A ground-based NUV secondary eclipse observation of KELT-9b

TL;DR

This study used a ground-based telescope to observe the secondary eclipse of exoplanet KELT-9b in the near-ultraviolet, constraining its atmospheric temperature and albedo despite not detecting the eclipse signal.

Contribution

First ground-based NUV secondary eclipse observation of KELT-9b, providing constraints on its atmospheric properties and demonstrating the feasibility of such studies from Earth.

Findings

No eclipse detection, upper limit of 181 ppm on depth.

Dayside temperature constrained to 4995 K at ~30 mbar pressure.

Albedo likely very low, consistent with similar hot Jupiters.

Abstract

KELT-9b is a recently discovered exoplanet with a 1.49 d orbit around a B9.5/A0-type star. The unparalleled levels of UV irradiation it receives from its host star put KELT-9b in its own unique class of ultra-hot Jupiters, with an equilibrium temperature > 4000 K. The high quantities of dissociated hydrogen and atomic metals present in the dayside atmosphere of KELT-9b bear more resemblance to a K-type star than a gas giant. We present a single observation of KELT-9b during its secondary eclipse, taken with the Wide Field Camera on the Isaac Newton Telescope (INT). This observation was taken in the U-band, a window particularly sensitive to Rayleigh scattering. We do not detect a secondary eclipse signal, but our 3$\sigma$ upper limit of 181 ppm on the depth allows us to constrain the dayside temperature of KELT-9b at pressures of ~30 mbar to 4995 K (3$\sigma$). Although we can place an observational constraint of $A_g<$ 0.14, our models suggest that the actual value is considerably lower than this due to H$^-$ opacity. This places KELT-9b squarely in the albedo regime populated by its cooler cousins, almost all of which reflect very small components of the light incident on their daysides. This work demonstrates the ability of ground-based 2m-class telescopes like the INT to perform secondary eclipse studies in the NUV, which have previously only been conducted from space-based facilities.

Tables

Table 1. Table 1: Parameters of the KELT-9 system

Parameter	Value	Ref.
Stellar Parameters
$R_{\ast }$ ($R_{\odot }$)	${2.362}_{-0.063}^{+0.075}$	A
$T_{\ast }$ (K)	10,170$\pm$450	A
$log\left(g\right)$	4.091$\pm$0.014	A
$[$Fe/H$]$	-0.03$\pm$0.20	A
Planetary Parameters
$R_P$ ($R_J$)	${1.888}_{-0.052}^{+0.062}$	A
$t_0$ (MJD)	57095.18572$\pm$0.00014	A
$P$ (days)	1.4811235$\pm$0.0000011	A
$a$ (au)	${0.03462}_{-0.0093}^{+0.0110}$	A
$i$ (∘)	86.79$\pm$0.25	A
$T_{day}$ (K)	4600$\pm$150	A
$F_{ecl,z^{\prime }}$ (ppm)	1006$\pm$97	B
Measured Parameters
$F_{ecl,U}$ (ppm)	-71$\pm$84	C
$F_{ecl,U}$ (ppm)	181 (3$\sigma$ limit)	C
$T_{day}$ (K)	4995 (3$\sigma$ limit)	C
${\sigma }_w$	(3.32$\pm$0.11)x10-4	C
${\sigma }_r$	(0.7$\pm$5.6)x10-4	C
References. A - Gaudi et al. (2017); B - Collins et al.
(2019, in prep.); C - This work.

Table 2. Table 2: Description of spectra in Figure 3

Colour	$\gamma$	$F_{ecl,U}$ (ppm)	$F_{ecl,z^{\prime }}$ (ppm)	$T_B$ (K)
Green	2.0	218	1307	5080
Purple	1.5	159	1146	4890
Yellow	1.1	116	1005	4700
Red	0.9	96	930	4600
Blue	0.2	39	641	4165

Full text

planets and satellites: atmospheres — stars: individual (KELT-9) — techniques: photometric — ultraviolet: planetary systems

1 Introduction

The measurement of the drop in flux of an exoplanet-star pair when the planet is occulted by its host star has established itself as an important tool to study the atmospheres of exoplanets. At near-infrared wavelengths and longer, thermal emission is the dominant source of flux from hot Jupiters (López-Morales & Seager, 2007). Measurements of thermal emission have led to the detection of atmospheric features such as global heat redistribution (Knutson et al., 2007), the presence of a temperature inversion (Evans et al., 2017) and atmospheric variability (Armstrong et al., 2016).

At optical wavelengths and shorter, the component of flux from hot Jupiters due to thermal emission drops off sharply, such that the dominant component of flux is expected to be due to light reflected from its host. Measurements of thermal emission for various hot Jupiters imply that they should have reflection signatures sufficiently large to be detectable with current instrumentation (e.g. Schwartz & Cowan, 2015; Schwartz et al., 2017). However, the vast majority of searches for reflected light from hot Jupiters at optical wavelengths—where their host stars typically emit most of their energy—have resulted in non-detections (e.g. Collier-Cameron et al., 2002; Leigh et al., 2003; Rowe et al., 2008; Gandolfi et al., 2013; Dai et al., 2017; Močnik et al., 2018). These results are consistent with predictions that scattering in the optical is suppressed by alkali absorption for cloud-free atmospheres (Sudarsky et al., 2000; Burrows et al., 2008).

To date, two studies have utilised the capabilities of HST/STIS to observe secondary eclipses of hot Jupiters in the NUV. This wavelength range is potentially more favourable than the optical for detecting reflected light from exoplanets orbiting hot stars, as alkali absorption is much weaker and the Rayleigh scattering cross-section is much higher. Whilst Bell et al. (2017) did not detect reflected light at NUV wavelengths for WASP-12b ($T_{eq}\sim$ 2,500 K), Evans et al. (2013) measured a geometric albedo ($A_{g}$) of 0.40$\pm$0.12 at 290-450 nm (a wavelength range overlapping with the U-band) for HD 189733 b ($T_{eq}\sim$ 1,200 K). These results support studies (e.g. Heng, 2016; Stevenson, 2016; Wakeford et al., 2017) suggesting that the most highly irradiated planets are less likely to have clouds in their atmospheres, as well as observational evidence for clouds in the atmosphere of HD 189733 b (Pont et al., 2008; Sing et al., 2011).

The 4050 K equilibrium temperature of the recently-discovered KELT-9b (Gaudi et al., 2017) is by far the hottest of any known exoplanet. Its 1.49 day orbit around HD 195689, a B9.5/A0-type star, means that KELT-9b is more heavily-irradiated at UV wavelengths than any other known exoplanet. Hoeijmakers et al. (2018) obtained high-resolution spectra of KELT-9b during its transit and detected Fe, Fe+ and Ti+ features with high significance, suggesting a temperature in excess of 4000 K at the terminator. The 4600$\pm$150 K dayside temperature (Gaudi et al., 2017) measured from its z’-band eclipse depth (Collins et al. 2019, in prep.) is comparable to that of a K4-type star. This high dayside temperature means that KELT-9b is the only known planet expected to have a U-band eclipse depth $>$ 50 ppm due to thermal emission.

In this letter, we present a photometric ground-based U-band secondary eclipse of KELT-9b, which allows us to constrain the energy budget of this unique exoplanet. In chapter 2 we summarise our observation, in chapter 3 we describe the steps taken to reduce the data, in chapter 4 we describe how we fitted the eclipse depth and in chapter 5 we summarise how we modelled the KELT-9b spectrum and discuss our result, along with the wider implications of this study.

2 Observation

We observed one secondary eclipse of KELT-9b on 2017 July 20 using the Wide Field Camera (WFC) with the U-band filter on the 2.5m Isaac Newton Telescope (INT) at the Observatorio del Roque de los Muchachos on the island of La Palma. The observations lasted $\sim$ 8.3 hours and started at 21:15:29 UT. During this time, 326 frames were obtained with an average cadence of 75.1 seconds and an exposure time of 45.1 seconds. 155 of these frames were taken when KELT-9b was fully or partially occulted by its host. The observations commenced and concluded in evening and morning twilights, respectively. 32 frames taken during twilight, when KELT-9b was out-of-eclipse, were removed as the increased sky brightness caused strong systematics in those sections of the light curves. For 17 frames that were randomly distributed through the night, we observed that the exposure time was only $\sim$44.6 seconds. However, due to our use of differential photometry, no visible correlation was observed between exposure time and flux once the target had been normalised with the comparison stars. Although the WFC consists of a 4 CCD mosaic — each with a pixel scale of 0.33” per pixel — only the central CCD (CCD4) was used, giving us a field-of-view of $\sim$22.7’ by 11.4’.

We performed the observations with the telescope defocused, which acts to reduce overheads, minimise errors associated with flat fielding and make the resulting PSFs less sensitive to variations in seeing. This resulted in a donut-shaped PSF with a diameter of 54 pixels (18”). Due to the defocusing, the telescope auto-guider was not used. Instead, a custom code that uses the science frames to account for telescope drift was used. Care was taken to ensure that the target and the most promising comparison stars were positioned on well-behaved parts of the detector and the drift throughout the night was less than 4 pixels (1.3”).

3 Data Reduction

Each of the images was overscan-subtracted on a row-by-row basis using the mean of the overscan regions at either side of the CCD. The row by row bias subtraction was used to correct for a known issue with the WFC that affected about half of the frames, in which the bias level present in the frames drops and corrects itself after a period of time. This was followed by a full-frame bias subtraction. The images were then each flat fielded using a master flat constructed from twilight flats. Finally, a second-order polynomial was fit to and subtracted from the entirety of each frame with the stars masked, to remove the small gradient in the sky background across the CCD.

We observed crosstalk between the 4 CCDs that make up the WFC mosaic, which caused bright stars from one CCD to be ghosted onto the same position on other CCDs. This caused the addition or subtraction of $<$ 3 ADU on a background of $\sim$ 2,000 ADU in the raw frames. However, the target and comparison stars did not fall on any of the affected regions.

Finally, we performed aperture photometry on the target and each of the two comparison stars using an aperture with a radius of 51 pixels, selected to maximise the flux and minimise the influence from the background. The annuli used to subtract the residual sky background from each star had inner and outer radii of 72 and 91 pixels, respectively.

4 Analysis & Results

The raw light curves for the target and each of the comparison stars are shown in the top panel of Figure 1 (a). The two comparisons (HD 195558 and BD+39 4224) are both A-type stars and have median fluxes of 0.27 and 0.12, respectively, relative to KELT-9 throughout the observation. The light curve in the bottom panel of Figure 1 (a) was created by dividing the target light curve by the sum of the two comparison light curves and normalising for a median value of 1. This step removed visible correlations with airmass and exposure time (top and middle panels of Figure 1 (b)) that were visible in the raw light curves. No correlations were observed with the x and y positions of the stars on the detector. There were small discontinuities visible in the sky background levels for the target (see the bottom panel of Figure 1 (b)) and comparison stars, but no such features were visible in the corresponding light curves. On initial inspection, a dip in the sky background at phase of $\sim$0.59 corresponded with a dip in flux in the normalised light curve (also see Figure 2). However, the latter feature was found to be much broader and no significant correlation was found. Despite our method of defocusing the telescope, a strong correlation with seeing was visible. There was also a low-order trend present through the time-series that was not removed by the normalisation with the comparison stars.

We fit for each of these trends simultaneously with an eclipse model using a Markov Chain Monte Carlo (MCMC) method with the Metropolis–Hastings algorithm and orthogonal stepping. To model the eclipse, we used the Mandel and Agol transit model (Mandel & Agol, 2002) with the limb darkening coefficients set to zero. The relevant stellar, planetary and orbital parameters were all fixed using values from Gaudi et al. (2017), which are shown in Table 1. The trend in the baseline was modelled using a second-order polynomial, which optimised the Bayesian information criterion with respect to higher and lower order models. The time-dependent and time-independent components of the noise associated with the residual flux were measured using the wavelet method (Carter & Winn, 2009). We ran an MCMC where the components of the model associated with the eclipse depth, a linear function of seeing, the polynomial baseline and the wavelet noise parameters were all allowed to vary. Firstly, we ran a “burn-in” phase of 105 steps, where the step sizes were recalculated every 104 steps to set proportionate step sizes for each parameter. We then used these step sizes in an MCMC chain of 106 steps to get the best fit values for each parameter, which are shown in Figure 1 (c). We verified convergence by checking the Gelman–Rubin criterion (Gelman & Rubin, 1992).

The detrended light curve is shown in Figure 2, with the best fit eclipse model shown in red. Whilst a negative value for a secondary eclipse depth is unphysical, we allowed this to avoid biasing the MCMC fit. The best fit depth of -71$\pm$84 ppm allowed us to place an upper limit on the secondary eclipse depth of 181 ppm at 3$\sigma$.

5 Discussion & Conclusions

We jointly interpreted our U-band upper limit and the z’-band eclipse detection (Collins et al. 2019, in prep.) by generating high-resolution emission spectra with wavenumber resolution of 0.03 cm*-1* (left panel of Figure 3; shown in 2 nm wavelength bins for clarity), which were calculated using a 4-stream discrete ordinate radiative transfer method. For the temperature-pressure (TP) profiles (Figure 3, right panel) we used the approximations for an irradiated atmosphere from Guillot (2010), with an infrared opacity of 0.03 cm2 g*-1*. The ratio of the shortwave to the infrared opacity ($\gamma$), which effectively controls the shape of the TP profile, is treated as a free parameter. Strong absorption of shortwave radiation in the upper atmosphere can result in large temperature inversions of several hundreds of Kelvin, which has previously been demonstrated for species such as TiO and VO (Spiegel et al., 2009). Atomic and ionic species expected to be present in the atmosphere of KELT-9b (such as \ceFe and \ceFe+; Hoeijmakers et al., 2018) are strong absorbers at optical and shorter wavelengths. Hence, the five spectra shown in Figure 3 were selected to explore a range of possible TP-profiles, ranging from a strong inversion ($\gamma$=2) to a rapid decrease in temperature with altitude ($\gamma$=0.2). As KELT-9b is tidally locked and expected to inefficiently redistribute heat from dayside to nightside, the temperatures in all five profiles are well above the equilibrium temperature of 4050 K from Gaudi et al. (2017). The chemical composition is calculated using the FastChem equilibrium chemistry code (Stock et al., 2018), assuming solar elemental abundances. As shown by Kitzmann et al. (2018), the assumption of chemical equilibrium is reasonable for the hot dayside of KELT-9b.

We account for about 50 different gaseous absorbers in the atmosphere. Cross-sections for \ceCO and \ceH2O were calculated with the opacity calculator HELIOS-K (Grimm & Heng, 2015), using the corresponding Exomol line lists. Atoms and ions, including \ceFe, \ceFe+, \ceTi, \ceTi+, \ceCa and \ceCa+, are incorporated with line list data from the Kurucz database. Continuum absorption of \ceH- is treated according to John (1988). Additionally, we include the collision-induced absorption of \ceH2-\ceH2, \ceH2-\ceHe, and \ceH-\ceHe pairs, based on data from HITRAN. Furthermore, Rayleigh scattering of \ceH2, \ceH, \ceHe, and \ceCO is incorporated in the radiative transfer calculations as well.

For KELT-9, we used a spectrum for a 10,000 K star with $log(g)$ = 4 and [Fe/H] = 0 (based on values from Gaudi et al., 2017, shown in Table 1) from the NextGen Model grid of theoretical spectra (Hauschildt et al., 1999).