PLANETARY CANDIDATES OBSERVED BY KEPLER. III. ANALYSIS OF THE FIRST 16 MONTHS OF DATA

For each KOI, the even-numbered transits and odd-numbered transits are modeled independently using the techniques described in Section 3. The depth of the phase-folded, even-numbered transits is compared to that of the odd-numbered transits as described in Batalha et al. (2010a) and Wu et al. (2010). A statistically significant difference in the transit depths is an indication of a diluted or grazing eclipsing binary system. A similar metric is computed by the DV pipeline module as described by Wu et al. (2010). Each uses a different methodology for detrending the light curves (i.e., filtering out stellar variability), and both proved useful and are tabulated in Columns 6 (modeling-derived statistic: O/E₁) and 7 (DV-derived statistic: O/E₂) of Table 5. In general, 3σ was the threshold for flagging a KOI as a false positive. However, when the two values disagreed, we deferred to the DV statistic owing to its more sophisticated whitening filters. One hundred forty-three KOIs in Table 5 have O/E₁ > 3 while only seven have O/E₂ > 3. Five have both O/E₁ and O/E₂ larger than 3σ but are otherwise clean candidates. Further inspection of their light curves suggested that stellar variability and/or instrumental transients were driving an anomalously high odd/even statistic, and the candidates were retained. In 121 cases, the DV model fitter failed, thereby precluding quantification of an odd/even statistic. In 18 of these cases, O/E₁ is larger than 3σ and the candidate was retained anyway. While most of these are marginal cases near the 3σ cutoff, users are cautioned that exceptions exist and should be examined independently on a case-by-case basis.

The modeling allows for the presence of a secondary eclipse (or occultation event) near phase = 0.5 as a means of identifying diluted or grazing eclipsing star systems. The Secondary Statistic (Column 8 of Table 5) is the relative flux level at phase 0.5, divided by the noise. As such, it can have positive as well as negative values. While its presence does not rule out the planetary interpretation, it acts as a flag for further investigation. More specifically, the flux decrease is translated into a surface temperature assuming a thermally radiating disk, and this temperature is compared to the equilibrium temperature of a low-albedo (0.1) planet at the modeled distance from the parent star. If the flux change is not severe enough to rule out the planetary interpretation (ascertained by the difference between the surface temperature and equilibrium temperature), the candidate is retained. Eight KOIs retained in the catalog have a statistic outside of (negative) 3σ. In each of these cases, it appears possible that the occultation signal is a result of stellar and/or instrumental flux changes. This statistic is relevant primarily for short-period orbits where circularization is expected since the search is only done at phase 0.5. The DV pipeline module checks to see if additional transit sequences were identified in the light curve at the same period but different phase. No such sequences were identified for the candidates reported here.

Table 2 lists KOIs (both old and new) that are V-shaped. The shape of a transit is not used as a diagnostic for rejecting planet candidates. The right combination of properties and geometry can, indeed, produce ingress and egress times that are a significant fraction of the total transit duration (e.g., grazing transits). However, a diluted eclipsing binary system is another possible interpretation and does not require such a narrow range of inclination angles (impact parameters). The false-positive rate amongst the V-shaped candidates is expected to be higher than the false-positive rate of the general population. A metric is constructed to flag such cases: 1 − b − (R_P/R_⋆). The objective is to alert the reader to populations with larger false-positive rates and also to avoid classification of a transit as V-shaped because of smear produced by the 30 minute cadence. Errors in the metric are dominated by the uncertainty of the modeled impact parameter which can be quite high. Therefore, all flagged light curves were visually inspected.

Table 2. KOIs Noted as V-shaped

KOI	P	$\frac{R_{\rm P}}{R_\star }$	$1-b-\frac{R_{\rm P}}{R_\star }$		KOI	P	$\frac{R_{\rm P}}{R_\star }$	$1-b-\frac{R_{\rm P}}{R_\star }$		KOI	P	$\frac{R_{\rm P}}{R_\star }$	$1-b-\frac{R_{\rm P}}{R_\star }$
	(days)					(days)					(days)
51.01	10.43	0.43	−0.632		886.03	21.00	0.03	−0.003		1793.01	3.26	0.30	−0.541
113.01	−387	0.57	−0.955		976.01	52.57	0.50	−0.791		1798.01	12.96	0.08	−0.050
138.01	48.94	0.12	−0.108		1020.01	54.36	0.37	−0.623		1799.01	1.73	0.47	−0.821
151.01	13.45	0.05	−0.027		1032.01	−650	0.09	−0.060		1829.01	22.84	0.33	−0.601
225.01	0.84	0.43	−0.810		1095.01	51.60	0.09	−0.020		1845.02	5.06	0.29	−0.540
256.01	1.38	0.44	−0.678		1096.01	−414	0.11	99.886		1872.01	30.52	0.08	−0.074
371.01	498.39	0.30	−0.556		1118.01	7.37	0.02	−0.011		1906.01	8.71	0.18	−0.323
403.01	21.06	0.40	−0.767		1192.01	−201292	0.32	−0.539		1935.01	15.44	0.14	−0.204
410.01	7.22	0.36	−0.656		1193.01	119.06	0.09	−0.037		1944.01	12.18	0.03	−0.005
417.01	19.19	0.12	−0.106		1209.01	272.07	0.08	−0.009		1968.01	10.09	0.03	−0.006
419.01	20.13	0.33	−0.549		1226.01	137.76	0.40	−0.351		2042.01	63.07	0.03	−0.000
466.01	9.39	0.08	−0.049		1227.01	2.16	0.31	−0.417		2128.01	24.26	0.24	−0.435
473.01	12.71	0.04	−0.001		1242.01	99.64	0.43	−0.797		2156.01	2.85	0.06	−0.034
601.02	11.68	0.23	−0.421		1359.02	104.82	0.07	−0.014		2189.01	33.36	0.47	−0.901
609.01	4.40	0.12	−0.123		1385.01	18.61	0.60	−0.935		2204.01	10.86	0.02	−0.005
611.01	3.25	0.10	−0.093		1387.01	23.80	0.44	−0.555		2259.01	12.19	0.23	−0.441
614.01	12.87	0.08	−0.039		1409.01	16.56	0.03	−0.003		2299.01	16.49	0.10	−0.162
617.01	37.87	0.41	−0.720		1426.03	150.03	0.15	−0.196		2363.01	3.14	0.02	−0.003
620.03	85.31	0.06	−0.029		1502.01	1.88	0.03	−0.001		2370.01	78.73	0.03	−0.019
625.01	38.14	0.18	−0.324		1540.01	1.21	0.38	−0.430		2380.01	6.36	0.04	−0.051
684.01	4.03	0.16	−0.280		1549.01	29.48	0.66	−1.208		2486.01	4.27	0.02	−0.012
698.01	12.72	0.11	−0.027		1560.01	31.57	0.16	−0.220		2512.01	15.92	0.25	−0.465
716.01	26.89	0.06	−0.036		1561.01	9.09	0.24	−0.431		2513.01	19.01	0.10	−0.183
728.01	7.19	0.10	−0.017		1582.01	186.40	0.08	−0.022		2519.01	4.79	0.08	−0.116
772.01	61.26	0.11	−0.107		1587.01	52.97	0.21	−0.343		2528.01	12.02	0.05	−0.058
797.01	10.18	0.09	−0.013		1591.01	19.66	0.04	−0.006		2538.01	39.83	0.04	−0.042
799.01	1.63	0.06	−0.042		1675.01	14.62	0.10	−0.144		2572.01	6.38	0.03	−0.001
815.01	34.84	0.35	−0.628		1684.01	62.82	0.06	−0.049		2573.01	1.35	0.06	−0.002
833.01	3.95	0.42	−0.791		1754.01	15.14	0.03	−0.008		2577.01	18.56	0.20	−0.372
838.01	4.86	0.12	−0.121		1761.01	10.13	0.08	−0.110		2578.01	13.33	0.38	−0.745
856.01	39.75	0.14	−0.039		1773.01	83.10	0.46	−0.783		2639.02	2.12	0.03	−0.024
882.01	1.96	0.20	−0.128		1783.01	134.48	0.08	−0.008

Notes. Negative, integer period values are intended to flag KOIs that have presented only one transit in the Q1–Q6 data. The complement of the impact parameter, b, minus the reduced planet radius, R_P/R_⋆, is listed in Columns 4, 8, and 12. This diagnostic serves as an indication of a grazing, or V-shaped, transit. This value is negative if the purported planet is not fully blocking the stellar disk at mid-transit. The closer this number is to −2 × (R_P/R_⋆), the more severely it is grazing. Grazing transit are required to model V-shaped light curves.

Download table as:  ASCIITypeset image

Negative values imply that the purported planet is not fully covering the stellar disk at mid-transit. The closer the metric is to −2 × (R_P/R_⋆), the more severely it is grazing. A grazing geometry is required to model V-shaped transits that are not caused by a large planet-to-star size ratio. The KOIs with light curves modeled as grazing transits are listed in Table 2 together with the orbital period and reduced radius.

A major source of false-positive planetary candidates in the Kepler data is a background eclipsing binary (BGEB) star within the photometric aperture of a Kepler target star. These BGEBs, when diluted by a target star, can mimic a planetary transit signal. Two methods are used to detect such BGEBs by using the Kepler pixels to determine the location of the object causing the transit signal: direct measurement of the source location via difference image analysis and inference of the source location from photo-center motion associated with the transits.

Photo-center motion is measured by computing the flux-weighted centroid of all pixels downlinked for a given star both in and out of transit. Fluxes associated with a transit event are identified using the ephemerides and transit durations listed in Table 2. The flux-weighted centroids are calculated in one of two ways. KOIs that were processed with the DV module in the SOC pipeline have a flux-weighted centroid measurement for every cadence. The in versus out of transit centroid offset is computed by performing a least-squares fit of the pipeline-generated transit model to the multi-quarter centroid time series, the idea being that the behavior of the flux-weighted centroid time series will mirror that of the flux time series. The centroid offset is the amplitude of the "transit" pulse in the model fit.

KOIs that were not processed by the DV pipeline module are treated slightly differently in that the flux-weighted centroid is computed for a pixel image constructed by taking the average of images in a single quarter when the KOI is not transiting. It is then compared to the flux-weighted centroid for a pixel image that is the average of images in the same quarter when the KOI is transiting. The centroid offset is the difference between the in and out of transit centroids. Quarterly offsets are averaged together as described below. Once the centroid offset is computed, the source location is inferred by scaling it by the inverse of the flux as described in Jenkins et al. (2010a).

Difference image analysis takes the difference between average in-transit pixel images and average out-of-transit images. Barring pixel-level systematics and field-star variability, the pixels with the highest flux in the difference image form a star image at the location of the transiting object, with flux level equal to the fractional depth of the transit times the original flux of the star. Performing a fit of the Kepler pixel response function (PRF; Bryson et al. 2010) to both the average difference and out-of-transit images gives the sky location of the transit source and the target star. The offset of the transit source from the target star is then defined as the transit source minus target star location. For most KOIs, the difference image offset is computed per quarter, and the quarterly offsets are averaged as described below. For a small number of KOIs with very low signal-to-noise ratio (S/N), a computationally expensive joint multi-quarter fit is performed, with uncertainties estimated via a bootstrap analysis.

In principle, both the photo-center motion and difference image techniques are similarly accurate for isolated stars and sufficiently high S/N transits, but the techniques have different responses to systematic error sources such as field crowding. The photo-center method is more sensitive to noise for low-S/N transits and crowding by field stars. In particular, the estimate of the transit source location by scaling the offset is highly sensitive to incompletely captured flux for either the target star or field stars in the aperture.

In the difference image method, the PRF fit to the difference and out-of-transit pixel images is biased by PRF errors described in Bryson et al. (2010), as well as errors due to crowding. Defining the centroid offset as the difference between the out-of-transit and difference images nearly cancels the PRF bias because both fits are subject to the same error. Bias due to crowding, however, is more of an issue. Excepting cases in which background stars are strongly varying, a difference image removes most point sources, leaving the transit signal as the predominant change in flux. The average out-of-transit image, on the other hand, contains all point sources. Consequently, the offset (formed by comparing a direct image and a difference image) may contain a bias.

Both bias types vary from quarter to quarter. A study of a large number of targets indicates that the combined biases have an approximately zero-mean distribution across quarters, so we can reduce their impact by averaging the quarterly centroid measurements across quarters. We compute this average by performing a robust χ² minimizing fit to the quarterly results. This approach produces an uncertainty that takes into account both the uncertainty of the individual quarterly observations as well as their scatter due to bias across quarters. An example set of quarterly measurements and the resulting average is shown in the left panel of Figure 1. This method works best for short-period candidates, where there are many transits in all quarters. This method is less effective in reducing the centroid bias for long-period candidates. In particular, single-transit candidates that appear in only one quarter may have unknown biases in their centroid positions that are not accounted for in the centroid uncertainties.

Zoom In Zoom Out Reset image size

The above centroid analysis was performed using data from Quarters 1–8. The transit source location offsets reported in Table 5 are from the difference image method since it is more reliable as evidenced by consistently smaller uncertainties compared to the photo-center motion method. A candidate passes the photo-center vetting step when its multi-quarter transit source offset is less than 3σ. There are, however, exceptions. KOIs having transit source offsets larger than 3σ were identified as having systematic errors (e.g., crowding biases) that are likely causing the large apparent offset. Modeling efforts to confirm these systematics are currently underway.

In some cases, the PRF fit algorithm failed, typically due to low quarterly S/N or to bright field stars that prevented reliable determination of the target star location via PRF fit. We retain these targets when visual inspection of the difference image indicates that the change in flux due to the transit is on the target star. Finally, difference imaging is very inaccurate for saturated targets, and we retain saturated candidates for which the difference images show no obvious indication of a background source. Offset values for slightly saturated stars (Kp between 10.5 and 11.5) are likely accurate to within 4'' (one pixel), while offset values for highly saturated stars (Kp <10.5) should be disregarded. Transit source offset values are set to −99 when we feel that the centroid measurement is unreliable.

KOIs were divided amongst more than 20 science team members for evaluation of the following metrics: (1) odd/even statistic, (2) occultation test, (3) quality of model fit, (4) long/short period comparison, (5) single-quarter photo-center motion, and (6) multi-quarter photo-center motion. An integer value of 0, 1, or 2 was assigned to each metric to indicate if the test clearly passed (0), was ambiguous (1), or clearly failed (2). A similar flag was ascribed to the visual appearance, where examples of suspicious characteristics would include markedly V-shaped transits, red noise and/or outliers in the time series calling into question the reliability of the transit signal, anomalously long transit durations, poor light curve fits, obvious secondaries, etc. Each candidate was then designated "yes," "no," or "maybe." Candidates flagged as ambiguous (maybe) were re-evaluated. In most cases, this required updated light-curve modeling, more detailed inspection of the software pipeline products, and/or further scrutiny of the pixel flux analysis yielding photo-center statistics.

Once every KOI was assigned a "yes" or "no" designation, the integer flags were summed and the distribution for the two populations was compared. This elucidated a small number (<10) of inconsistencies in which flags indicate a problem yet the KOI was assigned a "yes" designation. These were independently evaluated.

The period and location on the sky of every new KOI was cross-checked against the list of previously known KOIs and the catalog of eclipsing binaries (Prša et al. 2011; Slawson et al. 2011). This is a safeguard against redundancy. More importantly, though, the cross-check serves to identify flux contamination from bright eclipsing binaries that the photo-center analysis missed. Approximately 25 targets within 20'' of an existing KOI or eclipsing binary with commensurate (within a factor of 1, 2, 3) orbital period (to within 5 minutes) were identified. In this manner, we also identified two small swarms of spatially co-located stars, each with a transit-like event at a commensurate orbital period. The lack of sizable photo-center motion and the spatial extent of the contaminating flux suggests that scattered light (e.g., an optical ghost from a bright eclipsing binary at the focal plane anti-podal position) is responsible.

Every one of the 1 390 KOIs evaluated here will appear either in this contribution as a viable planet candidates, or in the associated false-positive catalog (S. Bryson et al. 2012, in preparation), or in future versions of the Eclipsing Binary Catalog.

Eight high-S/N (>30σ) single transit events were identified and included in the catalog. Their corresponding orbital periods are estimated from the transit duration and knowledge of the stellar radius assuming zero eccentricity. The periods are then rounded to the nearest integer and multiplied by −1 so as to distinguish them from the candidates that have reliable orbital ephemerides. Parameters requiring an accurate period (e.g., odd/even statistic, semimajor axis, etc.) are set to −99, which is the value adopted globally for unreliable, spurious, or unknown values.

The final list of viable planet candidates is presented in Table 4 with additional information listed in Table 5. In addition to the MES from the Q1–Q6 TPS run, the following vetting metrics are provided: (1) the odd/even statistic that tests for an eclipsing star system of nearly equal-mass components at twice the period (Columns 6 and 7), (2) the occultation (secondary) statistic that tests for a weak secondary event inconsistent with a planet occultation at phase 0.5 (Column 8), (3) the photo-center offsets in R.A. and decl. measured in arcsec and their associated uncertainties (Columns 9–12), and (4) the total photo-center offset position measured in units of the noise (Column 13).

Analysis is based on a blend of both quantitative metrics and manual inspection. Both the promotion from TCEs to KOIs and the promotion of KOIs to planet candidates has a human element that not only increases the reliability of the catalog but also reduces the number of false negatives that are discarded. The reader is cautioned, however, that reliability and efficiency of human scrutiny is difficult to quantify. Consequently, sample statistics derived from this sample will be difficult to interpret. This will improve with each subsequent catalog release as we work toward automation and uniformity of procedures and metrics.

For each KOI, a transit model was fit to the data. The transit model uses the analytic formulae of Mandel & Agol (2002) to model the transit and a Keplerian orbit to model the orbital phase. The model fits for the mean stellar density (ρ_⋆), center of transit (T₀), orbital period (P), scaled planetary radius (R_P/R_⋆), impact parameter (b), and occultation (secondary) depth (at phase = 0.5). In the case of multiple transiting candidates the orbits were assumed to be non-interacting and T₀, P, b and the occultation depth were fit for each candidate. The assumption is made that all candidates in the same system orbit the same star modeled by ρ_⋆ and that the total mass of the companion is much less than the mass of the central star. For a Jupiter-mass companion, an error of 0.02% will be incurred on the measurement of ρ_⋆, a 0.1 M_☉ companion would skew our estimate of ρ_⋆ by approximately 2%, and a 0.5 M_☉ companion would induce a systematic error of 41% on ρ_⋆. These assumptions give

$$\begin{equation} \left(\frac{a}{R_\star }\right)^3 \frac{\pi }{3 G P^2} = \frac{(M_\star +M_{\rm p})}{\frac{4 \pi }{3}R_\star ^3} \approx \frac{M_\star }{ \frac{4 \pi }{3}R_\star ^3}=\rho _\star , \end{equation} \tag{ 1 }$$

where a/R_⋆ is referred to as the reduced semimajor axis. Transits probe a small portion of an orbit giving little or no information about eccentricity. We assume circular orbits in our models. Consequently, the reduced semimajor returned by light curve modeling does not necessarily yield an accurate determination of the semimajor axis nor the stellar density. To emphasize this point, we purposely report the parameter as d/R_⋆ which, to first order, is equivalent to the ratio of the planet–star separation during transit to the stellar radius. It is equivalent to a/R_⋆ for zero-eccentricity orbits.

A full orbital solution is required to correctly determine stellar parameters from transit modeling. Great care must be taken when interpreting the fitted stellar parameters. A mismatch between the reported stellar parameters and a transit-derived stellar parameter can be interpreted as evidence of: an eccentric orbit, erroneous stellar classification, or a transiting companion with significant mass. One is not able to distinguish among these three cases using the model parameters provided. Best-fit model parameters were determined by computing a chi-square minimization search using a Levenberg–Marquardt algorithm (Press et al. 1992). Derivatives were numerically determined. Uncertainties are taken from the diagonal elements of the formal covariance matrix corresponding to each fitted parameter. They do not account for correlated errors between parameters.

Output parameters from light curve modeling can be used to compute the radius, semimajor axis, and equilibrium temperature of each planet candidate given knowledge of the stellar surface temperature, radius, and mass. These and other stellar properties for the host stars are listed in Table 3. Stellar coordinates and the apparent magnitude in the Kepler bandpass are obtained from the KIC at MAST (Brown et al. 2011).

Table 3. Host Star Characteristics

KOI	KICa	Kpb	CDPPc	α (J2000)	δ (J2000)	Teff	log g	R⋆/R☉	M⋆/M☉d	$f_{T_{\rm eff}}$e	fObsf
			(ppm)	(hr)	(deg)	(K)
5	8554498	11.665	220.7	19.31598	44.6474	5861	4.19	1.42	1.14	3	111111
41	6521045	11.000	105.5	19.42573	41.9903	5909	4.30	1.23	1.11	3	111111
46	10905239	13.770	56.2	18.88370	48.3552	5764	4.40	1.10	1.12	2	111111
70	6850504	12.498	198.7	19.17987	42.3387	5443	4.45	0.94	0.90	3	111111
82	10187017	11.492	322.8	18.76552	47.2080	4908	4.61	0.74	0.80	3	111111
94	6462863	12.205	98.8	19.82220	41.8911	6217	4.33	1.24	1.20	2	100110
108	4914423	12.287	93.9	19.26564	40.0645	5975	4.33	1.21	1.15	3	111111

Notes. ^aKepler Input Catalog number. ^bApparent magnitude in the Kepler bandpass. ^crms of Combined Differential Photometric Precision from Quarters 1–6 in units of parts per million. ^dStellar Mass is derived from surface gravity and stellar radius. ^eFlag indicates source of T_eff, log g, and R_⋆ as follows: (0) derived using KIC J−K color and linear interpolation of luminosity class V stellar properties of Schmidt-Kaler (1982); (1) KIC T_eff and log g are used as input values for a parameter search of Yonsei–Yale evolutionary models yielding updated T_eff, log g, and R_⋆; (2) T_eff, log g, and R_⋆ are derived using SPC spectral synthesis and interpolation of the Yale–Yonsei evolutionary tracks; (3) T_eff, log g, and R_⋆ are derived using SME spectral synthesis and interpolation of the Yale–Yonsei evolutionary tracks. ^fConcatenation of six integers, one for each of the six quarters of spacecraft data; the value of each successive integer indicates whether or not the star was observed for each of the successive quarters; a zero indicates the star was not observed that quarter; one indicates the star was observed the entire quarter; two indicates the star was observed only part of the quarter (relevant for stars on CCD Module 3 in Quarter 4); for example, a value of 000111 indicates the star was observed in Quarters 4, 5, and 6 but not in Quarters 1, 2, or 3.

Only a portion of this table is shown here to demonstrate its form and content. A machine-readable version of the full table is available.

Download table as:  DataTypeset image

The KIC has been a valuable resource for the stellar properties necessary for characterizing Kepler's planet candidates (e.g., T_eff, log g, and R_⋆). However, there are parameters in the KIC that imply populations that are inconsistent with stellar theory and observation. For example, G-type stars with log g ≈ 5 (i.e., well below the main sequence) are not uncommon in the KIC. While the uncertainty on the determination of log g shows that such a star is consistent with being near the main sequence, using such stellar parameters results in small stellar and planetary radii and an underestimation of incident flux on the planet candidate. We systematically correct such populations in the KIC by matching T_eff, log g and [Fe/H] to the Yonsei–Yale stellar evolution models (Demarque et al. 2004). We find the closest match based on minimization of

$$\begin{equation} \chi ^2 = \left(\frac{\delta T_{\rm eff}}{\sigma _{T_{\rm eff}}}\right)^2 + \left(\frac{\delta \log g}{\sigma _{\log g}}\right)^2 + \left(\frac{\delta \rm [Fe/H]}{\sigma _{\rm [Fe/H]}}\right)^2, \end{equation} \tag{ 2 }$$

where δ represents the difference in the KIC and Yonsei–Yale parameter and σ is the adopted uncertainty in the KIC parameter. We adopted $\sigma _{T_{\rm eff}}=200$ K, σ_log g = 0.3, and σ_[Fe/H] = 0.4 and required the modeled age to be less than 14 Gyr. For each star we adopt the model-determined mass and radius. Figure 2 shows T_eff and log g. The red dots show updated values that have been adopted for determination of the stellar mass and radius. Stellar properties derived in this manner are flagged in Table 3 by setting $f_{\rm T_{\rm eff}}$ equal to 1. The black lines point to the star's original location based on the KIC.

Zoom In Zoom Out Reset image size

There are three populations that show significant changes to the determined log g. (1) Stars that fall below the main sequence move to smaller values of log g. These stars have T_eff from roughly 4500 K to 6500 K. In general, estimates of the properties of these stars become larger and more luminous, which reduces the number of small stars and increases the amount of incident flux on the orbiting companion. (2) Most stars cooler than 4500 K see a substantial decrease in log g. However, modeled masses and radii are highly uncertain in this temperature range and should be used with caution. (3) There is a population of stars which have KIC log g near 4.1 and T_eff near 5000 K. Stars in this region have no match when we restrict model ages to ⩽14 Gyr. Such stars either move toward the red giant branch (lower values of log g) or toward the sub-giant branch (larger values of log g).

Forty-nine host stars listed in Table 3 have revised stellar properties determined from high-resolution spectroscopy taken as part of the Kepler ground-based follow-up observation program using the Shane 3 m telescope at Lick Observatory, the 1.5 m Tillinghast reflector at Whipple Observatory, the 2.7 m Harlan Smith telescope at McDonald, the 2.5 m Nordic Optical Telescope at La Palma, Spain, and the 10 m Keck telescope using HIRES. Fourteen of these were subjected to LTE spectral synthesis using Spectroscopy Made Easy (SME) (Valenti & Piskunov 1996; Valenti & Fischer 2005). The resulting surface gravity, effective temperature, and metallicity are used to identify the best-matching Yonsei–Yale stellar evolution model (Demarque et al. 2004) as described above for the KIC parameter adjustments. Stellar properties derived in this manner are flagged in Table 3 by setting $f_{\rm T_{\rm eff}}$ equal to 3.

The spectra of 45 host stars were compared to a synthetic library of spectra as part of the Stellar Parameter Classification (SPC) effort and analysis tool described by Buchhave et al. (2012). As in the case of the SME analysis, the resulting stellar parameters are compared to the Yonsei–Yale models to determine the stellar radius. The revised stellar properties are listed in Table 3 and flagged with $f_{\rm T_{\rm eff}}$ equal to 2.

Twenty-one host stars have neither KIC classifications nor spectroscopically derived stellar properties. In these cases ($f_{\rm T_{\rm eff}}$ equal to 0), the effective temperature is estimated from linear interpolation of the main-sequence properties of Schmidt-Kaler (1982) at the KIC J − K color. There is little justification for the assumption of a main-sequence luminosity class. False positives and large errors in the planet candidate radius should be expected amongst this sample.

Also included in Table 3 is the rms Combined Differential Photometric Precision (CDPP)—a measure of the photometric noise (including stellar and instrumental sources) on a 6 hr timescale after systematic-error correction and removal of strong sinusoidal features. A detailed description of the CDPP can be found in Jenkins et al. (2010b). CDPP values are crucial for statistical analyses of planet occurrence rates as they define the observational detection sensitivities. 3, 6, and 12 hr rms CDPP values for all observed stars are archived at MAST and can be obtained using the Data Retrieval Search form.

The objective of the Kepler mission is to determine planet occurrence rate as a function of planet radius and orbital period. This requires accurate stellar properties not only of the planet-hosting stars but also of the parent population of Kepler target stars. The latter is required for sensitivity corrections. By including corrections to the properties of planet-hosting stars (or subsets thereof) and not the parent sample, we are introducing a non-uniformity that can negatively impact statistical studies.

Independent analyses of systematic errors in the KIC have begun to appear in the published literature. Muirhead et al. (2012a), for example, utilize medium-resolution K-band spectroscopy to asses systematic errors in T_eff, log g, and [Fe/H] for late-K and M-type dwarfs amongst the sample of planet-hosting stars and report systematically overestimated stellar radii. Mann et al. (2012) perform a similar study of 382 target stars using medium-resolution optical spectra and show that the majority of bright (Kp <14) and cool (T_eff <4500 K) stars in the Kepler target catalog are giants. Pinsonneault et al. (2012) report on systematic errors in the griz photometry in the KIC used to determine T_eff, log g, and R_⋆ among other parameters. These errors lead to underestimates in effective temperature.

The star properties listed in Table 3 do not contain corrections from these independent analyses. However, there is a concerted effort to devise a sensible strategy affecting future catalog releases—one that produces accurate estimates of star and planet properties while preserving uniformity for the purpose of statistical studies. A Kepler Star Properties Working Group has formed for this express purpose.³²

In the meantime, we have checked for potential contamination of giants in the KOI sample using asteroseismology, which provides an effective means of identifying giant stars using Kepler data through the detection of solar-like oscillations (see, e.g., Gilliland et al. 2010). Oscillation frequencies scale with effective temperature and surface gravity, and generally range from periods of a few minutes for main-sequence stars to several hours and longer for red giants (e.g., De Ridder et al. 2009; Chaplin et al. 2010). Red giants with log g ≲ 3.5 and T_eff ≲ 5000 K oscillate at frequencies ≲ 300 μHz, and hence oscillations for these stars are readily detectable with LC Kepler data.

A preliminary analysis of all KOIs in the candidate catalog using Q1–Q10 LC data yields only one detection of giant-like oscillations in a KOI classified as a main-sequence star: KOI-2548. However, there is a nearby foreground star approximately 4 mag brighter that has the same oscillation pattern as KOI-2548 and is classified as a giant in the KIC. High-resolution spectroscopy of KOI-2548 acquired with the HIRES spectrograph on the Keck telescope is consistent with a dwarf. The analysis confirms that the majority of KOI host stars are on or near the main sequence.

Note that there may be giants that have escaped detection due to close binary interactions suppressing the oscillations (Derekas et al. 2011), but we suspect the number of such systems in the KOI catalog to be small. For stars with T_eff ≲ 4000 K, the oscillation periods of potential giants would be too long to be sufficiently resolved with the amount of data currently available. Hence, the small number of KOIs in this temperature regime, particularly those with Kp <14, should be viewed with caution. An in-depth asteroseismic study of KOIs using both short-cadence and long-cadence data will be presented in forthcoming papers.

The planet radius, semimajor axis, and equilibrium temperature of each planet candidate are computed using the estimated stellar properties and the parameters returned by the light curve modeling described in Section 5.1. Planet radius (Column 3 of Table 5) is the product of the reduced radius, R_P/R_⋆ (Column 11 in Table 4), and the stellar radius in Column 9 of Table 3. Planet radii are given in units of Earth radii.

Table 4. Planet Candidate Characteristics: Light Curve Modeling

KOI	tdur	Depth	S/Na	T0b	σT0	Periodb	σP	d/R⋆c	$\sigma _{d/R_\star }$	RP/R⋆	$\sigma _{R_{\rm P}/R_\star }$	bd	σb	χ2
	(hr)	(ppm)		(days)	(days)	(days)	(days)
5.02	3.6882	20	8.5	66.36690	0.01456	7.0518564	0.0002848	9.797375	0.558680	0.00428	0.00038	0.74970	0.29480	1.4
41.02	4.4764	76	36.8	66.17580	0.00321	6.8870994	0.0000617	6.177925	0.120880	0.00918	0.00016	0.86500	0.10100	1.2
41.03	6.1426	92	23.2	86.98394	0.00667	35.3331429	0.0006257	18.376942	0.359570	0.01042	0.00030	0.92010	0.09750	1.2
46.02	3.7909	58	9.7	65.51465	0.01139	6.0290779	0.0001918	6.892009	0.118930	0.00799	0.00059	0.83650	0.22630	1.2
70.05	3.6029	99	18.4	68.20094	0.00566	19.5778928	0.0002980	28.131385	3.635640	0.00998	0.00039	0.74920	0.23260	1.1
...	...	...	...	...	...	...	...	...	...	...	...	...	...	...
1612.01	1.2288	30	12.2	65.68197	0.00209	2.4649988	0.0000198	11.917724	4.152810	0.00531	0.00025	0.63910	0.50290	1.3
1613.01	4.2342	78	22.0	74.08600	0.00466	15.8662120	0.0001989	18.014821	4.615850	0.00933	0.00064	0.78970	0.35850	1.0
1615.01	1.7195	108	31.9	65.23335	0.00161	1.3406380	0.0000107	4.967419	1.056310	0.00961	0.00024	0.58290	0.28050	1.6
1616.01	2.4240	147	23.0	72.41072	0.00363	13.9328148	0.0001269	41.632554	20.251940	0.01117	0.00119	0.35180	0.91970	1.2
1618.01	3.5409	29	19.0	65.25658	0.00455	2.3643203	0.0000365	3.150859	0.318180	0.00547	0.00020	0.81200	0.17720	1.2

Notes. Invalid and/or missing data are given values of −99. Zero denotes a value smaller than the recorded precision. ^aS/N of the phase-folded transit signal computed from modeling of Quarters 1–8 data. ^bBased on a linear fit to all observed transits. Periods estimated from the duration of a single transit and knowledge of the stellar radius are rounded to the nearest integer and multiplied by −1. ^cTo first order, this parameter is equivalent to the ratio of the planet–star separation (at the time of transit) to the stellar radius. In the case of a zero-eccentricity orbit, it is equivalent to the reduced semimajor axis, a/R_⋆. ^dNote that there is a strong covariance between b and d/R_⋆.

Only a portion of this table is shown here to demonstrate its form and content. A machine-readable version of the full table is available.

Download table as:  DataTypeset image

The semimajor axis provided in Column 4 of Table 5 is derived from Newton's generalization of Kepler's third law given the orbital period (Column 2) and the stellar mass (Column 10 of Table 3). The stellar mass is derived directly from the surface gravity and stellar radius (Columns 8 and 9 of Table 3). Although the parameter d/R_⋆ (Column 9) is related to the reduced semimajor axis (a/R_⋆) as described in Section 5.1, the two are only equivalent for the case of a circular orbit.

The equilibrium temperature, T_eq (Column 5 of Table 5) is the temperature at which the incident stellar flux balances the thermal radiation. It is derived by assuming that the planet and star act as gray bodies in equilibrium and that the heat is evenly distributed from the day to night side of the planet (e.g., a planet with an atmosphere or a planet with rotation period shorter than the orbital period):

$$\begin{equation} T_{\rm eq}=T_{\rm eff}(R_\star /2a)^{1/2} [f(1-A_{\rm B})]^{1/4}, \end{equation} \tag{ 3 }$$

where T_eff and R_⋆ are the effective temperature and radius of the host star, the planet at distance a with a Bond albedo of A_B. The factor, f, acts as a proxy for atmospheric thermal circulation where f = 1 (assumed here) indicates full thermal circulation. The Bond albedo, A_B, is the fraction of total power incident on a body scattered back into space, which we assume to be 30%.

Section 4 describes the vetting procedures and the many metrics used to eliminate false positives. One of the statistics is a measure of the significance of odd/even depth differences. Aside from identifying eclipsing binaries, the metric can also be used as a warning that a low-S/N planetary period is a factor of two too low (as for KOI-730.03 and 191.04; Lissauer et al. 2011b). As an additional step outside the general procedures, we constructed a similar statistic for comparing the depths of every third or fourth transit. In this manner, the period of KOI-1445.02 was revised to be a factor of three larger than initial estimates (54 days compared to 18 days as determined the pipeline). Table 4 contains the updated period value. In this exercise we identified a handful of other planet candidates with low S/N, primarily based on one or two transits each,³³ with ephemerides predicting transits that are not detected. When only two transits are seen, it is possible that each is associated with a distinct planet candidate that only transits once, or, alternatively, that they are the primary and secondary eclipse of an eccentric, long-period, blended eclipsing binary. In some cases, there may be an alternative orbital period that phases the observed transits with a gap in the data take. More observations are required to determine a unique solution. These period aliases show that while the catalog is generally of high reliability, additional analyses on the light curves of low-S/N transits may generate revisions.

We compare the observed gain in the numbers of candidates to that expected from an increase in sensitivity afforded by three additional months of data. To do so, we develop a simplified model for the expected detection of transiting planets with Kepler data following Burke et al. (2006) whereby the detection probability is separated into two independent terms: probability for detection from a sensitivity (S/N) standpoint and probability for geometric alignment to transit.

The first term determines whether the data have sufficient photometric precision and number of observed transit events for detection above the σ_thresh = 7.1 significance threshold adopted by the mission (Jenkins 2002). We estimate the total S/N of a transit sequence of depth, Δ, and orbital period, P_orb as

$$\begin{equation} {{\rm S/N}}=\frac{\Delta }{\sigma _{\rm cdpp}}\sqrt{N_{\rm events}}, \end{equation} \tag{ 4 }$$

where σ_cdpp is the observed photometric noise for a given star on a timescale comparable to the transit duration. The duration is computed for a given stellar mass and assuming a circular orbit with period, P_orb, and an impact parameter b = (π/4) corresponding to the expectation value for an isotropic distribution of angular momentum vectors. The TPS pipeline algorithm measures σ_cdpp over a grid of transit durations from 1.5 to 15.0 hr (Jenkins 2002; Christiansen et al. 2012), and we interpolate to determine the photometric noise at the estimated duration expected. The expected number of transit events is given by

$$\begin{equation} N_{\rm events}=\eta \frac{T_{\rm obs}}{P_{\rm orb}}, \end{equation} \tag{ 5 }$$

where η = 0.95 is the duty cycle of the photometric time series and T_obs is the duration of the observations.

The total S/N of a transit event of a given depth, duration, and period associated with a star of specified characteristics is expressed as a detection probability. Since the detection statistic has unit variance (Jenkins 2002), the detection probability is the cumulative Gaussian distribution above σ_thresh = 7.1:

$$\begin{equation} P_{{\rm S/N}}=\frac{1}{2}+\frac{1}{2}\mathop {\mathrm{erf}}\nolimits {\left[\frac{({\rm S/N}-\sigma _{\rm thresh})}{\sqrt{2.0}}\right]}. \end{equation} \tag{ 6 }$$

We construct a grid of such probabilities as a function of transit depth and orbital period. Figure 5 shows a sample grid for a typical Kp = 12 magnitude star assuming five quarters of observations (407 days) and a 95% duty cycle. The grid is displayed as contours of equal detection probability, or "percent completeness."

Zoom In Zoom Out Reset image size

The TPS algorithm as implemented requires a minimum of three transit events for detection. To account for this in the detection probability we model the decrease in sensitivity toward longer P_orb by determining the P_{orb, 3} when N_events = 3 and P_{orb, 2} when N_events = 2. The detection probability then linearly decreases from 1.0 to 0.0 probability over the range P_{orb, 3} ⩽ P_orb ⩽ P_{orb, 2}. This modeling of requiring three transits in the detection probability results in the apparent "upturn" of the detection probability contours at the longest orbital periods in Figure 5.

The product of the detection probability and the probability of geometric alignment yields the total detection probability. The geometric alignment probability is proportional to the ratio of the stellar radius and planet–star separation normalized to 0.46% for an Earth–Sun analog. Figure 6 displays (logarithmic) completeness contours for the same Kp = 12 star after including the geometric alignment probability. From left to right, the contours are 5.0%, 2.0%, 1.0%, 0.5%, and 0.1%, or 0.7, 0.3, 0.0, −0.3, and −1, respectively, on the logarithmic scale. If nature produces planets that uniformly populate the radius/period plane with circular orbits, then the planets detected by Kepler will be distributed similarly to the contours shown.

Zoom In Zoom Out Reset image size

Instead of computing a grid of total detection probabilities for each of the >150,000 stars in the parent sample, we devise a representative sample. Stars observed by Kepler, in the range 4.0 < log g < 4.9 and R_⋆ < 1.4, are sorted into Kp = 0.25 mag bins, excluding those that do not have stellar classifications in the KIC. This results in 145,728 targets with 10.75 < Kp < 17.75. Table 6 lists the number of stars in each magnitude bin together with the median stellar radius for that bin. The associated log g and T_eff are interpolated of Tables 15.7 and 15.8 in Cox (2000). The photometric noise taken as the 30th percentile of the 6 hr CDPP values³⁴ of the targets in the magnitude bin. The stellar properties associated with the Kp = 12 mag bin are those used for the calculations that produced the results displayed in Figures 5 and 6. Though the majority of Kepler targets are G-type stars on or near the main sequence (Batalha et al. 2010b), it is evident from Table 6 that the median star type depends on magnitude. We assume that all stars in the magnitude bin can be represented by the properties tabulated. Though insufficient for computing Kepler's detection efficiency in an absolute sense, this assumption is a useful simplification for exploring the expected planet yield in a relative sense.

The total planet yield for a given magnitude bin is computed by taking the product of the alignment-corrected detection probability and the number of targets in that magnitude bin. Summing these results for all magnitudes yields the total expected planet yield for a given period/radius combination assuming all stars have such a planet. We can then integrate over period and radius to obtain the planet yields in specific bins. These sums are weighted by the power-law distributions defined in Howard et al. (2012) to account for the fact that certain areas of the radius/period domain are more heavily populated with planets than others. Table 7 contains the expected versus observed gains in the numbers of candidates (expressed as a ratio, N_Q6/N_Q5, where N_Q5 is the number of candidates in the B11 catalog and N_Q6 is the total number of candidates, old plus new) for various radius/period bins. In every case, the observed gains are significantly larger than the predicted gains. Even in areas of the parameter space where we predict no gains (e.g., Neptune-size planets with periods shorter than 125 days where N_Q6/N_Q5 = 1), we observe appreciably more candidates.

This simplified exercise is meant to serve as a cautionary example for those performing statistical studies of planet occurrence rates as it demonstrates the incompleteness of the B11 catalog.³⁵ The gains we observe cannot be explained by the longer observation window. More likely, the current batch of new planet candidates has benefitted from the growing sophistication of pipeline software, the most important of which is the multi-quarter functionality afforded by TPS for the first time in SOC 7.0. With multi-quarter functionality, data from different quarters are stitched together within pipeline modules before whitening and combing for transits. The analysis leading to the B11 catalog employed non-pipeline tools to do similar tasks via a modified BLS detection method. As such, the analysis did not make full use of the pipeline modules that carefully treat inter-quarter boundaries and perform the pre-search whitening of the light curves as described in Section 3. The absence of these modules has a significant impact on the detectability of shallow transits.

The SOC 7.0 pipeline also contains improvements to the module performing systematic-error correction (pre-search data correction, or PDC). Thermal transients occurring after monthly Ka-band downlinks and safe mode events introduce systematic errors in the light curves that are treated in the current pipeline. Left uncorrected, such transients have a negative impact of the detectability of transit events. Figure 5 of DRN 5 (KSCI-19045-001) at MAST illustrates an example of a thermal transient after a safe mode event that is properly corrected by the new version of PDC. The BLS analysis applied to multi-quarter data in B11 circumvented this by masking out data points associated with thermal transients (12 events totaling 54.7 days). Collectively, these pipeline improvements lead to increased detection efficiency compared to previous efforts.

An important question is the degree to which the current catalog also suffers from incompleteness. Figure 7 shows the distribution of Q1–Q6 detection statistics (MES) for the cumulative list of planet candidates (new plus old). A power-law increase in the planet radius distribution toward smaller radii translates to a power-law increase the MES distribution (toward smaller MES values). The turnover at an MES of 9.7 suggests that incompleteness is likely to still be an issue. Planned pipeline upgrades already underway are expected to yield another significant improvement. For example, there are known limitations to PDC whereby astrophysical features that are distorted both in amplitude and frequency thereby precluding the detection of shallow planet transits. DRN 4 and 5 provide numerous examples, though Figure 9 of DRN 5 is especially relevant. Future pipeline improvements to mitigate these distortions are already under development.

Zoom In Zoom Out Reset image size

There is a concerted effort to produce more uniform vetting procedures with each planet catalog release and to develop the tools that will allow for a quantitative assessment of catalog reliability, pipeline completeness, and planet occurrence rates. Such analyses are in progress.

The first multiple transiting planet systems were identified in Kepler data within the first year of science operations (Borucki et al. 2011a; Steffen et al. 2010; Holman et al. 2010). One third of the 1235 candidates in the B11 catalog, or 17% of the 997 unique stars are host to multiple planet candidates. Evidence of dynamical interactions in the form of transit timing variations (TTVs) has been identified for approximately 100 candidates (Ford et al. 2012), and dynamical modeling of those variations has successfully produced mass determinations for five of the planets associated with Kepler-11 (Lissauer et al. 2011a). TTVs have proven useful for confirming and characterizing planets in multiple systems. Moreover, statistical analysis of multiples indicates that nearly all (≳98%) are bona fide planets (Lissauer et al. 2012). That is, the false-positive rate for candidates transiting stars with multiple transiting planet candidates is significantly smaller than for single planet candidates. As such, the "multis" are a particularly reliable sample for future study.

As discussed in Section 3, all of the KOIs were inspected for evidence of additional transiting planet candidates. The search yielded 302 new candidates in multiple systems. Their properties are included in Tables 4 and 5. Combining the new candidates with the B11 catalog, we have 2338 candidates associated with 1797 unique stars. Three hundred sixty-nine of the 1797 stars host multiple candidates. Nine hundred ten of the 2338 candidates are part of multiple systems. The fraction of KOI host stars with multiple candidates has risen from 17% to 20%.

In Figure 8, we return to the cumulative list of candidates in the radius versus period plane, this time distinguishing candidates belonging to one (black), two (green), three (blue), four (yellow), five (cyan), and six (red) planet systems. The plot corroborates the observation made by Latham et al. (2011) in studying the multis in the B11 catalog: that there is a paucity of short-period giant planets in multiple systems. There is only one apparent exception to this—the green point near R_P = 17 R_⊕, P = 3 days (KOI-338.02). This is a new candidate associated with KOI-338, the first having been reported in the B11 catalog. Since that publication, spectroscopic observations have been acquired and subjected to spectral synthesis. The revised stellar parameters are T_eff = 4104 K and log g = 1.87 compared to T_eff = 4910 K and log g = 4.18 in the KIC. This results in a significant change in the estimated planet radius for KOI-338.01 reported in B11. Re-evaluation of KOI-338.01 with the new stellar radius consistent with the lower surface gravity yields R_P = 40 R_⊕. This transiting (eclipsing) object is too large to be planetary and is off the scale in Figure 8. The new candidate, KOI-338.02, is approximately Jupiter-size. A second spectroscopic observation from the same instrument (the TRES spectrograph on the Tillinghast Reflector at Whipple Observatory) was acquired, and yields a surface gravity that is more consistent with the KIC value. Additional observations are required to resolve the stellar properties. This target is discussed in more depth by Fabrycky et al. (2012).

Zoom In Zoom Out Reset image size

We note that Figure 1 of Latham et al. (2011) contains other short-period giant planet candidates: the candidates associated with KOI-961. This high proper M dwarf from the proper motion catalog of Lepine & Shara (2005) (LSPM J1928 + 4437) is unclassified in the KIC. In B11, the stellar radius was estimated by inferring an effective temperature from the J − K color and assuming a main-sequence luminosity class. Improved stellar properties and light curve modeling of Muirhead et al. (2012b) result in a considerable decrease in the star and planet radii thereby removing these candidates from the upper left corner of Figure 8 and strengthening the case for an observed paucity of short-period giant planets in multiple systems.

Application of the statistical arguments presented by Lissauer et al. (2012) to the current population of multis implies that there are over 880 (98% of 898) bona fide planets in this sample alone. A comprehensive study of the architecture of these multis is presented by Fabrycky et al. (2012).

Borucki et al. (2011b) identified 54 transiting planet candidates in the HZ, defined by an equilibrium temperature, T_eq (see Section 5.3), between the freezing and boiling point of water (273–373 K) at standard pressure. As pointed out by Kasting (2011a), this definition fails to take into consideration the warming effect of an atmosphere, thereby rendering many of the candidates too hot for habitability even under the most liberal assumptions about their climatic conditions. As the roster of small planets at long orbital periods grows, so has the attention paid to the issue of habitability. Recent modeling efforts suggest that the equilibrium temperature at the inner edge of the HZ might be closer to 270 K (Selsis et al. 2007), corresponding to rapid water loss via H escape or a runaway greenhouse effect, depending on the surface water content. The outer edge depends on the fractional cloud coverage and ranges from 175 K to 200 K (Kaltenegger & Sasselov 2011). Others have proposed defining the HZ in terms of insolation (the amount of stellar flux incident on the planet surface) in order to remove the built-in assumptions (e.g., Bond albedo) required to compute the equilibrium temperature (Domagal-Goldman 2012).

The star and planet properties provided in Tables 3, 4, and 5 can be used to compute the insolation and/or equilibrium temperature under different assumptions for the purpose of assessing questions of habitability on a case-by-case basis. Here, we use equilibrium temperature as defined in Section 5.3 to examine the population of planet candidates likely to be in or near the HZ. Figure 9 shows planet radius versus equilibrium temperature for the entire sample of planet candidates. For reference, we include a vertical dashed line (middle) to mark the equilibrium temperature of the Earth computed under the same set of assumptions. With each new catalog (blue to red to yellow points), we see a clear trend toward Earth-size planets at Earth's equilibrium temperature (i.e., toward the bottom left hand corner of the diagram).

Zoom In Zoom Out Reset image size

Figure 10 displays the same for the range 180 K < T_eq < 310 K. The dotted vertical lines (far left and far right) mark the (generous) HZ boundaries (185–303 K) proposed by Kasting (2011b). The intermediate dashed vertical line marks the equilibrium temperature of the Earth under the same set of assumptions. There are 46 candidates in this temperature range (compared to 22 in the B11 catalog). Nine are super-Earth-size (1.25 R_⊕ ⩽ R_P < 2 R_⊕) and one is Earth-size (R_P < 1.25 R_⊕). Table 8 lists the properties for the 24 new candidates in this temperature range that are plotted in Figure 10. Note that candidates with only one transit event in the Q1–Q6 period are excluded from this list and Figures 9 and 10.

Zoom In Zoom Out Reset image size

We have paid special attention to candidates that are in this temperature range and are also near Earth-size (see, for example, the discussion of KOI-326, KOI-364, and KOI-1026 in the Appendix). Figure 11 shows the relative flux time series of KOI-2124.01, the smallest viable candidate in this temperature range.

Zoom In Zoom Out Reset image size

Table 8 contains two sets of stellar parameters. They are identical when spectroscopic values are available (as indicated by the flag, $f_{T_{\rm eff}}$, in Table 3). They differ where KIC values were updated using a parameter search in the Yonsei–Yale stellar evolution models as described in Section 5.2. For the sample of stars listed in Table 8, the updates almost always lead to smaller stellar radii (and, hence, smaller and cooler planet candidates). Improved stellar characterization is required for a more reliable determination of the candidate location relative to the HZ.

PlanetHunters.org is a citizen science tool (Fischer et al. 2012), based on the Zooniverse platform (Lintott et al. 2008), that enables the search for transit events in the public Kepler data. The site serves up plots of Kepler light curves broken into 30 day segments, and, through a sequence of queries, leads the user through a high-level classification that sorts light curves by their qualitative properties, or appearance. Discerning eyes flag events that resemble transits, and the goal is to have every light curve examined by at least five independent users. Since its launch in 2010 December, over 10 million classifications have been made by over 100,000 users, underscoring the remarkable enthusiasm of the general public. The site affords one not only the opportunity to experience the scientific method but also the possibility of experiencing the gratification of discovery. With its power in numbers, the citizen science project is a welcome complement to the automated detection algorithms in Kepler's software pipeline.

The Planet Hunters science team combines the results from the multiple classifications for each 30 day light curve segment to identify potential planet candidates within the publicly released Kepler data. Transit-like events flagged by the public are assessed by the Planet Hunters science team. The Kepler project office then assists in vetting further for false positives. This process resulted in the identification of four potential planet candidates in the Q1 public data (Fischer et al. 2012), associated with stars KIC 10905746, KIC 8242434, KIC 6185331, and KIC 11820830, two of which (KIC 10905746 and KIC 6185331) were deemed viable candidates after inspection of vetting metrics. These four candidates were assigned KOI numbers (1725.01, 1726.01, 1727.01, and 1728.01, respectively), and blindly re-assessed together with the candidates identified here using Q1–Q6 data and the associated vetting metrics. All but 1728.01 survived as viable planet candidates. Their properties are listed in Tables 4 and 5. KOI-1728.01 was rejected due to the large radius of the companion (>5 R_J) and hints of ellipsoidal variations in the light curve indicative of a higher mass (i.e., stellar) companion. This target has been included in the Eclipsing Binary Catalog (Prša et al. 2011; Slawson et al. 2011).

More recently, inspection of the Q1 and Q2 public data by Planet Hunters led to another set of viable planet candidates associated with the stars KIC 4552729 and KIC 10005758 (Lintott et al. 2012). KIC 10005758 had already been assigned a KOI number due to the identification of a different transit event at a shorter period (KOI-1783.01). The new, longer period candidate identified by the Planet Hunters team has been assigned the KOI number 1783.02 while the candidate associated with KIC 4552729 has been assigned KOI number 2691.01. Detection statistics on KOI-1783.02 and KOI-2691.01 are not identified by the Q1–Q6 pipeline run due to (1) the long orbital period (KOI-1783.02) and (2) systematic noise sources that precluded identification of the correct orbital period (2691.01). The events are readily identified in a Q1–Q8 pipeline run that more recently became available, and all vetting statistics indicate that both are strong candidates. To maintain sample uniformity, we do not include these candidates in the tables presented here. They will, however, be included in future catalogs.

We note that nine of the candidates presented in Table 4 (1787.01, 1828.01, 1858.01, 1790.01, 1808.01, 1830.01, 1613.01, 1557.02, 1930.04) were independently identified by Planet Hunters (KIC 5864975, KIC 11875734, KIC 8160953, KIC 6504954, KIC 7761918, KIC 3326377, KIC 6268648, KIC 5371776, and KIC 5511081, respectively) as described by Lintott et al. (2012). This further illustrates the potential for contributions from the citizen science community.

The vetting statistics employed here and applied to the Planet Hunters candidates are quickly becoming an integrated part of a mature software pipeline. The Kepler team has worked to fine-tune these vetting metrics so that they are reliable and easily interpreted. The objective is to eventually make them publicly available so that users interested in identifying transits can also perform the vetting that is such an integral part of identifying viable planet candidates.