DISCOVERY OF A WIDE PLANETARY-MASS COMPANION TO THE YOUNG M3 STAR GU PSC

High-resolution optical spectroscopy was obtained with ESPaDOnS (Donati et al. 2006) at the Canada–France–Hawaii Telescope (CFHT). The data were reduced by the CFHT queue service observing team using the pipeline UPENA 1.0, that uses the Libre-ESpRIT software (Donati et al. 1997). The resulting spectrum goes from 0.37 μm to 1.05 μm (40 grating orders) with R ∼ 68,000.

High-resolution spectroscopy was also acquired for GU Psc with two different instruments in the NIR with the specific goal of measuring its precise radial velocity. With CRIRES on the Very Large Telescope (VLT; Kaeufl et al. 2004), the 04-wide slit was used in an order centered on 1.555 μm for a resulting R ∼ 50,000. With PHOENIX on Gemini-South (Hinkle et al. 2003), we used the 034 slit with the 1.547−1.568 μm blocking filter and obtained a resolving power of R ∼ 52,000. The NIR spectroscopic data were reduced using standard procedures with a custom IDL pipeline.

A 0.8–2.4 μm NIR spectrum of the primary was also obtained with SpeX, the medium-resolution spectrograph and imager at NASA InfraRed Telescope Facility (Rayner et al. 2003), using the cross-dispersing mode with the 08 slit, under good seeing conditions (∼1''). The reduction was done using SpeXtool (Cushing et al. 2004; Vacca et al. 2003), and telluric absorption was corrected with the A3 spectroscopic standard star HIP 5310. It was then flux-calibrated by adjusting the J − H and J − K_s synthetic colors with the Two Micron All Sky Survey (2MASS) photometry.

GU Psc was observed with NICI, the high-contrast imager on Gemini-South (Ftaclas et al. 2003; Chun et al. 2008), as part of a survey to find closer-in companion (M.-E. Naud et al., in preparation). It was observed twice (2011 October 21 and 2012 September 1) in two narrowband filters around 1.6 μm (CH₄ H 4% S centered at 1.578 μm and the CH₄ H 4% L at 1.652 μm). Each observing sequence is composed of 35 exposures of 3 coadd × 20.14  s, taken with the 032 focal plane mask. The reduction was carried using the method detailed in Artigau et al. (2008).

The companion was originally identified as part of a survey of young low-mass stars with Gemini-South/GMOS imaging in i and z bands (É. Artigau et al., in preparation). The original three 300 s exposures in i and three 200 s exposures in z were taken on 2011 September 22 (see Figure 1). The custom data reduction procedure included overscan and fringe subtraction and flat-field correction. Astrometry was anchored to the USNO-B1 catalog. The images were median combined and the magnitude zero point was determined through a cross-match with the Sloan Digital Sky Survey (SDSS). Among the 91-star sample of the survey, GU Psc's companion was the only credible candidate found for separations ranging between 5'' and 10'' (depending on the primary's brightness) and the edge of the GMOS 55 field of view (FoV). It was detected in the z-band image, but not in the i band. Follow-up observations with the same instrument and observational setup were made on 2011 October 18 to obtain a deeper i-band image: five 300 s i-band images were taken, as well as an additional 200 s z-band image. The z-band photometry was consistent with that of the discovery data set, confirming that this object was not a transient or artifact. The new i-band imaging still did not reveal the companion but provided a 3σ upper limit on the flux of i > 25.28, indicating a very red i − z color (>3.53, 3σ).

Zoom In Zoom Out Reset image size

Follow-up NIR photometry was carried at the CFHT with WIRCam (Puget et al. 2004). GU Psc b was first observed in the J band on 2011 October 10 for a total integration time of 14.2 minutes with single exposures of 50 s (see Figure 1). The target was centered on the North-East WIRCam detector and observed using a large dither pattern (15 positions or more) with the nominal 60'' amplitude. The images were preprocessed by the IDL Interpretor of WIRCam Images (`I`iwi)⁸ which performs the dark subtraction, flat fielding, bad pixel masking, and sky subtraction. The final stacks were produced with SExtractor, SCAMP, and SWarp (Bertin & Arnouts 1996; Bertin 2006; Bertin et al. 2002) and the zero point was determined using color-corrected 2MASS photometry converted to the MKO system with Leggett et al. (2006) color transformations.

NIR photometry follow-up was also made on 2011 November 5 at the 1.6 m telescope of the Observatoire du Mont-Mégantic, with the NIR camera CPAPIR (Artigau et al. 2004) in queue mode (Artigau et al. 2010). A set of 270 K_s-band images, each with two 10.1 s coadds were taken for a total exposure time of 91 minutes. A standard image processing (same pipeline as described in Artigau et al. 2011) was performed and yielded K_s = 17.10 ± 0.15 for the object. The resulting K_s band versus Wide-field Infrared Survey Explorer (WISE) colors (see Section 2.2.3) suggested a T dwarf spectral type and prompted both additional photometric observations at the CFHT/WIRCam and spectroscopic follow-up with Gemini-North/GNIRS.

The second and third epoch of photometry with WIRCam in J were thus acquired on 2011 December 26, 28, and 29 and on 2012 September 7. Images in Y, H, and K_s were also acquired on 2012 September 7. The observation strategy was the same as the one explained above, with total integration times of 45.8 and 30.4 minutes, respectively, in the two J-band epochs and 30.4, 19.0, and 8.3 minutes for the Y, H, and K_s stacks, respectively. Single exposures were 150 s, 50 s, 15 s, and 25 s for Y, J, H, and K_s, respectively. The Y-band zero point was determined through the observation of a spectrophotometric standard. The 2011 October and 2012 September J-band images allowed precise multi-epoch astrometric measurements: a linear astrometric solution was determined for each based on the 2MASS point source catalog (Cutri et al. 2003).

In the WISE All-Sky Source Catalog⁹ (Cutri et al. 2012), there is a source detected at the position of the companion with W1 = 15.818 ± 0.064 and W2 = 15.039 ± 0.120. However, the WIRCam, GMOS, and CPAPIR images also show a faint extended object, most likely an edge-on spiral galaxy, ∼3''  southeast of GU Psc b. The inset of Figure 1 shows the position of GU Psc b and of the galaxy at the epoch of the WISE measurements (red circles), overlaid on the WIRCam K_s-band image. The position of GU Psc b was deduced from WIRCam images and the proper motion in Table 1. The yellow and cyan plus signs show the WISE W1 and W2 fluxes barycenters, respectively. They both fall within 1σ on the line joining GU Psc b and the interloping galaxy, which confirms that the WISE photometry is a blend of the two objects. The position along the line allows one to derive the relative contribution of each object to the blended flux of the catalog. The galaxy contributes 71% ± 10% of the W1 flux and 46% ± 10% of the W2 flux. The resulting photometry for GU Psc b, listed in Table 1, is overall consistent with mid-T photometry for field objects (e.g., Kirkpatrick et al. 2011; Dupuy & Liu 2012).

The spectrograph GNIRS on Gemini-North was used in its cross-dispersed mode to obtain an NIR 0.9–2.4 μm spectrum with a resolving power of R ∼800. We used the 0675 slit, the short (015 pixel⁻¹) camera and the 31.7 l mm⁻¹ grating. The target observations, acquired on 2012 November 12, were followed by the observation of an A0 spectroscopic standard star (HIP13917) to calibrate the flux and correct for telluric absorption lines. A total of 1.2 hr of observation was taken, subdivided in five 180 s exposures. The reduction was carried with the pipeline presented in Delorme et al. (2008) and Albert et al. (2011). Wavelength calibration was made using bright OH sky lines (Rousselot et al. 2000). To perform the flux calibration, the spectrum was integrated over the WIRCam Y, J, H, and K_s bandpasses, respectively.¹⁰ For each filter, we then evaluated the factor by which one must multiply the integrated flux to get the WIRCam photometric measurements. A linear fit of the factors was used to rectify every wavelength of the spectrum. The calibrated spectrum is shown in Figures 6, 8, and 9, along with photometric points outside the range covered by the spectrum (z, W1, and W2).

On 2013 November 19, we used the sodium Laser Guide Star (LGS) AO system of the Keck II Telescope (Wizinowich et al. 2006; van Dam et al. 2006), located on the summit of Mauna Kea, in Hawaii, to verify if the companion is a resolved binary. The LGS constitutes the reference wavefront for the AO correction, and the tip-tilt was monitored using the close (264 away) star SDSS J011237.05+170456.9 (r_AB = 15.742 ± 0.004; SDSS DR 9; Ahn et al. 2012). We used the near-infrared camera NIRC2 in its narrow mode (FoV 10''× 10'', pixel scale of 9.942 mas pixel⁻¹) for the K-band images and in its wide mode (FoV 40''× 40'', pixel scale of 39.686 mas pixel⁻¹) for the H-band images. We obtained 12 images of two coadditions of 60 s in K band, for a total integration time of 24 minutes, and nine images of four coadditions of 30 s in H band, for a total integration time of 18 minutes. The images were obtained using a three-point dither pattern that avoided the noisy quadrant in the lower-left portion of the array. The positional offset between images varied between 15 and 3''. Conditions were photometric during the observation. Standard reduction techniques were used: images were divided by the dome flat and a median sky image constructed from all the observations of the night was subtracted. Then every image was shifted and stacked to produce a final image in each band (see Figure 2).

Zoom In Zoom Out Reset image size

GU Psc was originally identified as a highly probable M dwarf by Zickgraf et al. (2003). Then, Riaz et al. (2006) identified it as an M3 ± 0.5 using the TiO5 index (Reid et al. 1995) and obtained a visible spectrum that allowed the measurement of its Hα emission line equivalent width (EW), which is a proxy for its chromospheric activity, and thus youth.

Through Bayesian inference, GU Psc was then identified by Malo et al. (2013) as a highly probable (98%) member of the ABDMG. The Bayesian analysis makes use of a priori knowledge of known associations (Galactic position, space velocity, and I_C − J absolute photometry) and compares these properties with observables of a given candidate (sky position, proper motion, and I_C and J apparent magnitudes). The analysis gives as an output the membership probability of the star for every association considered as well as the probability that it is a field star unrelated to these associations. It also gives the most probable radial velocity the candidate should have if it were a true member of a given association (accurate to a few km s⁻¹) and the most probable statistical distance d_s it would have. Malo et al. (2013) showed that this statistical distance agrees with the trigonometric distance within ∼10% for bona fide members of the associations. For GU Psc, using the average proper motion shown in Table 1, they derived d_s = 48 ± 5 pc, which is the value we adopt for the distance hereafter and predicted a v_rad = −1.5 ± 1.9 km s⁻¹. Using any of the proper motion measurements available in the literature does not change these results significantly. As part of a comprehensive radial velocity follow-up program (Malo et al. 2014), multi-epoch radial velocity measurements of GU Psc through NIR and optical spectroscopy were secured.

The results, summarized in Table 2, yield a weighted average radial velocity of 〈v_rad〉 = −1.6 ± 0.4  km s⁻¹, in excellent agreement with the predicted radial velocity for membership in ABDMG. Adding the radial velocity observable to the Bayesian analysis yields an increased membership probability of 99.9% for ABDMG. These observations also show that GU Psc is a relatively fast rotator with a vsin i of 23 km s⁻¹.

Note that, as mentioned in Gagné et al. (2014), the probabilities mentioned here should not be interpreted as absolute. Even though any given star is a priori less likely to belong to a given association than to the field (there are much less stars in the association than in the field), the precise values of these prior probabilities are very uncertain so all hypotheses are considered as equally likely in the analysis of Malo et al. (2013). They estimate that for the candidates of ABDMG with a membership probability over 90%, the contamination rate (false positive) is about 14%.

We also used the analysis of Gagné et al. (2014) to evaluate GU Psc's membership. This analysis differs from that of Malo et al. (2013) in two major aspects. First, it uses a different method to outline the three-dimensional regions covered by the bona fide members of an association in the Galactic position space X, Y, Z and in the Galactic velocity space U, V, W. While Malo et al. (2013) uses ellipsoids with the three axes aligned on the Galactic coordinate system, Gagné et al. (2014) use a more realistic approach where the ellipsoids can be aligned in any direction. Second, Gagné et al. (2014) uses the knowledge of the distance of known members of a given association as an additional prior on the plausible distance that a candidate can have. In agreement with Malo et al. (2013), this analysis points toward a membership in ABDMG for GU Psc, albeit with a smaller probability (88%). It finds a compatible statistical distance of 47 ± 5 pc and a very similar predicted radial velocity (−1.8 ± 2.0 km s⁻¹). It also yields a non-negligible probability of 12% for the membership in the younger (12–22 Myr) β Pictoris Moving Group (βPMG), associated with a smaller statistical distance, d_s = 32 ± 3 pc.

Thus, both analyses suggest that GU Psc is a member of a young association, either ABDMG or βPMG, with a much higher probability for the former.

The age of ABDMG, first identified as a moving group by Zuckerman et al. (2004), is subject to debates. The comparative analysis of ABDMG and the open cluster IC 2391 in an M_V versus V − K color–magnitude diagram led Luhman et al. (2005) to derive an age between 75 and 150 Myr, roughly coeval with the Pleiades (for which they adopted an age of 100–125 Myr). Using an M_I versus V − I diagram and the lithium EW, Lopez Santiago et al. (2006) formulated the hypothesis that the group could be composed of two subgroups: one younger (30–50 Myr) and one older (80–120 Myr). The Li EW was also used in two other studies to deduce a lower limit on the age of 45 Myr (Mentuch et al. 2008) and an age of 70 Myr (da Silva et al. 2009). Recently, Barenfeld et al. (2013) studied the kinematics and the abundance of 10 different elements in 10 members of the "stream" (i.e., stars that are not among the nine stars considered as the "nucleus" by Zuckerman et al. 2004) and pointed out that many stars traditionally associated with ABDMG do not have a similar chemical composition and/or were not likely formed at the same position as the ABDMG nucleus. They concluded that the remaining members are at least 110 Myr, based on the fact that the group still has zero-age main sequence K stars members. Considering all these studies, we adopt a conservative age of 100 ± 30 Myr for ABDMG as a whole.

To better constrain the age of GU Psc, we consider here other age indicators. Table 3 summarizes all the information on the age of the GU Psc system.

Over time, the coronal activity that is induced by magnetic field is reduced, which causes the X-ray emission to decrease (Preibisch & Feigelson 2005). Using GU Psc's X-ray count rate and hardness ratio (HR1) measured by ROSAT (Voges et al. 1999; see Table 1) in the relation given in Schmitt et al. (1995) yields a value of 4.90 × 10⁻¹³ erg s⁻¹ cm⁻² for the X-ray flux, thus an X-ray luminosity of log  L_X = 29.1 ± 0.3 erg s⁻¹ at d_s = 48  ±  5 pc. This X-ray luminosity is very similar to that of single low-mass ABDMG members; if we use a similar procedure to evaluate the log  L_X of the six bona fide M dwarfs members listed in Malo et al. (2013), we obtain log  L_X = 29.03 erg s⁻¹, with a dispersion of 0.07 dex. The X-ray luminosity of GU Psc is also consistent with that of other candidate members of ABDMG, such as the M3.5 star J01225093−2439505 (hereafter 2M0122) that has log  L_X = 28.7 ± 0.2 erg s⁻¹ (Bowler et al. 2013), or the M2 star J235133.3+312720, that has a log  L_X = 29.3 ± 0.2 erg s⁻¹ (Bowler et al. 2012). GU Psc's X-ray luminosity is however significantly higher than that of field stars of similar mass. For example, the log  L_X of the 42 single field M dwarfs listed in Malo et al. (2014) has a mean of log  L_X = 27.6 erg s⁻¹ with a dispersion of 0.5 dex. At the statistical distance for the βPMG given by Gagné et al. (2014) analysis (d_s = 32 ± 3 pc), GU Psc's X-ray luminosity would be log  L_X = 28.8 ± 0.3 erg s⁻¹, which is also not consistent with that of βPMG members: the mean value computed for the nine single bona fide M dwarfs of the βPMG in Malo et al. (2013) is log  L_X = 29.63 erg s⁻¹, with a dispersion of 0.16 dex. The X-ray activity thus favors an ABPMG membership and not a βPMG membership.

The reduction of the magnetic activity occurring as the star evolves is also traceable by the diminution of the Hα emission line at 6562.8 Å. In our visible spectrum (Figure 3(a)), we measure EW_Hα = −3.96 Å. According to West et al. (2008), the activity lifetime of an M3 is 2 ± 0.5 Gyr, thus the presence of Hα in emission implies that GU Psc is likely younger than this. Also, according to the criteria developed in White & Basri (2003), the 10% width of the same Hα emission line (W10% = 125 km s⁻¹) is consistent with a star that is non accreting (W10% < 270 km s⁻¹), thus older than ∼10 Myr (Barrado y Navascués & Martín 2003). This lower limit is also consistent with the fact that no disk is seen in the form of a mid-infrared excess. Indeed, when the J, H, K_s bands are fitted with a model spectrum (BT-Settl AGSS2009; Allard et al. 2012), the WISE photometry in the four bands falls directly on the model spectral energy distribution (SED). These three indicators are consistent with a membership in either ABDMG or βPMG.

Zoom In Zoom Out Reset image size

The absence of the lithium absorption line at 6708 Å in the optical spectrum of GU Psc (EW_Li ≲ 18 mÅ, 3σ; see Figure 3(b)) also yields a minimum age for GU Psc. Indeed, the early M members of ABDMG show lithium measurements compatible with this upper limit (Mentuch et al. 2008; Yee & Jensen 2010). Our upper limit is also compatible with the wide range of lithium absorption that the low-mass stars show at the younger age (12–22 Myr) of βPMG, which varies from an upper limit of a few tenths of mÅ up to ∼500 mÅ (Mentuch et al. 2008; Binks & Jeffries 2014). Indeed, Figure 5 of Mentuch et al. (2008) shows that the youngest association for which early M stars show little or no lithium is βPMG; there is no early M star with EW_Li ≲ 350 Å in the TW Hydrae Association (TWA) nor in the η Chamaelontis Cluster, which are both ≲20 Myr (e.g., Fernández et al. 2008). We can thus estimate that GU Psc is likely older than ∼22 Myr.

For an object with a mass above the fully convective limit (≳0.35 M_☉, Chabrier & Baraffe 1997), the rotation rate increases as the star contracts toward the main sequence, reaches a plateau at an age comparable to that of ABDMG, and then slows down due to various interactions (Sills et al. 2000). The rotation period can thus help to constrain the age. According to the SuperWASP photometric survey, GU Psc is a variable star with a 1.0362 ± 0.0005 day period (Norton et al. 2007). That could be indicative of a relatively fast rotator, which is also suggested by its relatively large vsin i of 23 km s⁻¹. If GU Psc is not fully convective (>0.35 M_☉), the ∼1 day rotation period suggests an upper limit on the age of ∼650 Myr (see Figure 12 in Irwin et al. 2011). With the I_C − J listed in Table 1 and the age of ABDMG (100 ± 30 Myr), GU Psc's mass M_⋆ is estimated to be between 0.30 and 0.35 M_☉ (using the models of Baraffe et al. 1998), which is very close to the limit for a star to be fully convective. Without a parallax, it is challenging to determine whether or not GU Psc has a fully convective structure. If GU Psc is fully convective, the upper limit of the age we can set using the rotation period is much greater, since the spin-down time of fully convective objects is very long (>5 Gyr; Irwin et al. 2011).

It is interesting that HIP 17695, the only single bona fide member of ABDMG with a spectral type similar to GU Psc (M2.5; Malo et al. 2013), has a vsin i (18 km s⁻¹; da Silva et al. 2009) and rotation period (P_rot = 3.87 days; Messina et al. 2010) that are close to those of GU Psc. This object is probably at the low-age end of ABDMG (∼70 Myr), given its X-ray luminosity, Hα and Li EW (Zuckerman et al. 2004).

Considering that the Bayesian analysis favors a membership in ABDMG, and that other youth indicators suggest an age consistent with that association, we adopt hereinafter the age of ABDMG (100 ± 30 Myr) and the associated statistical distance (d_s = 48 ± 5 pc) for GU Psc's system.

We evaluated the metallicity of GU Psc using two metallicity calibrations developed recently, specifically for M dwarfs. One must bear in mind however that these calibrations were developed for field stars and are not necessarily appropriate for young stars. Newton et al. (2014) calibration is based on the Na line at 2.2 μm. For GU Psc, we obtain [Fe/H] = +0.10 ± 0.13. Mann et al. (2013) improves previous calibrations (notably Terrien et al. 2012 and Rojas-Ayala et al. 2012) and presents metallicity calibrations based on various features in optical and NIR. We obtained [Fe/H]_H = −0.14 ± 0.09 with the H-band calibration and [Fe/H]_K = 0.04 ± 0.08 with the K-band calibration.¹¹ While the value derived using the H band is slightly lower than the others, the ones derived with Mann et al. (2013) and Newton et al. (2014) using K band are consistent, within uncertainties, with each other and with the one derived by Barenfeld et al. (2013) for 10 ABDMG members, [Fe/H] = +0.02 ± 0.02.

The high-contrast imaging and high-resolution spectroscopy observations we made on the primary provide strong constraints on the mass ratio and separation of a possible planetary-mass, brown dwarf, or stellar companion (see Figure 4).

Zoom In Zoom Out Reset image size

First, high-resolution spectroscopy shows a single-line profile. Considering the measured projected rotational velocity of GU Psc (vsin i ∼ 23 km s⁻¹; Table 2), this excludes a double-line binary (SB2) with components showing Δvsin i > 23 km s⁻¹. Second, the multi-epoch radial velocity measurements obtained with ESPaDOnS and CRIRES (Table 2) are stable at the km s⁻¹ level and consistent with the value predicted for an ABDMG member in GU Psc's line of sight within <1 km s⁻¹, which excludes many cases of single-line binary (SB1). We adopt a conservative upper limit on GU Psc's radial velocity semi-amplitude of vsin i < 5 km s⁻¹, since a larger discrepancy between the measured radial velocity and the one predicted for GU Psc in ABDMG would correspond to a 3σ outlier. Note however that for both spectroscopic constraints (vsin i > 23 km s⁻¹ and vsin i > 5 km s⁻¹), a near-equal luminosity binary system would not be detected for nearly pole-on geometries. In the case of the second constraint (vsin i > 5 km s⁻¹), a companion with the same luminosity as GU Psc would not be detected either. The blue and cyan regions of Figure 4 show the mass ratio and separation ranges excluded by these two constraints, respectively, for an orbit orientation of i ∼ 90°.

At greater separations, no companions (other than GU Psc b) were found through imaging observations with NICI or GMOS. Inside NICI's semi-transparent mask, which absorbs ∼6 mag in the central 032 radius region (15 AU at d_s = 48 pc), no object is seen at separations greater than one FWHM of the point-spread function (008, or 4 AU), down to a flux ratio of 4 (1.5 mag). The NICI data were taken with the 4% CH₄ on and off filters within the H band; a companion 1.5 mag fainter than GU Psc would have M_H = 9.3 which corresponds to a ∼M6 spectral type (Dupuy & Liu 2012) and a temperature 400 K cooler than GU Psc (i.e., respectively ∼3300 K and ∼2900 K, the difference being more accurate than the absolute values, see Figure 5 in Rajpurohit et al. 2013). Using the BT-Settl model (AGSS2009; Allard et al. 2012) at an age of 100 Myr, a difference of 400 K for a primary star in the 3000–3500 K range leads to a maximal mass ratio of 32%–54%. The orange region of Figure 4 shows the excluded region from this observation, adopting 50% as a conservative upper limit on the mass ratio. NICI imaging beyond the edge of the mask shows no background companion out to a separation of 9'' (430 AU) at a 5σ contrast of ΔH ∼ 12, yielding an upper limit of M_H ∼ 18, or a mass limit of ∼5  M_Jup (green region of Figure 4). The GMOS imaging has a 5σ z-band limit at 23.2, which translates to a mass limit of ∼7  M_Jup and, thus, a mass ratio of 2.3% (red region of Figure 4).

Malo et al. (2013) mentioned the possibility that GU Psc might be a binary star. Indeed, the absolute magnitude GU Psc assuming d_s = 48 pc, M_J = 6.80, is about 0.78 mag brighter than the one predicted for a single ABDMG member with GU Psc's color (I_C − J = 1.44; see the M_J versus I_C − J color–magnitude diagram on Figure 3 in Malo et al. 2013). The magnitude dispersion along the ABDMG empirical sequence for that I_C − J (∼0.5 mag; Malo et al. 2013) is important. A limited number of bona fide low-mass members are known in this association and there is probably an intrinsic age spread among the members. Thus, this over luminosity may not be significant. If it is, it could be due to an unseen companion, but it could also be attributed to other factors (for example, to an important chromospheric activity; Riedel et al. 2011).

In conclusion, the various data sets we obtained pose stringent constraints on the presence of another companion around GU Psc. The near-equal luminosity binary scenario can be virtually excluded at all separations, unless a rather unlikely geometry is invoked, such as a near pole-on geometries for SB1 or SB2 cases or an alignment of a companion behind the star at the time of the high-resolution imaging observations. Besides, there is still a possibility of a stellar companion with a maximum mass of about half that of GU Psc's between 1 and 10 AU or of a brown dwarf companion inward of ∼10 AU.

Figure 5 shows the proper motion of GU Psc and of GU Psc b. For the primary, there are several proper motion measurements reported in the literature (Roeser et al. 2010, 2008; Ducourant et al. 2006; Ahn et al. 2012; Schlieder et al. 2012). Since these measurements are not independent, we adopt the mean value and use the average uncertainties as the error of the resulting mean: μ_αcos δ = 90 ± 6 mas yr⁻¹, μ_δ = −102 ± 6 mas yr⁻¹. For the companion, we used the two J-band epochs taken at CFHT/WIRCam in 2011 October and 2012 September (11 months apart) and found μ_αcos δ = 98 ± 15 mas yr⁻¹, μ_δ = −92 ± 15 mas yr⁻¹. The figure also shows the proper motions of stars in the WIRCam image field, computed like that of GU Psc b. The companion proper motion is clearly consistent with that of the star and inconsistent with that of the other stars in the field, thus confirming it is gravitationally bound.

Zoom In Zoom Out Reset image size

This conclusion is strengthened by the fact that the probability of finding an unassociated T dwarf with this proper motion in the vicinity of a young M dwarf is very low. To assess this probability, we first consider only proper motion. Using the 64 T dwarfs in the Dupuy & Liu (2012) sample (Table 9) that have a parallax and proper motion with errors smaller than 10%, we compute a median sky plane velocity, 35 km s⁻¹, which we assume to be typical of T dwarfs. Assuming an isotropic random Gaussian distribution of velocities, this velocity corresponds to a Gaussian dispersion per coordinate of 30 km s⁻¹, which translates back into a dispersion of 130 mas yr⁻¹ at 48 pc. A numerical integration of a two-dimensional Gaussian shows that only ∼1.5% of random T dwarfs would match GU Psc's proper motion at the <2σ level. Then, we consider the sky position. The local density of T0–T5.5 dwarfs is $1.4^{+0.3}_{-0.2}$ × 10⁻³ pc⁻³ (Reylé et al. 2010), there are therefore ∼730 early T dwarfs within 50 pc of the Sun. Our survey sampled a near-circular field with a 55 diameter around 91 stars, covering a total area of 0.6 deg² or 1.5 × 10⁻⁵ of the entire sky. There is thus only a ∼1% chance of finding a random single field early-T dwarf with properties comparable to GU Psc b within a GMOS FoV of one survey stars. The combined likelihood of a false positive match in both proper motion and position is therefore of the order of 2 × 10⁻⁴, and it is likely to be much lower as GU Psc and GU Psc b both display signs of youth.

The H- and K-band observations of the companion made with LGS AO at Keck show only one object. The companion is thus not a resolved binary. The K-band image excludes, at the >5σ level, any second object that would have a ΔK ≲ 4 for separations between 008 and 10'' (∼4–480 AU at 48 pc). Using M_K ∼ 14 for GU Psc b, this ΔK corresponds approximately to a T8 or later, according to the polynomial relations given in Dupuy & Liu (2012). For smaller separations, the K-band observations exclude a companion brighter than a typical T7.5 down to a separation of 004 (∼2 AU). The wider H-band image does not show any object (besides a few galaxies) down to a ΔH of ∼3.6, in a radius between 07 and the width of the field, 40'' (∼30–1900 AU at 48 AU). Considering M_H = 14.3 for GU Psc b, this excludes a companion earlier than ∼T7.5 in that region.

GU Psc b could still be a very tight binary object, but these observations largely exclude a T5–T6 companion in a wide range of distances and down to the typical separations of T dwarfs binaries (e.g., SDSS J153417.05+161546.1, a T1.5+T5.5 with a separation of 3.96 AU; Liu et al. 2006, SDSS J102109.69−030420.1, J1021 hereinafter, a T1+T5 with a separation of 5 AU; Burgasser et al. 2006b, or Indi B ab, a T1+T6 separated by 2.65 AU; McCaughrean et al. 2004).

Figure 6(a) shows the NIR spectrum and W1 and W2 photometry of GU Psc b compared with other objects. The upper panel of the figure shows the spectral standards T2, T3, and T4. The GU Psc b spectrum was smoothed over eight points to a final resolution of λ/Δλ ∼ 400 to ease comparison. The spectra are from the L and T dwarf data archive.¹² The T2 and T4 (J125453.90−012247.4, hereafter J1254, and J225418.92+312349.8) are the standards identified in Burgasser et al. (2006a). J120602.51+281328.7 was used as the T3 standard instead of J120956.13−100400.8, which was found to be a binary by Liu et al. (2010). All spectra are normalized to their value at the peak of the J band and are offset vertically for clarity.

Zoom In Zoom Out Reset image size

The global comparison of the SEDs suggests that GU Psc b is of a spectral type between T2 and T4. The absorption of CH₄ in H band is intermediate between that of a T2 and a T4, close to a T3, while the blue side of the H band is better reproduced by the T4 standard. The Y- and J-band flux are also closer to the T4, even though in both cases the side of each peak is slightly underluminous. The K band is not similar to any of the standards and is clearly brighter than the average T3 flux. That is probably explained by the collision-induced absorption (CIA) by H₂ that affects this region of the spectrum; CIA is expected to be reduced for objects with a lower gravity and/or greater metallicity (Saumon et al. 2012). The W1 flux is lower for GU Psc b than for any of the standards, and the W2 flux is stronger than the T4, closer to the T2.

We used the GU Psc b spectrum to compute the spectral indices defined in Burgasser et al. (2006a) and establish the spectral type. The position of the wavelength ranges used and the values derived for the indices are listed in Table 4, along with the associated spectral types for the five indices for which a quantitative scale exists. The K/J index defined by Burgasser et al. (2006a) measures the flux ratio between the K and the J band to evaluate the strength of the CIA-H₂ feature that is known to be sensitive to surface gravity. It has no quantitative scale, but the K/J of GU Psc b is stronger than for a typical T3.5, closer to that of a T2.5 (see K/J versus CH₄ −H diagram shown on Figure 10 of Burgasser et al. 2006a). The W_J index quantifies the width of the J-band peak. The value of GU Psc b is similar to that we computed for the T4–T5 objects present in the L and T dwarf data archive. The resulting spectral types vary between T3 and T5, with an average spectral type of ∼T4.

Altogether, the comparison of the spectrum and the indices suggest that GU Psc b is a T3.5 ± 1 spectral type, with indication of low gravity, and/or of high metallicity.

Even though GU Psc b is not a resolved binary object according to Keck NIRC2 observations (see Section 3.2.2), the values of its indices, within uncertainties, satisfy two to three of the six spectral index selection criteria developed by Burgasser et al. (2010a) to identify binary systems. GU Psc b would thus qualify as a candidate binary in that scheme (weak candidates satisfy two criteria, and strong ones satisfy three criteria or more). Indeed, as shown on Figure 6(b), the GU Psc b spectrum is very similar to that of J1021 (also shown on Figure 6(b)), which was the initial standard for the T3 spectral type (Burgasser et al. 2002), until it was confirmed through Hubble Space Telescope NICMOS imaging to be a 0172 ± 0005 (5.0 ± 0.7) AU binary composed of a T1+T5 (Burgasser et al. 2006b). The GU Psc b spectrum is also very similar to J121440.95+631643.4 (J1214 hereinafter), a T3.5 discovered by Chiu et al. (2006). Geißler et al. (2011) identified J1214 as a candidate binary because they obtained a better fit with a composite spectrum made of T2 and T6 templates than for a T3 template. Although GU Psc b remains significantly redder in J − K_s, the Y and J bands are closely matched, and the fit is much better in H band. In both cases, the only notable difference lies in the K_s-band spectrum. If GU Psc b is truly a single object, this suggests that it is slightly peculiar when compared to standards, perhaps due to the lower gravity expected for a relatively young object, or to some variation in metallicity or cloud properties.

To further constrain the physical properties of GU Psc b, we compared its NIR spectrum and its z, W1, and W2 photometry to the synthetic SEDs of two different sets of brown dwarf atmosphere models: the BT-Settl CIFIST model presented in Allard et al. (2013)¹³ and the model presented in Morley et al. (2012), which include low-temperature condensates (primarily sulfides). To determine quantitatively the best fit to this mix of photometric and spectroscopic data points, we used a method similar to that presented in Cushing et al. (2008) in order to compute a goodness-of-fit G_k for each model k that is minimized for the best fitting models:

$$\begin{equation} G_{k}=\sum _{i=1}^{n}W_i\left(\frac{F_{{\rm obs},i}-C_{k}F_{k,i}}{\sigma _{{\rm obs},i}}\right)^{2}. \end{equation} \tag{ 1 }$$

In this equation, F_{obs, i} is the flux observed in a given spectral range i. The associated uncertainty, σ_{obs, i}, is dominated by the ∼5% photometric zero-point uncertainty. The total number of spectral ranges is n: one per photometric filter or spectral point (after smoothing). F_{k, i} is the flux of the synthetic model over the same wavelength domain. The synthetic model spectra are convolved at the resolution of the spectrum and the synthetic photometric magnitudes are obtained using instrumental filter profiles.¹⁴ Rather than using W_i = Δλ = λ₂ − λ₁ for the weight (as done in Cushing et al. 2008), we chose to use W_i= Δlog λ = log (λ₂/λ₁) to purposefully not be biased by the arbitrary choice of working in wavelength space rather than frequency space. The scale C_k is equal to the dilution factor R²/d² for a source of radius R at a distance d. For each model, we constrained C_k using the central distance inferred from Bayesian statistical analysis for GU Psc in ABDMG, d_s = 48 pc, and the radius prescribed at the given T_eff and log g by evolution models (Saumon & Marley 2008 for the low-temperature cloud model and Baraffe et al. 2003 ¹⁵ for BT-Settl).

For the low-temperature cloud model described in Morley et al. (2012), we used a grid with temperatures between 700 K and 1300 K (ΔT_eff = 100 K) and log g between 4.0 and 5.5 (Δlog g = 0.5), at solar metallicity. We also tried different values for the sedimentation efficiency f_sed (1–5) and for K_zz, which quantifies departure from chemical equilibrium (0 and 10⁴ cm² s⁻¹). For the BT-Settl model, we used the CIFIST grid presented in Allard et al. (2013), computed with temperatures between 700 K and 1400 K (ΔT_eff = 50 K) and log g between 3.5 and 5.5 (Δlog g = 0.5), also at solar metallicity.

Figure 7 shows the goodness-of-fit map for both sets of models in the temperature/log g parameter space. The physical parameters that lead to the best fit between the observed spectrum and the models are the same for both sets, T_eff ∼ 1000–1100 K and log g ∼ 4.5–5.0. The observed SED of the object constrains the bolometric luminosity, there is thus a correlation between temperature and gravity for the best fits: they are achieved either at a low temperature (T_eff = 1000–1050 K) and low surface gravity (log g = 4.5) or at higher temperature (T_eff = 1100 K) and greater gravity (log g = 5.0). In the case of BT-Settl, even higher temperature (T_eff = 1200 K) and surface gravity (log g = 5.5) still give a good fit, although these physical parameters are not consistent with the age of ABDMG (see Section 3.2.5). Figure 8 shows the GU Psc b SED and the two best synthetic spectra for both sets of models.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

We also did the entire fitting process constraining the C_k with the extreme values of the distance range, 43 pc and 53 pc. The effective temperature and surface gravity we obtained did not vary significantly.

Both synthetic models match the SED of GU Psc b between 0.9 and 5 μm remarkably well, especially considering that there is no adjustment of the absolute flux of the models (C_k is constrained). Overall, BT-Settl fits the K band (especially its red side) and the Y band around 1 μm slightly better than the low-temperature cloud model, but the later one better reproduces the J − H color. The methane band at 1.6 μm is typically poorly matched by the models which tend to overestimate the flux in this wavelength range (see e.g., HN Peg b SED on Figure 4 of Leggett et al. 2008). Here, we observe an opposite trend: the flux redward of 1.7 μm is too low in both sets of models compared to GU Psc b. This excess flux, which is also seen in the comparison of GU Psc b to the T3 and T4 templates on Figure 6(a), could be indicative of a slight departure from solar metallicity or an inhomogeneous surface, or an indication that GU Psc b is a very tight, unresolved, binary object.

The good fit we obtain with Morley et al. (2012) model is somewhat surprising, considering our object is hotter than the targeted objects for this model. This model came as an attempt to explain why late T dwarfs (T_eff ≲ 900 K, primarily) were not perfectly fitted with models without iron and silicate clouds. Using Ackerman & Marley (2001) cloud model, they studied the influence of other condensates (sulfides mainly, e.g., Na₂S, MnS, and ZnS, but also KCl and Cr) that were previously ignored and realized they are important at low temperatures (between 400 and 1300 K). Since the targeted objects are late T dwarfs, Morley et al. (2012) models do not include the iron and silicate clouds that are important for L dwarfs since they are thought to be rapidly clearing at the L/T transition. In models including such condensates (but not the ones included in the Morley et al. 2012 model), early T such as GU Psc b are expected to be best described by thin clouds of larger particles (corresponding to a high f_sed parameter in the Ackerman & Marley 2001 model). This is the case, for example, for HN Peg b (T2.5), for which the best fit is obtained with f_sed = 3.5 (Leggett et al. 2008) or for J1254 (T2) or 2M J05591914−1404488 (T4.5), which were best fit with f_sed = 3 and 4, respectively, in Cushing et al. (2008). Alternatively, our best fits with Morley et al. (2012) model are obtained using f_sed = 1, thus very thick clouds of the less abundant sulfide condensates provide the moderate dust opacity evident in our T3.5 object. Figure 9 compares one of our best fits using f_sed = 1 (in blue) and a much poorer fit with the same temperature/log g but with thinner sulfide clouds (f_sed = 3, in green).

Zoom In Zoom Out Reset image size

The second parameter is the Eddy diffusion coefficient, K_zz that characterizes the vertical transport in the atmosphere. Vigorous vertical transport can bring molecular species from deeper, hotter layers of the atmosphere to the upper, cooler layers on a time scale faster than that of some chemical reactions, driving the molecular abundances away from their local equilibrium values. In particular, this results in increased CO and CO₂ abundances, and reduced CH₄, H₂O, and NH₃ abundances in the upper atmosphere (Lodders 2002; Saumon et al. 2006; Burningham et al. 2011). Figure 9 shows, for f_sed = 1, the two available values we tested for the Eddy diffusion coefficient: K_zz = 0 cm² s⁻¹ (chemical equilibrium, in blue) and K_zz = 10⁴ cm² s⁻¹ (in orange). It shows that K_zz has little impact on the Y and J bands, but that an increase in K_zz increases the flux in H, K and in the mid-infrared. The best fit at K_zz = 0 cm² s⁻¹ is obtained at T_eff = 1100 K and log g = 5.0 (G = 1.7, Figure 8, in blue) while at K_zz = 10⁴ cm² s⁻¹, the best fit is obtained at a slightly lower temperature and log g: T_eff = 1000 K and log g = 4.5 (G = 3.6, Figure 9, in orange). The depth of the 4.6 μm absorption band in W2 is reproduced a bit better with a higher values of K_zz. We must caution that we did not try a higher K_zz for values of f_sed different than 1.

The BT-Settl model treats condensation and sedimentation of all dust species as well as gas phase advection by relating to a single atmospheric velocity field derived from radiation hydrodynamical models. There are thus no adjustable cloud parameters in this model. The cloud sedimentation and Eddy diffusion are instead determined by direct comparison of the relevant timescales (for condensation, sedimentation, chemical reactions) to the mixing timescale derived from this velocity field. The cloud opacity is composed of a number of condensates, which are settling to various degrees. Although the current version of the model does not yet include the opacity contribution of all low-temperature condensates that are included in Morley et al. (2012) (most notably Na₂S), the BT-Settl model reproduces the observed spectrum of GU Psc b similarly well. This is because even if these low-temperature clouds start to form at high altitude for L/T transition objects like GU Psc b, they do not become optically thick in this effective temperature range. The continuum of the flux peaks is still shaped by the silicate clouds, even if they have receded relatively deep into the photosphere.

The excess absorption at 4–5 μm apparent for this model could be indicative of an overestimation of the diffusion efficiency, resulting in too much CO and CO₂ being mixed into the photosphere. The mixing derived at the transition from CO- to CH₄-dominated atmosphere regions yields a diffusion coefficient of ∼10⁵–10⁶ cm² s⁻¹. It thus produces more CO and CO₂ absorption at 4–5 μm than the K_zz = 0 or the K_zz = 10⁴ models of Morley et al. (2012). Alternatively, the excess flux in the CH₄ absorption band in W1 might also reveal the incompleteness of the currently used CH₄ opacities, which cover only a fraction of the lines relevant at temperatures above 1000 K (Yurchenko & Tennyson 2014).

We investigated the effect of leaving the scale C_k free in the fitting. For both synthetic models, for a given set of parameters, we then obtain a similar or slightly better fit than when imposing the scale. Nonetheless, the best fit occurs for the same physical parameters than with a constrained C_k. This strengthens our confidence both in the radii predicted by evolutionary models and in the distance inferred by the Bayesian statistical analysis.

Evolutionary models can be used to constrain the physical parameters (radii, surface gravity, bolometric luminosity, and mass) of GU Psc b. In the previous section, we showed that the best fit were obtained with T_eff/log g couples of 1000–1050 K/4.5 or 1100 K/5.0. Evolutionary models suggest that the lowest temperature and log g are the most likely; the age is then consistent with the age of ABDMG, 100 ± 30 Myr. The highest temperature and log g would imply ages older than ∼300 Myr. The age deduced from evolutionary models for 1200 K/5.5, which also led a good fit for BT-Settl, are greater than 1 Gyr and are clearly excluded for ABDMG age.

Thus, using ABDMG age (100 ± 30 Myr) and the range of plausible temperatures determined with atmosphere models (T_eff = 1000–1100 K) in evolutionary models, we obtain the ranges of values for the physical properties of GU Psc b presented in Table 5. We used the two different models presented in Saumon & Marley (2008): one with f_sed = 2, which is a good approximation for all cloudy models, and one without clouds. We present both results as limiting cases in Table 5, but since the atmosphere model fitting suggests a better match with clouds than without, the cloudy version is probably the most appropriate for GU Psc b. We also used the evolutionary model of Baraffe et al. (2003) and obtained similar results. In all cases, the values obtained for the bolometric luminosity are between log (L/L_☉) = −4.9 and −4.6. The surface gravity inferred (log g = 4.2–4.4) is consistent, albeit slightly lower, with the values derived from atmosphere model fitting (log g = 4.5–5). All models suggest a mass between 9 and 13  M_Jup. We thus find that GU Psc b is probably below the lower threshold of deuterium burning for its T_eff and age, unlike 2M0122 and several other brown dwarfs discussed in Bowler et al. (2013) that are possible deuterium burners.