EPIC 201702477b: A TRANSITING BROWN DWARF FROM K2 IN A 41 DAY ORBIT

The NASA Kepler telescope is a 0.95 m space-based Schmidt telescope with a 105 deg² (Borucki et al. 2010) field of view (FOV). The original mission monitored a single field in the northern hemisphere, and was designed to determine the frequency of Earth-like planets in the galaxy. After four years of operations two of the spacecraft's reaction wheels failed, ending the original mission. However, the telescope was re-purposed to monitor selected ecliptic fields, which optimizes the pointing stability, in a new mission called K2 (Howell et al. 2014).

K2 monitors pre-selected target stars in ecliptic fields for durations of approximately 80 days. While this duration is much shorter than the original Kepler mission, it is still a significant improvement over ground-based monitoring which must contend with interruptions from poor weather and the Earth's day–night cycle. The result of this is that K2 is currently the premier facility for finding long period transiting planets, and EPIC 201702477b is an example of such a discovery.

EPIC 201702477 was monitored by K2 as part of Campaign 1 between 2014 May 30 and August 21. The star was included as part of program GO1059 (Galactic Archaeology), which aimed to monitor red giant stars and selected targets based purely on a 2MASS magnitude and color cut. The 2MASS color of EPIC 201702477 is J − K = 0.502, right at the edge of the color cut for the program (J − K > 0.5). Given this and the magnitude of the target (V = 14.57), it was not likely EPIC 201702477 would be a giant star, and indeed our spectroscopy shows the star is a Sun-like dwarf (see Section 2.3).

EPIC 201702477b was first identified as a transiting exoplanet candidate in Foreman-Mackey et al. (2015), where a transit signal with a 40.7365 day periodicity was reported. The candidate was studied further by Montet et al. (2015b) using existing Sloan Digital Sky Survey (SDSS) imaging, and they noted the presence of a neighbor at 1211 with a Δr = 4.65 ± 0.09 mag. They concluded this neighbor was not sufficiently close to be responsible for the transit signal identified using a photometric aperture with a size of 10''. They also calculated the false positive probability (FPP) for EPIC 201702477b using the vespa algorithm (Morton 2012) to be 0.145, and therefore deemed it to be an "exoplanet candidate" (defined as 0.01 < FFP < 0.9).

Due to its long orbital period there are only two transit events in the K2 data, and at the K2 30 minute cadence this equated to just 16 in-transit data points. Such poor sampling of the transit event, even given the exquisite precision of K2, meant that the transit parameters were rather poorly defined. In such circumstances, further ground-based photometry is very important in order to help fully characterize the system.

Of the 37 candidates presented by Foreman-Mackey et al. (2015), EPIC 201702477 has the longest orbital period, with the exception of EPIC 201257461, which has been shown to be a false candidate (Montet et al. 2015b). The reported planet/star radius ratio of EPIC 201702477b is R_P/R_star = 0.0808, indicating a gas giant exoplanet assuming a solar-type host.

We downloaded the K2 pixel data for EPIC 201702477 from the Mikulski Archive for Space Telescopes (MAST)³³ and used a modified version of the CoRoT imagette pipeline (Barros et al. 2015) to extract the light curve. We computed an optimal aperture based on signal-to-noise of each pixel. The background was estimated using the 3σ clipped median of all the pixels in the image outside the optimal aperture and removed before aperture photometry was performed. We also calculated the centroid using the modified moment method by Stone (1989). For EPIC 201702477 we found that a 14 pixel photometric aperture resulted in the best photometric precision.

The degraded pointing stability of the K2 mission results in flux variations correlated with the star's position on the CCD. To correct for this we used a self-flat-fielding procedure similar to Vanderburg & Johnson (2014) that assumes the movement of the satellite is mainly in one direction. A full description of the pipeline given in Barros et al. (2016). The final light curve of EPIC 201702477 has mean out-of-transit rms of 293 ppm and the two transit events in the light curve are plotted in Figure 1.

Zoom In Zoom Out Reset image size

The LCOGT is a network of fully automated telescopes (Brown et al. 2013). Currently there are 10 LCOGT 1 m telescopes operating as part of this network, eight of which are in the southern hemisphere: three at the Cerro Tololo Inter-American Observatory (CTIO) in Chile, three at the South African Astronomical Observatory (SAAO) in South Africa, and two at Siding Spring Observatory (SSO) in Australia. Each telescope is equipped with an imaging camera; either a "Sinistro" or an SBIG STX-16803. The Sinistro is LCOGT's custom built imaging camera that features a back-illuminated 4 K × 4 K Fairchild Imaging CCD with 15 μm pixels (CCD486 BI). With a plate scale of 0387/pixel, the Sinistro cameras deliver a FOV of 266 × 266, which is important for monitoring a sufficient number of reference stars for high-precision differential photometry. The cameras are read out by four amplifiers with 1 × 1 binning, with a readout time of ≈45 s. The SBIG STX-16803 cameras are commercial CCD cameras which feature a frontside-illuminated 4 K × 4 K CCD with 9 μm pixels—giving a field of view of 158 × 158. These cameras are typically read out in 2 × 2 binning mode, which results in a read-out time of 12 s.

The Transiting Exoplanet CHaracterisation (TECH)³⁴ project uses the 1 m telescopes in the LCOGT network to photometrically characterize transiting planets and transiting planet candidates. A major focus of the TECH project is to characterize long period (>10 day) transiting planets or candidates which are difficult to monitor with single site or non-automated telescope systems. As such, EPIC 201702477 was selected as a good candidate for photometric monitoring, and was entered into the automated observing schedule in 2015 February.

The first transit event for EPIC 201702477b monitored by the TECH project was on 2015 March 15 from CTIO. We observed the target from 01:00 UT to 08:13 UT using a Sinistro in the r-band. The exposure times were 240 s, the observing conditions were photometric, and the airmass ranged from 2.3 to 1.2. We detected a full transit of EPIC 201702477b with a depth and duration consistent with that seen in the K2 data. The next transit event occurred 41 days later on 2015 April 28, and was observable from SAAO. EPIC 201702477 was monitored between 17:00 UT and 22:50 UT using an SBIG camera, again in the r-band. The exposure times were 180 s, the observing conditions were again photometric, and the airmass ranged from 1.8 to 1.2. These data show the first half of a transit event consistent with the previous events. The images for both observations were calibrated via the LCOGT pipeline (Brown et al. 2013) and aperture-photometry extracted in the standard manner as set out in Penev et al. (2013). The photometric data are provided in Table 1, and the phase-folded light curves are presented in Figure 1.

In order to determine the stellar parameters for EPIC 201702477, on 2015 March 2 we obtained a low-resolution (R = 3000) spectrophotometric observation with the Wide Field Spectrograph (WiFeS) on the Australian National University (ANU) 2.3 m telescope at SSO. The methodology for this spectral typing is fully set out in Bayliss et al. (2013). A spectrum of R = λ/Δλ = 3000 from 3500 to 6000 Å is flux calibrated according to Bessell (1999) using spectrophotometric standard stars. We determine stellar properties, particularly T_eff and $\mathrm{log}g$, via a grid search using the synthetic templates from the MARCS model atmospheres (Gustafsson et al. 2008). The results showed the star was a Sun-like dwarf star with T_eff = 5600 ± 200 K and $\mathrm{log}g$ = 4.5 ± 0.5 dex. Thus the transit depth was confirmed to be consistent with a planetary-size body.

To better determine the stellar properties we obtained a spectrum of the star with Keck/HiReS (Vogt et al. 1994) on 2015 June 30. The instrument was configured to the standard setup for the California Planet Search (Howard et al. 2010). We collected a single 7 minutes exposure using the C2 (14 × 0.861) decker for a spectral resolution of R ∼ 45,000 and signal-to-noise ratio (S/N) of ∼25 per pixel at 5500 Å. We used the software specmatch (Petigura 2015) to determine the stellar properties. The resulting parameters are listed as initial spectroscopic information in Table 2. Following the methodology described in Sozzetti et al. (2007) we used these initial spectral parameters from Keck as priors for the global fitting (see Section 3), determined a new $\mathrm{log}g$, and then used this as a prior for a second iteration of specmatch. The global fit was then run again with these updated parameters, and the final solution gave T_eff = 5517 ± 70 K and $\mathrm{log}g$ = 4.466 ± 0.058 for EPIC 201702477. The final set of stellar parameters is listed in Table 4.

Table 2.  Summary of Stellar Properties for EPIC 201702477

Parameter	Value	Source
Identification
R.A. (deg.)	175.2407940	K2 EPIC
Decl. (deg.)	+3.6815840	K2 EPIC
2MASS ID.	11405777 + 0340535	2MASS PSC
Photometric Information
Kepler (mag)	14.430	K2 EPIC
u (mag)	16.312 ± 0.005	SDSS DR12
g (mag)	14.871 ± 0.003	SDSS DR12
r (mag)	14.354 ± 0.003	SDSS DR12
i (mag)	14.189 ± 0.003	SDSS DR12
z (mag)	14.137 ± 0.004	SDSS DR12
J (mag)	13.268 ± 0.027	2MASS PSC
H (mag)	12.881 ± 0.028	2MASS PSC
K (mag)	12.766 ± 0.033	2MASS PSC
Space Motion
pmR.A. (mas yr−1)	−10.0 ± 3.6	PPMXL
pmDec (mas yr−1)	−9.8 ± 3.6	PPMXL
mean γRV (km s−1)	34.0	HARPS
Initial Spectroscopic Information
Teff (K)	5492 ± 60	Keck
$\mathrm{log}g$	4.12 ± 0.07	Keck
$[\mathrm{Fe}/{\rm{H}}]$	−0.20 ± 0.04	Keck
$v\sin i$(km s−1)	<2	Keck

Download table as:  ASCIITypeset image

We obtained a high-spatial resolution image with the instrument AstraLux (Hormuth et al. 2008), mounted on the 2.2 m telescope in Calar Alto Observatory (Almería, Spain), using the lucky imaging technique. The target was observed on 2015 November 18 under normal weather conditions. We obtained 60,000 frames with individual exposure times of 0.060 s, hence a total exposure time of 1 hour, in the SDSS i-band. The images were reduced using the observatory pipeline, which applies bias and flat-field correction to the individual frames and selects the best images in terms of Strehl ratio (Strehl 1902). The best 10% of the images are then aligned and stacked to compose the final image. The sensitivity limits are calculated following the process explained in Lillo-Box et al. (2014) and are presented in Figure 2.

Zoom In Zoom Out Reset image size

We observed EPIC 201702477 on 2015 December 27 using NIRC2 NGS-AO (PI: Keith Matthews) on Keck II. We used the Ks band and the narrow camera setting. We took a total of four images, each with 60 s of total integration time. We calibrated the images with a flat field, dark frames, and removed image artifacts from dead and hot pixels. We then created a single median-stacked image. We do not see any stellar companions in this image, and compute the contrast curve from the median stacked image. For every point in the image, we compute the total flux from pixels within a box with side length equal to the full width at half maximum (FWHM) of the target star's point-spread function (PSF). We then divide the image into a series of annuli with width equal to twice the FWHM. For each annulus, we determine the 1σ contrast limit to be the standard deviation of the total flux values for boxes inside that annulus. To convert from flux limits to flux ratios and differential magnitudes, we divide the computed standard deviation by the total flux of a similar box centered on the target star. Figure 2 shows the 5σ average contrast curve.

The clear conclusion from both the lucky imaging and the AO imaging is that the target appears to be an isolated star to within the limits presented in our contrast curves, and this indicates the transit is occurring on the target star rather than a nearby blended neighbor.

We performed radial velocity follow-up observations of EPIC 201702477 with the SOPHIE (Bouchy et al. 2009a) and HARPS (Mayor et al. 2003) spectrographs. Both instruments are high-resolution (R ≈ 40,000 and 110,000 for SOPHIE and HARPS, respectively), fiber-fed, and environmentally controlled echelle spectrographs covering visible wavelengths. We obtained three spectra with SOPHIE (OHP programme ID: 15B.PNP.HEBR) from 2015 June 12 to 2016 February 17 with exposure times of 1800 and 3600 s, reaching an S/N between 8 and 22 per pixel at 5500 Å. We obtained 10 other spectra with HARPS (ESO programme ID: 096.C-0657) from 2016 January 10 to February 15 with exposure times between 900 and 3600 s, corresponding to an S/N between 3 and 17 per pixel at 5500 Å.

All spectra were reduced with the online pipeline available at the telescopes. The spectra were then cross-correlated with a template mask that corresponds to a G2V star (Baranne et al. 1996). This template was chosen to be close in spectral type to the host star. Radial velocities, bisector span, and FWHM were measured on the cross-correlation function and their associated uncertainties were estimated following the methods described in Bouchy et al. (2001), Boisse et al. (2010), and Santerne et al. (2015). SOPHIE radial velocities were corrected for charge-transfer inefficiency (Bouchy et al. 2009b) using the equation provided in Santerne et al. (2012). The derived radial velocities are reported in Table 3 and plotted in Figure 3.

Zoom In Zoom Out Reset image size

Table 3.  SOPHIE and HARPS RVs of EPIC 201702477

BJD	RV	σRV	Vspan	${\sigma }_{{V}_{\mathrm{span}}}$	FWHM	σFWHM	Texp	S/N	Instrument
(2,400,000+)	(km s−1)	(km s−1)	(km s−1)	(km s−1)	(km s−1)	(km s−1)	(s)
57363.71073	37.566	0.025	−0.066	0.045	9.595	0.062	3600	21.7	SOPHIE
57399.62998	35.780	0.046	0.103	0.082	9.614	0.114	3600	13.7	SOPHIE
57436.62181	33.236	0.031	0.129	0.055	9.251	0.076	1800	8.2	SOPHIE
57397.85193	34.765	0.011	−0.031	0.016	6.744	0.022	3600	12.0	HARPS
57401.81118	36.943	0.007	0.002	0.010	6.709	0.013	3600	17.5	HARPS
57404.83131	37.670	0.050	0.033	0.075	7.004	0.100	900	3.0	HARPS
57407.80298	38.103	0.041	−0.117	0.061	6.802	0.082	1500	5.5	HARPS
57410.77375	37.918	0.056	−0.091	0.084	6.311	0.111	900	2.9	HARPS
57417.80853	36.254	0.041	0.108	0.061	6.574	0.081	900	3.9	HARPS
57424.79651	32.335	0.039	−0.080	0.058	6.912	0.078	900	4.2	HARPS
57427.78748	30.393	0.033	0.000	0.050	6.827	0.067	900	4.8	HARPS
57429.80114	29.672	0.053	0.079	0.079	6.797	0.106	900	3.1	HARPS
57433.79557	30.881	0.045	0.005	0.067	6.803	0.090	900	3.8	HARPS

Note. S/N is given per pixel at 550 nm.

Download table as:  ASCIITypeset image

Our radial velocity measurements show a large amplitude (K = 4.252 ± 0.028 km s⁻¹) variation in-phase with the photometric ephemeris and indicative of a brown dwarf mass companion in an elliptical orbit. We use these radial velocity data to determine the planetary parameters in Section 3.

We analyzed the radial velocity and photometric data of EPIC 201702477 with the Markov chain Monte Carlo (MCMC) algorithm of the pastis software, which is fully described in Díaz et al. (2014a). We modeled the radial velocities with a Keplerian orbit and the photometric data with the JKTEBOP package (Southworth 2011) and references therein. We chose as a prior for the stellar parameters the values derived from the Keck spectroscopy (Section 2.3). We used the Dartmouth stellar evolution tracks of Dotter et al. (2008) to derive the stellar fundamental parameters (mass, radius, age) in the MCMC, in particular the stellar density which was used to constrain the transit parameters given the eccentricity constrained by the radial velocities, as in Santerne et al. (2014). We ignored pre-main sequence solutions as there is no evidence that this is a young star and the pre-main sequence stage is extremely short in duration. We assumed uninformative priors for the parameters, except for the orbital ephemeris for which we used the ones provided by Montet et al. (2015b), the spectroscopic parameters that we took from our spectral analysis, and the orbital eccentricity for which we choose a Beta distribution as recommended by Kipping (2013). For the transit modeling, we used a quadratic law with coefficients taken from the interpolated table of Claret & Bloemen (2011) for both the Kepler and r bandpasses and changed them at each step of the MCMC.

We ran 20 chains of 3 × 10⁵ iterations each, with starting points randomly drawn from the joint prior. We rejected non-converged chains based on the Kolmogorov–Smirnov test (Díaz et al. 2014a). We then removed the burn-in of each chain before thinning and merging them. We ended with more than 3000 independent samples of the posterior distribution that we used to derive the value and 68.3% uncertainty of each parameter that we report in Table 4.

Table 4.  Parameters from the Global Fit for the EPIC 201702477 System

Parameter	Value
Brown Dwarf
P (days)	40.73691 ± 0.00037
T0 (BJD)	2456811.5462 ± 0.0011
T14 (hr)	4.04 ± 0.13
a/R⋆	54.0 ± 3.4
RBD/R⋆	0.0862 ± 0.0024
b	0.851 ± 0.023
bsec	0.752 ± 0.023
i (degrees)	89.105 ± 0.082
e	0.2281 ± 0.0026
ω (degrees)	195.9 ± 1.8
γRV (km s−1)	34.745 ± 0.020
K (km s−1)	4.252 ± 0.028
MBD(MJ)	66.9 ± 1.7
RBD(RJ)	0.757 ± 0.065
a (au)	0.2265 ± 0.0026
ρc (g cm−3)	191 ± 51
Star
$\mathrm{log}g$	4.466 ± 0.058
Teff (K)	5517 ± 70
$[\mathrm{Fe}/{\rm{H}}]$	−0.164 ± 0.053
R⋆ (R☉)	0.901 ± 0.057
M⋆ (M☉)	0.870 ± 0.031
ρ⋆ (ρ☉)	1.18 ± 0.24
age (Gyr)	8.8 ± 4.1
RV and Photometry
HARPS jitter (km s−1)	${0.035}_{-0.018}^{+0.031}$
SOPHIE jitter (km s−1)	${0.101}_{-0.070}^{+0.180}$
SOPHIE offset relative to HARPS (km s−1)	0.078 ± 0.081
K2 contamination	${0.0071}_{-0.0049}^{+0.0072}$
K2 flux out of transit	1.000022 ± 3.4e-05
K2 jitter	0.000253 ± 2.8e-05
SAAO contamination	${0.030}_{-0.021}^{+0.030}$
SAAO flux out of transit	0.99975 ± 2.7e-04
SAAO jitter	0.00039 ± 3.8e-04
CTIO contamination	${0.025}_{-0.018}^{+0.028}$
CTIO flux out of transit	0.99966 ± 2.0e-04
CTIO jitter	0.00089 ± 3.2e-04

Download table as:  ASCIITypeset image

We also modeled the system independently (but with the same datasets) using the exofast software (Eastman et al. 2013). We find parameters and uncertainties in close agreement with those that were derived using pastis, and therefore we only report the pastis results.

In order to test for transit timing variations (TTVs), we perform an independent fit of the K2 and LCOGT transit light curves. We fit for independent centroids T₀ for each transit, while forcing the transits to share the geometric parameters a/R_⋆, R_BD/R_⋆, and i. Since ground-based photometry suffers from instrumental systematics that can bias the centroid measurements, we simultaneously detrend the LCOGT light curves against a linear combination of the terms describing the time, X, Y pixel drift, airmass trend, sky background flux, and target star FWHM variations. No significant TTVs were detected at the 30 s level. The high cadence LCOGT light curves offer similar timing precisions as the long cadence K2 observations, and demonstrate the power of follow-up observations for long period candidates from K2. The variations in the transit centroid times are shown in Figure 4 and listed in Table 5.

Zoom In Zoom Out Reset image size

Table 5.  Summary of Photometric Observations for EPIC 201702477

Instrument	Epoch	Transit Centroid (BJD-TDB)	Filter
Kepler	0	$2456811.54499({}_{-60}^{+28})$	Kep.
Kepler	1	$2456852.28205({}_{-37}^{+61})$	Kep.
LCOGT 1 m+Sinistro	7	$2457096.70347({}_{-28}^{+34})$	sloan-r
LCOGT 1 m+SBIG	8	$2457137.44035({}_{-38}^{+35})$	sloan-r

Download table as:  ASCIITypeset image

We can place an upper limit on the companion's luminosity based on the secondary eclipse measurements. We checked for the presence of a secondary eclipse in the K2 light curves; the phase of the eclipse is constrained by a Gaussian prior on the e and ω orbital parameters, determined from the RV observations and presented in Table 4. No secondary eclipse is detected at a 2σ upper limit of 1.96 mmag, equating to a maximum blackbody temperature for the brown dwarf of T_eff < 3950 K.

Including EPIC 201702477b, there are just 12 known brown dwarfs (13M_J < M_BD < 80M_J) that transit main sequence stars—see Table 6 for a list and Csizmadia et al. (2015) for a detailed list of these systems. These systems are extremely important as they provide an independent check on the radial velocity statistics for brown dwarfs, in addition to giving us true masses and radii. While a full statistical analysis is beyond the scope of this paper, we note that from the K2 survey alone there have been five previously unknown hot Jupiter discoveries (NASA Exoplanet Archive on 2016 April 20), but EPIC 201702477b is the first brown dwarf discovery. Although this is in line with the relative statistics for these two populations presented in Santerne et al. (2016), we caution that the target selection process for K2 imprints a strong bias on the sample and makes robust statistics dependent on careful modeling of the selection effects. In addition, the detection of a large radial velocity variation may prompt follow-up efforts to be discontinued for some planet search programs.

Ma & Ge (2014) have suggested that there exist two populations of brown dwarfs. The first are brown dwarfs below ∼45 M_J that are formed in the protoplanetary disc via gravitational instability. The second are brown dwarfs above ∼45 M_J that are formed through molecular cloud fragmentation; essentially the very lowest mass objects of the star formation process. This division of the brown dwarf population at ∼45 M_J coincides with the minimum of the companion mass function derived by Grether & Lineweaver (2006) and the void in the mass range as derived from the CORALIE RV survey (Sahlmann et al. 2011). Under this division, EPIC 201702477b would clearly be classed in the second category as likely to be formed via molecular cloud fragmentation, as at 66.9 ± 1.7 M_J its mass lies well above the mass division.

Unlike pure RV detections, transiting brown dwarfs can have true masses determined, as opposed to minimum masses. We can also be fairly certain that these discoveries are free from a mass bias, as to first order the discoveries are made on the basis of the planet-to-star radius ratio alone, and radius of the companion is largely independent of the mass in the brown dwarf regime. Therefore while the numbers are still small, transiting brown dwarfs provide a critical test of the two-population model proposed in Ma & Ge (2014). As can be seen from Figure 5, we do indeed see evidence of a gap in the mass distribution between about 40 and 55 M_J, lending support to the two-population hypothesis.

Zoom In Zoom Out Reset image size

We also investigate if these populations of brown dwarfs correlate with the metallicity of the host stars they transit. In line with the well established relationship of giant planet frequency increasing with host metallicity (Gonzalez 1997; Santos et al. 2001; Fischer & Valenti 2005), we may expect the lower mass brown dwarfs, if formed by core-accretion, to orbit higher metallicity host stars than the higher mass brown dwarfs that formed by fragmentation. With the current sample of just 12 transiting brown dwarfs we do not observe any such correlation; see Figure 6.

Zoom In Zoom Out Reset image size

EPIC 201702477b lies at the minimum for brown dwarf radii, and with a density of 191 ± 51 g cm⁻³ it is the highest density object ever discovered in the regime from planets to main sequence stars—see Figure 7. To investigate the mass–radius relationship for brown dwarfs we take the known systems with precise (uncertainties <20%) mass and radius and compare the measured radius with the radius predicted from the cond03 models (Baraffe et al. 2003). We use the published masses and ages for each transiting brown dwarf (set out in Table 6), and compute a cond03 model radius for each object based on a two-dimensional (2D) linear interpolation of the model grid points. We plot the difference between the measured radius and these computed radii in Figure 8. For hot Jupiters, there exists a population of inflated radius objects at short periods where the insolation flux exceeds 10⁸ erg cm⁻² s⁻¹ (Demory & Seager 2011). However for brown dwarfs the radii do not appear to exhibit such a trend, and the radii appear to be uncorrelated with the insolation flux (or for that matter orbital period). This may be expected as most of the mechanisms proposed for giant planet inflation do not apply to these more massive brown dwarfs (Bouchy et al. 2011b). A possible exception may be KELT-1b (Siverd et al. 2012) which receives extremely high insolation of 7.81 × 10⁹ erg cm⁻² s⁻¹ and indeed appears to be inflated. However we do note that the higher mass population of brown dwarfs are in much closer agreement to the cond03 models than the lower mass population of brown dwarfs (see Figure 8).

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Of the 12 known transiting brown dwarfs, six have masses in the range of 59–67 M_J, as shown in Figure 5. The lack of higher mass objects is only because we restricted our sample to objects less than 80 M_J(the usual limit for what is considered a brown dwarf). Many transit and radial velocity surveys may also not report objects above this mass. However the lack of discoveries with masses below this group of high mass transiting brown dwarfs is interesting, and appears as a sharp lower mass edge to the high-mass brown dwarfs. While we caution that the sample size is still small, the edge is intriguing and may be related to the ejection process during formation. In the simulations of Stamatellos & Whitworth (2009) it was found that although the formation of brown dwarfs is approximately flat in the regime of 15–80 M_J, the subsequent ejection process, which results in the loss of over half of the companions, is strongly mass dependent. Primarily it is the lower-mass brown dwarfs that are ejected, leaving behind a higher-mass population. These simulations even show that companions around 70 M_J are among the least likely to get ejected (see Figure 15 of Stamatellos & Whitworth 2009). It is possible that it is these objects we find as the population of transiting brown dwarfs with masses from 60 to 70 M_J.