The NCORES programme: precise planetary masses, null results, and insight into the planet mass distribution near the radius gap

ABSTRACT

1 INTRODUCTION

The Transiting Exoplanet Survey Satellite (TESS) mission (Ricker et al. 2014) was launched in 2018 with the aim of searching for transiting exoplanets in a new, all-sky survey. The increased sky coverage of the mission over past efforts such as Kepler (Borucki et al. 2010) led to a substantial increase in the number of planets detected, particularly at Neptune and larger sizes (Guerrero et al. 2021). The brighter candidate host stars observed by an all-sky survey mean that radial velocity (RV) observations to confirm planetary mass and density are possible on a much wider scale than previously.

Many programmes exist to follow-up TESS transiting planet candidates, and that follow-up can take several forms. Photometric follow-up is used to detect the transit with alternative instruments (Collins 2019), confirming the nature of the signal, improving ephemeris determination and reducing the possibility that the transit occurs on a background star. High spatial resolution imaging is used to further identify unresolved stars blended with the target in the TESS pixels (Lillo-Box et al. 2024; Michel & Mugrauer 2024). Spectrographs are used to identify spectroscopic binaries and brown dwarfs, characterize the host stars, and, with enough precision, determine planetary masses (e.g. Turtelboom et al. 2022; Vines et al. 2023). Obtaining the planetary mass, along with radius from the transit observation, reveals the planetary density and allows us to build an understanding of possible internal structure, composition, and enable atmospheric measurements, all valuable information in determining the past formation and evolution of planetary systems.

The various forms of follow-up for the TESS mission are often coordinated by the TESS Follow-up Observing Program (TFOP; Collins et al. 2018), within which the NCORES programme described in Section 2 operates. Several TFOP-generated observations are published in this paper and the referenced publications. The coordination aims at reducing duplicated observations and enabling broader coverage of planetary candidates, and has helped generate a level and uniformity of follow-up coverage for TESS candidates rarely seen in previous transit surveys.

One particular characteristic of the small planet distribution is the radius gap (Fulton et al. 2017; Fulton & Petigura 2018; Van Eylen et al. 2018), an observed bimodal distribution in planet radius separating rocky worlds and ‘sub-Neptunes’. The gap is a predicted outcome of several models, arising either from thermal mass-loss after disc dispersal (Lopez & Fortney 2013; Owen & Wu 2017; Ginzburg, Schlichting & Sari 2018; Gupta & Schlichting 2019; Gupta, Nicholson & Schlichting 2022), differences in the formation process (Lopez & Rice 2018; Lee & Connors 2021), or a combination of the two (Venturini et al. 2020a; Burn et al. 2024; Venturini et al. 2024). These models are dependent on initial assumptions about the planetary properties, including planet core mass, envelope composition and mixing, and migration time-scales (e.g. Rogers et al. 2021). Conversely, the location, width, and presence or absence of planets in the gap can lead to constraints on those properties for the wider planet sample (Owen & Wu 2017; Jin & Mordasini 2018; Mordasini 2020). Investigating the mass of planets in and near the gap potentially allows us to constrain the processes that sculpt the radius valley, with Ho et al. (2024) and (Bonfanti et al. 2024) recently finding evidence for a mixture of scenarios.

2 THE NCORES PROGRAMME

The NCORES programme was a dedicated large programme on the ESO HARPS spectrograph aimed at studying planetary cores in detail, targeting their composition, formation, and mass distribution. The programme targeted transiting planet candidates from K2 and TESS, with radii approximately between 1.8 and 4|$R_{\rm{\oplus }}$|⁠, and orbital periods less than 10 d. The radius limits were chosen to take advantage of the precision of HARPS while allowing a substantial sample of targets to be observed. Candidates outside this range were occasionally included due to being in multiple systems with an in-range planet, or through particular scientific interest. The goal was to find planets exposed to intense irradiation where the outer atmosphere may have been evaporated, reducing the degeneracy in identifying the composition and internal structure of the remaining material. Additional goals included characterizing planets on either side of the radius gap, noting that the radius gap rises to larger radii for the most irradiated planets (Fulton et al. 2017), to investigate theories of how the gap relates to the core and composition of planets, and building a sample of small planets with precisely characterized masses from the TESS mission.

In total, 72 nights on the HARPS spectrograph were awarded over 2 yr, in periods P102–P105. Due to operational closures associated with the Covid-19 pandemic, additional time was granted and the programme continued through P107, overall covering October 2018 to September 2021. In this time period, over 1300 radial velocity measurements were taken. With this paper, results on 41 candidates in 23 systems have been published, including 32 precise mass measurements and 35 published planets across 18 planetary systems, several jointly with other programmes. The complete published output of the programme is summarized in Table 1. As this programme began at the start of the TESS mission, a target-of-opportunity strategy was used to select bright, plausible candidates for follow-up. Targets were selected on a rolling basis as TOIs were released, based on passing the vetting checks performed by the TESS team and falling within our targeted parameter space. Targets were prioritized based on observability, brightness, and scientific interest, for example multiplanet systems. In later efforts, a more quantifiable merit function will be utilized to improve statistical options with future results (see Section 7.4).

3 OBSERVATIONS

Planets with confirmed mass measurements have been published in various papers listed in Table 1. Readers should see those papers for details of data used and modelling for each case. Five additional targets where the planetary candidate could not be convincingly detected are published here, to ensure the programme’s null results are in the public domain, to allow other teams to use the mass constraints to inform their targeting, and to enable these constraints to be incorporated into mass-radius relations to avoid biases towards high-mass planets at small radii. These null result targets, namely TOI-271, TOI-510, TOI-641, TOI-697 and TOI-745, along with their identifiers are listed in Table 2. Additional observations undertaken by TFOP on the five new targets are detailed in this section. TOI-880 (Nielsen et al., in preparation) and TOI-1839 (Castro-González et al., in preparation) were also observed, but continued to receive multiple seasons of data after the NCORES programme finished, and will be published separately.

3.1 TESS photometry

Each additional candidate planet in this paper was first identified using TESS photometry. TESS observes in |$\sim$|27-d long sectors, and each target has at the time of writing been observed in multiple, different sectors, listed in Table 3 along with cadence and radial velocity data obtained. All targets were observed at a 2-min cadence, except TOI-641 which has a mixture of 30- and 2-min sectors.

TOIs 271.01, 510.01, 641.01, and 697.01 were identified by the TESS Science Processing Operations Center pipeline (SPOC pipeline; Jenkins 2002; Jenkins et al. 2010, 2016). TOI-745.01 was identified by the MIT Quick Look Pipeline (QLP; Huang et al. 2020). Although the TOI detection was performed on various fractions of the currently available data set, our joint model considers the full data available at the time of writing, listed in Table 3.

For the purpose of modelling the data to provide mass constraints, we utilized the TESS data provided by the SPOC pipeline for all targets, which performs light curve detrending via Presearch Data Conditioning Simple Aperture Photometry (PDCSAP; Smith et al. 2012; Stumpe et al. 2012, 2014). Variability is seen in the light curves of TOI-510, TOI-697, and TOI-745. Notably TOI-510 shows quasi-periodic variability at twice the candidate period. For these targets, the variability is later accounted for using Gaussian processes during the joint fit. Data from separate sectors were stitched together by dividing the sector by its median flux value to normalize each sector independently.

3.2 Light-curve follow-up

The TESS field of view near TOI-510 can be seen in Fig. 1, plotted using the tpfplotter package (Aller et al. 2020).¹ The field of view of TOIs-271, 641, 697, and 745 are shown in the Appendix. In each case some nearby stars can be seen, motivating further follow-up. We acquired ground-based time-series follow-up photometry of the stars in the fields around all targets in Table 2 as part of the TESS Follow-up Observing Program Collins (2019)² to attempt to rule out or identify nearby eclipsing binaries (NEBs) as potential sources of TESS detections at the times of transit predicted by the public SPOC and QLP ephemerides. We used the tess transit finder, which is a customized version of the tapir software package (Jensen 2013), to schedule our transit observations. The observations, facilities, and results are summarized in Table 4. The Las Cumbres Observatory Global Telescope (LCOGT; Brown et al. 2013) images were calibrated by the standard LCOGT banzai pipeline (McCully et al. 2018). The LCOGT and SLR2 photometric data were extracted using astroimagej (Collins et al. 2017). The Perth Exoplanet Survey Telescope used a custom pipeline based on c-munipack³ to calibrate the images and extract the differential photometry. The MuSCAT2 (Narita et al. 2019) and MEarth-South (Irwin et al. 2007) results were extracted using the custom pipelines described in Parviainen et al. (2019) and Irwin et al. (2007), respectively.

Figure 1.

TESS field of view around TOI-510 in Sector 6, generated with tpfplotter. Known stars from Gaia are shown as red circles. TOI-510 is marked by a white cross. The TESS photometric aperture is highlighted in red.

Open in new tabDownload slide

Since the TOI depths are generally shallower than can typically be reliably detected with ground-based follow-up observations, we initially checked for possible NEBs that could be contaminating the SPOC and QLP photometric apertures, which generally extend |$\sim 1\,{\rm arcmin}$|  from the target star. To account for possible contamination from the wings of neighbouring star PSFs, we searched for NEBs in all-known Gaia EDR3 and TIC version 8 nearby stars out to |$2{_{.}^{\prime}} 5$| from the target stars that are bright enough in TESS band (assuming a 100 per cent eclipse) to produce the TESS detection. To attempt to account for possible delta-magnitude differences between TESS-band and the follow-up filter band, we included stars that are up to 0.5 mag fainter. We consider a star cleared of an NEB within the above mentioned sensitivity limits if the RMS of its 10-min binned light curve is more than a factor of 3 smaller than the expected NEB depth in the star. We then visually inspect each neighbouring star’s light curve to ensure no obvious eclipse-like signal. All of our follow-up data are available on the EXOFOP-TESS website.⁴

3.3 High-resolution imaging

If an exoplanet host star has a close companion, that companion (bound or chance-aligned) can create a false-positive transit signal if it is, for example, an eclipsing binary (EB), or lead to underestimates of the exoplanet radius and incorrect stellar and planet properties (Lillo-Box, Barrado & Bouy 2014; Ciardi et al. 2015; Furlan & Howell 2017, 2020). Searching for nearby stellar companions is an especially important check to make for main-sequence stars with large planets, due to the possible confusion between a large planet and a stellar eclipse (Lillo-Box et al. 2021). The discovery of close, bound companion stars, which exist in nearly one-half of FGK type stars (Matson et al. 2018) and just under one-quarter of M class stars (Susemiehl & Meyer 2022), provides crucial information toward our understanding of exoplanet formation, dynamics, and evolution (Howell et al. 2021). Close companions can even cause non-detections of small planets residing within the same exoplanetary system (Lester et al. 2021). Thus, to search for close-in bound companions unresolved in TESS or other typical ground-based follow-up observations, we obtained high-resolution imaging speckle observations of the new targets.

These TOIs were observed with the Gemini North or Gemini South telescope between 2019 September and 2021 September⁵ using the speckle interferometric instruments ‘Alopeke and Zorro (Scott et al. 2021). The two instruments are duplicates of each other and both provide simultaneous speckle imaging in two bands (562 and 832 nm) with output data products including a reconstructed image with robust contrast limits on companion detections (e.g. Howell et al. 2016). Three to five sets of |$1000\times 0.06$| s exposures were collected for these stars as described in Howell & Furlan (2022) and subjected to Fourier analysis in the standard reduction pipeline (see Howell et al. 2011). Fig. 2 shows an example of final contrast curves and an 832 nm reconstructed speckle image, typical of what we obtained for each TOI. We find that TOI-271, TOI-641, and TOI-697 reveal one or more close companion stars and they have been taken to be binary or triple stars (see Lester et al. 2021). Individual cases are discussed in Section 6. TOI-510 and TOI-745 are single stars to within the angular (0.02–1.2 arcsec) and magnitude contrast limits (5–7 mag) achieved. Given that all of these TOIs are about 100 pc away, the angular limits above correspond to spatial limits of roughly 2–120 au at each star.

Figure 2.

Contrast curves for TOI-510 obtained by Gemini/Zorro at 562 and 832 nm. No companion star is detected within the contrast limits.

Open in new tabDownload slide

We also observed TOI-641 with NIRI at Gemini South on UT 2019-11-03 using the observatory’s adaptive optics system. We obtained 9 dithered images in the Br-|$\gamma$| filter with exposure times of 3.8 s to avoid saturating the primary star. The raw data products are available at the Gemini Data Archive; the processed products are at the ExoFOP website. The stacked observations reach a contrast level of |$\sim$|7.5 mag beyond |$\sim 0.4\,\rm {arcsec}$| and reveal two companions, with separation, position angle, and delta-magnitudes of 3.537 arcsec, 86.2|$^\circ$|⁠, and 4.16 |$\pm$| 0.03 mag and 3.627 arcsec, 80.7|$^\circ$|⁠, and |$3.90\pm 0.02$| mag, respectively.

3.4 HARPS

HARPS (High Accuracy Radial velocity Planet Searcher; Mayor et al. 2003) is an Echelle spectrograph mounted on the ESO 3.6 m telescope located at La Silla Observatory, Chile. In total, around 1300 spectra were taken as part of the NCORES programme (PI: Armstrong, ID 1102.C-0249) throughout periods P102–P107, 2018 October 1 to 2021 September 31. The majority of these spectra are published in other works listed in Table 1. Newly published analyses in this work are for TOI-271 (15 spectra, 900–1200 s exposure time), TOI-510 (36 spectra, 900 s), TOI-641 (9 spectra, 1800 s), TOI-697 (6 spectra, 1800 s), and TOI-745 (26 spectra, 1800 s). Data were obtained in HARPS High-Accuracy Mode with a 1 arcsec diameter fibre and standard resolution of R|$\sim$| 115 000. Raw data were reduced according to the standard HARPS data reduction software, with radial velocities extracted using a cross-correlation function with a G2 spectral template (Baranne et al. 1996; Pepe et al. 2002). The data table for these observations can be found in Table A1, which we use in our joint modelling (Section 5).

3.5 PFS

TOI-271 and TOI-641 were additionally observed by the Carnegie Planet Finder Spectrograph (PFS; Crane, Shectman & Butler 2006; Crane et al. 2008, 2010) is a custom designed precision RV echelle spectrometer covering 3900–6700 Å on the 6.5m Magellan Clay telescope at Las Campanas Observatory, Chile. With the exception of the focus, PFS has no moving parts and is embedded in an insulated box where the temperature is maintained at 27|$^{\circ }\pm$| 0.01|$^{\circ }$|C. As of January 2018, PFS uses a 10K |$\times$| 10K detector with 9|$\mu$|m pixels which produces a peak resolution of |$R \simeq$| 130 000 when using the 0.3|$^{\prime \prime }$| slit and default 1 × 2 binning, as was done for all observations of TOI-271 and TOI-641. PFS RVs are measured in the 5000–6200 Å region of the spectrum where an Iodine cell placed in the converging beam of the telescope provides a dense forest of sharp absorption lines that enable precise wavelength calibration (Marcy & Butler 1992). The PFS Iodine cell was scanned with the NIST Fourier Transform Spectrometer (Nave 2017) at a resolution of 1000 000. Raw PFS data are reduced to 1D, wavelength calibrated, spectra with a custom built raw reduction package and velocities are generated from an updated version of the Iodine-based, forward modelling approach outlined in Butler et al. (1996). TOI-271 was observed 94 times by PFS, with a consistent exposure time of 600 s. TOI-641 was observed 6 times. The data table for these observations can be found in Table A2, which we use in our joint modelling.

4 STELLAR PARAMETERS AND CHEMICAL ABUNDANCES

We co-added the individual exposures for each target in order to create a combined HARPS master spectrum with a higher signal-to-noise ratio. We used each master spectrum to derive the stellar spectroscopic parameters (⁠|$T_{\mathrm{eff}}$|⁠, |$\log g$|⁠, microturbulence, [Fe/H]), and the respective uncertainties following the ARES+MOOG methodology as described in Santos et al. (2013) and Sousa et al. (2014, 2021). The ares code⁶ (Sousa et al. 2007, 2015) was used to measure in a consistent way the equivalent widths (EW) of iron lines included in the line list presented in Sousa et al. (2008). Briefly ARES+MOOG performs a minimization process looking for the ionization and excitation equilibrium to find convergence on the best set of spectroscopic parameters. For the computation of the iron abundances we make use of a grid of Kurucz model atmospheres (Kurucz 1993) and the radiative transfer code MOOG (v2019; Sneden 1973). Following the same methodology as described in Sousa et al. (2021), we used the distance derived from the Gaia eDR3 parallaxes and estimated the trigonometric surface gravity for each target. We then estimated the stellar age, mass and radius using the derived spectroscopic parameters with the paramv1.3 tool (da Silva et al. 2006; Rodrigues et al. 2014, 2017) using the PARSEC isochrones (Bressan et al. 2012).

Under the assumption of local thermodynamic equilibrium, the classical curve-of-growth analysis technique was used. The stellar abundances of the elements were also determined using the same methods and models used to determine the stellar parameters. Although the EWs of the spectral lines were measured automatically with ARES, for some elements (Al, Mg, C, O) with only two-three lines available, we performed careful visual inspection of the EWs measurements. For the derivation of chemical abundances of refractory elements, we closely followed the methods described in Adibekyan et al. (2012, 2015) and Delgado Mena et al. (e.g. 2017), whereas for the derivation of C and O abundances we followed the methods described in Bertran de Lis et al. (2015) and Delgado Mena et al. (2021). The results of our analysis are shown in Tables 5 and 6. In a subsequent step, we used the derived abundances to obtain estimations of the stellar age by applying calibrations based on chemical clocks, i.e. certain abundance ratios with a strong correlation with age, such as [Y/Mg], [Y/Zn], or [Sr/Si] (see table 10 in Delgado Mena et al. 2019, with the 3D formulas employed here). The derived ages (weighted mean of different clocks) are in agreement with the values from paramv1.3. We note that the errors reported in the table are the weighted mean errors of the different clocks and their low values show the agreement between the different chemical clocks. A value of 1.5 Gyr should be considered as a more realistic age error for this method.

The chemical pattern of the observed stars suggest that, excepting TOI-510, all the stars belong to the Galactic thin disc. TOI-510 shows enhancement in |$\alpha$| elements (Mg, Si, Ti), which is typical for the chemically defined thick disk population (e.g. Adibekyan et al. 2011, 2013; Lagarde et al. 2021). TOI-510 also appears to be a somewhat older star than the others in the sample, consistent with thick disk membership.

To confirm TOI-510’s old evolutionary stage, we also followed another statistical treatment, by applying the isochrone placement routine (Bonfanti et al. 2015; Bonfanti, Ortolani & Nascimbeni 2016) that interpolates the input parameters along with their uncertainties (i.e. |$T_{\mathrm{eff}}$|⁠, [Fe/H], and |$\log {g}$|⁠) within pre-computed grids of parsec⁷ v1.2S (Marigo et al. 2017) isochrones and tracks. We further derived |$\log {R^{\prime }_{HK}}=-4.95\pm 0.02$| from the HARPS spectra to improve the interpolation scheme convergence as detailed in Bonfanti et al. (2016), Bonfanti & Gillon (2020), and obtained an age of |$12.5_{-2.5}^{+1.3}$| Gyr. Aware that the spectroscopic |$\log {g}$| we adopted as input in the previous derivation process might be poorly constrained, we performed other isochrone placement runs by replacing the input |$\log {g}$| with the Gaia parallax (offset-corrected following Lindegren et al. (2021) and different magnitude bands. In other words, the isochrone interpolation was done in the absolute magnitude–|$\log {T_{\mathrm{eff}}}$| diagram, rather than in the |$\log {g}$|–|$\log {T_{\mathrm{eff}}}$| (also known as the Kiel diagram). In detail, we performed one isochrone placement run per magnitude (namely the GaiaG, |$G_{\mathrm{BP}}$|⁠, and |$G_{\mathrm{RP}}$|⁠, the Johnson B and V, the Tycho|$B_T$| and |$V_T$|⁠, and the 2MASS J, H, and |$K_s$|⁠). From each run we derived a probability density function (pdf) for the age and finally we merged (i.e. we summed) the pdfs after checking their mutual consistency according to the |$\chi ^2$|-based criterion outlined in Bonfanti et al. (2021). We obtained an age of |$12.6_{-2.5}^{+0.8}$| Gyr, which is consistent with the outcome from our first approach, the chemical clocks, and the initial PARAM isochrone age, confirming the old evolutionary stage of TOI-510.

5 JOINT MODELS

The modelling applied to published cases is dependent on the particular system and described in the relevant papers listed in Table 1. Here, we describe the joint model applied in this work to new systems.

We model the data for each target using the exoplanet (Foreman-Mackey et al. 2021) code framework. This package also makes use of starry (Luger et al. 2019) and pymc version 5 (Salvatier, Wieckiâ & Fonnesbeck 2016).

We restrict ourselves to using TESS data in the joint model; although some other light curves are available from ground-based instruments, these are not of good enough precision to contribute to the transit fit. The whole TESS light curve is used.

The flux is normalized to zero for all of the photometry by dividing the individual light curves by the median of their out-of-transit points and subtracting one. To model the planetary transits, we used a limb-darkened transit model following the Kipping (2013) quadratic limb-darkening parametrization, and Keplerian orbit models. Relevant parameters are the stellar radius |$R_*$| in solar radii, the stellar mass |$M_*$| in solar masses, the orbital period P in days, the planetary radii |$R_p$|⁠, the time of a reference transit |$t_0$|⁠, the impact parameter b, the eccentricity e, and the argument of periastron |$\omega$|⁠. One Keplerian orbit is initialized for each planet in a system.

In all cases e and |$\omega$| were consistent with zero for a planet in initial model runs, and we fix e and |$\omega$| to 0 in the final joint fit model. We use values from the TESS pipelines to initialize our priors on the epochs, periods, transit depths and radii of the transiting planets, and the results of our stellar characterization to set priors on the stellar mass and radius. Prior distributions are given in Table 7. For the non-transiting TOI-510c, a wider normal distribution prior on period and epoch was used, with mean |$\mu$| of 10.5d and 1167.7 (BJD-2458000) and standard deviation |$\sigma$| of 4 and 10 d. The prior for |$P_c$| was bounded between 5 and 25 d to ensure stability in the fitting chains.

Typically, a simple jitter term was added in quadrature to the TESS data to account for additional noise. For TOI-510, TOI-697, and TOI-745, the TESS light curves show additional correlated stellar variability. The correlated variability was modelled with the SHOTerm Gaussian Process (GP) given in exoplanet,⁸ representing a stochastically driven, damped harmonic oscillator. We set this up using the hyperparameters |$\log {\rm (sigma)}$|⁠, |$\log {\rm (rho)}$|⁠, and |$\tau$|⁠. The prior on |$\log {\rm (sigma)}$| (where |$\rm sigma$| represents the standard deviation of the process) was set to a normal distribution with a mean equal to the log of the standard deviation of the TESS flux and a standard deviation of 3. The prior on |$\log {\rm (rho)}$| (where |$\rm rho$| represents |$2\pi /\omega _0$|⁠, with |$\omega _0$| being the frequency of the undamped oscillator) was also set as a normal distribution but with mean 0 and standard deviation 3. |$\tau$| represents the damping time-scale of the process and was fixed to 2 d. We then take the sum of our model planetary light curves and subtract these from the total TESS flux. The GP is trained on the resulting residual light curve simultaneously to remove the stellar variability.

We also fit separate systematic offset and jitter terms for each radial velocity instrument, and, in the case of TOI-510, a linear drift term. The means of the jitter priors are set to the log of the minimum error on the relevant dataset. No GP is used on any of the radial velocity data, as the activity indicators available from HARPS do not show clear signs of periodic variability. Note that given the relatively small number of HARPS points in some cases, particularly TOI-271, TOI-641 and TOI-697, this should not be taken as definitive information on the activity levels of the targets.

To fit the model, we use the pymc sampler, which draws samples from the posterior using a variant of Hamiltonian Monte Carlo, the No-U-Turn Sampler (NUTS). We allow for 1500 burn-in samples which are discarded, followed by 8250 steps for each of 8 chains.

Table 3 details the specific model setup for each system.

6 RESULTS

In this section, we present the upper limits on the planetary mass from joint models of the new TOIs where radial velocity observations could not detect the candidate planet. Joint fit results are presented in Tables 8 and 9, and phase-folded lightcurves for each target in Figs 3 and 4.

Figure 3.

Light curves phase-folded to the best-fitting orbital periods and zoomed to the transit region for new candidates considered in this paper. Binned flux is overplotted. Best-fitting models are shown as dashed lines. From top to bottom, TOI-641.01, TOI-510.01, and TOI-271.01 are shown.

Open in new tabDownload slide

Figure 4.

Light curves phase-folded to the best-fitting orbital periods and zoomed to the transit region for new candidates considered in this paper. Binned flux is overplotted. Best-fitting models are shown as dashed lines. From top to bottom, TOI-745.01 and TOI-697.01 are shown. Some discrepancy in the ingress and egress of TOI-745 can be seen, despite an eccentric model not being preferred.

Open in new tabDownload slide

6.1 TOI-510/HD52758

The TESS light curve of TOI-510 shows significant correlated variability at twice the period of the candidate transit (Fig. 5), which may be connected to stellar rotation, or the orbit of a blended eclipsing binary. Given the old age of the star (Table 5) it is highly unlikely that a 2.7 d periodic signal could be related to the stellar rotation of TOI-510 itself, lending more weight to a blended scenario. Although an apparently significant transit remains in the detrended light curve (Fig. 3), this transit is V-shaped, implying either a grazing planet, a blended stellar eclipse, or remnant stellar variability. We do not come to a conclusion on the nature of the TOI candidate TOI-510.01 but consider this ambiguity to cast doubt on a planetary origin. The variability has been accounted for using a Gaussian Process in the joint model, which can be seen in Fig. 5.

Figure 5.

Zoom of the Sector 61 TESS PDCSAP light curve for TOI-510, showing correlated variability on twice the transiting candidate period. Transits of TOI-510.01 are marked with vertical lines.

Open in new tabDownload slide

In radial velocity, TOI-510 shows a significant linear drift of |$0.785^{+0.070}_{-0.065}\, \mathrm{m\, s}^{-1}\mathrm{d}^{-1}$| (Fig. 6), a marginal 2|$\sigma$| signal |$K_b$| at the period of the transiting candidate of |$0.66^{+0.35}_{-0.33} \,\mathrm{m\, s}^{-1}$|⁠, and an additional signal on a period of |$10.0^{+1.3}_{-1.0}$| d with |$K=1.56^{+0.40}_{-0.41} \mathrm{m\, s}^{-1}$|⁠. The additional 10 d signal is only observed for approximately two cycles (Fig. 6) and only reaches a false alarm probability of 10 per cent, leading us to consider it as a candidate only. Periodograms of the RV data and activity indicators are shown in Fig. 7. Given the short baseline, aliases of the 10 d signal are also significant (at approx. 0.9, 0.47, and 0.32 d) and the true period is unclear, requiring more data points to clarify. No significant periodicity is seen in the indicators, although the CCF FWHM does show a spike at the period of the transiting candidate.

Figure 6.

Radial velocity timeseries for TOI-510, showing linear drift. The best-fitting model is shown in orange.

Open in new tabDownload slide

Figure 7.

Generalized Lomb–Scargle periodograms for TOI-510 RVs, residuals and activity indicators, after removing the linear trend. False Alarm Probability lines of 0.1 and 0.01 are shown. The transiting candidate period near 1.35 d and the non-transiting candidate near 10 d are highlighted with vertical dashed lines.

Open in new tabDownload slide

The corresponding mass of the inner transiting candidate b is |$1.08^{+0.58}_{-0.55}\,{\rm M}_{\rm{\oplus }}$|⁠, and |$m_c\sin i$| for the non-transiting candidate planet, assuming the 10 d peak is correct, is |$4.82^{+1.29}_{-1.26}\,{\rm M}_{\rm{\oplus }}$|⁠. The phased RV data and models for each planet are shown in Figs 8 and 9. The transit fit for planet b is grazing (Fig. 3) leading to a poorly constrained radius of approx. |$4.18^{+2.03}_{-0.96}\,{\rm R}_{\rm{\oplus }}$|⁠, assuming a planet led to the transiting signal. If real, the planet appears to have a low density compared to other planets of this mass, with |$\rho =0.08\,\mathrm{g\, cm}^{-3}$| using the nominal mass and radius values, but the density is strongly dependent on the poorly constrained radius. No stellar companions are detected by high resolution imaging, although the linear drift implies a larger mass planetary or stellar companion on a wide orbit. Corroborating this, the Gaia RUWE value for TOI-510 is 2.43, implying the star’s astrometric motion is inconsistent with a single-star solution. There is also a resolved nearby source in Gaia (TIC238086649) at a separation of 6|$^{\prime \prime }$|⁠, visible in Fig. 1. Dilution from this neighbour is removed by the SPOC pipeline, but that removal does not account for variability, and the nearby neighbour could be an eclipsing binary causing the apparent transit, and/or periodic variability. Ground-based transit observations capable of resolving both stars are recommended to clarify the nature of the candidate. We ran the TESSPositionalProbability tool (Hadjigeorghiou & Armstrong 2024) to determine probabilities for each star to be the true host of the candidate, given the observed centroid. We find a 41 per cent probability of TOI-510 being the true host, a 46  per cent probability for TIC238086649, and the remainder on more distant fainter sources. As such the centroid data is inconclusive but does not reject TOI-510 being the true host. Overall, this is a complex system which should be treated with caution.

Figure 8.

Radial velocity data for TOI-510, phase folded on the period of planet b from the joint model, with the best-fitting model of planet c removed. Random models from the Markov Chain Monte Carlo (MCMC) trace are shown in orange. The best-fitting model is shown in red.

Open in new tabDownload slide

Figure 9.

Radial velocity data for TOI-510, phase folded on the period of planet c from the joint model, with the best-fitting model of planet b removed. Random models from the MCMC trace are shown in orange. The best-fitting model is shown in red.

Open in new tabDownload slide

6.2 Other null results

TOI-271/HIP21952 was observed by both HARPS and PFS. TESS pixel data of the target is shown in Fig. A1. No significant signal is seen in the RVs at the transiting candidate period. Our joint fit finds a 95 per cent confidence upper limit on |$K_b$| of 1.02 |$\mathrm{m\, s}^{-1}$|⁠, corresponding to an upper limit on planetary mass of |$2.39\,{\rm M}_{\rm{\oplus }}$|⁠. The RV data and model are shown in Fig. 10. The planetary radius is found to be |$4.16^{+0.50}_{-0.52}\,{\rm R}_{\rm{\oplus }}$|⁠, but this does not consider dilution from the detected companion star from high spatial resolution imaging, which has a delta magnitude of 5, angular separation 0.15 arcsec (Lester et al. 2021) and is fully blended with the target star at the resolution of TESS, implying 1 per cent of the flux in the TESS aperture comes from the companion star. It is unclear which star the transiting object orbits, and hence what its true radius is accounting for dilution. If the object orbits the primary star, the implied planetary radius is |$4.24\,{\rm R}_{\rm{\oplus }}$|⁠, well within the error. If the object orbits the companion star, |$R_p/R_*$| becomes 0.301, implying an eclipsing stellar object depending on the unknown parameters of the companion star. Although the TOI disposition remains uncertain, given the stellar companion and null RV detection, this target has a high chance of being a false positive.

Figure 10.

Radial velocity data for TOI-271, TOI-641, TOI-697, and TOI-745, phase folded on the period from the joint model. Random models from the MCMC trace are shown in orange. The model best matching the 95 per cent confidence limit on semi-amplitude is shown in red to highlight the limit we place on candidate planet mass.

Open in new tabDownload slide

TOI-641 was observed by both HARPS and PFS. No significant signal is seen in the RVs at the transiting candidate period. Our joint fit finds a 95 per cent confidence upper limit on |$K_b$| of |$0.77\,\mathrm{m\, s}^{-1}$|⁠, corresponding to an upper limit on planetary mass of |$1.49\,{\rm M}_{\rm{\oplus }}$|⁠. The RV data and model are shown in Fig. 10. The candidate radius is found to be |$3.44^{+0.24}_{-0.22}\,{\rm R}_{\rm{\oplus }}$| under the planet model, not considering dilution. Two stellar companions are detected by high-resolution imaging from Gemini, with delta magnitudes of 3.9 and 4.2 and angular separations of 3.6 and 3.5 arcsec, respectively. The fraction of flux in the TESS aperture from these companions is therefore 4.5 per cent. Assuming the transiting object transits the primary star, this dilution in the TESS light curve would imply a planetary radius of |$3.60\,{\rm R}_{\rm{\oplus }}$|⁠, large considering the mass upper limit. If the transiting object orbits one of the other bound stars, the radius would be substantially larger and imply a stellar companion. There is additionally a resolved nearby source in Gaia, TIC744547537. We again ran the tesspositionalprobability tool (Hadjigeorghiou & Armstrong 2024), finding a 22  per cent probability that the star TOI-641 is the true host, and a 67 per cent probability that TIC744547537 is the source of the observed signal, with the remaining probability on fainter more distant stars visible in Fig. A2. These results combine with the tight upper mass limit to imply that the observed transit likely does not originate from TOI-641.

TOI-697 was observed with HARPS, with only 6 data points obtained which show variability out of phase with the transit ephemeris. Performing a simple Keplerian fit gives a 95  per cent confidence upper limit on |$K_b$| of 9.60 |$\mathrm{m\, s}^{-1}$|⁠, corresponding to an upper limit on planetary mass of |$30.1\,{\rm M}_{\rm{\oplus }}$|⁠. The RV data and model are shown in Fig. 10. TESS pixel data of the target is shown in Fig. A3. The candidate radius is found to be |$2.17^{+0.09}_{-0.08} {\rm R}_{\rm{\oplus }}$| under the planet model, not considering dilution. Several stellar companions are detected by high resolution imaging (Lester et al. 2021), implying that TOI-697 is a triple star with two lower mass companion stars (⁠|$m_*/m_1=0.41,0.38$|⁠). Transit analysis by Lester et al. (2022) showed that the primary star is the most likely host of the transiting object, with a probability of 87  per cent, which, if true, means that the impact of dilution on the estimated radius is approx. 1  per cent. Our mass limit is consistent with this hypothesis.

TOI-745 was observed with HARPS, with 26 data points obtained showing no variability in phase with the transit ephemeris. Our joint fit gives a 95  per cent confidence upper limit on |$K_b$| of 1.68 |$\mathrm{m\, s}^{-1}$|⁠, corresponding to an upper limit on planetary mass of |$2.48M_{\rm{\oplus }}$|⁠. The RV data and model are shown in Fig. 10. TESS pixel data of the target is shown in Fig. A4. The candidate radius is found to be |$2.56\pm 0.07R_{\rm{\oplus }}$| under the planet model. No stellar companions were detected by high resolution imaging, although there are multiple nearby resolved sources in Gaia. The transit model shows some discrepancy in transit duration, implying that either a non-zero eccentricity would be required to better fit the data, or that our spectroscopic stellar density, which is included in the fit, is incorrect, perhaps because the transiting object orbits a different star with different parameters. Given the short orbital period of 1.08 d, where the orbit would be expected to circularize rapidly, this casts further doubt on the planet hypothesis.

7 DISCUSSION

Having covered the less straightforward cases, here we move on to a broader interpretation of the whole NCORES sample, focused on confirmed planets published in our companion papers. Fig. 11 shows the complete NCORES sample, in orbital period, planet radius and planet mass. The characterized planets span in size from Earths to sub-Saturns, and cover a broad range of compositions. The planet locations on the mass–radius diagram are shown in Fig. 12.

Figure 11.

NCORES observed planets in orbital period–planet radius (left) and orbital period–planet mass (right). Confirmed planets are shown in purple and null results in orange. Planets detected in the radial velocities only and plotted with their |$M_p\sin {i}$| values and shown in blue. A Gaussian kernel density estimate showing the background population of planets from the NASA Exoplanet Archive is shown in green.

Open in new tabDownload slide

7.1 Individual highlights

Highlights of the NCORES programme include the TOI-125 system, containing three super-Earth transiting planets near the 2:1 orbital resonance. The planets have similar size, |$\sim 2.8\,{\rm R}_{\rm{\oplus }}$|⁠, but varying density and hence composition (Nielsen et al. 2020), providing an interesting laboratory of planet formation. The TOI-431 system (Osborn et al. 2021b) contains three detected planets, two transiting, with the inner, smaller planet on one side of the radius valley and an outer planet on the other, providing a chance to test theories of the valley’s origin without needing to consider different host stars (King et al. 2024). Another key highlight is the TOI-849 system (Armstrong et al. 2020), containing an ultrashort period Neptune-sized planet with a high mass, and density similar to the Earth. TOI-849b was the first ‘super-dense’ Neptune discovered, and is hypothesized to be the remnant core of a gas giant planet, exposed through collisions, photoevaporation, or an unusual formation process which avoided runaway gas accretion. TOI-849b was one of the first planets discovered populating the Neptunian Desert.

7.2 Planet core mass and composition distributions

Studying the masses of planets near the radius gap, which are expected to have experienced substantial photoevaporation, was one of the goals of the survey. Theoretical models of the gap use the distribution of planet core masses and compositions as an input to the models, and also constrain those distributions based on the depth and location of the gap (e.g. Rogers et al. 2021). This means that local exoplanet mass demographics are a key input to understanding the relevant physical processes. We supplement our NCORES sample by taking all confirmed planets from the National Aeronautics and Space Administration (NASA) Exoplanet archive as of 2024 January 11 with mass and radius determinations better than |$3\sigma$|⁠, and with masses and radii greater than 0.5 |${\rm M}_{\rm{\oplus }}$| and |${\rm R}_{\rm{\oplus }}$|⁠, respectively. We take the location of the gap in instellation (S)–radius (⁠|$R_p$|⁠) space, using the observationally derived relation from Ho & Van Eylen (2023) of |$\log (R_p/{\rm R}_{\rm{\oplus }}) = 0.07\log (S/{\rm S}_{\rm{\oplus }}) + 0.11$|⁠. Given this relation was determined for stars with |$M_\star \ge 0.7\,{\rm M}_\odot$|⁠, we cut the sample to planets with host stars above this mass. Using instellation to define the gap allows the effect of varying luminosity from different host star stellar types on the evaporation potential of different planets to be reduced. If the radius gap is driven primarily by mechanisms other than photoevaporation, instellation may play a less important role, but will still combine with those mechanisms to some degree to produce the final planet sample.

Fig. 13 shows the retrieved sample and gap locations in instellation–radius space. Fig. 14 shows the distribution of planet masses below and above the gap, limited to planets with mass less than |$20\,{\rm M}_{\rm{\oplus }}$| and instellation |$S\gt 10\,{\rm S}_{\rm{\oplus }}$| for clarity, giving a sample of 38 planets below the gap and 85 above for the FGK stars in our sample. The population of planets below the gap shows a distribution broadly consistent with the core mass distributions considered in Rogers & Owen (2021), showing a broad peak of masses for planets below the gap (i.e. ‘bare cores’) around |$4\,{\rm M}_{\rm{\oplus }}$| that falls off at higher masses. Notably, the distribution also shows a sharp cut-off at |$10\,{\rm M}_{\rm{\oplus }}$|⁠, sharper than seen in Rogers & Owen (2021).

Figure 12.

NCORES observed planets in mass–radius. Planets from the NASA Exoplanet Archive are shown for context in grey. Composition lines for pure iron (dark grey, Dorn et al. 2015), Earth-like (green), Steam-world (blue), and Condensed water-worlds (light blue) (Venturini et al. 2024) are shown, plus an additional line representing a planet with a 95 per cent water and a 5 per cent H–He envelope with |$S=50\,{\rm S}_{\rm{\oplus }}$|⁠.

Open in new tabDownload slide

Figure 13.

Planets with known mass and radius determined to better than |$3\sigma$| plotted in insolation–radius space, coloured and inversely scaled in size by their density. Only planets with host star mass |$M_*\gt 0.7\,{\rm M}_\odot$| are shown. The radius gap location determined in Ho & Van Eylen (2023) is shown as a black dashed line. The background is coloured by a kernel density estimate of the planets shown. The diversity of sub-Neptune composition above the gap can be seen in the range of densities observed in this region.

Open in new tabDownload slide