INDEPENDENT DISCOVERY OF THE TRANSITING EXOPLANET HAT-P-14b

HAT-P-14 is a bright (V_mag = 9.98) star in the constellation Virgo, located at α_J2000 = 17^h20^m27.87, δ_J2000 = +38°14'319 (GSC 3086-00152; 2MASS 17202788+3814317). It has been observed by SuperWASP (La Palma) since 2004 and a total of 25,474 photometric points have been collected. A periodic signature was detected in the observations obtained between March and August in 2007, 2008, and 2009. All the data were processed using the pipeline described in Collier Cameron et al. (2006), yielding a detection of a transit-like feature with a period of 4.6278 days in multiple cameras and successive seasons. The folded light curve is shown in Figure 1.

Zoom In Zoom Out Reset image size

The target underwent several consistency tests aimed at eliminating false positives (see Collier Cameron et al. 2007). Resolved blends were ruled out by inspecting Digitized Sky Survey images which showed the star to be isolated within the WASP photometric aperture. No significant ellipsoidal variability was measurable in the WASP light curve folded on the transit period. The transit depth and duration yielded a planet-like radius for the companion, and a stellar density appropriate to a main-sequence host star of the effective temperature derived from the Two Micron All Sky Survey (2MASS) colors. Having passed all these tests, the star was selected for follow-up observations.

We obtained six further light curves, shown in Figure 2, in order to refine the photometric parameters. All photometric data presented in this paper are available from the NStED database.¹⁵

Zoom In Zoom Out Reset image size

Two full transits were observed with FTN (Hawaii), on 2009 September 18 (SDSS i band) and 2010 May 21 (Pan-STARRS Z filter). The first transit was taken using the Merope camera (EM01)¹⁶ in binning 2 × 2 mode and defocused to enable an exposure time of 40 s. Aperture photometry was performed using an 18 pixel radius aperture and ensemble photometry using the two bright, non-variable comparison stars within the field of view. By 2010 May 21, the new Spectral camera (FS03)¹⁷ was available at FTN and defocused to obtain 20 s exposures. The 10.36 × 10.22 arcmin field of view included four bright comparison stars. These observations took advantage of the 4ag autoguider to maintain the locations of the stars to within an rms ∼ 2 pixel of the same location throughout the observations and led to a noticeable improvement in the photometric precision. All the FTN data were reduced using the pipeline¹⁸ written at Liverpool John Moores University, and photometry performed using the IRAF/DAOPHOT package.

A partial transit on 2010 March 13 was observed simultaneously using the RISE instrument (Gibson et al. 2008; Steele et al. 2008) mounted on the robotic 2.0 m Liverpool Telescope (LT) and an Andor CCD detector on an 18 cm Takahashi Astrograph telescope, both located on La Palma. Two further full transits were observed; on 2010 May 2 using the Astrograph, and on 2010 June 22 by the LT.

The Andor 1 × 1 K e2v CCD detector has a 5 s readout time. For the first transit, the Astrograph was defocused with FWHM = 20'' and 685 exposures of duration 20 s were obtained in the 4.8 hr period. For the second transit, the exposure time was increased to 30 s and FWHM = 28'' giving 456 observations in the 4.5 hr period. The data were reduced and differential photometry performed using the PYRAF/DAOPHOT package. The observations were subsequently binned with 3 and 2 points per bin, respectively.

The LT was defocused by −1.2 mm for both transit observations using the RISE instrument, giving FWHM = 18 and 21 pixels on the respective nights (0.54 arcsec pixel⁻¹). The CCD was binned 2 × 2 and exposures of 3.7 s taken with effectively no dead time. During the 3.0 hr and 2.3 hr periods, 3520 and 3290 frames were taken, respectively. The data were reduced and differential photometry performed using the ULTRACAM pipeline (Dhillon et al. 2007), and then binned with 10 points per bin.

We obtained follow-up spectroscopic observations to determine the planetary, orbital, and stellar parameters. The star was initially observed using the FIbre-fed Echelle Spectrograph (FIES) mounted on the 2.5 m NOT. In total, 24 usable spectra were obtained between 2009 July 12 and 2010 April 5. Two spectra were removed from the analysis due to Moon contamination affecting the radial velocity determination. FIES was used in medium-resolution mode (R = 46,000) with simultaneous ThAr calibration, and observations were conducted using exposure times of 1200 s. We used the bespoke data reduction package FIEStool¹⁹ to extract the spectra. An IDL cross-correlation routine was used to obtain radial velocities (RVs) by fitting Gaussians to the cross-correlation functions (CCFs) of 30 spectral orders and taking the mean. A template spectrum was constructed by shifting and coadding the spectra, against which the individual spectra were cross-correlated to obtain the final velocities. The template was cross-correlated with a high signal-to-noise spectrum of the Sun to obtain the absolute velocity to which the relative RVs were shifted. The RV uncertainty is given by σ = rms(v)/$\sqrt{N}$, where v is the RV of the individual orders and N is the number of orders. Line bisector spans (V_span) were computed using the difference in the position of the midpoint of the CCF at 25% and 75% of its depth.

A further nine spectra were taken between 2009 August 24 and 2009 September 5 with the stabilized echelle spectrograph SOPHIE at the 1.93 m telescope of Observatiore de Haute-Provence (Perruchot et al. 2008; Bouchy et al. 2009). The spectrograph was used in high efficiency mode (resolution R = 40,000) and the observations were taken with a signal-to-noise ratio (S/N) ∼ 30 to minimize the Charge Transfer Inefficiency (CTI) effect (Bouchy et al. 2009). Two 3 arcsec diameter optical fibers were used, the first centered on the target and the second on the sky to simultaneously measure the background to remove contamination from scattered moonlight. The spectra were extracted using the SOPHIE pipeline (Perruchot et al. 2008) and RVs computed from a weighted cross-correlation of each spectrum with a numerical mask of G2 spectral type, as described by Baranne et al. (1996) and Pepe et al. (2002). Cross-correlation with other masks (F0, K5, and M5) produced a consistent RV semi-amplitude, indicating that the variation is unlikely to be caused by a blended eclipsing binary system of unequal masses.

The RVs and line bisector spans are shown in Table 1 and plotted in Figure 3. A possible correlation between the bisector span and radial velocity was noted by T10 but was attributed to moonlight contamination rather than to a blended eclipsing binary. We find no correlation, supporting the signal's origin as a planetary companion rather than an astronomical false positive.

Zoom In Zoom Out Reset image size

In order to search for additional close stars which may cause blends, we obtained high-resolution H- and K-band imaging with the near-infrared camera INGRID, fed by the adaptive optics system NAOMI on the 4.2 m William Herschel Telescope, see Figure 4. Images were taken in two sky position angles (P.A., 0° and 30°) in natural seeing of 05, and the corrected point-spread functions (PSFs) have an FWHM of 012 and circular contours. No contaminating sources are detected to limiting magnitudes of H ∼ 17.1 ± 0.25 and K_s ∼ 16.7 ± 0.25 at distances >4 × FWHM from the center of the corrected profile. At 1.5 × FWHM the limiting magnitude is 2 mag brighter than this.

Zoom In Zoom Out Reset image size

We derived the fundamental stellar parameters of HAT-P-14 using the following techniques, and the results are shown in Table 2. First, we performed a spectral analysis using Spectroscopy Made Easy (see Valenti & Piskunov 1996), following the method of Valenti & Fischer (2005) and using the Kurucz model atmospheres (Kurucz et al. 1984). Individual NOT spectra were normalized and co-added to produce a single high signal-to-noise spectrum. The H_α, Na i D, and Mg i b lines were fitted simultaneously to determine the spectral parameters. This yielded the values T_eff = 6583 ± 100 K and log g_spec = 4.02 ± 0.1, which indicate that the star is of spectral type F5V (Gray 2008). Values for microturbulence (v_mic = 0.85 km s⁻¹) and macroturbulence (v_mac = 5.28 km s⁻¹) are taken from the prescription of Valenti & Fischer (2005). The projected stellar rotation velocity (vsin i) was determined to be vsin i = 8.4 ± 1.0 km s⁻¹ and the metallically [M/H] = 0.08 ± 0.10.

Table 2. Stellar Parameters for HAT-P-14

Parameter (Unit)	Value
Photometric and Spatial Properties
R.A. (J2000)	17:20:27.88
Decl. (J2000)	+38:14:31.7
V (mag)	9.98 ± 0.058
J (mag)	9.094 ± 0.021
H (mag)	8.927 ± 0.020
Ks (mag)	8.851 ± 0.019
μR.A. (mas yr−1)	1.6 ± 0.7
μdecl. (mas yr−1)	4.7 ± 0.7
U (km s−1)	6.5 ± 1.0
V (km s−1)	−10.6 ± 0.1
W (km s−1)	−6.0 ± 0.8
Galactic longitude (°)	62.6
Galactic latitude (°)	33.5
Spectroscopic Properties
Teff (K)	6583 ± 100
log g (cgs)	4.1 ± 0.12
vmic (km s−1)	0.85
vmac (km s−1)	5.28
vsin i (km s−1)	8.4 ± 1.0
[M/H]	0.08 ± 0.1
Derived Properties
M⋆ (M☉)a	1.32 ± 0.03
R⋆ (R☉)b	1.53 ± 0.10
ρ⋆ (ρ☉)c	0.371+0.056−0.058
L⋆ ($L_{\bigodot }$)d	3.91+0.54−0.37
Age (Gyr)d	1.6+0.4−0.3
Spectral type	F5V

Notes. The photometric and spatial properties are taken from or derived using the following sources: V (TASS; Richmond et al. 2000), J, H, K_s (2MASS; Skrutskie et al. 2006), proper motions (NOMAD; Zacharias et al. 2005), space velocities (calculated according to Johnson & Soderblom 1987, updated to J2000, for a right-hand coordinate system and corrected for solar motion using data from Dehnen & Binney 1998) and galactic coordinates NASA/IPAC Extragalactic Database. ^aDerived from the empirical relationship of Torres et al. (2010a). ^bDetermined using M_⋆ and ρ_⋆. ^cFound using the light curve geometry. ^dFrom stellar models.

Download table as:  ASCIITypeset image

The mean stellar density (ρ_⋆), found from the light curve geometry, can be used as a luminosity indicator for stellar evolutionary models and often provides a stronger constraint than the value of log g from spectral analysis (Sozzetti et al. 2007). We used a Markov Chain Monte Carlo (MCMC) approach to globally model the photometric and radial velocity data (see Section 3.2) and obtained a mean stellar density of 0.371^+0.056_−0.058 ρ_☉. We then compared T_eff, [M/H], and ρ_⋆ with the theoretical stellar evolutionary models of Girardi et al. (2000) to obtain the following stellar properties: M_⋆ = 1.36 ± 0.04 M_☉, Age = 1.6 ^+0.4_−0.3 Gyr, L_* = 3.91^+0.54_−0.37 L_☉, and log g_iso = 4.19 ± 0.03. The model isochrones are shown in Figure 5.

Zoom In Zoom Out Reset image size

The value of log g obtained from the isochrone fit is somewhat larger than that found from spectral analysis, which was also noted by T10. We investigated the effect this higher log g would have on the fit to the spectral line shapes by fixing log g_spec to the isochrone value and reassessing the spectral analysis. We found that in order to retain a good fit to the Mg i b lines, the magnesium abundance [Mg/H] must be reduced by 0.12 (in a similar fashion to WASP-1; see Stempels et al. 2007) due to the inherent anti-correlation between [Mg/H] and log g. The degeneracies between the spectroscopic parameters do not rule out the higher log g value as long as [Mg/H] is allowed to be underabundant, and we take the uncertainty in log g to include both values. Altering log g and [Mg/H] in this fashion had no significant effect on the derived effective temperature or metal abundance.

To determine the properties of the planet HAT-P-14b, we simultaneously modeled the light curves and RVs with a global MCMC fit. Details of this process are described in Collier Cameron et al. (2007) and Pollacco et al. (2008). The free parameters in the fit are: orbital period P, transit epoch T₀, transit duration T_dur, ratio of planet radius to star radius (R_P/R_⋆)², impact parameter b, RV semi-amplitude K, Lagrangian elements ecos ω and esin ω where e is the eccentricity and ω is the longitude of periastron, and the systematic offset velocity γ. In this particular case, two systematic velocities were fit to allow for instrumental offsets between the SOPHIE and FIES data sets. These were found to be: γ_SOPHIE = −20.4294 ± 0.0009 km s⁻¹ and γ_FIES = −20.3833 ± 0.0026 km s⁻¹. To normalize the light curves, a linear de-trending was applied and the normalization parameters were allowed to float during the fit.

The Markov chain length was 10,000 points and the resulting best-fit parameters and uncertainties are shown in Table 3. To account for red noise, the residuals of the non-WASP photometry were averaged into 6 minute bins before calculating χ². The photometric error bars were re-scaled so that χ² ∼ N_bin to guard against the possibility that a densely sampled photometric data set with correlated errors should dominate the result. We also ran the routine using 12 minute binning and this gave consistent values with those derived using the 6 minute binning. The Claret (2000, 2004) nonlinear limb-darkening coefficients were chosen for the appropriate stellar temperature and photometric passband appropriate for each light curve.

In order to estimate the stellar mass and properties derived from it, we used the Torres et al. (2010a) empirical calibrations from eclipsing binaries. This method, adapted by Enoch et al. (2010), computes the stellar mass as a function of T_eff, log ρ_*, and [M/H] and has the advantage of being independent of stellar models. We find M_⋆ = 1.32 ± 0.03 M_☉, similar to the stellar mass derived from the isochrone analysis undertaken here (M_⋆ = 1.36 ± 0.04 M_☉, Padova models) and in T10 (M_⋆ = 1.39 ± 0.05 M_☉, YY models).

We performed the MCMC fit and derived the following planetary parameters: M_p = 2.28 ± 0.07 M_J, R_p = 1.20 ± 0.15 R_J, and ρ_p = 1.32 ± 0.37 ρ_J. In addition, we analyzed the data solely from T10 and found that the fitted parameters were completely consistent with their results. The results are shown in Table 3 and are in good agreement.

In an initial fit to the SOPHIE and FIES RVs, the eccentricity was allowed to float, yielding e = 0.054 ± 0.017. We analyzed the data sets independently and found that a jitter term of 15 m s⁻¹ was required to be added in quadrature to the FIES RV uncertainties in order to obtain χ²_red = χ²/dof = 1 (dof = number of points − number of fitted parameters). This may due to astrophysical noise, as also noted by T10, and/or instrumental systematic noise which has not been accounted for. The FIES RVs do not significantly detect an eccentric orbit, yielding e = 0.04 ± 0.03, but equally do not rule one out.

In contrast, the SOPHIE RVs do not require any jitter term to reconcile the rms scatter with the measured uncertainties. Fitting the SOPHIE RVs alone, a more significant detection of eccentricity was found, e = 0.081 ± 0.022 with χ² = 2.0 for 9 RV points and 4 fitted parameters (K, ecos ω, esin ω, γ_SOPHIE). Fitting a purely circular orbit to the same data, we found χ²_circ = 22.6. Applying the Lucy–Sweeney test (Lucy & Sweeney 1971), the eccentricity is detected at the 3σ level.

A joint fit to both data sets (including the FIES jitter term) reveals e = 0.066 ± 0.023, χ² = 34.4 and χ²_circ = 49 for 33 RV points and 5 fitted parameters (including two systematic velocities, γ_SOPHIE and γ_FIES), indicating the eccentricity is non-zero at the 2.7σ level. We also apply the test to the T10 Keck RVs, including their 7.3 m s⁻¹ jitter term. We find the eccentricity is detected at the 4σ level, e = 0.104 ± 0.012, with χ² = 12.3, χ²_circ = 83.9, 14 RVs, and 4 fitted parameters.

The difference between the eccentricity derived from the NOT/SOPHIE and Keck data suggests that the uncertainties may have been underestimated. It has been shown that determining the significance and uncertainties of orbital eccentricity from sparse data is difficult, and there had been several initial detections which are now thought to be spurious, e.g., WASP-12 (Hebb et al. 2009; Husnoo et al. 2010; Campo et al. 2011); WASP-10 (Christian et al. 2009; Maciejewski et al. 2010). The FIES data are not regularly spaced enough for eccentricity measurements and the constraints mostly rely on the SOPHIE data. In fact, the SOPHIE value alone is more consistent with the Keck value.

The fact that for the SOPHIE data, χ²_red is near one around an eccentric-orbit fit can either mean that the eccentricity measurement is correct and there are no unrecognized sources of residuals, or that the eccentricity is different and the error budget is underestimated. Although care has been taken to keep the S/N constant to avoid the CTI effect, the values of the RV uncertainties vary and zero-point shifts could be present which may bias the eccentricity measurement.