TESS Delivers Its First Earth-sized Planet and a Warm Sub-Neptune^*

TESS will survey nearly the entire sky over two years by monitoring contiguous overlapping 90 × 24° sectors for 27 days at a time (Ricker et al. 2015). The primary mission will complete the southern ecliptic hemisphere in its first year, and the northern hemisphere in its second. Toward the ecliptic poles (i.e., higher ecliptic latitudes), there is overlap between sectors and targets can be observed for more than 27 days. This is the case for HD 21749, which was observed in four TESS sectors. We used the publicly available four-sector TESS data in our analysis.

The first transit of HD 21749b was identified by both the MIT Quick Look Pipeline (which searches for planet candidates in the 30 minutes Full Frame Images) and the Science Processing Operations Center (SPOC) pipeline based at the NASA Ames Research Center (Jenkins et al. 2016). No other matching transits were found in the publicly released data from sectors 1 and 2. After TOI 186.01 was alerted, we searched for archival spectroscopy of this very bright star and found 59 High-accuracy Radial-velocity Planet Searcher (HARPS) radial velocities (RVs) in the European Southern Observatory (ESO) archive (see Section 2.4). A periodogram of these RVs showed a clear signal at 35.57 days, but the TESS photometry and the R_HK index (Boro Saikia et al. 2018) indicate a stellar rotation period of around 35 days, calling for caution. If the strongest period in the RVs did correspond to the planet, then we expected to see additional transits in sectors 3 and 4. Once the sector 3 data were released, we discovered that a momentum dump²⁹ occurred approximately 35.6 days after the sector 1 transit (see Figure 1). We did not let this unexpected turn of events foil our search efforts, and upon close inspection of the light curve we succeeded in recovering a partial transit (including egress) immediately following the momentum dump. Finally, we observed a third transit in sector 4, thus allowing for a robust ephemeris determination (see Section 3.3). Serendipitously, when applied to the first three sectors of the HD 21749 light curve, the SPOC pipeline yielded an additional planet candidate with a period of 7.9 days (TOI 186.02).

Zoom In Zoom Out Reset image size

We used the 2 minute target pixel data for our analysis. The target was on the edge of the camera, where the point-spread function (PSF) is triangular in shape. We tried improving the light curve precision by extracting light curves from the publicly available target pixel stamps using different photometric apertures (circles as well as irregular pixel boundaries). We then detrended the raw light curve by fitting a basis spline with knots spaced by 0.3 days, after excluding both 3σ outliers and data obtained during and immediately surrounding transits (Vanderburg & Johnson 2014). The final 2 minute cadence light curve has an rms of 240 ppm.

Care must be taken to rule out false positives that could masquerade as planet candidates. Both planet candidates in the HD 21749 system pass the false-positive tests performed on the TESS photometry: there is no evidence of secondary eclipses, and no detectable motion of the centroid of the star on the detector during the transit events. While a giant star (HIP 16068/TIC 279741377; R_* = ${3.15}_{-0.09}^{+0.12}$ R_⊙) is present 22'' from HD 21749, the excellent period match between the transits and the HD 21749 RVs rule out this neighboring star as the source of the TOI 186.01 transit signals. If TOI 186.02 transits the giant star, it would have to be a Jupiter-sized object as the transit events have a low impact parameter (and are thus unlikely to be due to a grazing eclipsing binary). The probability that the giant star blended with HD 21749 would coincidentally also host a transiting planet is very low, but we nevertheless explored this scenario in Section 3.2. The presence of the neighboring star also warrants correcting the transit depths for dilution (see Section 3.3).

To test for nearby stellar companions that could dilute the light curve and bias the measured planetary radius, we used archival data from the VLT/NaCo instrument (Lenzen et al. 2003; Rousset et al. 2003) collected in 2005. These data were initially collected to search for planets and low-mass stellar companions, and as such used common direct imaging techniques to maximize the achievable contrast. Images were collected with a 6 s exposure time, causing the central pixels of the star to saturate. Beyond  50 mas, where the images are not saturated, the longer integration time increases the observing efficiency and reduces read noise. To complement these saturated images, a small number of images with a shorter exposure (1.9 s) were collected in which the target star is unsaturated and its flux can thus be calibrated. In addition, images were collected simultaneously at multiple wavelengths, and sequentially at multiple telescope rotations, such that variations in the PSF could be used to separate stellar speckle noise and companion flux (Marois et al. 2006). For simplicity, in this Letter we used only the 1.6 μm data in the upper-left quadrant of the chip, and simply co-added the aligned data. Although this reduces the absolute contrast reached, it is sufficient for the purposes of this analysis.

We used the first set of 10 saturated frames collected on 2005 November 29 (each 28 × 6 s)—the conditions degrade in the second set of 10 frames—and two unsaturated frames (16 × 1.8 s) taken immediately before and after this. We followed a standard data reduction process: we subtracted a dark/sky background, cropped the data to the relevant part of the chip, flat-fielded, aligned images based on the stellar position, and derotated frames to align the north image before combining the frames into a single reduced image. To determine the sensitivity of this image, we inserted scaled copies of the unsaturated PSF, and recovered these with standard aperture photometry. A sensitivity curve and an image of the target are shown in Figure 2. The target appears single to the resolution of these images, with no companions detected within the field of view, which extends to at least 2'' from the target in every direction. Beyond this separation Gaia is sensitive to companions that are ∼6 mag fainter than the host, and a companion of this magnitude would change the measured planet radius by 0.2%. We conclude that the measured radii of HD 21749b and HD 21749c are unbiased by nearby stars.

Zoom In Zoom Out Reset image size

We used the Las Cumbres Observatory (LCO; Brown et al. 2013) Network of Echelle Spectrographs (NRES; Siverd et al. 2016) to obtain two spectra of the neighboring star TIC 279741377, and analyzed them with SpecMatch (Petigura et al. 2017). The two spectra had a signal-to-noise ratio (S/N) of 59 and 69, and show RVs consistent at the 100 m s⁻¹ level, thus ruling out an eclipsing binary on TIC 279741377.

We obtained additional spectroscopy with the CORALIE spectrograph (Queloz et al. 2000) on the Swiss Euler 1.2 m telescope at La Silla Observatory in Chile. The spectra were collected at phases 0.18 and 0.82, near the expected RV maximum and minimum. Cross correlations were performed with a weighted K5 binary mask from which telluric and interstellar lines were removed (Pepe et al. 2002).

The NRES and CORALIE RVs are shown in Figure 3.

Zoom In Zoom Out Reset image size

Prior to this work, we had collected 59 observations of HD 21749 with HARPS (Mayor et al. 2003) on the ESO 3.6 m telescope at La Silla Observatory in Chile. Of these, 55 were obtained between 2003 and 2009 by the original HARPS guaranteed time observation (GTO) programs to search for planets, and four measurements were obtained by another program in 2016. These observations are publicly available on the ESO Science Archive Facility.

We extracted precision RVs from these R ∼ 115,000 spectra and utilized them to both rule out an eclipsing binary as the origin of the transit signal, and to measure the mass of HD 21749b. The data taken in 2016 have slightly different coordinates and proper motion, and the RVs were extracted using a template for a different spectral type compared to the data obtained by the HARPS GTO. We homogenized all the observations to account for these differences, and re-reduced all the data using the latest HARPS pipeline.³⁰ In addition, the 2016 observations were gathered after an upgrade to the instrument that involved a change of some optical fibers (Lo Curto et al. 2015). This modification induced a few m s⁻¹ offset in RV that is dependent on the stellar spectral type. It is difficult to model this offset, and the best option so far is to fit for an offset between the HARPS data taken before and after this upgrade (see Section 3.3).

The second PRV data set presented in this work comes from the iodine-fed Planet Finder Spectrograph (Crane et al. 2010) on the 6.5 m Magellan II telescope at Las Campanas Observatory in Chile. HD 21749 was observed with PFS, with somewhat irregular sampling, as part of the long-term Magellan Planet Search Program between 2010 January and 2018 October, for a total of 48 RVs (45 epochs). We then began a high-cadence observing campaign purposely for TESS follow up in 2018 December, adding 34 more velocities (nine more epochs). The iodine data prior to 2018 were taken through a 05 slit resulting in R ∼ 80,000, and those from 2018 were taken through a 03 slit, resulting in R ∼ 130,000; the iodine-free template was taken through the 03 slit. The PFS detector and observing strategy changed in 2018 February, and the jitter in the 2018 observations decreased significantly. Exposure times ranged from ∼150 to ∼600 s, resulting in S/N of ∼100–200. All PFS data are reduced with a custom IDL pipeline that flat fields, removes cosmic rays, and subtracts scattered light. Further details about the iodine-cell RV extraction method can be found in Butler et al. (1996), and the PFS RVs used in this Letter can be found on ExoFOP-TESS.³¹

We performed a fit to the broadband spectral energy distribution (SED), following the methodology described in Stassun & Torres (2016). We adopted the fluxes published in all-sky photometric catalogs: Tycho-2 B_TV_T, 2MASS JHK_S, and WISE1–4. These flux measurements span the wavelength range 0.4–22  μm. We assumed solar metallicity based on values in the PASTEL catalog (Soubiran et al. 2016), and we fit Kurucz atmosphere models (Kurucz 2013) with the free parameters being the effective temperature, the overall flux normalization, and the extinction, the latter limited to the maximum line-of-sight value from the dust maps of Schlegel et al. (1998). The resulting SED fit has a reduced χ² = 3.4, T_eff = 4640 ± 100 K, ${A}_{V}={0.14}_{-0.04}^{+0.00}$, F_bol = (2.43 ± 0.11) × 10⁻⁸ erg s⁻¹ cm⁻². With the Gaia second data release parallax (Gaia Collaboration et al. 2018) corrected for systematic offset from Stassun & Torres (2018), this gives ${R}_{\star }=0.695\,\pm 0.030\,{R}_{\odot }$. Note that we have used here the simple 1/π distance estimate with symmetric errors; in this case the parallax is so large and the relative uncertainty so small that this simple estimate is nearly identical to the Bayesian estimate with asymmetric errors (Bailer-Jones et al. 2018). We can estimate the stellar mass from the empirical relations of Torres et al. (2010), which gives M_⋆ = 0.73 ± 0.07 M_⊙. Together with the stellar radius, this provides an empirical estimate of the stellar mean density, ρ_⋆ = 3.09 ± 0.23 g cm⁻³.

We also derived the stellar age, employing the PARAM online Bayesian interface (da Silva et al. 2006), which interpolates the apparent V-band magnitude, parallax, effective temperature, and metallicity onto PARSEC stellar evolutionary tracks (Bressan et al. 2012). We employed the initial mass function from Chabrier (2001) and a constant star formation rate, obtaining an age of 3.8 ± 3.7 Gyr.

The top section of Table 1 lists the stellar parameters.

Table 1.  Median Values and Uncertainties for the HD 21749 System

Parameter		Value
Catalog Stellar Information
R.A.	Right Ascension (h:m:s; J2015.5)	03:27:00.045
Decl.	Declination (d:m:s; J2015.5)	−63:30:00.60
λ	Ecliptic Longitude (deg)	352.8828
β	Ecliptic Latitude (deg)	−73.8346
HD ID	Henry Draper Catalog ID	21749
TIC ID	TESS Input Catalog ID	279741379
TOI ID	TESS Object of Interest ID	186.01
Photometric Stellar Properties
V mag	Apparent V-band Magnitude	8.1
TESS mag	Apparent TESS-band Magnitude	6.95
Stellar Parameters
D	Distance (pc)	16.33 ± 0.007
M*	Mass ($\,{M}_{\odot }$)	0.73 ± 0.07
R*	Radius ($\,{R}_{\odot }$)	0.695 ± 0.030
L*	Luminosity ($\,{L}_{\odot }$)	0.20597 ± 0.00016
${\rho }_{* }$	Density (cgs)	${3.03}_{-0.47}^{+0.50}$
$\mathrm{log}g$	Surface Gravity (cgs)	${4.613}_{-0.061}^{+0.052}$
${T}_{\mathrm{eff}}$	Effective Temperature (K)	4640 ± 100
[Fe/H]	Metallicity (dex)	0.003 ± 0.060
Age	Stellar Age (Gyr)	3.8 ± 3.7
Markov Chain Monte Carlo (MCMC) Fit TESS Bandpass Wavelength Parameters
u1	Linear Limb-darkening Coeff.	0.492 ± 0.026
u2	Quadratic Limb-darkening Coeff.	${0.170}_{-0.025}^{+0.026}$
ADa	Dilution from Neighboring Stars	0.121 ± 0.012
MCMC Fit Telescope Parameters
$\dot{\gamma }$	RV Slope (m s−1 day−1)	${0.00157}_{-0.00068}^{+0.00070}$
${\gamma }_{\mathrm{rel},1}$	Relative RV Offset HARPS 1 (m s−1)	${59608.99}_{-0.64}^{+0.66}$
${\gamma }_{\mathrm{rel},2}$	Relative RV Offset HARPS 2 (m s−1)	${59619.1}_{-3.3}^{+3.2}$
${\gamma }_{\mathrm{rel},3}$	Relative RV Offset PFS 1 (m s−1)	−4.6 ± 2.0
${\gamma }_{\mathrm{rel},4}$	Relative RV Offset PFS 2 (m s−1)	$-{5.3}_{-3.1}^{+3.0}$
${\sigma }_{J,1}$	RV Jitter HARPS 1 (m s−1)	${4.17}_{-0.41}^{+0.47}$
${\sigma }_{J,2}$	RV Jitter HARPS 2 (m s−1)	${1.7}_{-1.7}^{+4.9}$
${\sigma }_{J,3}$	RV Jitter PFS 1 (m s−1)	${4.63}_{-0.53}^{+0.65}$
${\sigma }_{J,4}$	RV Jitter PFS 2 (m s−1)	${1.27}_{-0.29}^{+0.42}$
		HD 21749b	HD 21749c
MCMC Fit Planetary Parameters
P	Period (days)	${35.61253}_{-0.00062}^{+0.00060}$	${7.78993}_{-0.00044}^{+0.00051}$
T0	Optimal Conjunction Time (${\mathrm{BJD}}_{\mathrm{TDB}}$)	${2458385.92502}_{-0.00055}^{+0.00054}$	${2458371.2287}_{-0.0015}^{+0.0016}$
i	Inclination (Degrees)	${89.33}_{-0.11}^{+0.15}$	${88.90}_{-0.37}^{+0.50}$
${R}_{P}/{R}_{* }$	Radius of Planet in Stellar Radii	${0.0350}_{-0.0010}^{+0.0011}$	${0.01196}_{-0.00052}^{+0.00051}$
K	RV Semi-amplitude (m s−1)	${5.51}_{-0.33}^{+0.41}$	<1.43b
$e\cos {\omega }_{* }$	⋯	−0.025 ± 0.051	⋯
$e\sin {\omega }_{* }$	⋯	${0.179}_{-0.085}^{+0.078}$	⋯
Derived Planetary Parameters
$a/{R}_{* }$	Semimajor Axis in Stellar Radii	${60.1}_{-3.3}^{+3.2}$	21.8 ± 1.2
RP	Radius (R⊕)	${2.61}_{-0.16}^{+0.17}$	${0.892}_{-0.058}^{+0.064}$
a	Semimajor Axis (au)	${0.1915}_{-0.0063}^{+0.0058}$	${0.0695}_{-0.0023}^{+0.0021}$
Teq	Equilibrium Temperature (K)	${422}_{-14}^{+15}$	${701}_{-23}^{+25}$
e	Eccentricity	${0.188}_{-0.078}^{+0.076}$	⋯
${\omega }_{* }$	Argument of Periastron (Degrees)	${98}_{-17}^{+21}$	⋯
TP	Time of Periastron (${\mathrm{BJD}}_{\mathrm{TDB}}$)	${2458350.8}_{-1.1}^{+1.6}$	${2458332.2789}_{-0.0027}^{+0.0026}$
MP	Mass (M⊕)	${22.7}_{-1.9}^{+2.2}$	<3.70b
τ	Ingress/egress Transit Duration (hours)	${0.163}_{-0.034}^{+0.043}$	${0.0358}_{-0.0046}^{+0.0055}$
T14	Total Transit Duration (hours)	${3.230}_{-0.038}^{+0.041}$	${2.515}_{-0.086}^{+0.077}$
b	Transit Impact Parameter	${0.587}_{-0.15}^{+0.097}$	${0.42}_{-0.18}^{+0.11}$
${\rho }_{P}$	Density (cgs)	${7.0}_{-1.3}^{+1.6}$	<31.93b
log gP	Surface Gravity	${3.514}_{-0.069}^{+0.067}$	<3.68b

Notes.

^aA_D is defined as $F2/(F1+F2)$, where F2 is the flux of the target and F1 is the flux of the neighboring contaminating star. ^bLimits denote the 99.73% (3σ) upper confidence interval.

Download table as:  ASCIITypeset image

In this section we rule out the remaining false-positive scenario for HD 21749c. The non-grazing impact parameter of the TOI 186.02 transits and the NRES and CORALIE RVs of TIC 279741377 (Figure 3) rule out that the candidate is an eclipsing binary on this star. If TOI 186.02 were a planet transiting TIC 279741377, it would have a radius of 1.3 R_J, for which the mass–radius relations of Ning et al. (2018) suggest a mass of ≈1 M_J. We used the T_eff, [Fe/H] and logg values from the SpecMatch analysis of the NRES spectra together with isochrone modeling (as described in Johnson et al. 2017) to obtain a stellar mass for TIC 279741377 of 1.86 pm 0.17 ${M}_{\odot }$. We determined that a 1 M_J planet around this star would result in an RV semi-amplitude of 65 m s⁻¹, which is ruled out by the CORALIE RVs. Moreover, the equatorial transit duration in this scenario (assuming a circular orbit) would be 9.2 hr–3.7 times longer than the transit of TOI 186.02. To match the transit duration of TOI 186.02, the planet would have to transit TIC 279741377 with an impact parameter of 0.96, which is inconsistent with the value we find for TOI 186.02.

We further detrended the spline-corrected light curves (described in Section 2.1) by training a Gaussian Process (GP; Foreman-Mackey et al. 2017) on the out-of-transit data, and subsequently evaluating it on the full light curve (Günther et al. 2018). We conservatively discarded the last transit of HD 21749c because it occurred during a momentum dump. We then performed two independent analyses that jointly fitted the TESS photometry with the HARPS and PFS RVs, using a two-planet model with a long-term trend ($\dot{\gamma }$). We used EXOFASTv2 (Eastman et al. 2013; Eastman 2017) and allesfitter (M. Güenther & T. Daylan 2019, in preparation). EXOFASTv2 is based on a differential evolution Markov Chain Monte Carlo (MCMC) algorithm that uses error scaling. allesfitter is an inference framework that unites the packages ellc (light curve and RV models; Maxted 2016), dynesty (static and dynamic nested sampling; https://github.com/joshspeagle/dynesty), emcee (MCMC sampling; Foreman-Mackey et al. 2013), and celerite (GP models; Foreman-Mackey et al. 2017) to model systematic noise in the photometry. We used allesfitter with nested sampling instead of MCMC.

We fitted for constant offsets and jitter terms (added in quadrature to the instrumental uncertainties) for each of the four RV data sets that we used: HARPS 1 (pre-upgrade), HARPS 2 (post-upgrade), PFS 1 (pre-upgrade), PFS 2 (post-upgrade). We used the stellar mass, radius, metallicity, and temperature values listed in Section 3.1 as Gaussian priors for both analyses. We also fit for dilution of the transit signal due to the neighboring star, for which we used the contamination ratio from TIC V7 (Stassun et al. 2018) and 10% of its value as a Gaussian prior. We did not use any other Gaussian priors. We note that EXOFASTv2 determines and uses quadratic limb-darkening coefficients interpolated from the Claret tables for TESS (Claret 2017), while allesfitter uniformly samples the quadratic limb-darkening coefficients following Kipping (2013).

The best-fit parameter values are consistent between our two independent analyses. To make it easier to reproduce our results, we report in this work the values from the EXOFASTv2 fit (see Table 1). The transit and RV observations for each planet, together with the best-fit models, are shown in Figures 1 and 4, respectively. We do not measure a statistically significant RV semi-amplitude for planet c, on which we instead set a 3σ upper limit of ∼1.43 m s⁻¹.

Zoom In Zoom Out Reset image size

An additional independent analysis of the RV measurements using RadVel (Fulton et al. 2018) with the periods fixed to 35.608 and 7.7882 days and T_c for each planet fixed to the TESS-determined values, gives results for K, e and ω that are fully consistent with those in Table 1.

Finally, as an experiment before the sectors 3 and 4 data became available, we performed a joint analysis of the sector 1 transit together with the HARPS RVs. The period determination of this fit was driven by the RVs, and the transit only poorly constrained the time of inferior conjunction because of the long baseline between the RV and TESS measurements. Notably, the fit preferred a higher eccentricity (which is no longer allowed when including the second and third transits, and the PFS RVs), and a time of periastron passage offset by 10 days from the true value. The best-fit period was also shorter and the best-fit mass higher than the values obtained when including all transits.

Our findings underline the need for at least two transits (not necessarily all from TESS) to confirm a TESS planet and determine its properties adequately. A single transit is insufficient for most systems, even if RVs are available.

The $\mathrm{log}({R}_{{HK}})$ value for HD 21749 suggests a stellar rotation period of 34.5 ± 7 days (Mamajek & Hillenbrand 2008). We investigate this further by extracting S_HK and H_α indices from the PFS (pre-upgrade only) and HARPS spectra. Figure 5 shows Lomb–Scargle periodograms of these indices (second and third panels from the top). In the 10–100 day range, the highest peak in both of these HARPS activity indicators corresponds to 37.2 days, which we attribute to the rotation of the star. These peaks also do not overlap with the orbital period of HD 21749b (marked by the red lines).

Zoom In Zoom Out Reset image size

We also obtained Kilodegree Extremely Little Telescope (KELT; Pepper et al. 2004) photometry of HD 21749. The star has been monitored by KELT as part of its long-term transit survey of bright stars. The KELT light curve for HD 21749 spans ∼3.3 yr, contains 7848 individual points (taken between 2010 February and 2013 June), and has an rms of ∼0.0098 mag. A Lomb–Scargle periodogram of the KELT photometry finds the most significant peak at a period of 38.954 days (see the bottom panel of Figure 5).

More data will help to better determine the stellar rotation period, but the existing photometric and spectroscopic data sets suggest that it is longward of planet b's period. Importantly, the RV periodogram (top panel of Figure 5) shows the strongest peak at 35.6 days (the period of HD 21749b) but does not show significant (above 0.01% false alarm probability) power in the 37–39 day period range.

Nevertheless, we employed Equation (2) of Vanderburg et al. (2016) to estimate the magnitude of systematic errors due to the stellar variability (${\sigma }_{s,\mathrm{RV}}$) that could affect the RV signals of HD 21749b and HD 21749c. We used the standard deviation of the raw TESS light curve (${\sigma }_{s,{\text{}}{TESS}}=0.0013$) rather than its peak-to-peak amplitude (${{Fpp}}_{s,{\text{}}{TESS}}$), and vsini = 1.04 km s⁻¹ (derived from the HARPS spectra) to obtain ${\sigma }_{s,\mathrm{RV}}\approx 1.3$ m s⁻¹.

Therefore, while we do not expect the RV signal of HD 21749b to be strongly affected by stellar variability, the uncertainties on the K values reported in Table 1 could be somewhat underestimated.