NGTS 15b, 16b, 17b, and 18b: four hot Jupiters from the Next-Generation Transit Survey Free

ABSTRACT

1 INTRODUCTION

The field of exoplanet discovery has uncovered a cosmic zoo of planetary types that extends far beyond those of our Solar system. As some of the first exoplanets ever detected, a particularly immediate revelation was the existence of Jupiter-sized planets on extremely short orbits (P < 10 d; e.g. Mayor & Queloz 1995; Charbonneau et al. 2000; Henry et al. 2000). In spite of their apparent rarity, comprising only <1 per cent of systems (Mayor et al. 2011; Wright et al. 2012; Fressin et al. 2013; Hsu et al. 2019), these ‘hot Jupiters’ are some of the most easily detectable exoplanets, as their large radii produce transits that are well above typical telescope noise limits, and their large masses and short orbital periods yield large radial velocity signals.

A distinctive yet poorly understood feature of the hot Jupiter population is the observation that many of these planets have radii that are larger than expected from theoretical models (e.g. Guillot & Showman 2002; Anderson et al. 2011; Hartman et al. 2012; Espinoza et al. 2016; Raynard et al. 2018). While this feature appears to correlate with incident flux (Laughlin, Crismani & Adams 2011; Weiss et al. 2013; Thorngren & Fortney 2018), this alone does not provide a sufficient explanation for inflation, and thus the underlying driving mechanisms continue to be debated (see e.g. Spiegel & Burrows 2013, for a comprehensive review). Indeed, evolutionary models that incorporate stellar irradiation and heavy metals can explain the unexpected radii of some hot Jupiters (Fortney, Marley & Barnes 2007; Baraffe, Chabrier & Barman 2008), but are yet to adequately describe the high-irradiation regime.

Efforts to characterize the relationship between inflation and incident flux have nevertheless provided valuable insight into the topic. In particular, the physical processes that cause inflation are believed to become effective in planets that are irradiated at fluxes in excess of ∼2 × 10⁵ W m⁻² (Demory & Seager 2011; Miller & Fortney 2011). Furthermore, a recent study by Sestovic, Demory & Queloz (2018) suggests that above an incident flux of ∼1.6 × 10⁶ W m⁻², all hot Jupiters in the mass range of 0.37–0.98 M_J appear inflated. For statistical studies such as this, increasing the sample of well-characterized hot Jupiters is crucial, and allows us to further constrain and parametrize inflation mechanisms, which remain poorly understood.

In this paper, we report the discovery of four new hot Jupiter planets from the Next-Generation Transit Survey (NGTS), three of which appear inflated. In Section 2, we describe the discovery photometry from NGTS and the subsequent photometric and radial velocity follow-up observations. In Section 3, we present our analysis of these data, including the determination of both the stellar and planetary parameters from spectral analysis, spectral energy distribution (SED) fitting, and global modelling. An investigation into the inflation of each planet is covered in Section 4, with the subsequent results being discussed in Section 5. Finally, our conclusions are laid out in Section 6. NGTS-15b, NGTS-16b, NGTS-17b, and NGTS-18b bring the total planet count from NGTS to 17 (note that NGTS-7Ab is a brown dwarf).

2 OBSERVATIONS

2.1 NGTS discovery photometry

NGTS-15 to NGTS-18 were initially identified as transiting exoplanet candidate systems from photometry from the NGTS (Wheatley et al. 2018), a ground-based wide-field exoplanet survey. Based at ESO’s Paranal observatory in Chile, NGTS consists of an array of 12 independently mounted 20-cm Newtonian telescopes, each equipped with a 2K×2K deep-depleted Andor IKon-L CCD camera. The custom NGTS 520–890-nm filter optimizes the survey for studies of K and M dwarf stars that, due to their small radii, provide the best opportunity for discovering small transiting planets.

NGTS-15, NGTS-16, and NGTS-17 were observed during the 2017 observing campaign for 160, 137, and 160 nights, respectively, between 2017 August 16 and 2018 March 18. NGTS-18 was observed the following season for 165 nights over the period of 2017 December 10 to 2018 August 7. Over 188 000 images were collected for each object using a single NGTS telescope with 10-s exposures (see Table 1 for an excerpt of the NGTS photometry for each object; see Table 2 for further details).

The data were reduced and aperture photometry performed via the casutools package,¹ before being detrended using an optimized version of the SysRem algorithm (Tamuz, Mazeh & Zucker 2005) that has been adapted for the NGTS pipeline. The data were then searched for transit-like events by orion (Wheatley et al. 2018), a custom implementation of the BLS fitting algorithm (Kovács, Zucker & Mazeh 2002). orion identified eight partial and three full transits for NGTS-15; four partial and four full transits for NGTS-16; fifteen partial transits for NGTS-17; and sixteen partial and four full transits for NGTS-18. Light curves of the NGTS detections, phase folded on the best-fitting period, are shown in Fig. 1. The initial fits to the NGTS data provided by orion revealed that the depths, widths, and shapes of the transits for each object were compatible with transiting hot Jupiter planets. In addition, a convolutional neural network applied to the NGTS data found that the probabilities of each light curve containing a transiting exoplanet were all greater than 0.95, consistent with previous confirmed NGTS planet discoveries (Chaushev et al. 2019).

Figure 1.

NGTS discovery photometry (grey points) and follow-up photometry (TESS: green points, SAAO 1.0-m telescope I band: dark and light blue points, Lesedi I band: pink points, Lesedi V band: purple points) for each planet. Note that, for clarity, we exclude the unbinned NGTS data from the plots. The red lines show 20 light-curve models generated from randomly drawn posterior samples of the allesfitter fit. Residuals are shown to the right of the light curves. (a) Photometry for NGTS-15b. The NGTS discovery light curve is phase folded at the best-fitting period of 3.276 23 ± 0.000 01 d. The light curves from the 1.0-m telescope at SAAO show a detection of egress from 2018 November 21 and a detection of ingress from 2019 December 6. Note that the in-transit scatter during the 2018 observation is a result of poor atmospheric conditions, and that the transit depths from global modelling for the NGTS and SAAO light curves agree to within errors. (b) Photometry for NGTS-16b. The NGTS and TESS light curves are phase folded on the best-fitting period, 4.845 32 ± 0.000 02 d. For this object, we obtained two light curves with the SAAO 1.0-m telescope on 2018 December 21 and 2019 December 5, both of which include egress. As with NGTS-15, the large in-transit scatter of the SAAO light curve from 2018 is due to poor atmospheric conditions and the true errors of these data points are likely underestimated. All four transit depths from the global modelling agree to within errors. (c) Photometry for NGTS-17b. The NGTS discovery light curve and TESS follow-up light curve are phase folded at the best-fitting period of 3.242 53 ± 0.000 01 d. The SAAO light curve consists of one data set from 2019 February 5. (b) Photometry for NGTS-18b. The NGTS discovery light curve and TESS follow-up light curve are phase folded at the best-fitting period of 3.051 25 ± 0.000 01 d. The SAAO 1.0-m telescope light curve consists of one data set from 2019 March 21; although egress is seen, due to the poor quality of the data, we omit it from our global modelling. The two Lesedi data sets in the I and V bands comprise of single observations from June 24 and 30, respectively.

Open in new tabDownload slide

In order to rule out the possibility that any of our detected companions were stellar rather than planetary, we performed additional checks on the NGTS data. The phase-folded light curves for each object were searched for any evidence of a secondary eclipse around phase 0.5 that would indicate the presence of a second star. Furthermore, we compared the transit depth of consecutive odd and even transits to check for a depth difference consistent with an eclipsing binary star system that had been mis-folded on half the true period. For all four targets, we find no indication in the NGTS photometry to suggest that the companions were not planets.

2.2 Additional photometry

In order to confirm that each stellar companion was indeed a planet, and to constrain the transit parameters, we obtained additional photometry with the 1.0-m and Lesedi telescopes at the South African Astronomical Observatory (SAAO). For three of the candidates, we were also able to use data from the Transiting Exoplanet Survey Satellite (TESS; Ricker et al. 2014). The details of this additional photometry are outlined below.

2.2.1 TESS

TESS is a space-based NASA survey telescope that searches for transiting planets around bright stars (Ricker et al. 2014). It has a wide field of view, with four 24 × 24^○ cameras, each equipped with four 2k×2k CCDs. Its typical observing baseline of 27 d makes it well suited to detecting short-period transiting exoplanets.

We searched for our candidates in the TESS full-frame images (FFIs) using the TESSCut² tool and found that NGTS-16 and NGTS-17 were observed in TESS Sectors 4 and 5, respectively, each for 27 consecutive nights between 2018 October 19 and 2018 December 11. In addition, NGTS-18 was observed between 2019 March 26 and 2019 April 22 in TESS Sector 10. NGTS-15 falls on a CCD in TESS Sector 5, but unfortunately into the overscan region of the camera rather than the science pixels, and therefore there is no TESS FFI observation for this star. A summary of this information can be found in Table 2.

Each star in the TESS FFIs was observed at 30-min cadence, and for each of our candidates we used an automatically determined optimal aperture to exclude neighbouring stars. These aperture sizes ranged from three to eight pixels. However, the large pixel scale of TESS (21 arcsec) meant that there was still some slight contamination of the FFI light curves, which we account for in our analysis (see Section 3.2).

For each candidate, we found that the best BLS period from TESS was consistent with the orion value for the NGTS photometry. TESS detected three full and two partial transits for NGTS-16, eight full partial transits for NGTS-17, and seven full transits for NGTS-18. Phase-folded light curves of the TESS data can be found in Fig. 1.

2.2.2 SAAO

Between 2018 November and 2019 December, we took at least one set of follow-up photometry for each target using the 1.0-m telescope at SAAO. The telescope was equipped with the full-frame transfer CCD Sutherland High-speed Optical Camera (SHOC), ‘SHOC’n’awe’, and all observations were taken using the I-band filter. The 2.85 arcmin × 2.85 arcmin field of view of SHOC’n’awe on the 1.0-m telescope allowed us to simultaneously observe at least one comparison star of similar brightness for each object.

For NGTS-18, we also had the opportunity to obtain two light curves using the newly commissioned 1.0-m Lesedi telescope at SAAO, which was equipped with the ‘SHOC’n’disbelief’ optical camera. The instrument’s slightly wider field of view (5.7 arcmin × 5.7 arcmin) allowed for an increased selection of comparison stars.

Two light curves of NGTS-15 were obtained using the 1.0-m telescope on the nights of 2018 November 21 and 2019 December 6 with exposure times of 60 and 30 s, respectively. We obtained 285 images for the 2018 data, and truncated the 2019 data from 543 to 276 frames due to extremely poor conditions towards the end of the night. Similarly, additional photometry was also obtained using the same telescope on the night of 2020 February 3, but a combination of poor atmospheric conditions (with seeing reaching 6.5 arcsec) and load shedding (scheduled electrical power shutdowns) at SAAO caused data gaps and fluctuations in the light curve of the order of the size of the transit signal, and so we omit these data from our analysis.

Load shedding also tainted the observations for NGTS-18 in 2019 March, resulting in a noticeable in-transit data gap of about 20 min in the final light curve. Despite being able to identify transit egress in these data, we omitted it from our final modelling process, as the data quality was too poor to obtain a reliable fit. However, we include the light curve in Fig. 1(d). We instead utilize follow-up photometry from the Lesedi telescope, taken on the nights of 2020 June 24 and 30 in the I and V bands, respectively. The I-band data consist of 426 × 15 s exposures; although it also includes a data gap of about 14 min long, this time due to an autoguider failure, the in-transit data are stable, and we thus include it in the modelling. Fortunately, the final data set in the V band was taken continuously over 2.29 h, and consists of 550 × 15 s exposures.

NGTS-16 and NGTS-17 were both observed using only the 1.0-m telescope at SAAO. For NGTS-16, we obtained two follow-up light curves on the night of 2018 December 21 (200 × 60 s exposures) and the night of 2019 December 5 (540 × 30 s exposures), while NGTS-17 was observed once on the night of 2019 February 5 (1104 × 15 s exposures).

A full summary of the photometric observations for each object is detailed in Table 2.

Each light curve was bias and flat-field corrected using the local python-based SAAO SHOC pipeline, which uses iraf photometry tasks (pyraf) and facilitates the extraction of raw and differential light curves. We used the Starlink package autophotom to perform aperture photometry on both our target and comparison stars, and chose apertures that gave the maximum signal-to-noise ratio. Background apertures were adjusted to account for changes in apparent star size over the night as the atmospheric conditions varied. Finally, the measured fluxes of the comparison stars for each object were used for differential photometry of our targets.

The SAAO light curves of each candidate are shown in Fig. 1. For NGTS-15 and NGTS-16, which each has two SAAO detections from the 1.0-m telescope, the data have been phase folded on the best-fitting period.

We detected a clear egress for NGTS-15 in 2018 November, as well as an ingress in 2019 December. Although seeing reached 6.4 arcsec near the beginning of the 2018 observations, resulting in large in-transit scatter for this light curve, our fitting procedure reveals that the transit depth is consistent with the NGTS data to within errors. Additionally, we note that these errors are likely underestimated for the weather-affected parts of the SAAO light curve.

Analysis of the 2018 December data for NGTS-16 revealed the majority of a transit, just missing ingress, while the 2019 December data are primarily out of transit with a few data points in egress. As with the 2018 data for NGTS-15, the 2018 data for NGTS-16 were affected by varying atmospheric conditions, with the full moon and high cloud causing in-transit scatter. However, again the transit depths between telescopes are in agreement.

Similar to the 2018 light curve for NGTS-16, we detected close to a full transit for NGTS-17, again just missing ingress. Finally, the data for NGTS-18 from the Lesedi telescope contained an egress on the night of 2020 June 24 and a full transit, including some points in egress, from the night of 2020 June 30.

For all four objects, the transits observed with SAAO were consistent with planetary companions. Subsequently, each was flagged for spectroscopic follow-up to enable mass determination.

2.3 Spectroscopic follow-up

In order to constrain the masses of the planetary companions, we obtained multi-epoch spectroscopy for radial velocity measurements. Three different fibre-fed Échelle spectrographs were used, all located at the La Silla Observatory in Chile. The HARPS spectrograph (Mayor et al. 2003) is mounted on the ESO 3.6-m telescope, CORALIE (Queloz et al. 2001a) on the Swiss 1.2-m Leonard Euler telescope, and FEROS (Kaufer & Pasquini 1998) on the 2.2-m MPG/ESO telescope. The spectrographs have a range of spectral resolutions of R = 115 000, 60 000, and 48 000, respectively. Due to the faintness of the targets (14.3 < V < 14.6; see Tables 5, 6, 7, and 8), all HARPS spectra were collected with the high-efficiency fibre link (EGGS), which uses a fibre size of 1.4 ″ instead of the usual 1.0 ″ mode. The HARPS-EGGS mode trades spectral resolution for approximately twice the photon count, depending on seeing. We used the second HARPS-EGGS fibre to monitor the sky simultaneously with science observations. All four objects were observed by FEROS and at least one other spectrograph, and we obtained a minimum of 12 data points for each (see Table 3 for details; see Tables A1 and A2 for a complete list of radial velocity data for each object).

The HARPS data were reduced via the offline data reduction pipeline (DRS) before being cross-correlated with a binary G2 mask to extract radial velocities (Baranne et al. 1996). This procedure was also employed for the CORALIE data, for which the spectra were reduced using the standard CORALIE DRS. For the data collected with FEROS, the CERES reduction pipeline (Brahm, Jordán & Espinoza 2017) performed a radial velocity extraction by the same method.

Two HARPS spectra of NGTS-18 obtained on BJD 2458916.780 and 2458918.843 were contaminated by moon light. We corrected the radial velocity measurements by subtracting the signal of the simultaneous sky-fibre from the science fibre in cross-correlation function (CCF) space. The radial velocities are then extracted as per usual on the star-sky CCF. Additionally, the CORALIE radial velocities for NGTS-17 were computed while excluding the first 30 spectral orders, in which the signal-to-noise ratio was <1. Phase-folded plots of the radial velocity data for all four targets are shown in Fig. 2.

Figure 2.

Phase-wrapped radial velocity measurements for all four stars from the HARPS (cyan points), FEROS (orange points), and CORALIE (purple points) spectrographs. The red lines show 20 radial velocity models generated from randomly drawn posterior samples of the allesfitter fit.

Open in new tabDownload slide

For all four targets, the radial velocity measurements were consistent with a Jupiter-mass planet, with semi-amplitudes of the order of K ∼ 100 m s⁻¹ (see Table 9) and variations in-phase with the periods derived from the photometric data. To confirm that the signals did not arise due to cool stellar spots or a blended eclipsing binary, we checked for any correlation between the radial velocity measurements and the line bisector span of the CCF (Queloz et al. 2001b). Correspondingly, no evidence was found in the spectroscopic data that contradicted the existence of a planetary companion.

3 ANALYSIS

3.1 Stellar properties

In order to produce reliable fits to our photometric and radial velocity data for NGTS-15, NGTS-16, NGTS-17, and NGTS-18, we must first determine the stellar properties of each system. Constraining these is vital as the accuracy of the measured planetary parameters is dependent on how well we can characterize the host star. Below, we outline our stellar analysis procedure and results.

3.1.1 Gaia DR2

For all four objects, we obtained additional photometric and astrometric data from Gaia DR2 (Gaia Collaboration 2016, 2018; see Tables 5, 6, 7, and 8). Part of these data, including the stellar radius and parallax, was utilized as priors in our SED fitting procedure (see Section 3.1.3).

The Gaia parallaxes, proper motions, and absolute radial velocities of each star were used to determine the Galactic velocity components (U_LSR, V_LSR, W_LSR), assuming a Local Standard of Rest of UVW = (11.1, 12.14, 7.25) km s⁻¹ (Schönrich, Binney & Dehnen 2010). By using the selection criteria for kinematically thin disc (V_tot < 50 km s⁻¹) and kinematically thick disc (70 < V_tot < 180 km s⁻¹) objects (Gaia Collaboration 2018), we conclude that all four stars belong to the thin-disc population (see Table 4).

Finally, our host stars show no evidence of being unresolved binaries, with all four targets having an astrometric excess noise of zero as well as a low (<0) astrometric goodness of fit in the along-scan direction (GOF_AL) in Gaia DR2 (Evans 2018).

3.1.2 Stellar analysis

Although we used SED fitting to derive our final stellar parameters (see Section 3.1.3), we first fit the stacked spectra for each object in order to obtain suitable priors for the key parameters with which to constrain the SED fits. For this, we used ispec (Blanco-Cuaresma et al. 2014; Blanco-Cuaresma 2019), an open-source framework for spectral analysis. ispec users can choose from a list of models for spectral synthesis; we employ the spectrum (Gray & Corbally 1994) radiative transfer code and the solar abundance model from Asplund et al. (2009). Additionally, we use the Gaia-ESO Survey (GES) line list (version 5.0; Heiter et al. 2015), which covers the full wavelength range of our spectra (from 420 to 920 nm), and the MARCES.GES model atmosphere (Gustafsson et al. 2008).

We used our high-resolution HARPS data for the spectral analysis of NGTS-15, NGTS-16, and NGTS-18. Since NGTS-17 was not observed with HARPS, we instead used the FEROS spectra. For each object, we shift the spectra to the laboratory frame of reference and co-add them to produce a single, high-SNR, combined spectrum. These SNRs are reported in Table 3. We note that, although NGTS-17 has the highest SNR, the data were affected by correlated red noise, which hampered spectral analysis.

We determined the stellar effective temperature, T_eff, and the surface gravity, log g, via fits to the the H α, Na i D, and Mg ib lines. Individual Fe i and Fe ii lines were used as diagnostics for metallicity ([Fe/H]) and the rotational line-of-sight broadening (vsin i). Values for each parameter were obtained by synthesizing model spectra until we found an acceptable fit to the data, and uncertainties were estimated by varying the fit until the models no longer matched the stacked spectra of each object.

From our spectral analysis, we find all four stars to be metal rich, which aligns with observations that short- and intermediate-period gas giants tend to be found around more metal-rich stars than longer period gas giants (Jenkins et al. 2017; Maldonado et al. 2019).

3.1.3 SED fitting

To determine the stellar parameters of each system, we performed a fit to their SEDs. For this, we employed the python tool ariadne (Vines & Jenkins, in preparation), which fits catalogue photometry to different atmospheric model grids [Phoenix V2 (Husser et al. 2013), BT-Settl, BT-Cond, BT-NextGen (Hauschildt, Allard & Baron 1999; Allard, Homeier & Freytag 2012), Castelli & Kurucz (2004), Kurucz (1993)] that are then convolved with a range of filter response functions.

The SEDs were modelled by interpolating the model grids in T_eff–log g–[Fe/H] space, with radius, distance, and extinction in the V band used as additional model parameters. Any underestimated uncertainties in the photometry were accounted for by including an individual excess noise term for each photometric data set.

Priors for T_eff, log g, and [Fe/H] were taken from our stellar analysis with ispec, while priors for the stellar radius and distance were obtained from the Gaia DR2 values, although we used an inflated value for the radius error to account for modelling errors. In addition, A_V was limited to the maximum line-of-sight value taken from the re-calibrated SFD galactic dust map (Schlegel, Finkbeiner & Davis 1998; Schlafly & Finkbeiner 2011). The priors for the excess noise parameters were set to a normal distribution with a mean of zero and a variance equal to five times the associated uncertainty.

Note that, by averaging over several posterior distributions, ariadne is able to achieve a higher precision than would typically be obtained with a single atmospheric model. Subsequently, some uncertainties may be underestimated. We therefore calculated an additional systematic error for these parameters by following the approach taken in Southworth et al. (2015), wherein each SED was fitted with the individual stellar atmosphere models before being compared to the overall Bayesian model averaging solution. By taking the largest difference between the model values and the averaged value from all posterior distributions, we were able to obtain the systematic uncertainty.

Finally, the mass value was estimated using MIST isochrones (Choi et al. 2016). The derived stellar parameters are listed in Tables 5, 6, 7, and 8. A detailed overview of the ariadne fitting procedure, as well as the accuracy and precision of the tool, can be found in Vines & Jenkins (in preparation).

3.2 Global modelling

We determined the physical parameters of the systems, including planet radius and mass, via a simultaneous fit to the photometric and spectroscopic data for each object. For this, we used the publicly available open-source astronomy software package allesfitter (Günther & Daylan 2020, 2019), which unites the packages ellc (light curve and RV models; Maxted 2016), emcee (MCMC sampling; Foreman-Mackey et al. 2013), and celerite [Gaussian Process (GP) models; Foreman-Mackey et al. 2017]. The combination of these packages allows allesfitter to model a variety of signals, including multistar systems, star-spots, stellar variability, and transit-timing variations.

Initial MCMC fits required millions of steps to converge, so we chose the nested sampling approach to produce our global fits. Our priors for transit epoch and transit depth (R_p/R_s) were refined from orion by generating quick MCMC fits to the NGTS data using bruce, an open-source binary star and exoplanet analysis package.³ The results from the ariadne SED fits were used as priors on the stellar parameters, specifically R_s, M_s, and T_eff, although we used the T_eff error from ispec as this is underestimated by ariadne. Because all four host stars show no out-of-transit variability or trends, GP modelling was not necessary; indeed, an initial GP fit to the 2019 SAAO data of NGTS-18 incorrectly adjusted a data gap between the mid-transit and egress data points, resulting in a smaller transit depth than the true value. Furthermore, we accounted for instrumental offsets in the radial velocity data, as each object was observed with at least two different spectrographs. We adopted a quadratic limb-darkening law as parametrized in Kipping (2013), and fit for the limb-darkening coefficients.

We ran two fits for each planet: one in which the orbital eccentricity, e, was fixed at 0, and one for which e was allowed to vary freely. Each of the latter fits resulted in non-zero values of e, at |$0.102\, ^{+0.061}_{-0.068}$|⁠, |$0.105\, ^{+0.130}_{-0.079}$|⁠, |$0.168\, ^{+0.093}_{-0.101}$|⁠, and |$0.035\, ^{+0.033}_{-0.024}$| for NGTS-15b, NGTS-16b, NGTS-17b, and NGTS-18b, respectively. However, Lucy & Sweeney (1971) showed that many small values of e are spurious, and define a probabilistic test to determine whether a small e is statistically significant from 0. By adopting a 5 per cent level of significance, they find that if the condition e > 2.45σ_e is satisfied (where σ_e is the observational uncertainty on e), then one can be confident that the measured eccentricity is real. We find that all of our measurements fail to meet this criterion, and we therefore adopt the results from the global modelling fits in which e is fixed at zero.