A PHOTOMETRIC SURVEY FOR VARIABLES AND TRANSITS IN THE FIELD OF PRAESEPE WITH THE KILODEGREE EXTREMELY LITTLE TELESCOPE

After initial image processing with ISIS, we found that a large number of variable-star candidates had double-valued light curves, with the amplitude of many sinusoidal light curves being larger on some nights and smaller on others. This variation in amplitude was correlated in time, and appeared to apply mostly to stars in a horizontal zone across the upper part of the chip. We traced the origin of this effect to changing FWHM of the stellar images across our field from night to night. The FWHM does not change significantly horizontally across the field, but it has significant structure in the Y-direction, with the FWHM being large (∼2.8 pixels) at the top and bottom of the chip, and decreasing linearly toward the middle of the chip in a V-shape, with a minimum FWHM of ∼2.0 pixels at about one third of the distance from the top of the chip.

Had the size and shape of this pattern remained constant throughout the observations, it would not have presented a major problem, since ISIS is able to work with a changing FWHM across the field. However, the vertical-axis position of the FWHM minimum changed significantly between different nights, ranging from the middle of the chip to the top edge. That is, the bottom of the V-shape of the FWHM distribution moved up and down the chip over different nights. The position of the FWHM minimum correlated with the change in amplitude of the light curves.

The reason for the double-valued light curves relates to the way ISIS works. ISIS requires a reference image that has the smallest FWHM of all images in the set, since it convolves the reference image with the kernel of each of the individual images. With the changing shape of the FWHM pattern in our data, no single image or set of images has a smaller FWHM across the entire field for all the nights. Any given choice for a reference image will contain a horizontal region for which there are other nights where the stars have smaller FWHM values. Regions with smaller point-spread functions (PSFs) than the reference image on certain nights require deconvolution of the PSFs (i.e. adjusting the stellar images from the reference frame to have smaller PSFs rather than larger). This creates a problem since ISIS works to convolve an image, even with a varying degree across the field, but ISIS is not equipped to correctly deconvolve an image.

We have not been able to absolutely determine the origin of the time-varying FWHM pattern. We believe it could be related to temperature changes from night to night, which affected the 200 mm lens we used for these observations. Since the problem was not detected until after we stopped using the lens, we were unable to test this hypothesis. Instead, our objective is to mitigate the impact of this effect as much as possible and get the best measurements we can out of the data set.

The procedure we adopted is as follows. We first register the images to the same coordinate system, then divide them into four horizontal sections. For each image section we identify a different reference image that has the smallest FWHM out of all images in that section. We then treat the different sections as if they were four separate images, and run each section though the data reduction pipeline with their own reference images to obtain light curves (with an additional subdivision step described in Section 4.2). We also convolve all of the images (but not the reference image) with a Gaussian smoothing function (using a Gaussian with σ = 0.6 pixels) to slightly broaden the PSFs, and thus ensure that the reference image has a smaller FWHM. With this procedure, we are able to eliminate most of the effects of the time-varying FWHM. We are not able to completely eliminate the effect, which can be seen in the slightly double-valued light curves of some variables (e.g., see plots of KP300133 and EF Cnc in Figure 8, where the scatter in the light curve is more pronounced at the fainter phase of the periodic cycles. Also see the excessive scatter in the light curves of KP407282 and KP118899 in Figure 10, with significant scatter upward when the stars brighten and downward when the stars dim).

We perform image subtraction on each section of the 3137 images with ISIS, as described in Section 4.1. We use a feature of ISIS to further subdivide each section into a grid of subfields, each of which can take on different values for the parameters that are used to convolve the reference image with the kernel for image subtraction. This step is particularly advantageous for large fields of view, in which cloud patterns can be smaller than the size of the field. We subdivide each section into grids from 1 × 5 to 5 × 5, depending on the size of the section, and proceed with the image subtraction.

We use DAOPHOT (Stetson 1987) to identify all of the objects in the field of the reference image of each section, yielding a list of 69,337 stars on all four sections. Stars along the image edges have particularly bad light curves, and so we remove all stars within 50 pixels of the edge from further analysis, leaving 66,638 stars for which we generate light curves. The ISIS photometry program calculates the flux from a star and its error in each image. In some situations, such as when the star is located at the edge of the chip, or on the edge of one of the of subregions of the subtracted image, the reported flux and error values are not indicative of the true flux. We clean such points from the data by removing data points where the reported flux or error is unphysically high or low. We also remove the two highest and lowest flux measurements from each lightcurve. Removing such a small number of points should not affect detection of variability or transits, but it does help remove spurious points and reduces the number of false positives when searching for variable sources.

We use the Astrometrix⁵ program to derive astrometric solutions for the reference image, using the Tycho-2 catalog (Høg et al. 2000) to select reference stars. Because of high-order distortions in the corners of the field, we first subdivide the full images into 25 subsections and find separate astrometric solutions for each subimage. The astrometry is good to within an arcsecond, or ∼0.1 pixel.

We match our data set to two catalogs. We first match our stars to the Two Micron All Sky Survey (2MASS) catalog (Skrutskie 2006), using a search radius of 9.5 arcseconds or about one KELT pixel. We find that 58,620 out of 66,638 of our KELT stars are in the 2MASS catalog, with 1559 of them matching to more than one 2MASS source.

We also match our star catalog to known members of the Praesepe cluster. We compiled a catalog from the WebDA website, identifying 832 likely cluster members. After matching to the KELT data using a search radius of 9.5 arcseconds, we find matches to 333 Praesepe member stars. However, many of the stars in the WebDA database are too faint for KELT to detect. If we consider only the 210 WebDA sources with known V magnitudes of 6.8 < V < 16.4, we find matches to 147 stars, although the brighter stars are mostly saturated and unusable in the KELT images.

Our goal for KELT is to obtain highly precise relative photometry, so we do not attempt to achieve extremely precise absolute photometry. The KELT bandpass is an approximate wide R band. We define a KELT magnitude R_K to which we calibrate our observations, such that

$$\begin{equation} R_K \equiv -2.5\,{\rm log(ADU\, s^{-1})} + R_{K,0}\, \end{equation} \tag{ 1 }$$

where the instrumental ADU s⁻¹ is measured using aperture photometry with IRAF, R_K is defined as an approximate KELT R magnitude, and R_K,0 is the zero-point. We find that the R_K magnitudes are within a few tenths of a magnitude of standard R band photometry, with the uncertainty dominated by the color term. Since we do not have V − I colors for all our stars, we quote R_K, which can be considered equivalent to Johnson R, modulo a color term defined by

$$\begin{equation} V = R_K + C_{VI}(V-I), \end{equation} \tag{ 2 }$$

where C_VI is the (V − I) color coefficient, and V/I are in the Johnson/Cousins system. Since we do not have previously measured R magnitudes of large numbers of stars in our fields in our magnitude range, we relate R_K to known magnitudes by matching stars from our observations to the Hipparcos catalog, selecting only stars with measured V and I colors in Hipparcos. We take the mean instrumental magnitude from a set of high-quality images, and match the known magnitudes to the mean instrumental magnitudes, using Equations (1) and (2).

We find a magnitude zero-point of R_K,0 = 16.38 ± 0.06 mag, and C_VI = 0.55 ± 0.2, and that a fiducial R = 10 mag star at the field center with (V − I) = 0 has a flux of 356 counts per second. Since the Hipparcos stars we use to calibrate our data have (V − I) colors mostly between 0 and 1, we expect our calibrations to be less accurate for redder stars. Our observed magnitudes are listed in R_K, which again is equivalent to Johnson R modulo a color term which is typically 0.2 magnitudes, but can range from −0.3 magnitudes for very blue stars to 0.8 magnitudes for very red stars. See Pepper et al. (2007) for additional details about the calibration process.

One feature of ISIS that has been noted by others is that the formal reported errors tend to be underestimated for brighter stars. Since the errors on individual points are important in the variable selection process, we rescale the errors following the procedure of Kaluzny et al. (1998) and Hartman et al. (2004). We first compute the reduced χ² for every star,

$$\begin{equation} \chi ^2/N_{\rm dof} = \frac{1}{N-1} \sum _{k=1}^{N} \frac{(m_k - \mu )^2}{\sigma _k^2}, \end{equation} \tag{ 3 }$$

where the sum is over N observations, m_k is the instrumental magnitude with error σ_k, and μ is the weighted-average instrumental magnitude. We then plot χ²/N_dof versus magnitude and fit a curve to the bottom edge of the heaviest concentration of points. We then multiply the formal errors by the square root of the function of that curve, so that the least variable light curves in our data set have χ²/N_dof close to 1 for all magnitudes.

A common method for describing the photometric precision of transit searches is to plot the magnitude root-mean-squared (rms) values of all the light curves as a function of magnitude. Because of the FWHM changes described in Section 4.1, the long-term photometric precision has been degraded. However, since the intranight FWHM pattern appears to be stable, we plot the overall rms and the rms for one of the nights for all 66,638 stars in Figure 1. Panel (a) of Figure 1 shows the rms plot for a single night of data, while panel (b) shows the rms plot for the full 34 nights of data. The dashed horizontal lines show the 2% and 1% rms limits, which generally define the required sensitivity for detecting Hot Jupiter transits. Two features stand out in these plots. The first is that there are significantly fewer stars with rms <1% in plot (b) than plot (a), although there are about the same number of stars in each plot with rms <2%. This behavior shows that the inter-night systematics, of which we believe the FWHM changes to be the most significant, become most important at the sub-1% level. The second feature to note is the greater rms of stars brighter than R_K ≈ 9. The light curves of the brightest stars in our sample are dominated by systematic saturation and/or nonlinearity effects that are present during a single night but are more severe over the entire run.

Zoom In Zoom Out Reset image size

For each of the 66,638 stars, we compute the Stetson J statistic, using the implementation from Kaluzny et al. (1998). This statistic identifies coherent variable stars by selecting for photometric variations that are correlated in time. After visually examining a number of light curves, we define a cutoff of J = 0.7 to select variable candidates, deliberately choosing a liberal cut on J since we have several more tests to filter out non-variables. We eliminate those stars with R_K < 9, since the light curves of the brightest stars in our data set are dominated by systematics due to saturation effects. We also eliminate any star that is less than 10 pixels away from stars with R_K < 9, since extended wings and bleed trails from the bright stars create false variability. We finally remove any candidate that is less than 13 pixels away from a variable candidate with a higher rms in flux. This eliminates false positives due to constant stars close to true variables. After these cuts we are left with 3430 candidate variable stars.

Among the 3430 variable candidates that pass the cuts described above, most of the stars are clustered around a few of the subsection boundaries, mostly between subsections along the field edges where the fitting parameters vary most, and adjacent subsections closer to the field interior. We assume that all spatial clustering of variables is due to artifacts in the reduction process, and we suspect that the reason for the clustering has to do with how large changes in PSF size and shape across the field affect ISIS.

As described in Section 3, there was an intranight drift in the telescope pointing. One effect of this drift is that stars undergo slight changes in the PSF shape and size during each night. When ISIS convolves the reference image, it uses different parameters for each subsection (see Section 4.2). Subsections at the edges of the field experience the strongest optical distortions due to the wide field, and ISIS has the most difficulty fitting the convolution parameters in those areas. At the edges of those subsections the assumptions used to compute the fitting parameters break down. For stars in those areas, the intranight drift means that any consistently inadequate convolution will show up as photometric variability on timescales comparable to the drift rate.

To eliminate false variability due to this effect, we want to remove any candidates that appear in areas with high spatial clustering. However, these areas are not precisely defined—they result from the combination of the ISIS subsection grid, the direction and speed of the drift, and the nature of the optical distortions. We therefore devised an algorithm to remove from our list any stars that are in clustered areas. We divide the entire field into boxes 100 pixels on a side, and count the number of variable candidates in each box. We perform this process four times, with each grid offset from the previous one by 20 pixels in both the X and Y directions. We thus have four staggered grids with which to measure the clustering of the variable candidates. We classify all candidates that appear in a box with more than two other candidates in any of the grids as spurious and eliminate them from our sample, leaving 1101 variable star candidates.

There are some stars that pass the cut on J that do not exhibit periodic variability. Many are long-period variables (LPVs) that show monotonically increasing or decreasing brightness during our campaign, some of which may vary periodically but on timescales longer than our campaign. We identify such objects and remove them from our later analysis, which focuses on identifying periodic variables.

To identify LPVs, we use the method described in Section 4.3 of Hartman et al. (2004), in which a star is defined as an LPV when a parabola fits the lightcurve much better than a horizontal line. We fit a parabola to the 1101 remaining variable candidates and calculate the χ²/N_dof for the fit, which we call χ²_N−3, along with the χ²/N_dof for the fit to the mean, χ²_N−1 (i.e. a fit to a horizontal line). There are 52 stars for which χ²_N−3/χ²_N−1 > 0.8, and therefore identified as candidate LPVs, which we eliminate from our remaining variable candidate list. We show the light curves of three of our LPVs in Figure 2.

Zoom In Zoom Out Reset image size

A much smaller fraction of our stars are LPVs (52 out of 66,638 stars, or 0.078%), than the 1535 LPVs out of 98,000 stars, or 1.6%, found by Hartman et al. (2004). We attribute this difference partly to our greater observational time baseline (74 nights versus 30 nights), since we end up classifying stars with periodic behavior in that range as regular periodic variables rather than as LPVs. Also, the Praesepe field is located well out of the Galactic plane, and we therefore expect many fewer background giants than in the field that was observed by Hartman et al. (2004). Since background giants are one of the main types of LPVs (e.g. Mira variables), we would therefore expect fewer LPVs when observing at higher galactic latitudes.

We use the period search algorithm of Schwarzenberg-Czerny (1996) to select periodic variables, adopting the implementation by J. Devor. The Schwarzenberg-Czerny algorithm reports a periodicity likelihood statistic analysis of variance (AoV), and the Devor method analyzes that information to estimate σ_AoV, a measure of the confidence of the light curve's periodicity. We apply the algorithm to the 1049 stars remaining in our catalog after removal of the LPVs.

The plot of the best-fit period P versus σ_AoV (Figure 3) shows significant aliasing effects at periods that are integer multiples or fractions of one day. We thus need to make further cuts to select the true variables from among our list of candidates. First, we construct three histograms of the 1049 periods in log(P). Each histogram describes the same set of periods, but the edges of the bins are shifted in log(P) from the other two histograms by 1/3 of the width of a bin. All objects that appear in bins in any of the three histograms that have more than seven objects in them are rejected, with that number determined empirically by searching for the optimal rejection number while avoiding rejecting too many true variables. Using more than three histograms did not significantly change the set of rejected candidates. Any remaining objects with σ_AoV > 3.3 are retained. To ensure that true variables were not accidentally eliminated by the binning procedure, any stars that have σ_AoV > 4.0 are automatically retained, even if they are caught by the aliasing identification algorithm.

Zoom In Zoom Out Reset image size

After these cuts, we are left with 182 variable star candidates. We then examine the light curves of the remaining candidates by hand, and find that 70 of these are false variables, all of which slightly missed the clustering or alias filters. We also find that six of the candidates, all of which have estimated periods longer than 20 days, are in fact LPVs.

Wide-field transit surveys are susceptible to a number of systematic errors. Observing objects for long stretches of the night requires the telescope to cycle through significant changes in airmass. A wide field of view allows differential cloud patterns to complicate the process of obtaining accurate relative photometry. Temperature changes can affect the optics or detector performance. With so many sources of systematic error, it can be prohibitive to attempt to identify, measure, and compensate for all of these effects. Instead, several methods have been developed to identify all generic systematic effects in data sets of this type. The method we choose to implement is the SYSREM algorithm (Tamuz et al. 2005; Mazeh et al. 2007). In brief, this algorithm identifies and subtracts out linear trends that appear in a large portion of light curves in a given data set.

We apply SYSREM to the KELT light curves that have been reduced using the procedures described in Section 4. Since light curves with very large variations can create problems for detrending programs, and are not useful for the identification of low level systematics, we only apply SYSREM to the 15,012 stars from our data set with rms <5%. We recalculate the rms for the detrended light curves and display the results in Figure 4. For most of the light curves, detrending improves the rms by about ∼10%, although for a few stars, especially toward the bright end, the rms improves by a factor of 3 to 4.

Zoom In Zoom Out Reset image size

We suspect that the reason that detrending does not improve the rms to a greater degree is related to the FWHM problems described in Section 4.1. Algorithms such as SYSREM are designed to remove linear systematics, but it is probable that the systematic errors caused by the FWHM changes are higher order, and so are not rectifiable by linear detrending. Even so, detrending manages to improve the rms by a small amount, and in some cases works very well, as shown in Figure 5. For the periodic variables we identify with non-detrended rms below 5%, detrending improves σ_AoV on average by 10%.

Zoom In Zoom Out Reset image size

To detect transits, we apply a version of the box-fitting least-squares (BLS) transit-search algorithm (Kovacs et al. 2002) implemented Burke et al. (2006). This algorithm cycles through a range of periods and phases searching for a transit-like event and selects the one with the highest significance. For each light curve, we calculate the Δχ² between a constant flux and the best-fit transit model, the Δχ²₋ for the best-fit antitransit (a brightening rather than dimming), the fraction of Δχ² that results from a single night f, and the transit period P_t. We then examine five parameters: Δχ², Δχ²/Δχ²₋, f, P_t, and the transit depth. We also examine Digital Sky Survey (DSS) images of the stars which have much higher resolution than KELT to check whether the single KELT sources consist of blends of more than one star.

Because of the unique characteristics of this data set, we have elected not to construct a rigorous transit selection algorithm using cuts. Instead we sort the light curves based on each of the five parameters described above, and examine by eye the 100 best light curves identified by each criterion. We classify the light curves as either possible transits, variables, or neither. We find four stars with transit-like behavior that are unblended and that are identified with known 2MASS stars. We also find 38 additional variable stars that were not found with the steps described in Section 5. Of those objects, 31 had J < 0.7, which is not surprising since the transit-like signals that BLS searches for would not necessarily show the coherent variations characteristic of a variable star with a large J value. Of the seven other stars, one was eliminated by the cluster-rejection routine, and the remaining six had σ_AoV < 3.3.

There are 168 known variables within our field of view combining The General Catalog of Variable Stars (GCVS4.2; Samus & Durlevich 2004) and the New Catalog of Suspected Variable Stars (NSV; Kukarkin et al. 1982),

The magnitudes of the variables are reported in either Johnson V or U, or photographic magnitude p. The range of magnitudes for which we found variables in the KELT data using all the methods described above are 9 < R_K < 16. Out of the 168 variables, 72 have reported magnitudes outside that range, and an additional 17 are within 15 < V < 16. We only try to match known variables with reported magnitudes between 9 and 15.

We check whether we detect the remaining 79 variables in our data. We search for KELT counterparts using a matching radius of two pixels (19'') and identify 63 KELT stars that match to known variables. Of the 16 known variables we do not detect, 3 appear either saturated or blended with saturated stars in our data, and 13 do not appear to have counterparts within two pixels of the reported positions, suggesting that the reported positions, proper motions, or magnitudes may be incorrect.

We then compare our list of 208 variables with the 63 detected known variables and find 14 matches. Of the 49 known variables we do not classify as variable through our tests, 5 have counterparts that are either brighter than R_K = 9 or blended with a bright star. Another 38 are irregular, eruptive, or unspecified variables, which we would not expect to detect with our periodicity-based search algorithms. Three are eclipsing variables of unspecified type with no periods listed in the catalogs. Upon investigation of their KELT counterparts, no eclipses are seen, indicating that either our observations missed the eclipses, or that the eclipse depths were too small to appear in our data. One is listed as an RR Lyrae variable with no period given, for which the KELT lightcurve shows no periodic variation. The remaining two are eclipsing binaries which are removed in the clustering filter, but can be seen in the KELT light curves. We add those two stars to our list of variables, giving a total of 210.

We thus have KELT light curves for 16 known variables. We find that one KELT LPV is the semi-regular variable GV Cancri. The lightcurve for this object is shown in Figure 2. We find three RR Lyrae variables, CQ Cancri, AN Cancri, and EZ Cancri, the first two of which have periods reported in the catalogs that we confirm. One more variable, EF Cancri, is listed as a W UMa contact eclipsing binary, but the light curve, despite the low-quality photometry, appears to indicate that it, too, is an RR Lyrae variable.

We detect six known eclipsing binaries: EH Cancri, GW Cancri, FF Cancri, RU Cancri, NSV 04207, and NSV 04158, in addition to the two eclipsing binaries that were caught by the clustering filter: TX Cancri and RY Cancri. We confirm the reported periods of RY Cancri and TX Cancri, but we find that RU Cancri has a period of 10.0591 days rather than the reported period of 10.172988 days. The remaining variable stars do not have previously reported periods.

The star NSV 04269 is listed as a semiregular variable in the NSV catalog, but our light curve shows it to be an eclipsing binary. The star NSV 04069 is not listed as any variable type; we classify it as an eclipsing binary. Lastly, the star FR Cancri is listed as a BY Draconis variable, consistent with our light curve.

We list the properties of our variable stars in Table 1. For each star we list the KELT ID number, mean R_K magnitude, coordinates in right ascension and declination (J2000.0), JHK colors from 2MASS and 2MASS ID for those that matched to 2MASS objects, period (for non-LPVs) in days, type of variable based on our classification, and the GCVS or NSV ID and classification for those that we have identified with previously known variables.

We plot the 2MASS infrared color–magnitude diagram (CMD) for all our stars in Figures 6 and 7. In Figure 6 we show Praesepe cluster members along with all background stars. The cluster members form a coherent main sequence. The background stars form three populations in J − K color. The largest group, at J − K ∼ 0.4, consists of main-sequence stars. The group at J − K ∼ 0.6 consists of red giants, and the last group, at J − K ∼ 0.8 − 0.9, consists of nearby late-type stars, clearly overlapping the same stellar population at the low end of the Praesepe cluster.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

We plot the light curves of the periodic variables we identify with the methods described above, along with the two previously known periodic variables that missed our cuts. We classify 48 variables as pulsators, and we plot their light curves in Figures 8 and 9. We classify 104 variables as eclipsing binaries, and we plot their light curves in Figures 10–14.

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

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

The four transiting-planet candidates reported in Section 6.2 were followed up spectroscopically using the CfA Digital Speedometer (Latham 1992) on the 1.5 m Tillinghast Reflector at the F. L. Whipple Observatory atop Mount Hopkins, AZ. This instrument has been used extensively for the initial spectroscopic reconnaissance of transiting-planet candidates identified by Vulcan (Latham 2003) and by TrES and HAT (Latham 2007). Single-order echelle spectra centered at 5187 Å  were recorded using an intensified photon-counting Reticon detector with a spectral resolution of 8.5 km s⁻¹ and typical signal-to-noise ratio of 10–20 per resolution element. After rectification to intensity versus wavelength, the observed spectra were correlated against extensive grids of synthetic spectra drawn from a library calculated by Jon Morse using Kurucz (1992) model atmospheres and codes. This allowed us to estimate the effective temperature and surface gravity of the star, assuming solar metallicity, as well as the rotational and radial velocities. For the initial spectroscopic reconnaissance we normally obtain at least two spectra, so that we can look for velocity variations down to the level of about 1 km s⁻¹ for slowly-rotating solar-type stars.

In the case of KP200924 the first CfA observation revealed that this candidate has a composite spectrum. Plots of the one-dimensional correlation functions clearly show two peaks corresponding to the two stars in a double-lined spectroscopic binary, with a velocity separation of about 200 km s⁻¹ and rotational broadening of about 65 km s⁻¹. This must be a grazing eclipsing binary.

For KP102791 we obtained 14 CfA spectra spanning 88 days. The spectroscopic analysis yielded an effective temperature of T_eff = 7000 K, a surface gravity of log(g) = 4.5 cm s⁻², and rotational velocity of V_rot = 27.5 km s⁻¹. The 14 radial velocities, listed in Table 3, allowed us to derive a single-lined spectroscopic orbit with period P = 3.0227 ± 0.0011 days, eccentricity e = 0.033 ± 0.026, orbital semi-amplitude K = 64.9 ±  1.0 km s⁻¹, and mass function f(m) = 0.00856 ±  0.00039 solar masses; see Figure 16. Note that the spectroscopic period is twice the photometric period, which often happens when the secondary eclipses look similar to the primary eclipses. The eccentricity is indistinguishable from circular, suggesting that the orbit has been circularized by tidal forces. Thus, it is not unreasonable to assume that the rotation of the two stars has been synchronized and aligned with the orbital motion. In this case the observed spectroscopic line broadening can be used to estimate the radius of the primary star, which comes out to about 1.6 solar radii. This in turn implies that the primary has not evolved very much, which is consistent with the surface gravity derived from the spectra. Adopting a mass of 1.5 solar masses for the primary, the mass of the unseen secondary implied by the mass function is about 0.75 solar masses. The ephemeris for future primary eclipses based on just the radial velocity data is 2454170.445 ± 0.069 + (3.0227 ± 0.0011) × E. Observations with KeplerCam on the 1.2 m reflector at the Whipple Observatory revealed an ingress starting at heliocentric Julian date 2,454,165.84. Unfortunately the event started 6 h later than predicted by the photometric ephemeris available at that time from data that were already two years old, and the ingress was still in progress and already 0.025 magnitudes deep when the telescope reached its pointing limits. It turns out that this particular transit event must have been a secondary eclipse, with its center almost exactly 1.50 cycles after the spectroscopic epoch for primary eclipses quoted above.

Zoom In Zoom Out Reset image size

For KP102662 we obtained five CfA spectra spanning 33 days. The spectroscopic analysis yielded T_eff = 6500 K, log(g) = 4.5V_rot = 3.5 km s⁻¹, and mean radial velocity 〈V_rad〉 = −20.83 ± 0.37 km s⁻¹ rms. The χ² probability that the observed velocity residuals are consistent with Gaussian errors and constant velocity is P(χ²) = 0.82, so if the transit-like light curve observed for the visible star in this system is caused by an orbiting companion, the companion mass must be less than just a few Jupiter masses. On this basis KP102662 survives as a viable transiting planet candidate. However, the transit light curve looks V-shaped and rather too deep to allow a Jupiter-sized planet. Nevertheless, this candidate deserves further follow-up observations. Obtaining a high-quality lightcurve would probably prove to be time consuming, because the ephemeris is based on data that are already two years old, so highly precise radial velocities may be the best way to proceed.

For KP103126 we obtained six CfA spectra spanning 58 days. The spectroscopic analysis yielded T_eff = 6250 K, log(g) = 4.5 cm s⁻², V_rot = 0.5 km s⁻¹, 〈V_rad〉 = −11.49 ± 0.79 km s⁻¹ rms, and P(χ²) = 0.035. The velocity residuals are a bit larger than expected, but still consistent with an orbiting companion that is no more than several Jupiter masses. Thus this candidate also survives as a viable transiting-planet candidate, one that might reward highly precise radial-velocity observations.