Kepler Eclipsing Binary Stars. V. Identification of 31 Candidate Eclipsing Binaries in the K2 Engineering Dataset

For the engineering run of K2, only calibrated pixel data were made available, in contrast to the datasets released for the original mission, which also included extracted light curves. For this work, we have extracted light curves for all 1951 targets from the pixel data, using the tools used and presented in, e.g., Pápics et al. (2013). We have removed the background flux in the pixels using a low-order spline fit to all available pixels around the targets. The light curves were then constructed by adding up all flux in the pixels around the central pixel that have more than 100 counts as their 99.9 percentile value for flux. This creates a constant mask that includes all pixels that ever, with the exception of outliers, contribute significant flux. The number of pixels in the masks for each EB is listed in Table 1. We find this to be a near-optimal choice, given that including pixels with less flux would increase the noise, and limiting the pixel selection to pixels with higher count levels would increase systematic trends.

The extracted light curves are detrended to remove any trends, instrumental or astrophysical, not related to the EB signal. This is done using an iterative sigma-clipping technique to divide the light curve by a polynomial fitted to the baseline of the data (see Prša et al. [2011] for details). In addition, the first 2 days of engineering data are particularly noisy for many of the targets, likely due to instrumental and satellite pointing issues, and were removed from all subsequent analysis (see also Vanderburg & Johnson 2014).

The scatter for each of the extracted 1951 light curves as a function of pixel mask size and magnitude is shown in Figures 1 and 2, respectively. The candidate EBs identified in this paper (§ 3) are depicted by square markers, with their sigmas computed about an analytical fit to the EB signal (see § 3.2). As is evident from these figures, the K2 engineering light curves overall exhibit a strongly bimodal distribution of sigma values, and this bimodal character is largely independent of pixel mask size or stellar magnitude. The reason for this bimodal distribution of light-curve sigmas is not clear.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Within each of the two groupings of stars in these figures, there is the expected trend of decreasing sigma with increasing brightness and increasing pixel mask size. The candidate EBs identified in this paper roughly sample the underlying distribution of light-curve noise properties, however, these do appear to exhibit somewhat higher sigma relative to the other stars of comparable magnitude and pixel mask size, since the analytical models we use to represent them do not fully capture their intrinsic light-curve variations (see § 3.2).

A more thorough investigation of the K2 engineering light-curve noise properties is beyond the scope of this paper. However, as this paper represents one of the first investigations utilizing the K2 engineering dataset, the noise properties presented in Figures 1 and 2 may provide the community with a means for estimating the expected light-curve precision of the observed K2 engineering targets as functions of pixel mask size and magnitude. We note that Vanderburg & Johnson (2014) have recently reported that improvements in the K2 photometric precision can be achieved through an optimal, pixel "self flat-fielding" approach. It is possible that the K2 engineering data noise properties described here can be ameliorated through such an approach.

In the K2 engineering target list, nine objects (60017809, 60017810, 60017812, 60017814, 60017815, 60017816, 60017818, 60017821, 60017822) were identified as previously known EBs. One of these (60017818) did not show a clear EB signature in the 12 days of K2 data, so it was excluded, but the remaining eight were all recovered independently.

Through manual inspection of all 1951 long-cadence light curves, we identified a total of 37 candidate EBs in the K2 engineering dataset. As in the original Kepler dataset, we cannot claim with absolute certainty that these are all true EBs, and some of these signals (e.g., 60017887, 60018031, 60018081, 60019244) could actually be due to pulsating or spotted stars rather than an eclipsing binary or ellipsoidal variable. To confirm their nature as an EB they would likely need to be validated by photometry or radial velocity follow-up, so we instead report these as candidate EBs based on their light-curve characteristics. However, see § 3.3 for evidence from automated classification that these are indeed bona fide EBs.

Some light curves also exhibit signals in addition to the EB signal (either one of the EB members is a variable or a background variable is blending with the EB). Several (e.g., 60018241, 60018343, 60020058, 60024040, 60021220, 60023653) show residuals on top of the EB signal, which could be indicative of some other short-period variation. It is difficult to tell whether any of these individual remaining variations are astrophysical, due to some sort of pulsations, or instrumental.

In the original mission, we identified EBs through a variety of methods (Prša et al. 2011), but since there were no threshold crossing events (TCEs—automated detection of transit and eclipse-like features) released by the Kepler science office for the engineering dataset, manually inspecting each light curve was a necessary step in order to test the feasibility of automated detection of eclipsing binary signals in K2 data. Candidate EBs were identified if they showed clear periodic ellipsoidal variation or eclipses in the light curves that repeated at least 3 times in the 12-day baseline of the data. If a light curve showed one or two single eclipse events, the source is included in the list, but ephemerides could not be determined. In the few cases in which only two cycles were observed, the period is still reported if it could be confirmed in the ∼2 days of data that had been trimmed prior to analysis. Additionally, 60018343 is known to be a transiting planet false-positive (Poleski et al. 2010), and so is included despite having a period too long to observe three full cycles. The unphased light curves for the objects can be seen in Figure 3. Planet Hunters¹¹ (Fischer et al. 2012) had independently detected and identified several of these EBs as well.

Zoom In Zoom Out Reset image size

Of these 37, there were six sets of nearby targets that exhibited the same period and shape in their light curves. It is likely that we are seeing the same EB signal from a single source bleeding into both apertures. Due to the large pixel and aperture sizes, it is difficult to determine the true source of this potential EB signal, and there is no direct mapping from Kepler identification to stellar objects. In these cases, the target with the larger amplitude signal was considered the true source, and the other target was marked as a blend (false-positive) and removed from the list. After removing these cases, there are a total of 31 manually detected candidate EBs (Table 2). The removed targets that were determined to be a duplicate source are listed in Table 3. This only removes the same EB signal from appearing twice—without follow-up the exact source cannot be confirmed and there may still likely be contamination from other sources in addition to the EB in the aperture.

The K2 engineering target list, unlike the KIC (Kepler Input Catalog) used for the original Kepler mission and the EPIC (Ecliptic Plane Input Catalog) used for the K2 campaigns, does not include target object names. All identified candidate EBs were cross-matched against known sources by their target coordinates with a radius of 1^'. These nearby sources and their previous characterizations are listed in Table 4. We have thus identified 20 previously unknown EBs as candidate EBs.

Kepler identification 60017806 was also initially identified as a candidate EB, but is actually a known extrasolar planet (WASP-28b) and was removed from the catalog.

Ephemerides for the candidate EB systems that exhibited at least three subsequent eclipse events are determined by computing a periodogram for each detrended light curve using BLS (Kovács et al. 2002), manually adjusting the correct period if necessary, and setting BJD₀ so that the deeper eclipse is placed at zero phase. The ephemerides for all 31 candidate EBs (BJD₀ for all 31 and periods when possible) are listed in Table 2 and are available online at http://keplerEBs.villanova.edu/k2. Despite such a small sample size, the distribution in EB orbital periods is consistent with that found from the original mission (Fig. 4), with a total detected candidate EB occurrence rate of 1.6%.

Zoom In Zoom Out Reset image size

The light curves are phase-folded and fit by a chain of four quadratic functions that describe the shape of the phased light curve (Fig. 5), as described in Prša et al. (2011). This analytic function is then used to determine the morphology, a value between zero (detached) and one (overcontact), using locally linear embedding (Matijevič et al. 2012). These values are listed in Table 2 under the "morph" column. The detrended and phased light curves for all candidate EBs with determined periods are shown in Figure 6, and the standard deviation of the residuals around these analytic functions are listed in Table 1.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

The candidate EBs identified in the K2 dataset provide an initial benchmark set for newly developed pipelines intended for automated discovery of EBs from large datasets such as those that will be provided by the ongoing K2 mission. We applied the Eclipsing Binary Factory (EBF) pipeline (Paegert et al. 2014; Parvizi et al. 2014) to the K2 light curves to test its ability to correctly recover these EBs. The EBF correctly recovered 92% of the manually identified candidate K2 EBs with at least 90% confidence in the classification, where the confidence here represents the posterior probability for one of ten variable types derived from priors associated with EBF's neural network training set (Paegert et al. 2014). This recovery rate is similar to that obtained by the EBF from the original Kepler dataset (Parvizi et al. 2014), suggesting that automated methods such as the EBF are capable of identifying a large sample of EBs in the upcoming K2 campaigns with good completeness.

The probability-based confidence level provided by EBF for each of its output classifications can also be used as a quantitative assessment of the reliability of each EB. Four objects returned a primary classification from EBF of "misc" (miscellaneous), which is a vestige of the original All Sky Automated Survey (ASAS) classification scheme. Since in all of these cases the next most likely classification returned by EBF is "EB," in these four cases we renormalize the confidence level, effectively ignoring the "misc" classification; these classification probabilities should be regarded as suggestive. For all of the other candidate EBs (24 cases possessing periods), the EBF classification probabilities cleanly represent the most likely quantitative classification confidence level. These confidence levels are reported in Table 2 and serve as a quantification for how likely each candidate may be a real EB, relative to all the other classes of periodic variable stars recognized by EBF. In all but two cases the confidence level is ∼99%, and in the other two cases is 70%–80%.

The authors gratefully acknowledge everybody who has made Kepler, and especially the K2 mission, possible. K. E. C. and K. G. S. gratefully acknowledge support from NASA Astrophysical Data Analysis Program grant NNX12AE22G. A. P. gratefully acknowledges support from the NASA Kepler Participating Scientist Program grant NNX12AD20G. S. B. is supported by the Foundation for Fundamental Research on Matter (FOM), which is part of the Netherlands Organisation for Scientific Research (NWO). This research has made use of the SIMBAD database, operated at CDS, Strasbourg, France.