Five New Hot Jupiter Transits Investigated with Swift-UVOT

All Swift data reduction was performed with HEAsoft v6.28 and Swift-UVOT CALDB files released on 2020 October 26. With the aim of achieving <2% photometric precision, we used the event files to select time intervals to have a minimum of 3000 counts bin^–1 from the target star. As a result, the light-curve bin size is different for each source, depending on the count rate of each host star. Table 1 lists the average count rate for each target along with the median bin size used. We used the barycorr tool to convert the time bin values to barycentric time. Then we used uvotscreen and uvotimage to create images for each time bin.

To align the images, we ran uvotdetect on a single long-exposure observation to create a point-source catalog for the field. This catalog was fed into uvotskycorr to perform aspect corrections before running uvotimsum on image files with multiple image extensions. For targets with only a few stars within the Swift-UVOT field of view, aspect corrections were not always successful. Summing the image extensions then led to extended point-source images. We correct as much as possible for this effect by performing photometry on reference stars, described below, which will also suffer from aspect-correction artifacts. In general, poor aspect corrections reduced the precision for some time bins, but calibration provided by the reference stars preserved the relative accuracy of the photometric measurement.

Telescope jitter caused the targets to move on arcsecond scales among each time-binned image. We accounted for this by running uvotcentroid on every time-binned image. We then ran uvotsource to perform photometry on the target star with a 3'' radius circular source and a 275–35'' annuluar background region by default. ⁴ The background annuli were adjusted for WASP-121, WASP-62, and HAT-P-6 in order to avoid contamination from nearby point sources. All background regions were chosen to cover approximately the same area on the sky as the default background annulus, around 1500 arcsec². We also ran uvotdetect on every image, which provided the photometry for our calibration sources.

Next, we defined a set of reference sources to correct for instrumental effects. Using the catalog of aspect-correction sources described above, we searched for field sources within $8^{\prime}$ of our target that were also bright enough to yield high-precision photometric results. We filtered for these sources by limiting our consideration to measurements that had <3% error and within $7^{\prime}$ of the target to avoid the edges of the detector. This process yielded about three to six reference sources per target field (Table 1). We modified this list by hand to account for alternate telescope pointing positions and roll angles so that every reference source in the final list could be found in every observation. We identified reference sources in each uvotdetect output file using coordinate matching. The final reference light curves were created by computing the error bar–weighted average flux for all reference sources in each time bin. We calibrated the target light curves by dividing them by the respective reference light curves, producing a relative photometric measurement that is corrected for instrument systematics. The target, reference, and final corrected light curves are discussed in more detail in the Appendix.

We normalized each visit light curve for intrinsic stellar variability to first order. For each visit light curve, we fit a one-dimensional line to the out-of-transit data using the transit ingress and egress times derived from the optical data sets and excluding data points that deviated >3σ from the out-of-transit mean. We then normalized each light curve by the linear model and phase-folded all visits using the orbital information from the references listed in Table 2, which were supplied to the NEA ephemeris calculator.

Table 2. Transit Model Fit Results for All Targets in This Study

Target	Opt.	σ(ΔT0)	ΔT0	uvm2	χ2 Ratio	p-value	NC Lim. a
	(RP /R*)	Prior (s)	(hr)	(RP /R*)	(Transit/Null)		(RP /R*)
WASP-121 b	${0.12454}_{-0.00048}^{+0.00047}$	25	0 ± 0.006	0.15 ± 0.01	0.856	<10−5	0.10
XO-3 b	0.09057 ± 0.00057	900	−0.37 ± 0.08	0.18 ± 0.01	0.940	0.0007	0.15
KELT-3 b	0.0939 ± 0.0011	1700	${0.14}_{-0.18}^{+0.16}$	0.16 ± 0.02	0.967	0.006	0.16
WASP-3 b	0.1051 ± 0.0124	200	−0.04 ± 0.06	0.17 ± 0.02	0.953	0.05	0.20
HAT-P-6 b	0.09338 ± 0.00053	500	−0.01 ± 0.14	0.0 ± 0.11	0.996	⋯	0.16
WASP-62 b	0.1109 ± 0.0009	200	−0.003 ± 0.06	0.0 ± 0.07	0.997	⋯	0.11

Note. References for optical transit depth and transit midpoint ephemeris: XO-3b, Winn et al. (2008); KELT-3b, Pepper et al. (2013); WASP-3b, Christiansen et al. (2011); HAT-P-6b, Noyes et al. (2008); WASP-62b, Hellier et al. (2012); WASP-121b, Delrez et al. (2016).

^a Null confidence limit, the maximum R_P /R_* value that is 1σ distinguishable from the noise in each data set.

Download table as:  ASCIITypeset image

We used the transit simulator, BATMAN (Kreidberg 2015), to fit the normalized, phase-folded light curves. The free parameters in the transit model are the planet radius relative to the stellar radius (R_p /R_*) and the transit midpoint (T₀). The latter free parameter was included to account for potential systematic errors in the absolute timing of UVOT, ephemeride uncertainty, or unknown transit timing variations. The remaining BATMAN parameters—orbital period, semimajor axis, inclination, eccentricity, and longitude of periastron—were held fixed to the literature values (references in Table 2). We used the quadratic limb-darkening law and fixed the limb-darkening coefficients to the values obtained from the Exoplanet Characterization Tool Kit (ExoCTK) Limb Darkening Calculator using the Phoenix ACES model for the GALEX UV bandpass. We also modified the BATMAN code to allow for R_p /R_* < 0, which is nonphysical but signals the code to model an "inverse transit," yielding a light curve that increases in flux instead of decreasing during the transit. Allowing for this possibility reduces the likelihood of spuriously detecting a transit from noise.

We used the Markov Chain Monte Carlo (MCMC) analysis package emcee (Foreman-Mackey et al. 2013) for fitting. We applied a Gaussian prior to the transit midpoint that was equal to the uncertainty on the ephemeris prediction obtained from the NEA ephemeris calculator (third column of Table 2). No constraint was applied to the value of R_p /R_*. The χ² statistic was used to characterize the model likelihood. We ran emcee with 100 walkers, conducted 250 steps for the burn-in phase, then ran the MCMC sampler for 1000 steps to arrive at a final a posteriori probability distribution containing 10⁵ sample points. Table 2 (fourth and fifth columns) lists the median parameter values with 1σ confidence intervals, derived from the posterior probability distribution. Figures 1–3 show each transit light curve (gray circles) with the best-fit model (blue solid curve) and 1-sigma contours (blue dashed curves). For visual purposes, a coarser 1 ks (16.7 minute) binned light curve is overplotted (black squares). The binned flux values were computed using the error bar–weighted average of all data points within that bin.

Zoom In Zoom Out Reset image size

To quantify the probability that our measured transits arise from natural statistical variation, we performed null-hypothesis testing on all targets. All light curves were fit with a null-hypothesis straight-line model with the line position as a free parameter. We then calculated the ratio of the best-fit χ² value from the transit model to the best-fit χ² value from the null hypothesis (sixth column of Table 2). Values <1 indicate that the transit model fits better. In order to characterize whether a better-fitting transit is due to chance, we derive the χ² ratio for a large number of simulated light curves. For each target, we simulated a light curve under the null hypothesis by drawing from a normal distribution with the same standard deviation of the real light curve. Error bars were drawn randomly from the set of real light-curve error bars. Then we fit the simulated null data sets with a transit model and a null-hypothesis model. This process was repeated 100,000 times. We then calculated the probability of measuring the observed χ² ratio from a light curve that did not contain a transit. This is the p-value (seventh column of Table 2), which can be thought of as a false-positive probability for the best-fit transit.

Finally, we used simulated data sets to compute the 1σ null confidence limit on the transit depth for all of the target light curves. This process provides an estimate for the largest transit that can hide in a noisy light curve. We simulated 1000 transit light curves using the same standard deviation and error bar distribution as the true light curve for a grid of R_p /R_* values ranging from zero to 0.3. Each simulated transit light curve was refit with a transit model, producing a distribution of best-fit transit radii. The null confidence limit was found by selecting the input R_p /R_* value that produced a fit distribution that deviated by 1σ from the fit distribution derived from the null (R_p /R_* = 0) simulations. The results are listed in the last column of Table 2. A null confidence limit that is lower than the best-fit transit depth is expected for a positive transit detection. In the case of no transit detection, or if the best-fit transit yielded an R_p /R_* that is less than the null confidence limit (as is the case for WASP-3b), we adopted the null confidence value as the upper limit for R_p /R_*.

We applied the analysis techniques described above to the Swift-UVOT data set for WASP-121b, which has an NUV transit reported in the literature (S19). The left panel of Figure 1 shows the uvm2 light curve with the best-fit transit model, and the right panel shows the posterior distribution from emcee (blue contours) plotted against the best-fit transit models from the simulated null-transit data sets (gray contours). The contours demonstrate that, despite the large scatter and poor visual quality of the Swift-UVOT light curves, the instrument is capable of measuring ∼1% transits at a >3σ level. Our transit depth measurement for WASP-121b, 2.25% ± 0.30%, is within 1σ of the transit depth reported by S19.

We used all available Swift data to characterize the X-ray and UV (XUV) properties of the exoplanets' host stars. The XRT on Swift ran concurrently in photon-counting mode throughout all of our UVOT observations. We stacked all available XRT data, including observations that did not cover exoplanet transits, summarized in Table 3. We used XSELECT to extract an image and a spectrum from the stacked event file after filtering for the 0.2–2.4 keV range. We then ran the source detection algorithm XIMAGE to search for point sources coincident with the coordinates of our target exoplanet host stars.

Table 3. XUV Properties of the Host Stars

Target	X-Ray Exp.	$\mathrm{log}{F}_{{\rm{X}}}(\mathrm{abs})$ a	NH b	$\mathrm{log}{L}_{{\rm{X}}}$ a	UV
	(ks)	(erg s−1 cm−2)	(1021 cm−2)	(erg s−1)	AB mag
XO-3	13.8	<−14.6	1.7	<28.6	13.5 ± 0.02
KELT-3	33.9	<−14.9	2.2	<28.4	14.2 ± 0.02
WASP-3	12.4	<−14.1	0.58	<29.0	14.7 ± 0.02
HAT-P-6	16.0	<−14.6	2.3	<28.9	14.0 ± 0.02
WASP-62	5.2 (XMM)	$-{13.86}_{-0.06}^{+0.09}$	0.53	${28.96}_{-0.06}^{+0.09}$	15.0 ± 0.02 c

Notes.

^a 0.2–2.4 keV band. ^b Using the ISM scaling relations of Foight et al. (2016). ^c NUV magnitude derived from total 12.5 ks Swift-uvm2 data set.

Download table as:  ASCIITypeset image

The only host star detected in the Swift-XRT data sets was WASP-62, with a 2.5σ significance, registering 5 × 10⁻³ counts s^–1. Because there was not a sufficient number of counts to fit a spectrum, we used the CXC webtool, PIMMS v4.10, to convert the Swift-XRT count rates to a flux value. Using the Gaia g-band extinction value of 0.1843 (Gaia Collaboration et al. 2016, 2018) and the scaling relations for ISM column versus visual extinction (Foight et al. 2016), we arrive at a value of N_H = 5.3 × 10²⁰ cm⁻² for WASP-62. We supplied this N_H value to model the extinction of a 0.3 keV temperature APEC model in PIMMS, yielding an absorbed flux of (1.0 ± 0.5) × 10⁻¹⁴ erg cm⁻² s⁻¹ and an unabsorbed flux of (1.7 ± 0.8) ×10⁻¹⁴ erg cm⁻² s⁻¹.

We also examined the publicly available XMM-Newton data for WASP-62 (PI: Sanz-Forcada). Obtaining a 46 count spectrum for the source, we fit a single-temperature APEC model (Smith et al. 2001) together with a TBABS model for interstellar absorption (Wilms et al. 2000). We binned the spectrum to a minimum of 5 counts bin^–1 and used the C-statistic (Cash 1979) as our fit statistic. The errors on the fitted values are evaluated using Xspec's built-in MCMC sampler. We obtain a 0.2–2.4 keV absorbed (unabsorbed) flux value of ${1.38}_{-0.20}^{+0.28}\ ({2.51}_{-0.36}^{+0.51})\times {10}^{-14}$ erg cm⁻² s⁻¹. With the absorption correction, we find an intrinsic luminosity that is 50% larger than that reported in Alam et al. (2021). Because there is significantly more X-ray information in the XMM-Newton data set, we use this value in Table 3 and subsequent analysis.

For the remaining exoplanet host stars, no point sources were detected. We used the background count rates, scaled to a 10'' radius region, to estimate the upper limit to the absorbed flux for each point source. We again used PIMMS to convert the count rates to flux, modeling the stellar coronae with a 0.3 keV APEC model with solar abundances. As with WASP-62, we used the Gaia g-band extinction and ISM scaling relations to estimate N_H. The ISM columns were used to compute an upper limit to the unabsorbed flux and, subsequently, luminosity for each exoplanet host star. The results are summarized in the last three columns of Table 3.

Finally, we ran uvotsource on every uvm2 image produced by the standard Swift pipeline and computed the error-weighted average. The results are listed in the last column of Table 3. Nearly all of the measured magnitudes are similar to those available in the GALEX catalog (listed in Table 1), except for XO-3, which appears to be a factor of 3 brighter at the time of the Swift measurements.

Main-sequence F5V star XO-3 has one of the most unique planetary systems ever measured. It is particularly massive (11.7 ± 0.4 M_J) and has a large eccentricity (e = 0.28) and an orbital plane that is highly inclined (λ = 373) relative to the equatorial plane of its host star (Hébrard et al. 2008; Johns-Krull et al. 2008; Winn et al. 2008, 2009; Bonomo et al. 2017). The high eccentricity and orbital inclination, thought to arise from planet–planet interactions early in the system's history, make XO-3b an important subject for understanding the role of tidal forces in planetary evolution.

Swift conducted two visits of the XO-3b transits 3 months apart. Despite a large systematic variance in the light-curve amplitude, the average appears to decrease around the expected transit time. Null-hypothesis testing indicates that this is the most likely transit in our sample, with a very low likelihood of being a false detection. Based on the p-value and contours in Figure 2, we consider this a 3σ detection.

Zoom In Zoom Out Reset image size

It is notable that the modeled XO-3b transit midpoint is more than 3σ away from the midpoint predicted from the Winn et al. (2008) ephemeris. However, the 22 minute offset is close to the propagated ephemeris uncertainty of 0.01 days (14 minutes). We experimented with a few different ephemeris values from the literature (Wong et al. 2014; Bonomo et al. 2017) and found that all of them returned an early NUV transit for XO-3b (ΔT₀ = −0.47 and −0.49 hr, respectively). Most importantly, none of them significantly altered the transit depth of XO-3b; both alternative ephemerides returned R_p /R_* = 0.20 ± 0.01. Future simultaneous optical observations of an XO-3b transit can be used to determine whether or not the NUV transit is centered with respect to the optical transit.

Past studies of the XO-3b system suffered from large uncertainties due to the unknown distance to XO-3, which can now be resolved. Using the Gaia DR2 values for the stellar radius, the transit depth measured by Winn et al. (2008) implies that XO-3b is 1.280 ± 0.087 R_J. Using the stellar distance provided by Gaia DR2, the optical flux of the star on the planet surface (insolation) is 2.8 × 10⁹ erg s⁻¹. We use these values with the most up-to-date mass and orbital parameters inferred from radial velocity measurements (Bonomo et al. 2017) to estimate the ratio of tidal heating versus energy available to photoabsorptive heating, ${\dot{E}}_{\mathrm{tide}}/{\dot{E}}_{\mathrm{insolation}}$, described in Machalek et al. (2010, and references therein). We find ${\dot{E}}_{\mathrm{tide}}/{\dot{E}}_{\mathrm{insolation}}=0.44{({Q}_{p}/{10}^{5})}^{-1}$, where Q_p is the dimensionless heating parameter. To explain a radius of 1.28 R_J, XO-3b would require Q_p ≈ 10⁶ (Liu et al. 2008), implying a nonnegligible amount of tidal heating: ∼5% of the luminosity incident on the planet surface. This formula does not take into account the efficiency of photoabsorption in the planet's atmosphere. Assuming a nominal value of 10% heat transfer efficiency, one can conclude that the atmospheric heating due to tidal forces on XO-3b is roughly equal to the heating from stellar irradiation.

With a transit depth of 3.2% ± 0.4%, our measurement of XO-3b also has the largest value for the NUV radius relative to the host star in our sample. The particularly high mass of XO-3b should prevent it from experiencing significant mass loss. The NUV radius, R_p (uvm2) = 2.54 ± 0.22 R_J, is well within the planet's Roche lobe of 9.4 R_J. Given the upper limit of the X-ray flux, we find that the total extreme-UV (EUV) luminosity of the star is <7.5 × 10²⁹ erg s⁻¹, and the predicted mass loss for the system is <2 × 10¹⁰ g s⁻¹. This is the lowest mass-loss rate calculated for the planets in our sample.

The 1.5 M_J hot Jupiter KELT-3b has an optical radius of 1.3 R_J (or 1.5 R_J using the stellar radius from Gaia DR2) and orbits an F7V-type star (Pepper et al. 2013). The insolation received by the planet, 10⁹ erg s⁻¹ cm⁻², places it in the class of hot Jupiters observed to have "inflated" radii (Demory & Seager 2011).

We obtained two visits of the KELT-3b transit in 2019. Preliminary analysis of that work hinted at a transit, but an anomalous flare appeared to contaminate the light curve (Ravi et al. 2020). However, KELT-3 does not exhibit strong Ca ii H and K lines, which are positive indicators of stellar activity (Pepper et al. 2013). In this work, we found that normalizing the KELT-3 light curve by the averaged light curve of our reference sources removed the apparent flare, demonstrating that our techniques are sufficient to remove instrumental variations from the light curve (see Appendix). To compliment this conclusion, we obtained a subsequent transit observation in 2020 (visit 3) to fill sample gaps in the original light curve.

Our best-fit transit depth is 2.6% ± 0.6%, which is 70% deeper than the optical transit. Our p-value of 0.7% implies a 2.5σ detection, but visual inspection of the fit contours (Figure 2, right) suggest that it is only distinguishable from the null hypothesis by about 1.5σ.

The Roche lobe size for KELT-3b is 4.3 R_J, which is nearly a factor of 2 larger than its NUV radius of 2.5 ± 0.3 R_J. A search for Hα emission as a signature for an extended or escaping atmosphere from KELT-3b yielded a nondetection (Cauley et al. 2017). They placed an upper limit on the size of the Hα absorbant atmosphere that is 10% larger than the optical planet disk. However, the lack of an Hα signature is most likely a sign that <0.1% of the hydrogen in the KELT-3b atmosphere is in an excited state, which can be expected in planetary systems older than 1.2 Gyr (Allan & Vidotto 2019). The EUV flux incident on the planet is <5 × 10²⁹ erg s⁻¹, and the estimated mass-loss rate is <1.5 × 10¹¹ g s⁻¹. This places KELT-3b in a moderate mass-loss regime for our sample, even though it does not have a visibly escaping atmosphere.

The 1.9 M_J planet WASP-3b has a 1.4 R_J radius and orbits a type F7-8V star with a very short period of 1.85 days, making it one of the the most irradiated hot Jupiters at the time of its discovery (Pollacco et al. 2008; Bonomo et al. 2017). Eclipse measurements with Spitzer revealed dayside temperatures of 2200–2400 K, placing it in the category of ultrahot Jupiters, and best-fit models over the 2–10 μm spectral energy distribution imply that WASP-3b has an inverted temperature profile with ineffcient heat exchange between the day- and nightsides of the planet (Zhao et al. 2012; Rostron et al. 2014). The ultrahot status of WASP-3b makes it similar to WASP-121b, which exhibits a range of dayside temperatures from 2300 to 2800 K (Bourrier et al. 2020; Garhart et al. 2020). By virtue of its similarity to WASP-121b, WASP-3b is the top planet in our sample from which one might expect the deepest NUV transit.

We searched for an NUV transit from three visits of Swift taken over the course of 1 month. Our best-fit transit depth, 3.2% ± 0.7%, has a p-value of 5%, implying a marginal <2σ detection. This conclusion is also supported by the contours in Figure 2 (right). However, in examining the light curve in Figure 2 (left), it appears that the transit fit is biased by three data points showing a 5%–10% lower flux measurement, all of which arise from a single exposure within transit visit 3 (see the Appendix). We are unable to attribute the anomalously low flux measurement to any obvious artifact in the image file. Based on the lack of obvious transit signatures from the remaining light curve, we adopt the 1σ null confidence value as the upper limit to the WASP-3b transit depth.

The corresponding upper limit for the radius of WASP-3b is 2.5 R_J, which is smaller than its Roche lobe radius, 3.6 R_J. The upper limit to the X-ray flux of WASP-3 is consistent with, and more restrictive than, a previous X-ray brightness limit reported by Kashyap et al. (2008). We calculate an EUV luminosity of <8 × 10²⁹ and a corresponding evaporative mass-loss rate of <2 × 10¹¹ g s⁻¹. This is consistent with the evaporation rate of 4 × 10¹¹ g cm⁻¹ predicted by Bourrier et al. (2015), who assumed η = 0.3 for the heating efficiency.

Object HAT-P-6b has an optical radius of approximately 1.5 R_J, a mass of 1.1 M_J, and an optical insolation of 2 × 10⁹ erg cm⁻² s⁻¹, placing it in the class of inflated hot Jupiters (Noyes et al. 2008; Demory & Seager 2011; Bonomo et al. 2017). Swift made two visits of the HAT-P-6b transit, but the planet was not detected (Figure 3, right). The NUV transit upper limit of R_p /R_* < 0.16 implies a radius R_NUV < 2.5 R_J. The Roche lobe for this system is approximately 7 R_J, substantially larger than the NUV radius. The observational limit on the X-ray flux implies an EUV luminosity of <10³⁰ erg s⁻¹, which yields an estimated mass-loss rate of <3 × 10¹¹ g s⁻¹.

Zoom In Zoom Out Reset image size

Object WASP-62b orbits a late F-type star and exhibits the largest optical transit depth of all targets in this work. Its large radius of 1.4 R_J, low mass of 0.6 M_J, and optical insolation of 10⁹ erg cm⁻² s⁻¹ put WASP-62b in the curious category of low-density inflated planets (Hellier et al. 2012; Bonomo et al. 2017; Brown et al. 2017). The young age of this planetary system, ∼0.4–1.3 Gyr based on gyrochronology (Hellier et al. 2012), suggests that heating via XUV radiation, which dominates younger systems (King & Wheatley 2021), might also be important for explaining this planet's low-density appearance.

Swift visited WASP-62 three times over the course of a month. However, the first visit missed the transit of WASP-62b by about 4 hr. Visits 2 and 3, unfortunately, sparsely sampled the transit. Relatively few data points are available outside of the transit window, making it particularly difficult to obtain a baseline light curve for the host star. We attribute these factors to our null result, demonstrated in Figure 3 (right). This data set provides us with an upper limit of R_p /R_* < 0.11 (R_NUV <1.4 R_J), which is consistent with the optical radius of WASP-62b.

The Roche lobe radius for WASP-62b is 6 R_J, a factor of 4 larger than the upper limit to the NUV radius. As described in Section 2.3, WASP-62 was the only host star in our sample with a positive X-ray detection. The estimated EUV luminosity of the host star is (1.4 ± 0.2) × 10³⁰ erg s⁻¹, and the estimated mass-loss rate is 5 × 10¹¹ g s⁻¹. This is a factor of 2 larger than the value predicted by Bourrier et al. (2015).

Here we test the hypothesis set forward by S19 that NUV transit depth provides a probe of atmospheric escape. One of the leading theories to explain atmospheric mass loss is photoevaporation, which is dominated by the EUV (40–912 Å). In Section 2, we utilized the X-ray luminosity of the host star to estimate the EUV component and corresponding mass loss by photoexcitation processes. However, the X-ray luminosity of a host star might also correlate with other evaporative processes, such as mechanical erosion by stellar winds and coronal mass ejections. In this section, we look for correlations in R_NUV/R_opt for the observed (X-ray irradiation) and derived (photoevaporative mass-loss rate) characteristics. We calculated the X-ray insolation of each planet by scaling the unabsorbed luminosity to the flux at the semimajor axis of the planetary orbit. In four out of our five cases, we are only able to derive upper limits.

In this examination, we also include WASP-80b (King et al. 2018), WASP-121b (S19), and HD 189733b (King et al. 2021), for which positive NUV transit detections have been derived and X-ray measurements of the host stars exist. The X-ray insolation for HD 189733b is derived from the X-ray luminosity measurement of Poppenhaeger et al. (2013). We treat the X-ray measurement of WASP-121 as the upper range of the stellar luminosity, because S19 reported that X-ray emission from the star was only detected in a subset of the Swift observations. The measurement likely characterizes the star in a flare state, not its baseline level of activity. We also plot the WASP-121b NUV measurement derived in this work (Table 2), rather than the S19 value. Finally, note that the NUV transit depth for WASP-80b was obtained with the XMM-Newton Optical Monitor (OM) instrument, which has the same filters as the Swift-UVOT. The WASP-80b data point differs from the other measurements in two significant ways. First, the light curve was obtained using a different instrument operation mode (fast window) and analysis techniques. Second, the data were taken with the uvw1 filter, which covers a redder wavelength range than uvm2.

Figure 4 (left) shows the ratio of the NUV to optical planet radius as a function of X-ray insolation. The estimated photoevaporative mass-loss rate is plotted in Figure 4 (right). For the host stars with positive X-ray detections, WASP-80 and WASP-62, we injected a 50% uncertainty on the value of η to account for the uncertainty in both the XUV flux derived from the 0.2–2.4 keV measurement and the mass-loss efficiency. The energy-limited mass-loss rate for HD 189733b is taken from Kubyshkina et al. (2018), assuming a 30% uncertainty. The radius derived from NUV transit measurements does not appear to be positively correlated with the mass-loss rate.

Zoom In Zoom Out Reset image size

It is difficult to interpret the large NUV transit depths observed for XO-3b and potentially KELT-3b under the current paradigm of atmospheric escape. Rudimentary cloud-free models for the UV transmission spectra of exoplanet atmospheres have demonstrated that Fe ii is the dominant gas-phase absorber in the 200–250 nm range, which is covered by the uvm2 filter (Turner et al. 2016; S19). The Fe ii absorption is generally predicted to increase with increasing irradiation—as gauged by equilibrium temperature—but the predicted NUV transit depth relative to the optical (600–700 nm) does not exceed 20% and peaks at T_eq ≈ 2200 K (Lothringer et al. 2020). If Fe ii absorption is responsible for the deep NUV transit of XO-3b, which has an estimated mass-loss rate that is nearly 1000 times smaller than that of WASP-121b, then this work challenges the assumption that the strength of Fe ii absorption correlates with mass loss. This conclusion is supported by the theoretical work of Young et al. (2020), who predicted significant Fe ii absorption using a CLOUDY model for the exosphere of HD 209458b, a well-studied hot Jupiter orbiting an F-type star with a relatively small estimated mass-loss rate (10¹⁰ g s⁻¹). More work is needed, in both modeling individual star–planet systems and examining the transit signal under the relevant instrumental resolution, to say whether exospheric Fe ii can explain the NUV transit depths in this work.

Another explanation for a larger NUV radius relative to the optical could be the presence of aerosols with particle sizes <1 μm, which are highly efficient at attenuating NUV light. In general, clouds are expected to form more easily in environments with T < 2000 K, at which point aluminum oxide, iron, and silicate condensates can form (e.g., Morley et al. 2012; Sing et al. 2016; Wakeford et al. 2017). This motivates an examination of NUV transit depths, as they depend on planet equilibrium temperature.

We plot the ratio of the NUV and optical planet radius versus equilibrium temperature (Figure 5, top). We find that, for T ≳ 1700 K, R_NUV/R_opt > 1 and can be as large as 2. For T ≲ 1700 K, we have only two data points, exhibiting R_NUV/R_opt ≤ 1. In a study of the XMM-OM transit of HD 189733b, which has T_eq = 1209 ± 11 (Addison et al. 2019), the NUV radius is equal to the optical radius within 1σ uncertainty for the uvm2 filter and 2σ uncertainty for the uvw2 filter (King et al. 2021). This holds with the general observed trend.

Zoom In Zoom Out Reset image size

Around 1700 K, a transition is also observed in the strength of the 1.4 μm water feature measured from hot Jupiters (Fu et al. 2017). One-dimensional models of hot Jupiter atmospheres by Gao et al. (2020, hereafter G20) demonstrated that this could be explained by the onset of cloud formation, which can obscure the water signatures. The G20 results for cloud particles sizes and atmospheric pressures contributing to the near-IR (NIR) opacity near the water features of interest are plotted in Figure 5 (bottom). In this scenario, water features become obscured as cloud particles—mostly silicates—form around 10 mbars. At lower temperatures, the cloud particles can reach sizes as large as 8 μm but sink to 100 mbars, allowing some water in the upper atmosphere to be visible. Below 1000 K, hydrocarbon hazes begin to form, dominating the opacity at the higher-altitude 1–10 mbar atmosphere and obscuring the water features again.

The G20 models might also explain the observed trends in R_NUV/R_opt. The dashed vertical line in Figure 5 (T_eq = 1700 K) marks the highest temperature at which water-obscuring cloud particles >1 μm can form. This value happens to coincide with the peak R_NUV/R_opt value. Even though 1 μm is larger than the grain size expected to maximize NUV attenuation, numerical models of cloud formation generally yield a high-density cloud deck full of large particles, with smaller particles extending into higher layers of the atmosphere. Powell et al. (2018) demonstrated that assuming a single mean particle size underestimated cloud opacities by a factor of 3–5 compared to using the full particle size distribution. Additionally, hotter atmospheres had cloud tops with smaller particles that extended to higher altitudes, leading to a steeper slope in the optical transmission spectrum (Powell et al. 2018, 2019). Thus, at T_eq > 1700 K, we expect a population of cloud particles with radii <1 μm to contribute significantly to the NUV transit depth.

Cloud particles with a ≥ 1 μm are gray in opacity, obscuring NUV and broadband optical light equally (Figure 6). Thus, the NUV and optical transits of hot Jupiters in the T_eq ≲ 1700 K regime could probe similar atmospheric regions—the atmosphere just above the cloud deck—leading to NUV transit depths that are roughly equal to the optical transit depths. WASP-62b, which is near the temperature range where G20 predicts the largest cloud particles, exhibits pressure-broadened Na I absorption features and has a high-resolution transmission spectrum that can be fit without clouds (Alam et al. 2021). This suggests that any clouds that may be present on WASP-62b have sunk well below the pressure at which the atmosphere becomes optically thick.

Zoom In Zoom Out Reset image size

At 800 K, where WASP-80b sits, the G20 model is at a transition point between a silicate cloud and a haze-dominated atmosphere. Because hazes are likely to be efficient NUV absorbers, one might expect the apparent NUV radius of the planet to be larger than the optical. However, the observed WASP-80b NUV transit is shallower than the optical transit. This could be because the WASP-80b observation was conducted with the uvw1 filter, which is redder than uvm2 and extends significantly into the U band. Figure 6 plots the extinction efficiency of polycyclic aromatic hydrocarbons (PAHs; Li & Draine 2001), which we treat as a proxy for large hydrocarbon molecules, along with the optical depths for samples of Titan-like and oxidized hazes produced in the lab (Gavilan et al. 2018). Because the uvw1 filter misses the peak of these absorption profiles, it is possible that the WASP-80b measurement happens to probe a relatively transparent window for typical atmospheric hazes. Measuring transits of cooler hot Jupiters such as this one in different NUV filters could be important for examining the prevalence and composition of hazes in exoplanet atmospheres.

Despite the parallels between the NUV transit depth and the current state of cloud formation simulations, none of the transmission spectrum models presented for T_eq > 1800 K hot Jupiter atmospheres suggest that NUV transit depths could be more than 10% larger than the optical (Powell et al. 2019). However, lab experiments have demonstrated that hazes can form even in hot atmospheres around 1500 K (Fleury et al. 2019; He et al. 2020). To fully investigate the hypotheses that aerosols could contribute to NUV transit depths, modeling the dual populations of cloud condensates and photochemically produced hazes is necessary. Helling et al. (2020) recently simulated these two physical mechanisms of aerosol production for the case of WASP-43b, which has T_eq ≈ 1400 K (Esposito et al. 2017). Using the photochemical models of Kawashima & Ikoma (2018), they predicted that hazes in WASP-43b could extend to atmospheric pressures that were 1–3 dex lower than the pressure levels occupied by cloud particles. However, their simulated hazes contributed far less to the NUV atmospheric opacity compared to cloud particles. In addition to using a single particle size to estimate the haze opacity, Helling et al. (2020) used the optical properties of Titan hazes, which are less absorbant in the NUV range than oxidized hazes (Gavilan et al. 2018, demonstrated in Figure 6).

Hot Jupiter HD 189733b is well known for having a steep optical slope in its transmission spectrum, which might be explained by atmospheric aerosols (Sing et al. 2011; Pont et al. 2013; Barstow 2020; Sánchez-López et al. 2020). Using three-dimensional circulation models of photochemically produced hazes, Steinrueck et al. (2021) demonstrated that enhanced vertical mixing, or optical properties of hazes that have stronger reddening properties than soot, are one of a few options for explaining the steep slope of the optical transmission spectrum of HD 189733b. The error bars from broadband NUV transit measurements obtained by King et al. (2021) are too large to place strong constraints on the aerosol type and are consistent with the planet radius in the optical. However, it is worth noting that the best-fit value for R_p /R_* from the bluer uvw2 filter (measured from 16 separate transit visits, resulting in a 5% level precision) is larger than that obtained with uvm2 (measured from a single transit, resulting in a 15% level precision). Such behavior might be expected from some species of aerosols or silicate grains smaller than 0.1 μm (Figure 6). Because of the broad diversity in planetary properties and radiation fields, detailed modeling of individual systems will be important to investigate whether aerosols can produce the NUV transit depths found in this work.

Heterogeneity on the stellar surface, such as starspots and faculae, is a source of systematic bias for the transmission spectra of exoplanets. At optical to infrared wavelengths, transit depths are only expected to be affected on a ≲5% level for F stars such as those in this study (Rackham et al. 2019). At shorter wavelengths, the magnetic activity that produces dark starspots in the photosphere (optical) will appear as bright patches in the chromosphere (UV). If the exoplanet happens to pass in front of these active areas, the short-wavelength transit will appear deeper than the true planet radius at that wavelength. However, the magnitude of the effect is inversely proportional to wavelength (Llama & Shkolnik 2015), and the 200–250 nm range probed by the uvm2 filter is expected to be the least affected UV band.

Averaging over multiple transits also reduces the likelihood that a transit depth will be under- or overestimated. Llama & Shkolnik (2016) showed that, at Lyα wavelengths (121 nm), combining the data from two transit visits was sufficient to recover the true planet radius 80% of the time. Because the effect is less pronounced for longer wavelengths, we expect our two to three visit light curves to yield a reliable transit depth. Assessing the effect of starspots and stellar variability on our targets would require additional transit visits and stellar activity monitoring.

Finally, we note that the object in our study with the highest R_opt/R_NUV value, XO-3b, is also the planet that is most subject to heating by tidal dissipation. The fact that the observed NUV transit occurs earlier than predicted optical transits hints that the NUV absorption might arise from other gas in the system, such as a compressed bow shock formed as the planet's atmosphere interacts with the stellar wind (e.g., Llama et al. 2013). Thus, the uniqueness of this result might be a symptom of the uniqueness of this planetary system.