 24 equilibrium temperatures from 300 to 2600 K in steps of 100 K, six metallicities (0.1–200 × solar), four planetary gravities (5–50 m s⁻²), four C/O ratios (0.35–1.0), two condensation regimes (local and rainout condensation), and four parameters each describing scattering hazes and uniform clouds. The haze parameter is defined as the multiplicative factor to the wavelength-dependent Rayleigh scattering due to small particles, while the cloud parameter is defined as the multiplicative factor to uniform gray scattering, simulating the effects of a cloud deck, where a factor of 0 indicates no clouds, while a factor of 1 indicates extremely cloudy. Further details can be found in Goyal et al. (2019).

7.1.1. Forward Model Results and Interpretation

The best-fit forward model is found by scaling each model in the grid to the parameters of HAT-P-23b and finding the model with the lowest χ² throughout the entire grid. Figure 6 shows this best-fit forward model spectrum with the χ² value of 34.25 and reduced χ² value of 2.14 with 16 degrees of freedom (17 data points minus 1). We also note that while fitting the forward model grid to the observed spectrum, a wavelength-independent constant vertical offset (transit depth) between model and observed spectra is a free parameter. The best-fit model is consistent with a temperature of 2500 K, ten times solar metallicity, and C/O ratio of 0.35, without any haze or clouds. For both the condensation regimes (local and rainout), we find the best-fit model is the same, which is driven by the very high best-fit temperature (2500 K), where rarely any condensation occurs. Figure 18 shows the χ² map of all the models in the grid when fitted to observations, along with contours of confidence intervals. There is a clear preference for high-temperature models. There is also a preference for low C/O ratio models within 1σ, due to lack of spectral features of any carbon-based species. However, the other parameters are less constrained. The rise in the transit depth that begins after 4000 Å in this model spectrum is due to TiO/VO opacity. Thus the best-fit forward model for HAT-P-23b hints toward the presence of TiO/VO in its atmospheric limb.

Zoom In Zoom Out Reset image size

With this key insight from our forward modeling analysis, we next performed a retrieval analysis with exoretrievals similar to the one described in Weaver et al. (2020), with the key difference being now holding the spot parameters fixed to values informed from the host star's photometry. We explored a suite of models for the bounding spot parameters (T_sp,lower, f_sp,lower) = (2200 K, 0.022) and (T_sp,upper, f_sp,upper) = (3800 K, 0.026), and atmospheric models including Na i, K i, and TiO, and assuming: (i) a clear atmosphere (clear), (ii) an atmosphere with clouds (clear+cloud), (iii) an atmosphere with hazes (clear+haze), and a clear atmosphere in the presence of a starspot (clear+spot).

We ran these various models both fitting for the mean reference radius, R₀, and leaving it fixed to our weighted mean white-light value (using the R_p/R_s values in Table 3), to investigate its degeneracy with clouds. All models used the updated value for the stellar radius (R_s = 1.152 ${R}_{\odot }$ ) from Stassun et al. (2019). In total, we explored 224 possible retrievals: 8 combinations of the model types (clear, cloud, haze, spot) × 7 combinations of the three species × 4 combinations of the fixed spot temperatures/covering fractions with and without fitting for R₀.

The suite of retrieval models described above all returned consistent results marginally favoring TiO being present in the atmosphere of HAT-P-23b, with the most favored model corresponding to a subsolar to supersolar TiO abundance of 10^−7.1±4. We share a representative summary of the Bayesian log-evidence of each model relative to the model with the lowest log-evidence in Figure 7 and associated priors in Table 4. We also show a selection of the retrieved transmission spectra in Figure 8, and the summary statistics for the corresponding posterior distributions in Table 9 of the Appendix.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Table 4. Priors Used to Produce Figure 7 and 8

Parameter	Definition	Prior (Uniform)
f	reference planet radius normalization (fR0)	[0.8, 1.2]
$\mathrm{log}{P}_{0}$	log10 cloud top pressure (bar)	[−6, 6]
Tp	planet atmospheric temperature (K)	[100, 3000]
$\mathrm{log}$(species)	trace species mixing ratio	[−30, 0]
$\mathrm{log}\,{\sigma }_{\mathrm{cloud}}$	log10 gray opacity cross section	[−50, 0]
$\mathrm{log}\,a$	log10 haze amplitude	[−30, 30]
γhaze	haze power law	[−4, 0]
Tstar	occulted stellar temperature (K)	[5420, 6420]
Thet	starspot temperature (K)	[2200]
fhet	spot covering fraction	[0.022]

Download table as:  ASCIITypeset image

Although models including TiO in the atmosphere of the planet without contributions from stellar activity tended to be preferred in all retrieval combinations examined, the preference over stellar activity was not as large when fitting for R₀ (Figure 7, right panel). For that case, the log-evidence for every model was below 5, making the differences between specific models less statistically significant. New data collected in the UV and IR would allow for more robust retrieval results in this case. Finally, we note that because these retrievals are a 1D analysis, our retrieved terminator planet temperatures are consistently lower than the expected equilibrium temperature of HAT-P-23b, as predicted by MacDonald et al. (2020) and Pluriel et al. (2020).

The trend between exoplanets with known surface gravity and equilibrium temperature versus water feature strength in the near IR, proposed by Stevenson (2016), would predict that HAT-P-23b has a clear atmosphere. In the reanalysis of this relationship by Alam et al. (2020) using new data from the Hubble Space Telescope's Wide Field Camera 3 (HST/WFC3) (Wakeford et al. 2019), they note that this divide may not be as distinct as initially suggested. Because of its high surface gravity (g ≈ 30 m s⁻²) and relatively high equilibrium temperature (T_eq = 2027 K), HAT-P-23b tests a region of parameter space previously unexplored (see Figure 9). Measurements of its transmission spectrum in the J band (1.36–1.44 μm) and baseline region (1.22–1.30 μm) to determine the planet's water feature signal would provide insight into this potential trend in exoplanet atmosphere types. For example, a single transit in the 7500–18000 Å IR range with the G141 grism aboard HST/WFC3 would provide the novel data needed for such an analysis. In addition, a second transit in the UV/visible channel with the 2000–8000 Å grism (G280) (Wakeford et al. 2020) would provide the wide wavelength coverage necessary to perform a robust retrieval of the combined transmission spectrum and test our preliminary result.

Zoom In Zoom Out Reset image size