Main

We observed one transit of the ultra-hot Jupiter exoplanet WASP-178b/KELT-26b (refs. 7,8) (WASP-178b hereafter) with Hubble Space Telescope (HST)’s Wide Field Camera 3 (WFC3) ultraviolet–visible channel (UVIS) using the G280 grism (0.2–0.8 µm, R = λ/Δλ ≈ 70, where λ is the wavelength of light) on 5 September 2020 as part of Program 16086 (principal investigator: J.D.L.). WASP-178b is unique among known exoplanets for its especially hot host star: at A1 IV-V and Teff = 9,360 K, WASP-178 is second only to KELT-9 as the hottest planet-hosting star. Time-series spectra were obtained over 7.5 h centred around the transit event (Extended Data Figs. 1 and 2), and they were used to extract the transmission spectrum, which probes the middle and upper atmosphere around the day–night terminator. Further details regarding the data reduction and observational setup are given in the Methods.

The resulting spectrum is shown in Fig. 1 (also see Extended Data Fig. 3 and Extended Data Table 1). A steep rise in transit depth is seen towards short wavelengths, beginning at about 0.35 µm. The difference in the near-ultraviolet (NUV) transit depth (0.2–0.28 µm) compared with the optical transit depth (0.35–0.8 µm) is 2,500 ± 138 ppm, which is an 18.0σ significance. In terms of the equilibrium atmospheric scale height (calculated at the equilibrium temperature and assuming an H2-dominated atmosphere), the NUV transit depths rise nearly 20 scale heights above the optical continuum, making this one of the largest known spectral features seen in an exoplanet atmosphere. A lack of transit asymmetry indicates that this absorption is present on each limb (Extended Data Fig. 4 and Methods). We are able to rule out a scattering slope from, for example, photochemical hazes, and stellar inhomogeneities as causes for this feature (Methods). We are thus confident that the rise in the transit depths at NUV wavelengths is due to absorption by gaseous species in the atmosphere of WASP-178b.

Fig. 1: WASP-178b NUV-optical transmission spectrum.
figure 1

a, WFC3/UVIS G280 transmission spectrum of WASP-178b (with 1σ uncertainties) compared to the contribution of various important opacity sources in the retrieved best-fit spectrum. SiO dominates the contribution at short wavelengths. b, The 1σ constraint on the pressure–temperature profile (shaded region) with the median retrieved profile (solid line) on the same radius scale as the left plot, indicating that the minimum pressures probed by our observations are less than a microbar.

Full size image

We ran a series of retrievals to obtain constraints on the atmospheric properties from the observations (Table 1). The retrievals included free parameters for the abundance of major NUV and optical absorbers, including SiO, TiO, VO, Fe i, Fe ii, Mg i and Mg II, plus a general [Fe/H] parameter for the abundance of all other atmospheric species. We also used a five-parameter temperature structure parameterization (described in Methods). Figure 1 shows the retrieved best-fit spectrum from our fiducial retrieval with all opacity sources present and the contribution from these various opacity sources. Also plotted are the constraints on the temperature structure. We show the two-dimensional posterior distributions from the fiducial retrieval in Extended Data Fig. 5 along with the retrieved parameters in Extended Data Table 2. The maximum transit depths of about 1.5% correspond to a radius of about 2RJupiter, where RJupiter is the radius of Jupiter. This corresponds to a pressure of about 1 µbar, similar to the strong lines of Na and K at high spectral resolution.

Table 1 Retrieval results summary
Full size table

In all the retrievals we tested, two scenarios were able to fit the data: (1) an atmosphere with an approximately solar abundance of SiO or (2) an atmosphere with a super-solar abundance of Mg i and Fe ii but with no SiO. SiO absorbs throughout the 0.2–0.35 µm range, enabling a good fit to the data. On the other hand, bound-free opacity from Mg i absorbs shortwards of 0.255 µm (refs. 9,10), whereas Fe ii absorbs between 0.24 and 0.3 µm. Thus, the combined absorption from Mg i and Fe ii is also able to provide an adequate fit to the observations, albeit at super-solar abundances (Extended Data Fig. 6). Taken in tandem, our analysis indicates that there is strong evidence (with a difference in the Bayesian information criterion ∆BIC = 8.39) that SiO or Mg must be present in the atmosphere of WASP-178b to explain our observations. As both Mg and SiO are the major constituents of silicate condensates such as enstatite (MgSiO3) and forsterite (Mg2SiO4), we can say with equally high confidence that silicates have not rained out at the terminator of WASP-178b. Owing to the lack of transit asymmetry mentioned above, this result holds for both the evening and morning terminator.

This result is in line with theoretical expectations. Previous studies have pointed out that, above 2,000 K, SiO is expected to be a major absorber shortwards of 0.4 µm at its chemical equilibrium abundance11,12 and so we would expect to see it in the transit spectrum a priori (also see Extended Data Fig. 7). Similarly, we expect many neutral and ionized atomic species to be present, as indicated by chemical equilibrium calculations and high-resolution observations (for example, ref. 13). Although SiO and Mg i with Fe ii independently provide good fits to the data, our previous expectation based on chemical equilibrium is that each of these opacity sources are probably present.

The non-detection of neutral Fe is somewhat surprising as the species has been detected in planets of similar temperature, such as for KELT-20b/MASCARA-2b14, WASP-76b15,16 and WASP-121b13,17,18,19. The apparent absence of significant Fe i absorption in WASP-178b’s spectrum could be explained by the high temperatures and high ultraviolet (UV) flux from the A-type host star ionizing most of the Fe i. Ground-based high-resolution studies of WASP-178b could provide a comprehensive survey of neutral and ionized refractory species, as has been done for other ultra-hot Jupiters13,20,21.

To spectrally resolve potentially escaping neutral and ionized Fe and Mg features, we additionally analysed high-resolution NUV transit observations of WASP-178b taken with HST’s Space Telescope Imaging Spectrograph (STIS) E230M grating (Methods). The STIS data show no evidence for either Fe ii or Mg ii (Extended Data Fig. 8), even though both elements were easily detected in similar data of WASP-121b17. The STIS E230M transmission spectrum is in good agreement with the broadband UVIS spectrum, indicating unresolved escaping Fe ii and Mg ii lines are not the cause of the NUV absorption feature in the UVIS spectrum, with continuum level absorption by SiO, Mg i and Fe ii the most likely scenario.

Only a handful of observations exist that are precise enough to measure the continuum of exoplanets shortwards of 0.35 µm: HAT-P-41b (Teq = 1,950 K) has been observed with a similar setup to our observations of WASP-178b (that is, HST/WFC3/UVIS G280)22, whereas WASP-121b (Teq = 2,350 K) has been observed at high resolution with HST/STIS E230M with four binned points between 0.23 and 0.31 µm (ref. 17). A clear difference between the spectrum of HAT-P-41b and those of the hotter WASP-121b and WASP-178b is apparent (Fig. 2). Although WASP-121b’s transit spectrum indicates a similar level of absorption at NUV wavelengths to that of WASP-178b, HAT-P-41b’s spectrum shows a definite absence of absorption on both limbs at these same wavelengths. This dichotomy indicates that whereas gaseous refractory species such as SiO, Mg and Fe are abundant in the atmospheres of WASP-121b and WASP-178b, such species have rained out of the gaseous phase in HAT-P-41b. Because some of the NUV absorption in WASP-121b is from escaping exospheric metals17 and the difference between the morning and evening terminator has not been well-constrained as in WASP-178b, we choose to define the equilibrium temperature of WASP-178b as the empirical upper limit for the onset of silicate condensation in hot Jupiters. Therefore, silicate condensation at the terminators must begin between equilibrium temperatures of 1,950 K and 2,450 K.

Fig. 2: Comparison of NUV-optical transmission spectra.
figure 2

WFC3/UVIS G280 transmission spectrum of WASP-178b (with 1σ uncertainties) compared to the UV and optical spectra of similar giant planets HAT-P-41b22 and WASP-121b33,34, normalized by each planet’s equilibrium temperature scale height, Heq. Substantial UV absorption is seen at the shortest wavelengths in WASP-121b and WASP-178b.

Full size image

This empirical constraint on the onset of condensation is consistent with theoretical predictions4,23. Figure 3 compares pressure–temperature profiles from theoretical one-dimensional atmosphere models of HAT-P-41b, WASP-121b and WASP-178b to condensation curves of silicate and iron species. In equilibrium, silicates and iron will condense between about 1,500 and 2,000 K between 1 mbar and 10 bar for atmospheric metallicities between one times and ten times solar metallicity. Throughout the atmosphere, HAT-P-41b is much closer to the silicate condensation curve than WASP-121b and WASP-178b and will almost certainly cross it on the nightside. If WASP-121b and WASP-178b do reach temperatures cool enough to condense silicates on the nightside, it also appears they are both able to avoid rainout on either limb through rapid evaporation, vertical lofting, insufficiently rapid nucleation and condensation, or some combination of these and other hydrodynamic and microphysical processes24,25,26. At depth, where the temperatures in WASP-121b and WASP-178b are the closest to the condensation curves, the higher internal temperature in hot and ultra-hot Jupiters may also help such planets to avoid condensation27.

Fig. 3: Atmospheric structures and condensation curves.
figure 3

Pressure–temperature profiles of three ultra-hot Jupiters from atmosphere models (Methods) compared with the condensation curve of Fe and Mg2SiO4 in a 1–10 times solar metallicity atmosphere23. Clouds are expected to form in equilibrium where the profile of the planets intersects the condensation curve, as in the case of HAT-P-41b. We also show the retrieved median and 1σ confidence interval for the 1D pressure–temperature profile of WASP-178b for comparison, demonstrating agreement with self-consistent atmosphere model expectation.

Full size image

As noted above, few NUV transit spectra exist for hot Jupiters. Future low- and high-resolution observations, combined with multi-dimensional theoretical modelling24,25,26 and laboratory studies of aerosols28,29 in hot and ultra-hot Jupiters, could provide more detailed constraints on the beginning of cloud formation in these atmospheres, while taking into account the myriad processes that promote or inhibit cloud formation, such as nightside cold-trapping, rainout and vertical mixing and other potentially confounding variables such as surface gravity and host star type. We estimate that about 20 Jovian exoplanets can be characterized with HST/WFC3/UVIS G280 with four or less transits30,31. When combined with STIS E230M observations to disentangle the effects of atmospheric escape as we have done here, these planets can reveal the conditions for, and sequence of, condensation in exoplanet atmospheres. Observation, modelling and retrieval analysis of brown dwarfs at similar effective temperatures will also shed light on these questions3,32.

Methods

HST/WFC3 observations

One transit was observed of WASP-178b with the HST/WFC3/UVIS instrument using the G280 grism (0.2–0.8 µm). The transit is covered over five HST spacecraft orbits, with the transit approximately centred in the third orbit. Exposure times of 40 s were used, along with a 590 × 2,250 pixel detector subarray that reduced the readout overheads providing 123 total exposures, with 23 to 25 exposures per HST orbit. The spectrograph is slitless, and we centred the subarray such that both the +1 order and −1 order spectra were recorded and could both be fully analysed. Using both orders provides two independent transmission spectra of the same transit event, although the +1 order provides higher signal-to-noise ratio given a higher throughput (Extended Data Fig. 3). Also see ref. 22 for more information on the instrument mode and analysis.

Data reduction

The raw data were processed with the STScI CALWF3 pipeline (v.3.5.1), which applies reduction steps including bias subtraction, dark correction and flat fielding. The target flux was subsequently extracted starting from the pipeline FLT files. From each image we used the mode to measure and subtract the background flux. We then removed cosmic rays following a two-step process. First, we identified and removed cosmic rays using the time series counts of each pixel. Outlier cosmic rays were flagged and replaced with a 5σ clipping algorithm. We then removed cosmic rays spatially, using a Laplacian edge detection algorithm on each image separately35. We then extracted the one-dimensional spectral flux for each image on both the −1 and +1 orders separately using the IRAF software apall with an eighth-order Legendre polynomial fit to the spectral trace. A large range of aperture sizes were extracted, between 10 and 28 pixels, with a 14-pixel aperture found to be optimal in the subsequent light curve fitting stage. The wavelength solution was determined from the spectral trace detector position following ref. 36.

UVIS light curve analysis

The general light curve fitting followed the procedures detailed in ref. 17 and previously used on WFC3 G280 observations in ref. 22, which we refer the reader to for subsequent details. The flux measurements over time, f (t), were modelled as a combination of a theoretical transit model37, T(t, θ) (which depends on the transit parameters θ), the total baseline flux of the star, F0, and an instrument systematics model S(x) giving

$$f(t)=T(t,\theta )\times {F}_{0}\times S({\bf{x}}).$$
(1)

As in ref. 17, we explored a wide range of models for S(x), investigating detrending variables including a fourth-order polynomial in HST orbital phase and linear terms in spectral position as measured from the spectral extraction, wavelength shift measured from cross-correlation of each spectra and the spacecraft jitter detrending vectors, which are products of HST’s Engineering Data Processing System. For both the +1 and −1 orders, we used the Akaike information criterion (AIC) with a correction for small sample sizes to determine the optimal detrending variable parameters to include from the full set without overfitting the data and while minimizing the red noise. The light curve error bars were derived from the residual scatter of the best fit. In addition, residual systematic noise, σr, was measured along with the white noise, σw, using the binning technique38, with the final fit parameter errors inflated by a factor β if red noise was present39 (Extended Data Table 1). Overall, the effect of red noise was minimal as the noise was comparable to the binned photon noise at typically several hundred parts per million, and the errors in only 5 of 124 light curves required an increased scaling by more than 10%.

To model the effects of limb-darkening and centre-to-limb differences in the star during transit, we calculated a custom PHOENIX stellar model40 using the parameters of WASP-178. For each wavelength bin of interest, we then used the stellar intensity profile to fit for limb-darkening coefficients using the non-linear four-parameter limb-darkening as described in ref. 41, which were subsequently used in the light curve transit fit.

As a test of the stellar models, we additionally fitted for the limb-darkening assuming a linear law between the wavelengths of 0.255 and 0.285 µm. At these NUV wavelengths, the stellar limb-darkening is strong, but is also predicted by models to be nearly linear in intensity across the limb, enabling a direct comparison largely free of complex degeneracies between transit-fit coefficients. We found the transit data fitted with a linear coefficient of u = 0.778 ± 0.028, which matches very well to the PHOENIX model prediction of u = 0.7758. The limb-darkening was subsequently fixed to the PHOENIX model values for all the transit light curve fits.

We first fitted the white-light curve, which integrates the entire spectra (Extended Data Fig. 1). The planet’s orbital system parameters, including inclination, i, semi-major axis in units of stellar radii, a/Rstar, and time of transit, T0, were fitted along with the planet-to-star radius ratio, Rpl /Rstar, F0 and the systematics model using the +1 order. Given the good phase coverage, these system parameters were generally more accurate than previous literature values, and we found i = 84.41 ± 0.20 degrees, a/Rstar = 6.588 ± 0.091 and T0 = 2459097.869279 ± 0.00014 days. We kept the period fixed to the literature value of 3.3448285 ± 0.0000012 days7.

To derive the planetary transmission spectrum, we fixed the system parameters to the white-light curve best-fit values and used the optimal systematics model S(x) as determined by the white-light curve fit. Although the white-light curve did not require a term to model wavelength shifts in the time-series spectra, this term was found to be needed in the spectral light curve fits with ten fit parameters overall for each light curve. The residual fit scatter in the time-series spectral bins achieved a level that was on average 1.2 times the theoretical photon-noise limit scatter and typically ranged between 1.1 times and 1.4 times the scatter (Extended Data Fig. 2). A variety of spectral bin locations and resolutions were measured, with the adopted spectra chosen to balance the resolution and signal-to-noise ratio. In each case, the overall shape of the transmission spectrum was consistent between different resolutions. We measured the transmission spectrum of both the +1 and −1 orders, finding good agreement between both orders (Extended Data Fig. 3). We calculated the weighted mean value of the spectra to report our final derived spectrum (Extended Data Table 1) as seen in Fig. 1.

We also independently verified the transmission spectrum using the marginalization method described in ref. 22. This method resulted in a spectrum that was consistent with the spectrum analysed with jitter detrending, both showing large NUV absorption. This method used limb-darkening coefficients from the Kurucz stellar model grid42, again indicating our results are robust against the details of limb-darkening.

STIS E320M light curve analysis

To help resolve possible Fe ii and Mg ii features in the NUV spectrum of WASP-178b, we also analysed a transit observed on 30 July 2020 by HST with the STIS E230M instrument. As with the UVIS data, these observations were also taken as part of HST Program 16086. The STIS observations were observed with the NUV-MAMA detector using the Echelle E230M STIS grating and a square 0.2" × 0.2" aperture. The E230M spectra has a resolving power of R = λ/(2∆λ) = 30,000 and we set the grating to 2,707 Å to cover the wavelength ranges from 2,280 to 3,070 Å across 23 orders.

Our analysis closely follows that of ref. 17, which we refer to for additional method details. We used the jitter detrending method to correct for time variable slit losses seen in the white-light curve photometry, and fitted the light curves using the system parameters given in Extended Data Table 2. We found a band integrated white-light curve planet-to-star radius ratio (Rp /Rs) of 0.1244 ± 0.0050, which matches well (0.6σ significant difference) when compared to the same wavelength region as measured by UVIS (Rp /Rs = 0.12133 ± 0.00054). The main difference in our method from that of ref. 17 was the use of a common-mode analysis when analysing the spectroscopic channels, in which the best-fit transit model was removed from the white-light curve raw photometry and used to remove common instrument trends. In this case, large slit losses repeating every spacecraft orbit are seen in the photometry. The common-mode analysis removes the majority of the instrument trends seen, with the remaining trends modelled with a second-order polynomial in the HST orbital phase. The spectroscopic channels each reach a residual scatter that is consistent with the photon noise level. As we did for WASP-121b, we divided the E230M spectra into 196 spectroscopic channels each with a 4 Å bandpass. The resulting spectrum can be seen in Extended Data Fig. 8. Compared with WASP-121b, WASP-178b does not show strong Fe ii or Mg ii absorption features, with the high-resolution E230M spectrum consistent with the broadband NUV spectrum of UVIS.

Atmosphere models

Self-consistent 1-D PHOENIX atmosphere models43 of HAT-P-41b, WASP-121b and WASP-178b were computed to compare the expected atmospheric temperatures in each planet to refractory species condensation curves. The model setup was similar to past ultra-hot Jupiter studies with PHOENIX12,44,45 that are computed on a 64-layer optical-depth grid from τ = 10−10 to 102, which corresponds to similar magnitudes in pressure. We ran models for two internal temperatures, 200 K and 700 K, which correspond to the lowest and highest internal temperatures expected for a hot Jupiter27. In addition we assumed full planet-wide heat redistribution to approximate temperatures at the terminator, consistent with similar investigations4. The temperature structure from the self-consistent model of WASP-178b is quite similar to the retrieved temperature profile from PETRA (Fig. 3), indicating the full heat redistribution assumption appears to be a good approximation of the average conditions at the terminator.

The model includes opacity from 130 molecular species and neutral and ionized atomic species up to uranium. TiO and VO are important visible-wavelength opacity sources46,47, but observations indicate that they do not always appear to be present in the atmosphere48. Consistent with retrievals of observations from these planets33,49, the HAT-P-41b model did not include TiO or VO, the WASP-121b model only included VO and the WASP-178b model only included TiO but at a reduced abundance (Extended Data Fig. 5 and Extended Data Table 2). All other abundances, including SiO and Fe, were treated in local chemical equilibrium. In addition, we used the modelled HAT-P-41b transmission spectrum to rule out the presence of gaseous refractories on one or both limbs, further supporting the fact that these elements have rained out.

Atmosphere retrievals

We used PETRA50 to retrieve atmospheric properties from the observations. PETRA uses a differential-evolution Markov Chain statistical framework51 to sample the posterior distribution of the parameter space. Our retrieval setup was similar to previous transmission retrievals with PETRA52,53. We parameterized the temperature structure using the five-parameter approach of ref. 54. We also included a necessary reference radius parameter. The abundances of major UV and optical opacity sources were treated as free parameters with uniform vertical abundance. These included Fe, Fe ii, Mg, Mg ii, TiO, VO and SiO. We also included a free parameter for the metallicity ([Fe/H]) of the rest of the atmosphere, which was treated in chemical equilibrium. This enabled the effect of other potential absorbers expected to be of lesser importance (for example, H−, Ca, Ni and FeH) to be taken into account while reducing the number of free parameters to explore. Continuous opacity from H− and scattering from hydrogen and helium were also included.

Uniform priors between volume mixing ratios of 10−12 and 10−1 were placed on each of the opacity sources. Priors were also placed on the temperature structure to avoid extremely low (less than 500 K) and extremely high (greater than 8,000 K) temperatures. We ran a total of 120,000 iterations among 30 chains reaching a Gelman–Rubin statistic55 of less than 1.025. Extended Data Figure 5 shows two-dimensional cross sections of the retrieved posterior distribution along with the one-dimensional marginalized distribution for each of the retrieved parameters. A summary of the retrieved atmospheric properties is included in Extended Data Table 2.

We also ran a series of retrievals without certain opacity sources to compare the ability of different atmospheric species to fit the observed data. The scenarios we tested are listed in Table 1. We computed the Bayesian information criteria for each scenario, taking the retrieval with all opacity sources included as our fiducial scenario (Full) to calculate a ∆BIC that quantifies whether there is statistical evidence to include a given parameter, in this case an opacity source, in the retrieval. Generally, a ∆BIC between 2 and 6 indicates positive evidence for the inclusion of a given parameter, whereas ∆BIC above 6 indicates strong evidence. Our retrieval analysis indicates that there is strong evidence for the inclusion of SiO or Mg i, and thus the gaseous precursor species to silicate condensates. We also ran a retrieval that included a haze parameterization56 to account for photochemical or other high-temperature aerosols, but found that they did not improve the fit.

In our fiducial scenario, the retrieval did find a highly super-solar abundance of Mg ii. This result is being driven by the single data point at 0.28 µm (Fig. 1). A retrieval without Mg ii provides a similarly good fit to the data (Table 1), demonstrating that the inclusion of Mg ii is not necessary to fit the data. We therefore do not choose to interpret the retrieved Mg ii abundance as unambiguously physical. This is supported by the lack of any Mg ii signal in the HST STIS E230M observations (Extended Data Fig. 8).

Stellar activity

Starspots or faculae cause stellar inhomogeneities that can potentially contaminate transmission spectra57,58. Even though magnetic activity (and thus starspots and faculae) are most relevant for low-mass stars, higher-mass stars may also show some degree of activity59, and some transmission spectra of hot Jupiters around early-type host stars may be consistent with unocculted stellar inhomogeneities60. The problem is most acute at the shortest wavelengths at which the flux contrast between the nominal stellar photosphere and the active region is the greatest.

We examined whether the transit spectrum of WASP-178b could be caused by unocculted stellar activity. We found that, because of the magnitude of the spectral feature, extreme spot covering fractions and temperatures would be required. To fit the magnitude of the NUV feature, a starspot 1,500 K cooler than the nominal photosphere would require a spot covering fraction of 60%. For a starspot 2,500 K cooler than the photosphere, a spot covering fraction of 50% is required. Towards longer wavelengths, at which the observed spectrum is flatter, the contaminated model spectrum would continue sloping to small transit depths, which is not seen in the data. In the end, these factors, combined with the satisfactory fit to the stellar SED without activity and low activity levels found in the star (see below and ref. 7), indicate that stellar activity cannot be responsible for the large feature seen in the transit spectrum of WASP-178b.

The 2019 Transiting Exoplanet Survey Satellite (TESS) light curve of WASP-178 shows a consistent 0.115% variable photometric signal with a 0.185 day period8, which is also easily visible in 2021 high-cadence TESS photometry. This variability was speculated to be from δ Scuti pulsations in WASP-178, and possible gravity-darkening light curve asymmetries were reported from the TESS data as well8. However, our HST transit light curves show no evidence of any variability at the 0.115% level, with the raw UVIS photometry showing variations of less than 0.02% over a 0.3 day window. On further inspection of the TESS field of view and near-by faint contaminant stars from the ASAS-SN photometry database61,62, we determined the origin of the photometric variations to be ASASSN-V J150908.07–424253.6, which is a nearby 14.5 magnitude W Ursae Majoris-type binary star with a reported period of 0.369526 days, which is an alias of the reported 0.185 day period. Both the period and magnitude of variations match that of the signal that is seen to be diluted in the TESS data by the brighter WASP-178b. As such, we conclude that there is no evidence for WASP-178 to have any photometric variations larger than 0.02%. In addition, we find no evidence for reported transit asymmetries due to possible gravity-darkening effects8 in the HST data either. As a transit asymmetry signal in the TESS data could have also been influenced by the binary star, we analysed the 2021 TESS photometry taken at a higher cadence. When removing the contaminating binary star variables, the TESS light curve shows no transit asymmetries, in agreement with the HST data. Thus, there is no evidence that either gravity-darkening or significant photometric stellar activity are an issue with WASP-178.

Scattering

An alternative mechanism for producing large short-wavelength transit depths is through scattering. Small particles tend to scatter short-wavelength light more effectively than longer-wavelength light, leading to slopes towards greater transit depths at shorter wavelengths in transmission spectra63. If we describe the scattering cross section as

$$\sigma ={\sigma }_{0}{(\lambda /{\lambda }_{0})}^{\alpha }$$
(2)

where, σ0 is the cross section at a reference wavelength, λ0, then the slope in the transmission spectrum can be expressed as

$$\frac{1}{H}\frac{{\rm{d}}{R}_{{\rm{p}}}}{{\rm{dln}}(\lambda )}=\alpha $$
(3)

where H is the atmospheric scale height. Rayleigh scattering, the limit in which the particle is smaller than the wavelength of light, has a characteristic α = −4. Given our observed feature magnitude of approximately 20 scale heights over 0.2 µm, we calculate that α = −28.99 is required to match the data. Although super-Rayleigh slopes in transmission spectrum are possible due to a vertical gradient in opacity64, a slope with α = −28.99 would still be difficult to create, especially with non-purely scattering particles such as a photochemical haze. Additionally, we do not expect aerosols to survive to the pressures or temperatures that we probe with our observations (around 1 µbar and approximately 4,000 K).

Limb asymmetries

Limb asymmetries can potentially complicate the interpretation of a transmission spectrum. In particular, because of atmospheric advection from the hotter dayside to the cooler nightside, the morning terminator can potentially be cooler with increased condensate clouds, whereas the hotter evening terminator can be cloud-free. This effect may be evident in WASP-76b15, and theoretical models have investigated the effect by coupling cloud formation to atmospheric dynamics24,25,26,65. For WASP-178b, if the limb asymmetries were prevalent, we would expect the NUV transmission spectrum to be strongly affected, as features such as SiO could be in gaseous form on one limb but condensed into aerosols on the other. The combined effect would be to potentially bias the interpretation towards the hotter, clearer limb, albeit with a reduced signal.

As our data cover both ingress and egress, for which limb asymmetries have large observable effects, we searched the UVIS data between 0.18 and 0.28 µm for terminator asymmetries using catwoman66. In a scenario in which Mg and SiO are condensed on the leading morning terminator and has a transit radius ratio consistent with the optical (Rp /Rs = 0.11133 ± 0.0005), the trailing evening terminator would require Rp /Rs = 0.12924 to match the transit radius ratio of Rp /Rs = 0.12062 ± 0.00067 measured in the NUV with the +1 order. The NUV +1 order light curve and the magnitude of limb asymmetries are shown in Extended Data Fig. 4. In ingress and egress, such asymmetries are detectable in the data as they are found to have a 500 ppm effect on the transit light curve, which is comparable to the 1σ error bars (450 ppm).This simplified model is ruled out at the 5σ level by the +1 order alone (Extended Data Fig. 4). To place additional constraints, we used catwoman to fit the +1 and −1 order NUV data (0.18 to 0.28 µm) for the two hemisphere planetary radii, Rp,1 and Rp,2, as well as the terminator inclination angle φ. We found φ to be unconstrained and Rp,1 and Rp,2 to be consistent at 1σ, favouring a scenario without limb asymmetry. We also fixed φ to strictly assume an east/west limb asymmetry. In this case, we also found that both hemispheres fitted to nearly the same radii, Rp,1 /Rs = \({0.1195}_{-0.0021}^{+0.0020}\) and Rp,2 /Rs = \({0.1211}_{-0.0020}^{+0.0019}\), which are both larger than the optical radius at more than 3σ confidence.

This indicates that the NUV transmission spectral features (SiO/Mg) occur on both the leading and trailing limbs. With no indications of either limb being cloudy, potential silicate condensates are confined to the nightside of the planet on WASP-178b. However, WFC3 phase-curve observations of the similar planet WASP-121b67 show that the nightside temperatures do not generally drop low enough to be conducive for silicate material condensation as substantial heat is transported.

We also note that the fact that both hemispheres are the same radius to within 1σ could point to implications for the atmospheric circulation and heat transport at the low pressures probed in transit at NUV wavelengths. If the atmospheric circulation were dominated by super-rotation at these pressures, one would expect the morning terminator to be much colder, and thus have a smaller radius at a given optical depth, than the evening terminator. Because this does not seem to be the case, our observations might be an indication that the circulation at microbar and less pressures is dominated by a day-to-night flow, so that the morning and evening terminators would be more similar in temperature, and thus radius. This behaviour is in line with theoretical expectations68.