ACCESS and LRG-BEASTS: A Precise New Optical Transmission Spectrum of the Ultrahot Jupiter WASP-103b

WASP-103b, discovered by Gillon et al. (2014), is an ultrahot Jupiter ($M=1.51\pm 0.11$ M_J, $R={1.623}_{-0.053}^{+0.051}$ R_J, ${T}_{{\rm{eq}}}=2484\pm 67$ K; Delrez et al. 2018). Its high equilibrium temperature owes to the fact that it orbits its F8V host star in 0.93 days (Gillon et al. 2014; Delrez et al. 2018). Because of its very high temperature, WASP-103b has a large scale height of 600 km, which contributes 152 ppm to the transit depth (assuming the parameters from Delrez et al. 2018 and a mean molecular weight of 2.3 amu).

WASP-103b has been the target of atmospheric studies on several previous occasions. Southworth et al. (2015) presented a transmission photometry study of the planet using transits acquired in seven different photometric bands, from $g^{\prime}$ to $z^{\prime}$. They found a steep slope (gradient, $\alpha (\lambda )=19.0\pm 1.5$) in their transmission spectrum rising toward their bluest wavelength bins and extending over ∼10 atmospheric scale heights. Southworth et al. (2015) concluded unocculted starspots were unlikely to cause the observed slope due to the lack of the host star's photometric variability (Gillon et al. 2014) and the temperature of the host (${T}_{\mathrm{eff}}=6110$ K), which they suggested was too hot for significant spot-induced activity.

Southworth & Evans (2016) subsequently revisited the earlier transmission spectrum of Southworth et al. (2015) after new high-resolution photometry revealed the presence of a blended background star (Wöllert & Brandner 2015; Ngo et al. 2016). While correcting for this third light source did reduce the gradient of the slope in their data to $\alpha (\lambda )=11.2\pm 0.9$, it remained significantly steeper than expected for Rayleigh scattering ($\alpha (\lambda )=4$).

Turner et al. (2017) extended the observations of WASP-103b to the near-UV with ground-based observations of the planet in the U band. The transit depth they derived was consistent with the bluest bin of Southworth & Evans (2016), indicating that the optical slope may not continue to rise into the near-UV.

The presence of a slope in the optical transmission spectrum of WASP-103b was challenged by Lendl et al. (2017), who presented results using the multiobject spectrograph GMOS on the Gemini-North telescope. They found no signs of the strong slope reported by Southworth et al. (2015) and instead found the atmosphere to be cloud-free with absorption from sodium and potassium.

Following these studies in transmission, Cartier et al. (2017) presented Hubble Space Telescope (HST)/WFC3 secondary eclipse observations of WASP-103b. Their near-IR emission spectrum was featureless to a precision of 175 ppm, finding the planet's atmosphere to be approximately isothermal over the wavelength range probed (1.1–1.7 μm). However, they also noted that a thermally inverted atmosphere and a monotonically decreasing atmosphere with C/O $\gt 1$ were consistent with their data.

Delrez et al. (2018) presented 16 new ground-based secondary eclipses of WASP-103b in the $z^{\prime}$ and K_S bands, and combined these with the HST/WFC3 results of Cartier et al. (2017). They found the $z^{\prime}$ and HST/WFC3 observations to be best fitted with a vertically isothermal atmosphere. However, they detected an anomalously deep eclipse in the K_S band, which suggested the presence of a thermal inversion, but they noted that the K_S band is not expected to contain strong spectral features and so could not explain this feature. They additionally reanalyzed the transit data from Southworth et al. (2015), finding consistent results, indicating that the slope seen by Southworth & Evans (2016) is inherent to the data and not dependent on the data reduction method.

Kreidberg et al. (2018) presented phase-curve measurements of WASP-103b observed with HST/WFC3 and Spitzer/IRAC. They found the phase curves to show large amplitudes and negligible hot spot offsets, indicating poor circulation of heat from the day side to the night side of the planet. In the WFC3/IR/G141 bandpass (1.1–1.7 μm), Kreidberg et al. (2018) found the phase-resolved spectra to be consistent with blackbodies with brightness temperatures ranging from 1880 K on the night side to 2930 K on the day side. However, they observed a significantly higher brightness temperature in the Spitzer 4.5 μm band that is likely due to CO in emission, indicating a thermal inversion in the planet's atmosphere. As a result, Kreidberg et al. (2018) speculated that either TiO, VO, or FeH may be present in the planet's upper atmosphere. By running atmospheric retrievals on their infrared data, they found WASP-103b to have a ${23}_{-13}^{+29}\times$ solar metallicity atmosphere but did not detect water, which they attributed to its dissociation.

Most recently, Wilson et al. (2020) presented new VLT/FORS2 data of WASP-103b covering the wavelength region of 4000–6000 Å. By running retrievals on the combined VLT, Gemini, and HST transmission spectrum, they found the optical transmission spectrum to be featureless, with evidence for Na at 2σ, but they did detect H₂O at 4σ confidence.

Given the contrasting results regarding WASP-103b's optical transmission spectrum and the potential for a previously undetected species at optical wavelengths giving rise to the planet's temperature inversion, we present here a new optical transmission spectrum of WASP-103b. Our new transmission spectrum combines five transits from the ACCESS survey on Magellan/Baade (Jordán et al. 2013; Rackham et al. 2017; Bixel et al. 2019; Espinoza et al. 2019; Weaver et al. 2020, 2021; McGruder et al. 2020) and two transits from the LRG-BEASTS survey on the William Herschel Telescope (Kirk et al. 2017, 2018, 2019; Louden et al. 2017; Alderson et al. 2020). We combine these new transit observations with a reanalysis of the published Gemini/GMOS data (Lendl et al. 2017) and VLT/FORS2 data (Wilson et al. 2020) and include these with the published HST/WFC3/IR/G141 and Spitzer/IRAC transmission spectrum of Kreidberg et al. (2018) for our retrieval analyses.

This paper is organized as follows. We describe the observations in Section 2, the data reduction in Section 3, our light-curve fitting procedures in Section 4, and our corrections for third-light and night-side contamination in Section 5. We present the transmission spectrum of WASP-103b in Section 6 and interpret it using the petitRADTRANS and POSEIDON retrieval frameworks in Sections 7 and 8, respectively. We discuss our results in Section 9 and, finally, summarize our conclusions in Section 10.

We observed five transits of WASP-103b in the 2015–2017 observing seasons with the Inamori-Magellan Areal Camera and Spectrograph (IMACS; Dressler et al. 2011) on the 6.5 m Magellan Baade Telescope at Las Campanas Observatory. Our observation design largely mirrored that of previous ACCESS observations (Jordán et al. 2013; Rackham et al. 2017; Bixel et al. 2019; Espinoza et al. 2019; Weaver et al. 2020, 2021; McGruder et al. 2020). For each observation, we used the $f/2$ camera of IMACS in multiobject spectroscopy mode with 2 × 2 binning (04 pixel⁻¹). Conditions were clear for all observations except the first two nights ("ACCESS n1" and "ACCESS n2"), for which passing clouds impacted the data quality during the 2.6 hr transit for ∼1 hr and ∼0.25 hr, respectively. Further details of these observations are provided in Table 1.

Table 1. Log of Magellan/IMACS Observations

Transit	Date	Time	Airmass	Seeing	Grism	Mask	Filter	Frames	Exposure	Readout +
	(UTC)	(UTC)							Times (s)	Overhead (s)
ACCESS n1	5 Jun 2015	03:31–08:00	$1.28\to 1.24\to 2.19$	0.7–1.2	300 line mm−1 (175)	3215	Spectroscopic	261	33	29
ACCESS n2	6 Jun 2015	01:30–06:28	$1.74\to 1.24\to 1.47$	0.7–1.4	300 line mm−1 (175)	3215	Spectroscopic	289	33	29
ACCESS n3	2 Jul 2015	00:30–04:05	$1.49\to 1.24\to 1.34$	0.7–1.0	300 line mm−1 (175)	3215	WB5600–9200	210	33	29
ACCESS n4	5 Apr 2017	05:45–10:13	$1.65\to 1.24\to 1.40$	0.7–1.7	150 line mm−1 (188)	3397	WB5600–9200	348	10–25	31
ACCESS n5	1 May 2017	03:34–08:43	$1.89\to 1.24\to 1.45$	1.0–1.8	300 line mm−1 (175)	3397	GG495	262	40	31

Note. Mask numbers may be used to query information about the masks via the IMACS Slit Mask Manager (http://masks.lco.cl/search/).

Download table as:  ASCIITypeset image

We used a custom multislit mask with large slits ($10^{\prime\prime}$ wide by $21^{\prime\prime}$ long) for collecting spectra of WASP-103 and six comparison stars simultaneously. After the first three observations, we redesigned the mask so that each spectrum would only be dispersed across a single chip. The second mask includes six comparison stars as well, four of which are repeated from the first mask (Table 2). We designed the two science masks for use with the 300 line mm⁻¹ and 150 line mm⁻¹ grisms, respectively, though the second is also compatible with the 300 line mm⁻¹ grism over a narrower wavelength range. For each science mask, we also made a mask with narrow (05) slits for wavelength calibration. After the first two unfiltered observations, we began using the WB5600–9200 filter to block any potential contribution from second-order light. For our final Magellan/IMACS observation, we used the GG495 filter for increased sensitivity at bluer optical wavelengths.

We selected exposure times to provide a maximum count rate of ∼35,000 analog-to-digital units (ADU) in order to remain within the linear regime of the IMACS CCDs (see Section 3.8 of Bixel et al. 2019). We used the "turbo" readout mode for the first three transits and opted for the slightly slower and less noisy "fast" readout mode for the latter transits.

For each observation, we collected a series of quartz lamp flats with the same mask, disperser, and binning as the science frames. Immediately before or after the science observations, we also collected a series of exposures of HeArNe lamps through the narrow-slit calibration mask to aid with wavelength calibration.

The LRG-BEASTS transits of WASP-103b were observed on the nights of 2016 June 1 (referred to as "LRG-BEASTS n1" hereafter) and 2016 June 26 (referred to "LRG-BEASTS n2"). Both transits were observed with the low-resolution ($R\approx 300$) grism spectrograph ACAM (Benn et al. 2008) on the 4.2 m William Herschel Telescope (WHT) in La Palma. This is the same instrument as has been used in the previous five LRG-BEASTS studies (Kirk et al. 2017, 2018, 2019; Louden et al. 2017; Alderson et al. 2020).

Unlike IMACS, ACAM is a long-slit spectrograph. For both of the LRG-BEASTS nights, we used the wide 76 × 40'' slit, allowing us to simultaneously observe the spectrum of WASP-103 and a comparison star through the same slit. While a wide slit is useful to avoid differential slit losses between the target and the comparison, it does mean that other stars can fall within the slit and whose light can contaminate both the target and comparison. ²⁰ We therefore had to balance our desire to have a comparison star with a similar magnitude and color to the target while avoiding both possible contaminating stars and the edges of the detector. For our LRG-BEASTS observations of WASP-103, we selected the bright HD 149891 as our comparison star, which has a V magnitude of 8.7 and B − V color of 1.1. By comparison, WASP-103 has a V magnitude of 12.1 and B − V color of 0.6.

Given the brightness of the comparison, we had to use short, 9 s exposures to avoid saturation. In order to reduce the overhead, we used a fast readout mode and we windowed the CCD into two windows ([530:800, 1100:3100] for the target and [1302:1572, 1100:3100] for the comparison), which reduced the overhead to 5 s.

On the first night, we acquired 1400 spectra of the target with an airmass varying between 1.40 $\to$ 1.08 $\to$ 1.43. The moon did not rise during our observations on this night. On the second night we took 1068 spectra of the target with an airmass varying between 1.09 $\to$ 1.08 $\to$ 1.70, with a moon illumination of 58% at a distance of 108° from the target. Biases, tungsten lamp flats, sky flats, and CuAr plus CuNe arc spectra were taken at the start and end of both nights.

Three transits of WASP-103b were observed with the Gemini-North GMOS multiobject spectrograph on 2015 June 27 ("GMOS n1"), 2015 July 10 ("GMOS n2"), and 2016 May 3 ("GMOS n3"). A single transit of WASP-103b was observed using VLT/FORS2 on 2017 May 1. This was the same transit as observed by ACCESS ("n5"). For all GMOS and FORS2 transits, two comparison stars were observed in addition to WASP-103. We refer the reader to Lendl et al. (2017) and Wilson et al. (2020) for a description of these observations.

We reduced the Magellan/IMACS data using our custom Python-based pipeline described previously (Jordán et al. 2013; Rackham et al. 2017; Bixel et al. 2019; Espinoza et al. 2019; Weaver et al. 2020, 2021; McGruder et al. 2020). The pipeline performs standard procedures for multiobject spectroscopy, including bias and flat calibration, bad-pixel and cosmic-ray correction, sky subtraction, spectral tracing and extraction, and wavelength calibration. It outputs for each observation sets of wavelength-calibrated 1D spectra for the target and comparison stars and arrays of state parameters useful for detrending systematics. These include time, airmass, x and y detector positions of the spectra, sky background level, and the FWHM of the spectra. As with previous ACCESS studies (Rackham et al. 2017; Bixel et al. 2019; Espinoza et al. 2019; Weaver et al. 2020, 2021; McGruder et al. 2020), we found that flat-fielding introduced additional noise into our high-signal-to-noise-ratio science images, and so we ultimately chose not to apply a flat-field correction. To extract the 1D spectra, the ACCESS pipeline uses standard aperture photometry. For all nights, we used a 15 pixel-wide aperture, corresponding to 6'', which we selected to limit the impact of seeing variations while also leaving adequate space in the spatial direction for estimating the sky background.

The extracted, wavelength-calibrated mean-averaged spectra for each ACCESS observation of WASP-103 are shown in Figure 1.

Zoom In Zoom Out Reset image size

To reduce the LRG-BEASTS long-slit data, we used the same custom-made Python scripts as used in Kirk et al. (2017, 2018, 2019) and Alderson et al. (2020), and we refer the reader to those papers for a more in-depth description of the process.

In brief, the pipeline performs bias removal, tracing of the target and comparison spectra, and standard aperture photometry using a linear polynomial interpolated across the stellar spectra (in the spatial direction) to estimate the sky background, cosmic-ray removal, and wavelength calibration.

Master biases were created for each night and subtracted from the science data before extraction. For the first night 257 bias frames were median-combined to create a master bias, and 301 bias frames were combined for the second night. As with the ACCESS data and our previous LRG-BEASTS studies, we chose not to apply a flat-field correction. In a test-run on LRG-BEASTS n1, we found that the use of a flat-field led to an identical transmission spectrum but with uncertainties ∼10 % larger than without the use of the flat-field.

We experimented with our choice of aperture width to extract the stellar spectra, finding an aperture width of 26 pixels to be optimal for night 1 and 40 pixels for night 2 (owing to the poorer seeing conditions). The optimal aperture width was determined as the aperture width that produced the minimum standard deviation following a fit of an analytic transit light curve with a cubic in time polynomial to the resulting white light curve. We note that the plate scale of ACAM is 025 pixel⁻¹.

The extracted, wavelength-calibrated LRG-BEASTS spectra of WASP-103 are shown in Figure 1.

For the Gemini/GMOS and VLT/FORS2 transits, we used the stellar spectra as reduced by Lendl et al. (2017) and Wilson et al. (2020). Importantly, we followed the approach of both Lendl et al. (2017) and Wilson et al. (2020) in using only one of the two comparison stars observed in those studies. This was due to the second comparison star leading to an increase in the noise in the data.

Given the number of different transits used in our analysis and the subsequently different wavelength coverage, we opted for a straightforward binning scheme comprising 150 Å-wide wavelength bins from 3900 to 9450 Å. The bins and nightly mean-averaged spectra of WASP-103 from all 11 transits are shown in Figure 1. We note that we excluded the four bins of the Gemini/GMOS data that included chip gaps.

Given Lendl et al.'s (2017) finding of strong sodium and potassium in the atmosphere of WASP-103b in 20 Å-wide bins, we additionally constructed separate 20 Å-wide bins centered on the sodium doublet and each of the two absorption lines in the potassium doublet.

Using our 150 Å- and 20 Å-wide wavelength bins, we then created the spectroscopic light curves by integrating the spectra of the target and comparison stars over these wavelength ranges. The white light curves were created by integrating over the full wavelength range used for each night.

For the LRG-BEASTS, GMOS, and FORS2 data, the target's light curves were divided by the comparisons' light curves at this point to correct for telluric absorption. However, the ACCESS light curves were treated differently due to the principal component analysis (PCA) used in the ACCESS fitting routine (Section 4.1).

The white light curves for all 11 transits are shown in Figure 2.

Zoom In Zoom Out Reset image size

4.1.1. Model Setup

4.1.2. Application of the Model

4.2.1. Model Setup

4.2.2. Application of the Model

To fit the Gemini/GMOS and VLT/FORS2 light curves, we used the same gppm_fit code as we used to fit the LRG-BEASTS light curves. For the GMOS data we again used five squared-exponential GP kernels to fit the systematics in the data, with each taking one of airmass, FWHM, position angle, y position of the trace, and sky background as input variables. For the VLT data, we followed the approach of Wilson et al. (2020) in modeling the systematics with a single GP kernel, taking time as the input variable, combined with a linear in time polynomial. The white light-curve fits to the GMOS and FORS2 data are shown in Figure 2.

As with the ACCESS and LRG-BEASTS transits, we removed the common noise systematics model from the spectroscopic light curves prior to fitting. As we show in Section 6.3, our resulting GMOS and FORS2 transmission spectra are in good agreement with the initial analyses of Lendl et al. (2017) and Wilson et al. (2020), giving us confidence in the repeatability of our results.

As an additional test of our light-curve-fitting routines, we generated a set of ACCESS differential light curves by dividing each night by the single "best" comparison star (as shown in Figure 2). The best comparison star was chosen as the comparison star that leads to the smallest median absolute deviation in the out-of-transit flux. We then ran these light curves through the same procedure as described in Section 4.2. The results of this test are presented in Section 6.5 and show excellent agreement between GPTransSpec and gppm_fit.

The results of our fits to the white light curves are shown in Figure 2. We note that we also include the differential ACCESS light curves in Figure 2, which correspond to the target's light curve divided by the best comparisons' light curve, for ease of comparison with the other six transit light curves. However, we did not actually fit the differential ACCESS light curves but instead fit the raw stellar light curves as described in Section 4.1.

As the number of transits and wavelength bins resulted in over 322 spectroscopic light curves, we only include example fits to our ACCESS and LRG-BEASTS data in the Appendix (Figures 16 and 17). The rest of our light-curve fits can be found online. ²²

Table 3 lists the system parameters we derive from the "free system parameter" fits to our 11 white light curves. We note again that these results were not the ones used for our final transmission spectra, where we instead used the parameters of Southworth et al. (2015), as used by Kreidberg et al. (2018).

Table 3. The Resulting System Parameters from Our White Light-curve Fits with these as Free Parameters

Data set	TC (BJDTDB)	$a/{R}_{* }$	i	${R}_{P}/{R}_{* }$
ACCESS n1	${2457178.747991}_{-0.000248}^{+0.000215}$	${2.97}_{-0.06}^{+0.06}$	${85.14}_{-1.44}^{+1.8}$	${0.11535}_{-0.00306}^{+0.00258}$
ACCESS n2	${2457179.673866}_{-0.000128}^{+0.000123}$	${2.97}_{-0.05}^{+0.04}$	${86.2}_{-1.5}^{+1.89}$	${0.10947}_{-0.00261}^{+0.00234}$
ACCESS n3	${2457205.588929}_{-0.000108}^{+0.000119}$	${3.03}_{-0.01}^{+0.01}$	${87.26}_{-0.23}^{+0.23}$	${0.11135}_{-0.00116}^{+0.00114}$
ACCESS n4	${2457848.843293}_{-0.000164}^{+0.000191}$	${2.96}_{-0.05}^{+0.04}$	${85.4}_{-1.61}^{+2.44}$	${0.11754}_{-0.00327}^{+0.00311}$
ACCESS n5	${2457874.758341}_{-0.000089}^{+0.000087}$	${3.00}_{-0.02}^{+0.01}$	${87.86}_{-1.32}^{+1.11}$	${0.11562}_{-0.00101}^{+0.00099}$
LRG-BEASTS n1	${2457541.562056}_{-0.000129}^{+0.000136}$	${3.02}_{-0.05}^{+0.03}$	${87.08}_{-1.83}^{+1.87}$	${0.11155}_{-0.00115}^{+0.00108}$
LRG-BEASTS n2	${2457566.551617}_{-0.000106}^{+0.000117}$	${3.01}_{-0.04}^{+0.02}$	${87.71}_{-1.82}^{+1.52}$	${0.11203}_{-0.00182}^{+0.002}$
FORS2	${2457874.758372}_{-0.000343}^{+0.000310}$	${2.97}_{-0.07}^{+0.05}$	${86.25}_{-2.2}^{+2.43}$	${0.11567}_{-0.00168}^{+0.00167}$
GMOS n1	${2457200.961020}_{-0.000372}^{+0.000325}$	${2.93}_{-0.07}^{+0.07}$	${84.15}_{-1.39}^{+1.82}$	${0.11095}_{-0.00333}^{+0.00344}$
GMOS n2	${2457213.918548}_{-0.000351}^{+0.000348}$	${2.82}_{-0.06}^{+0.06}$	${82.13}_{-1.08}^{+1.22}$	${0.11194}_{-0.00451}^{+0.00287}$
GMOS n3	${2457511.944288}_{-0.000091}^{+0.000085}$	${3.00}_{-0.03}^{+0.01}$	${88.08}_{-1.51}^{+1.31}$	${0.11456}_{-0.00094}^{+0.00081}$
Combined	2456836.296374 ± 0.000040	3.01 ± 0.01	87.0 ± 0.21	0.1136 ± 0.00045
Southworth et al. (2015)	2456836.296445 ± 0.000055	2.999 ± 0.031	87.3 ± 1.2	0.1127 ± 0.0009
Delrez et al. (2018)	2456836.296427 ± 0.000063	${3.010}_{-0.013}^{+0.008}$	${88.8}_{-1.1}^{+0.8}$	${0.1150}_{-0.0014}^{+0.0020}$

Note. In all cases we held the period fixed to 0.9255456 days, as found by Southworth et al. (2015) and used by Kreidberg et al. (2018). We note that these parameters were not used in the fitting of our spectroscopic light curves and hence our transmission spectrum. We also include the weighted mean of our results, labeled "Combined," and the results of Southworth et al. (2015) and Delrez et al. (2018) for comparison.

Download table as:  ASCIITypeset image

Table 4. The Weighted-mean Transmission Spectrum of WASP-103b from All 11 Ground-based Transit Light Curves

Bin Center	Bin Width	${R}_{P}/{R}_{* }$	$\sigma ({R}_{P}/{R}_{* })$	${F}_{\mathrm{cont}}/{F}_{{\rm{W}}103}$	Bin Center	Bin Width	${R}_{P}/{R}_{* }$	$\sigma ({R}_{P}/{R}_{* })$	${F}_{\mathrm{cont}}/{F}_{{\rm{W}}103}$
(Å)	(Å)				(Å)	(Å)
3975	150	0.11415	0.00248	0.0086	6825	150	0.11512	0.00052	0.0573
4125	150	0.11238	0.00145	0.0172	6975	150	0.11657	0.00059	0.0565
4275	150	0.11291	0.00088	0.0118	7125	150	0.11660	0.00050	0.0585
4425	150	0.11402	0.00089	0.0234	7275	150	0.11549	0.00058	0.0579
4575	150	0.11330	0.00108	0.0205	7425	150	0.11419	0.00056	0.0653
4725	150	0.11375	0.00066	0.0270	7575	150	0.11434	0.00069	0.0665
4875	150	0.11415	0.00069	0.0283	7665	20	0.11440	0.00098	0.0664
5025	150	0.11438	0.00068	0.0242	7699	20	0.11574	0.00085	0.0674
5175	150	0.11458	0.00063	0.0229	7725	150	0.11442	0.00059	0.0680
5325	150	0.11436	0.00060	0.0298	7875	150	0.11552	0.00057	0.0711
5475	150	0.11356	0.00057	0.0336	8025	150	0.11532	0.00059	0.0720
5625	150	0.11300	0.00056	0.0368	8175	150	0.11459	0.00060	0.0696
5775	150	0.11372	0.00052	0.0432	8325	150	0.11411	0.00074	0.0733
5893	20	0.11751	0.00107	0.0361	8475	150	0.11586	0.00069	0.0756
5925	150	0.11439	0.00051	0.0450	8625	150	0.11555	0.00063	0.0761
6075	150	0.11500	0.00050	0.0473	8775	150	0.11554	0.00070	0.0774
6225	150	0.11519	0.00051	0.0467	8925	150	0.11579	0.00068	0.0795
6375	150	0.11530	0.00053	0.0497	9075	150	0.11523	0.00084	0.0786
6525	150	0.11527	0.00058	0.0524	9225	150	0.11641	0.00105	0.0778
6675	150	0.11546	0.00072	0.0560	9375	150	0.11369	0.00142	0.0804

Note. These results have been corrected for the third-light contamination and the planet's night-side flux. We note that we also include the third-light correction factors (${F}_{\mathrm{cont}}/{F}_{{\rm{W}}103}$) in this table.

Download table as:  ASCIITypeset image

The ${R}_{P}/{R}_{* }$ values in Table 3 are those after correcting for the third-light contamination and planetary night-side flux (Section 5). We plot the ${R}_{P}/{R}_{* }$ derived from each of our transits from both the "fixed" and "free" system parameters fits, along with the values of Southworth & Evans (2016), Lendl et al. (2017), Delrez et al. (2018), and Wilson et al. (2020) in Figure 5. This demonstrates that while there is some variation in ${R}_{P}/{R}_{* }$, each of our 11 transits are within 1σ–2σ of the weighted mean. We note that ACCESS n5 and VLT/FORS2 are observations of the same transit epoch.

Zoom In Zoom Out Reset image size

In Figure 6, we show the transmission spectra resulting from all 11 of our ground-based transits, following the correction for the third-light contamination and planetary night-side flux. The nightly transmission spectra in the top panels of Figure 6 have had offsets applied to ensure they have the same nightly mean ${R}_{P}/{R}_{* }$ as the weighted mean ${R}_{P}/{R}_{* }$ from all 11 nights, as discussed in Section 6.2. The nightly transmission spectra for all instruments are given in Tables 6 and 7 in the Appendix.

Zoom In Zoom Out Reset image size

Figure 6 shows there is good overall agreement between each of the transits' transmission spectra. The most notable difference comes in the LRG-BEASTS transmission spectra, where the spectra diverge at blue wavelengths (although they are still consistent within 1σ, other than in the bluest bin).

We also note that the Na signal is much deeper in the first LRG-BEASTS night. As an additional check, we performed light-curve fits to five 30 Å-wide bins centered on Na to see whether this was caused by systematics related to the 20 Å bin being too narrow. In this case we found the central Na bin remains deep but a neighboring bin is anomalously shallow, giving us reason to question the authenticity of the deep LRG-BEASTS n1 bin. This behavior did not occur for any of the 11 other transits we analyzed. We are unsure of the causes of this deep Na bin, however; we tried different treatment of the stellar limb darkening (fixed versus free coefficients), GP inputs and GP kernels (Matérn 3/2 versus Squared Exponential), and the use of a flat field, and found these made no difference in the depth of the bin.

Notably, however, our final 11-transit-combined transmission spectrum changes by ≪1σ regardless of whether the second LRG-BEASTS night or the LRG-BEASTS Na result is included. This is the advantage of combining 11 data sets. As a result, we do not exclude the LRG-BEASTS data from our final transmission spectrum.

In Figure 6, we also plot the Gemini/GMOS and VLT/FORS2 transmission spectra as derived by Lendl et al. (2017) and Wilson et al. (2020). This shows the good agreement between our reanalysis of these data sets and the previously published analyses. We believe that the small differences between our analysis of the Gemini/GMOS data compared to the analysis of Lendl et al. (2017) could be due to the different system parameters used and different systematics modeling approaches used.

In Figure 7, we show the weighted-mean transmission spectra for each instrument and the 11-transit weighted-mean combination. In this figure, all the spectra have been corrected for the third-light contamination as described in Section 5. Our combined transmission spectrum, along with the third-light correction factors, is tabulated in Table 4.

Zoom In Zoom Out Reset image size

Figure 7 shows an excellent agreement between all four instruments. This also highlights the precision we have achieved in our combined transmission spectrum with a median uncertainty on the transit depth of 148 ppm. The transmission spectrum shows considerable structure, with an upward slope from ∼4000 to 5200 Å, before dropping toward 5600 Å and rising again toward 6200 Å. There is also a bump in the spectrum at around 7000 Å and a deeper transit centered on the Na doublet at 5893 Å, which we discuss in more detail in the following subsection.

In order to estimate the significance of the sodium detection in our combined transmission spectrum (Figure 7), we fitted a straight line across the neighboring bins between 5625 and 6075 Å where the continuum is rising. This is the same process used to estimate the significance of sodium as used in, e.g., Nikolov et al. (2016), Carter et al. (2020), and Alderson et al. (2020). Comparing the difference between our Na bin and the fitted continuum, we find the Na bin is higher at $3.1\sigma$ confidence.

At face value, this appears significant. However, when we exclude the LRG-BEASTS Na bin from our combined transmission spectrum, which we believe could be affected by systematics (Figure 6), the significance of Na by this method drops to $2.8\sigma$. While the inclusion of the LRG-BEASTS Na bin changes the transmission spectrum by <1σ, owing to the fact we are combining 11 data sets, we favor our conservative $2.8\sigma$ evidence for Na, considering the possible systematics affecting the LRG-BEASTS Na bin. Wilson et al. (2020) detect Na at $2.0\sigma$ confidence while Lendl et al. (2017) find signs of strong Na absorption. Based on our analysis, we cannot claim a significant detection of Na in WASP-103b's atmosphere.

As an additional check of our results, we also fitted the ACCESS light curves with the LRG-BEASTS fitting code (gppm_fit, Section 4.2). Figure 8 shows the comparison between our final transmission spectra from both methods, prior to applying the third-light correction. This shows the excellent agreement between the two fitting routines despite the different approaches to the stellar limb darkening, systematics models, sampling algorithm, and the priors on the transit and systematic models' parameters. Figure 8 demonstrates our results and conclusions are not dependent on the different fitting methods used.

Zoom In Zoom Out Reset image size

As described in Section 5, we had to correct our transmission spectrum for the flux of the blended contaminant (4400 ± 200 K; Cartier et al. 2017). In Figure 9 we plot the pre- and post-third-light-corrected transmission spectra. We additionally include (in dotted lines) the impact of the contamination assuming the effective temperature of the contaminant ± the 1σ and 2σ uncertainties reported by Cartier et al. (2017), which shows that the uncertainty in the contaminant does not result in a significant change in the transmission spectrum. This was also found by Lendl et al. (2017).

Zoom In Zoom Out Reset image size

Figure 9 demonstrates that while the third-light correction has a significant impact over a broad wavelength range, changing the transmission spectrum from gently sloping down toward red wavelengths to sloping upwards, it has a minor impact on the more localized features, such as the dip at ∼5600 Å, the Na bin, and the bump at ∼7000 Å. We explore the impact this correction has on our retrievals in Section 8.

Our retrieved spectra for both the third-light-corrected and -uncorrected optical data are shown in Figure 11 (top panels). Prior to the third-light correction (Section 5), there is a gentle upwards slope toward blue wavelengths, which is best fit by unocculted cool active regions. Following the third-light correction, the spectral morphology changes into a steep downwards slope best explained by unocculted hot active regions, which we interpret as faculae. Quantitatively, the evidence for stellar activity increases from 1.5σ to $4.0\sigma$ following the third-light correction (see Table 5). Clearly the third-light correction makes a significant impact on the conclusions about stellar activity.

Zoom In Zoom Out Reset image size

We additionally infer tentative evidence for atmospheric TiO absorption from our optical data. This TiO inference persists for both the uncorrected and third-light-corrected data, changing from 2.3σ to $1.7\sigma$ following the third-light correction. The TiO mixing ratio, however, is poorly constrained from the optical data alone. No other chemical species are favored by our Bayesian model comparisons with POSEIDON, with $2\sigma$ upper limits on their mixing ratios reported in Table 5. The nondetection of Na, despite the 20 Å bin centered on the Na D lines being ∼3σ deeper than the surrounding continuum (Section 6.4), is likely due to the lack of pressure-broadened Na wings in the surrounding data points (e.g., Nikolov et al. 2018; Alam et al. 2021).

The tentative evidence for TiO from POSEIDON is an interesting contrast with our earlier petitRADTRANS retrieval (Section 7), which instead inferred VO without the need for TiO. To better understand this discrepancy, we ran additional retrievals with both codes varying properties such as the prior ranges on the VMRs and the pressure at the top of the planet's atmosphere. Ultimately this made little difference. The most notable difference we identified is that POSEIDON uses updated ExoMol line lists for VO and TiO, resulting in considerably different optical band shapes compared to the petitRADTRANS VO and TiO cross sections (see Figure 10). However, as we discuss in Section 9.1, we cannot be certain that this is the cause of the disagreement. Indeed, our 4σ preference for unocculted faculae with POSEIDON (for the corrected data) automatically considered VO as an alternative explanation via the Bayesian evidence computations. Therefore, we believe that the inclusion of stellar heterogeneity in our POSEIDON retrievals is the more likely cause of the disagreement.

Our IR-only retrievals on the HST/WFC3 and Spitzer/IRAC data from Kreidberg et al. (2018) weakly favor unocculted faculae and H₂O. We find a lower detection significance for faculae than that from the optical-only corrected spectrum ($2.4\sigma$ versus $4.0\sigma$). This is expected, since the influence of unocculted stellar heterogeneities is largest at short wavelengths (e.g., Rackham et al. 2018, 2019). The retrieved heterogeneity parameters from the IR data are consistent with those from our "Ground Corrected" retrieval (Table 5), giving us added confidence in this interpretation. Our inference of H₂O is marginal at best ($1.7\sigma$) and notably less significant than that found by Wilson et al. (2020, $4.0\sigma$). We believe this discrepancy is due to our consideration of stellar heterogeneities, which were not included in the retrieval analyses of Wilson et al. (2020). Finally, we note that TiO is not inferred from our IR-only retrievals, though the upper limit on its abundance is consistent with the "Ground Corrected" retrieval. In all, these consistent findings between each portion of the transmission spectrum strengthen the conclusions derived from the combined optical and IR data set.

Our retrieval of the combined optical+IR data, shown in Figure 11 (bottom panel), identifies strong evidence of unocculted faculae ($4.3\sigma$) and tentative evidence for TiO ($2.1\sigma$), H₂O ($1.9\sigma$), and HCN ($1.7\sigma$). The inferred faculae are ∼350 K hotter than the stellar photosphere with a coverage fraction of ${22}_{-9}^{+12}$%. The fitted offset between the optical and IR data is ${210}_{-85}^{+83}$ ppm, consistent with the 1σ uncertainties in our optical transmission spectrum. Due to the weak evidence for individual molecules, we also derived the combined significance of TiO + H₂O + HCN from a Bayesian model comparison with all three species excluded. We find that the significance of at least one of TiO, H₂O, or HCN, is $2.8\sigma$. Therefore, we conclude there is moderate evidence for molecular opacity from WASP-103b alongside the stellar faculae contamination.

We illustrate the contributions of each chemical species and the unocculted faculae to our best-fitting POSEIDON model in Figure 12. This figure demonstrates that unocculted faculae still induce a slight downward slope between the red and blue ends of the HST/WFC3/IR/G141 bandpass, explaining why our IR-only retrievals also favor unocculted faculae. Unlike our retrievals with petitRADTRANS, we see that POSEIDON does not fit the dip in the transmission spectrum around 0.55 μm, which may be due to the ExoMol line lists used (Figure 10) or the inclusion of faculae–we discuss this more in Section 9.1. We note that the "minimal" model shown in this figure uses only those parameters required to fit the combined optical + IR data (i.e., without unconstrained chemical species or cloud parameters). This nine-parameter model achieves a minimum reduced chi-square of ${\chi }_{\nu }^{2}=1.36$, which we consider an excellent fit to our transmission spectrum of WASP-103b. The full posterior corresponding to this best-fitting model is shown in Figure 13, with the parameter constraints also given in Table 5. We caution, however, that our recommended atmospheric and stellar properties are those from the "full" optical + IR retrieval (see Table 5) because the full retrieval accounts for marginalization over the entire parameter space.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Our inferred atmospheric properties for WASP-103b are summarized in Table 5. We find a TiO mixing ratio of $-{7.31}_{-2.06}^{+2.14}$ dex, broadly consistent with expectations for a solar abundance atmosphere at WASP-103b's equilibrium temperature (Woitke et al. 2018). The retrieved H₂O mixing ratio, $-{3.28}_{-4.08}^{+1.46}$ dex, spans a wide range from supersolar to subsolar abundances. The latter is consistent with the suggestion by Kreidberg et al. (2018) of H₂O dissociation in WASP-103b's atmosphere. We suggest that Wilson et al.'s (2020) tighter constraint on the H₂O abundance of ∼40× solar, without subsolar solutions, may arise from the absence of marginalization over unocculted faculae in their retrievals. A hint of HCN near 1.6 μm (see Figure 12) emerges only from our combined optical + IR retrievals, though the abundance is poorly constrained with the present data. Future observations of WASP-103b can serve to confirm the presence of TiO, H₂O, and/or HCN while strengthening their abundance determinations.

As shown in Section 7, our petitRADTRANS retrievals that do not include stellar activity favor VO, albeit with an implausibly high volume mixing ratio (Figure 10). However, retrievals that do include the impacts of stellar activity significantly favor stellar-activity contamination with tentative evidence for TiO (Section 8). This may be due to the different opacity at bluer wavelengths; the Plez (1999) VO line list used by petitRADTRANS leads to a steeper drop off than the ExoMol line list used by POSEIDON (Figure 10). As a result, the POSEIDON retrievals with the ExoMol line list necessitate unocculted faculae to fit the downwards trend we observe at blue wavelengths (Figure 12).

In order to place WASP-103b into context, we compare our transmission spectrum with that of another ultrahot Jupiter, WASP-121b (Delrez et al. 2016). WASP-121b's equilibrium temperature of 2358 ± 52 K is 130 K cooler than WASP-103b's, and it orbits an F6 dwarf, compared to WASP-103, which is an F8 dwarf.

WASP-121b's low-resolution HST/STIS transmission spectrum (Evans et al. 2018) is shown in Figure 14. There is an excellent agreement between the transmission spectra of WASP-121b and WASP-103b, particularly between $\sim 4000$ and 6500 Å. Figure 14 additionally demonstrates our ability to achieve HST-quality transmission spectra from the ground.

Zoom In Zoom Out Reset image size

Figure 14 also includes the near-UV photometric data point from Turner et al. (2017). We did not analyze that data set in the current work nor include it in our retrieval analysis due to the different system parameters that Turner et al. (2017) used. However, at face value, this point also seems to follow the subsequent rise in the transmission spectrum toward the UV, as seen in WASP-121b.

9.1.1. On the Potential for VO and TiO

9.1.2. On Stellar-activity Contamination

Figure 15 shows the comparison between our newly derived transmission spectrum and the previously published results of Southworth & Evans (2016), Turner et al. (2017), Lendl et al. (2017), and Wilson et al. (2020). In all cases the transmission spectra have been corrected for the third-light contamination. We note that in this figure we applied an offset in ${R}_{P}/{R}_{* }$ of 0.00065 and 0.0021 to the transmission spectra of Lendl et al. (2017) and Wilson et al. (2020), respectively, so that their median ${R}_{P}/{R}_{* }$ is equal to the median of our newly derived transmission spectrum. We believe the larger offset for the Wilson et al. (2020) data is because those authors used a Gaussian prior on ${R}_{P}/{R}_{* }$ and took the mean and standard deviation from Southworth et al. (2015), who did not account for the third-light contamination (Section 5). We have not applied an offset to the transmission spectrum of Southworth & Evans (2016), where a steep slope is found, nor to the single wavelength bin of Turner et al. (2017).

Zoom In Zoom Out Reset image size

This figure demonstrates the improved precision that we achieved by combining our 11 transits. It also shows the slight disagreement between our new transmission spectrum and some of the Gemini/GMOS and VLT/FORS2 bins of Lendl et al. (2017) and Wilson et al. (2020), which highlights the benefits of combining many transits to form our transmission spectrum.

Additionally, we do not see the blueward rise as observed by Southworth & Evans (2016). However, other than Southworth & Evans's (2016) bluest photometric data point, our results are roughly consistent to 1σ. Additionally, Turner et al.'s (2017) near-UV photometric band indicates a rise toward wavelengths bluer than our spectroscopic coverage. However, we caution against reading too much into this due to the different system parameters used and the offsets in ${R}_{P}/{R}_{* }$ needed to bring the spectroscopic data in agreement.