The K2-HERMES Survey. I. Planet-candidate Properties from K2 Campaigns 1–3

A little over two decades ago, the first planets were discovered orbiting Sun-like stars (Mayor & Queloz 1995; Butler & Marcy 1996; Marcy & Butler 1996), and humanity entered the exoplanet era. Those first planets revolutionized our understanding of planet formation and offered a tantalizing hint to the uniqueness of the architecture of our solar system.

In the years that followed, the number of known exoplanets grew—and the surprises continued to come. Some systems contained planets moving on highly eccentric orbits (e.g., Tamuz et al. 2008; Wittenmyer et al. 2009; Kane et al. 2016). Others had multiple planets on highly compact orbits, far closer to their host stars than the distance between Mercury and the Sun (e.g., Lissauer et al. 2011; Campante et al. 2015; Gillon et al. 2017), or planets on highly inclined orbits (e.g., Triaud et al. 2010; Addison et al. 2013; Huber et al. 2013).

However, the techniques used to discover these myriad planets are all strongly biased—toward more massive planets, and typically toward planets with short orbital periods (e.g., O'Toole et al. 2009; Kipping & Sandford 2016; Vanderburg et al. 2016b). To begin searching for true solar system analogues requires either a search for massive long-period planets (e.g., Bedell et al. 2015; Endl et al. 2016; Wittenmyer et al. 2016, 2017) or a search for small worlds analogous to the terrestrial planets (e.g., Wittenmyer et al. 2011; Howard et al. 2012; Swift et al. 2015).

The first real information on the frequency of planets that resemble the solar system's rocky worlds came from the Kepler spacecraft (Borucki et al. 2010), which stared continuously at a single patch of the sky for just over four years. By observing over 100,000 stars, Kepler discovered an unprecedented number of planets (Coughlin et al. 2016)—including objects as small as the planet Mercury (Barclay et al. 2013). From Kepler's great census, it has become clear that small planets are common—which might be the first hint that our solar system is far from unique.

In 2014, the second of Kepler's four reaction wheels, used to orient the spacecraft and keep it pointing at its target field of view, failed. As a result, the first phase of Kepler's mission came to an end, and the spacecraft was repurposed to carry out the K2 survey (Howell et al. 2014). The telescope was re-oriented to point in the plane of the ecliptic—a position it could maintain without requiring the use of the broken reaction wheels.

To avoid pointing directly at the Sun, the K2 mission points at any given patch of the ecliptic for a period of about 80 days. At the end of that observing cycle, it pivots further away from the Sun, to point at a new field, and repeat the process. Where the original Kepler survey searched for planets out to orbital periods of around one year, K2 can find only those planets with the shortest orbital periods. However, to counterbalance this, through the full K2 mission, the spacecraft will be able to observe a diverse multitude of community-selected stars, and thereby yield a new treasure trove of short-period planets to add to the original survey's grand census (e.g., Crossfield et al. 2016; Vanderburg et al. 2016a).

To fully understand the nature of the planets found orbiting those stars, it is critically important that the stars themselves are well characterized and understood (e.g., Huber et al. 2014; Ren et al. 2016; Petigura et al. 2017). As a result, there is a need for the target stars to be observed spectroscopically from the ground. With such a large number of stars to be targeted, it would be grossly inefficient to observe them one at a time—but fortunately, in Australia, the 2dF/HERMES instrument on the Anglo-Australian Telescope is ideally suited to such a survey.

Built to perform "Galactic Archaeology" (e.g., Kos et al. 2017; Martell et al. 2017; Traven et al. 2017), the 2dF/HERMES instrument allows observers to obtain high-quality and high-resolution (R ∼ 28,000) spectra of several hundred stars in a single observation—typically of around an hour's duration. Observations with HERMES allow the abundances of a number of species in the target stars to be determined, as well as enabling us to obtain relatively precise values for the stellar parameters (mass, radius, and age)—information critical to the understanding of the plethora of planets that will be found by the K2 mission.

In this paper, we present the first observations from the K2-HERMES survey. In Section 2, we describe the observing setup for our survey and give more detail on the 2dF/HERMES instrument. In Section 3, we describe how the stellar parameters have been calculated from the HERMES spectra, before detailing the physical properties of the K2 planet candidates orbiting those stars in Section 4. Finally, in Section 5, we present our conclusions and discuss our plans for future work.

The observations were obtained with the 3.9 m Anglo-Australian Telescope located at Siding Spring Observatory in Australia. We use the High Efficiency and Resolution Multi-Element Spectrograph (HERMES), which can obtain spectra of up to 360 science targets simultaneously (Barden et al. 2010; Brzeski et al. 2011; Heijmans et al. 2012; Sheinis et al. 2015).

2.1. Observational Strategy

A single 2dF/HERMES exposure covers most of a Kepler CCD module (Figure 1), enabling relatively efficient observations of K2 fields. Our objective is simply to gather spectra for as many K2 targets as possible, without introducing biases driven by the relative probability of hosting a planet. As with the related TESS-HERMES program (Sharma et al. 2018), we favor bright stars to obtain targets of most interest for asteroseismic and exoplanetary science. Owing to fiber "cross talk" in the instrument, we follow the procedure implemented by the GALAH survey (Martell et al. 2017), where the fields are chosen such that the brightest and faintest stars observed do not differ in brightness by more than three magnitudes. Balancing this against the desire for the most efficient use of the 360 HERMES science fibers results in a strategy whereby we observe each Kepler CCD module twice: once as a bright visit (10 < V < 13) and again as a faint visit (13 < V < 15). Bright visits consist of a single 30-minute exposure, while faint visits consist of three 30-minute exposures. In the bright visits, HERMES typically observes all of the available bright stars in a single pass. The total number of stars per 2dF field in this range is typically less than 360, so we can observe all of them in one 2dF pointing. In the faint visits, the total number of stars per 2dF field is greater than 400 but the number of K2 targets is 210 on average. This means all K2 targets lying within a 2dF circle can be observed in just one telescope pointing (consisting of a "bright" and a "faint" exposure).

Zoom In Zoom Out Reset image size

2.2. Raw Reduction

The K2-HERMES survey uses the same instrument as the GALAH survey (De Silva et al. 2015; Martell et al. 2017) and follows a similar observing strategy. Hence, we use the same reduction pipeline as GALAH to perform the data reduction from the raw CCD images to the final calibrated spectra. The procedure, described fully in Kos et al. (2017) and Sharma et al. (2018), is in brief: (1) raw reduction is performed with a custom IRAF-based pipeline, (2) four basic parameters (T_eff, log g, [Fe/H], and radial velocity) and continuum normalization are calculated with a custom pipeline "GUESS" by matching the observed normalized spectra to synthetic templates. A grid of AMBRE synthetic spectra is used for this purpose de Laverny et al. (2012). Figure 2 shows a histogram of the signal-to-noise (S/N) for our K2-HERMES spectra in each of the four HERMES bandpasses.

Zoom In Zoom Out Reset image size

The spectroscopic stellar parameters have been estimated with a combination of classical spectrum synthesis for a representative reference set of stars and a data-driven approach to propagate the high-fidelity parameter information with higher precision onto all of the stars in the K2-HERMES survey. The method is identical to that used by the TESS-HERMES survey (Sharma et al. 2018) and is briefly outlined as follows. First, we use the spectrum synthesis code Spectroscopy Made Easy (SME) by Piskunov & Valenti (2017) to analyse the reference set. This training set includes samples of stars with external parameter estimates, Gaia benchmark FGK stars, and stars with asteroseismic information from K2 Campaign 1 (Stello et al. 2017). Next, we use these SME results as training labels of the training set as input for The Cannon (Ness et al. 2015) to propagate the analysis to all stars. As shown by Sharma et al. (2018), the comparison with benchmark stars shows trends and systematic offsets for the hottest/coolest dwarfs and turnoff/subgiant stars of our survey, which are however below 250 K (for T_eff) and 0.25 dex (for $\mathrm{log}g$). Similar shortcomings have been noted by Torres et al. (2012) as well as Bensby et al. (2014). With new data provided by Gaia, we will however be able to overcome such shortcomings of purely spectroscopic analyses in future data releases for the joint GALAH/K2/TESS pipeline for HERMES. A more detailed explanation will be given by S. Buder et al. (2018, in preparation), who include, e.g., Gaia parallaxes to constrain $\mathrm{log}g$ for the training set of the GALAH Data Release 2.

Derived stellar properties such as mass, radius, and age are then computed via the Bayesian Stellar Parameters estimator (BSTEP), described fully in Sharma et al. (2018). The input observables J, J − K_s, T_eff, log g, [Fe/H] are brought to bear on a grid of about 5 × 10⁶ points in [Fe/H], age, and initial mass, outputting the intrinsic parameters [Fe/H], age, initial mass, distance, E(B − V), and their probabilities. The output is then used to compute other derived parameters, like stellar mass and radius, which are functions of the intrinsic parameters. In Table 1, we report our results for the 46 planet-candidate host stars observed by K2-HERMES during C1–C3. Figure 3 gives a comparison of our derived radii and masses with those obtained by the empirical method of Torres et al. (2010). There is good agreement within uncertainties, and no systematic trends are evident.

Zoom In Zoom Out Reset image size

While none of the planet-candidate host stars discussed here are a priori known to be binaries, five are flagged as possible binaries from the t-SNE classification as described in Traven et al. (2017). Closer inspection of the spectra reveals that four of these stars display evidence for a second set of lines: EPIC 201407812, 202634963, 202688980, and 203753577. These stars have been marked as binaries in Table 1, and the presence of a weak secondary set of lines may have affected The Cannon analysis. Hence, we caution that those stellar parameters, while not obviously erroneous, are potentially unreliable. The fifth, EPIC 206024342, is flagged but the spectrum S/N is too low to visually detect any features of a second set of lines.

Huber et al. (2016) presented a catalog of stellar parameters for 138,600 stars in K2 campaigns 1–8. For the vast majority of those stars, parameters were derived from photometry; here, we compare those results with our spectroscopically derived parameters. Figure 4 shows the difference in log g derived here with that from Huber et al. (2016), as a function of the difference in T_eff from the two works. The center and right panels of Figure 4 show similar comparisons, but for [Fe/H] and the derived stellar radii, respectively. No systematic trends are apparent, apart from the expected anticorrelation between log g (panel a) and the stellar radius (panel c). That is, stars for which we obtain a smaller log g will have a larger derived radius. The median parameter offsets, in the sense of (this work—H16), are as follows: Δ T_eff = −39 ± 287 K, Δ log g = −0.06 ± 0.54 dex, Δ [Fe/H] = −0.025 ± 0.392, and Δ R_* = 0.04 ± 2.38 R_⊙.

Zoom In Zoom Out Reset image size

The right panels of Figure 4 give a comparison of our derived stellar radii with those of H16. Some stars do exhibit significant differences in derived radii, which is mainly due to the difference in log g. Huber et al. (2016) measured log g from photometry and proper motions, introducing substantial uncertainty (as evidenced by the large error bars in Figure 4). We find four stars with log g (H16)—log g (this work) < −0.7. These are all red giants in H16 but we classify them as dwarfs. We also find two stars with large differences in the opposite direction: log g (H16)—log g (this work) > 0.7. They are EPIC 201516974 and EPIC 203070421. The first was classified as a low-luminosity giant in H16, but our results place it in the red clump region with log g = 2.66 ± 0.12. The second was a hot dwarf but now sits in an odd position in the (log g, T_eff) plane. In our spectroscopic pipeline, it is flagged as being too far away from the training set, and so we caution the reader that the results for that star may be suspect.

Table 2 gives the properties of the 57 planet candidates from C1–C3 for which the K2-HERMES program has obtained spectra of their host stars. The planet data (orbital period and R_p/R_*) have been obtained from the NASA Exoplanet Archive. We derived the planetary radii by multiplying R_p/R_* by the stellar radii obtained by our grid-based modeling as described above. Uncertainties in the planetary radii result from the propagated uncertainties in R_* and R_p/R_*; for those planet candidates from Vanderburg et al. (2016a) without published uncertainties in R_p/R_*, we adopted the median fractional uncertainty of 0.0025 obtained by Crossfield et al. (2016). Figure 5 shows our newly derived planet radii against published values (Table 2).

Zoom In Zoom Out Reset image size

For the majority of planets, our newly derived radii agree with the published values, though we do find that for six planet candidates, our results show radii that are more than >3σ larger than the published values. Of these, five orbit somewhat evolved stars with radii in the range 1.9–8 R_⊙, resulting in larger inferred planetary radii, turning some potentially terrestrial worlds into gas giants. The most dramatic change is candidate super-Earth EPIC 203070421.01 (Vanderburg et al. 2016a), now approaching three Jupiter radii. Given that the largest inflated planets are ∼2 R_Jup, our revised host-star parameters suggest that this candidate is a false positive.

The various catalogs of K2 planet candidates contain some fraction of stars with spectroscopically derived parameters and others with photometrically derived parameters only. In Figure 6, we reprise Figure 5 comparing our results with published values, but showing only those for which the published values were derived from spectroscopic measurements. Notably, all of the candidates identified in Vanderburg et al. (2016a) had only photometric host-star radii, and that catalog was the source of all of the discrepant results discussed above. The remaining planets shown in Figure 6 are in excellent agreement with the published spectroscopic results.

Zoom In Zoom Out Reset image size

As noted in Section 3, four stars in this sample are found to be binaries. We thus caution that the following (as-yet unconfirmed) planet candidates may be false positives: EPIC 201407812, 202634963, 202688980, and 203753577.

Most of the 57 planet candidates examined here remain within the range of reasonable planet radii (i.e., smaller than a few tens of Earth radii). However, EPIC 202843107.01 now has a radius of 216.6 R_⊕ (approximately 2 solar radii). That candidate has an exceptionally deep transit (R_p/R_* = 0.6032), again unphysically large for a planet, particularly given that the host star appears to be a main-sequence A star (Table 1). We thus strongly suspect EPIC 202843107.01 is a false positive. Similarly, EPIC 201407812 hosts a candidate 84.7 R_⊕ planet, (nearly eight Jupiter radii); given that the host is now confirmed as a binary, this candidate also appears to be a false positive.

Our revised stellar parameters bring to light some interesting individual planets in this sample. One metal-poor star hosts a candidate giant planet: EPIC 206311743 ([Fe/H] = −0.42 ±0.10). A second star, EPIC 202634963, also hosts a candidate planet but we confirm it to be a double-lined spectroscopic binary and hence is likely to be a false positive. Such planets are rare by virtue of the well-known planet-metallicity correlation (Gonzalez 1997; Fischer & Valenti 2005), whereby giant planets have difficulty forming by core accretion from metal-poor protoplanetary disks. While both remain candidates, if they were to be confirmed, they would be extremely interesting counterexamples.

Close-in planets orbiting evolved stars are also known to be rare, with only 12 known within 0.5 au from the various radial velocity surveys of so-called "retired A stars" (e.g., Bowler et al. 2010; Jones et al. 2015; Wittenmyer et al. 2015). Our sample contains two giants (log g < 3.5), each hosting one planet candidate. They are EPIC 201516974 and EPIC 203070421, neither of which is flagged as a binary star. Caution is warranted, however, as giant stars are intrinsically more noisy (due to granulation), and hence the false-positive rate of detecting transit signals is higher (Sliski & Kipping 2014).

The California-Kepler Survey (CKS) team have noted that Kepler planets exhibit a gap in their radius distribution (Fulton et al. 2017), with planets of 1.5–2.0 R_⊕ apparently depleted by more than a factor of two. In light of this finding, we examine the radius distribution for the 38 small (R_p < 5 R_⊕) K2 planets featured herein (Figure 7). As perhaps expected with the low numbers involved, no compelling pattern is yet evident; future papers extending this work to further K2 campaigns will provide the necessary data to fill in this distribution. Figure 8 shows the planetary radii versus orbital period, with lines connecting the updated radii to their previously published counterparts.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

The incident flux levels and equilibrium temperatures of the planet candidates are also shown in Table 2. For completeness, we give two values of equilibrium temperature: "hot dayside" and "well-mixed," corresponding to re-radiation over 2π and 4π steradians, respectively. The former would be most suitable for close-in planets presumed to be tidally locked. Figure 9 shows the relation of the planet radii and their incident fluxes for the 57 planet candidates considered here. Lundkvist et al. (2016) identified a "hot super-Earth desert": a lack of planets between 2.2–3.8 R_⊕ receiving incident flux more than 650 times that of Earth. The envelopes of these planets have been stripped by photoevaporation. From the Kepler planet sample for which the host stars have been characterized by asteroseismology, Lundkvist et al. (2016) found no planets in that range; in our sample of K2 planet candidates, we find one object falling within the dashed rectangle in Figure 9. That candidate, EPIC 203518244b, orbits a star we characterize as a slightly evolved F subgiant (T_eff = 6205 ± 125 K, log g =3.8 ± 0.2, and M_* = 1.37 ± 0.21 M_⊙). Our parameters for this star are in excellent agreement with those of Huber et al. (2016), who derived T_eff = 6349 K, log g = 3.9, and M_* = 1.42 M_⊙.

Zoom In Zoom Out Reset image size

Though none of the planet candidates discussed here are even remotely considered habitable, for completeness we show the stellar habitable zone boundaries in Table 3, which may prove relevant for any outer planets subsequently found to reside in these systems.

We note that K2 Campaigns 1–3 have more than 100 identified planet host stars (Adams et al. 2016; Crossfield et al. 2016; Vanderburg et al. 2016a), whereas this work currently presents spectroscopically derived stellar properties for 57 planets orbiting 46 stars. The "missed" planet hosts can be accounted for primarily by (1) targets falling on the corners of the Kepler CCD modules and hence not currently covered by the K2-HERMES survey, and (2) host stars fainter than the faint limit of the K2-HERMES survey. The 2dF field is two degrees in diameter, for an area of 3.14 square degrees, while a single CCD module has an area of five square degrees. This means that K2-HERMES only observes ∼54% of (assumed uniformly distributed) stars falling on K2 detectors. K2 targets are skewed in favor of cooler, fainter stars (Huber et al. 2016), but our observed sample contains very few stars cooler than ∼4300 K, and our spectroscopic analysis pipeline is currently not equipped to handle very cool stars. Hence, we expect to miss the coolest targets, accounting for about 25% of the total planet candidates (Crossfield et al. 2016). Those stars are most likely to be characterized by surveys using near-infrared spectroscopy (e.g., Muirhead et al. 2014; Dressing et al. 2017a, 2017b).

The precision of our planetary radii is comparable to the published values. The CKS results permitted a significant improvement in the precision of radii for Kepler prime mission KOIs (Johnson et al. 2017), primarily because those objects, as a sample, had very few prior spectroscopic observations. On the other hand, the K2 planet candidates described here are small enough in number, and their host stars are sufficiently bright that the discovery teams have obtained spectra of sufficient resolution and S/N to derive reasonably precise stellar (and hence planetary) radii. The primary value of our work is that we have presented a fully self-consistent set of spectroscopic and model-derived parameters for the K2 planet-candidate sample. Applying this uniformly derived set of host-star parameters to the planet-candidate sample thus yields a set of self-consistent planetary properties, which is critically important for informing population studies of small exoplanets. Future papers in this series will extend our analysis to additional K2 campaigns. The K2-HERMES first data release paper (S. Sharma et al. 2018, in preparation) will provide precise stellar parameters for thousands of K2 targets, including those without planet detections, facilitating studies of occurrence rates and planetary properties as functions of host-star properties and their relative position in the Galaxy.

D.S. is supported by Australian Research Council Future Fellowship FT1400147. S.S. is funded by University of Sydney Senior Fellowship made possible by the office of the Deputy Vice Chancellor of Research, and partial funding from Bland-Hawthorn's Laureate Fellowship from the Australian Research Council. S.L.M. acknowledges support from the Australian Research Council through DECRA fellowship DE140100598. L.C. is supported by Australian Research Council Future Fellowship FT160100402. This research has made use of NASA's Astrophysics Data System (ADS), and the SIMBAD database, operated at CDS, Strasbourg, France. This research has also made use of the Exoplanet Orbit Database and the Exoplanet Data Explorer at exoplanets.org (Wright et al. 2011; Han et al. 2014). We thank the Australian Time Allocation Committee for their generous allocation of AAT time, which made this work possible.

Software: SME (Piskunov & Valenti 2017), The Cannon (Ness et al. 2015).