First spectroscopic observations of the sub-stellar companion of the young debris disk star PZ Telescopii ^††thanks: Based on observations made with ESO Telescopes at the La Silla Paranal Observatory under programme ID 087.C-0109(A)

In 2010 two groups performing high-contrast direct imaging surveys around young nearby stars (Mugrauer et al. 2010; Biller et al. 2010) independently reported the discovery of a sub-stellar companion around the young solar analog pre-main-sequence star PZ Tel (HD 174429, HIP 92680), a likely member of the 12${}^{+8}_{-4}$ Myr old $\beta$ Pic moving group (Zuckerman et al. 2001; Torres et al. 2006) having a spectral type of G5 – K8 (Spencer Jones & Jackson 1936; Messina et al. 2010) at an age of 12.8 $\pm$ 2.2 Myr (Tetzlaff et al. 2011).

While not detectable in 2003 because of its proximity to the host star (Masciadri et al. 2005), the sub-stellar companion was first observed on its orbit in 2007 at a projected separation of 0.255 arcsec (13 au at 51.5 $\pm$ 2.6 pc (van Leeuwen 2007)), identified by Mugrauer et al. (2010) in archival data of Chauvin et al. (2010) after detection at 0.337 arcsec in 2009 by Mugrauer et al. (2010). Rebull et al. (2008) detected excess emission in the spectral energy distribution of PZ Tel at 70 $\mu$m with MIPS/Spitzer, indicating the star is surrounded by a low-mass, cold ($\sim$41 K) debris disk. On the basis of further astrometric data Mugrauer et al. (2012) could show that the preliminary determined orbital parameters of the PZ Tel companion are compatible with this disk being circumbinary (around both components), but not with being circumstellar, and that a semimajor axis of the nearly edge-on orbit not exceeding 25 au is likely.

Both Biller et al. (2010) using NICI at Gemini-South and Mugrauer et al. (2010) using NACO at ESO VLT concluded from photometric data in their discovery papers that the imaged object is likely a brown dwarf companion to the $\beta$ Pic moving group member PZ Tel. Biller et al. (2010) find best fitting parameters of 2702 $\pm$ 84 K and log g of 4.20 $\pm$ 0.11 dex, hence 36 $\pm$ 6 M_Jup, while Mugrauer et al. (2010) find 2500–2700 K, hence 28${}^{+12}_{-4}$ M_Jup, both using evolutionary models (Chabrier et al. 2000). Neuhäuser & Schmidt (2012) noted that given this lower mass limit, the PZ Tel companion could even be a planetary mass object, considering the planet definition by Schneider et al. (2011). Recently, Jenkins et al. (2012) derived an age of the system of 24 $\pm$ 3 Myr, based on evolutionary models and chromospheric activity of the primary, contrasting with their own age determination of 7${}^{+4}_{-2}$ Myr obtained from lithium depletion. This higher age estimate yields a higher effective temperature 2987 $\pm$ 100 K, surface gravity ($\log{g}$) 4.78 $\pm$ 0.10 dex and mass 62 $\pm$ 9 M_Jup. They moreover determined the metallicity [Fe/H] of the system to be 0.05 $\pm$ 0.20 dex.

In order to determine the basic parameters of the PZ Tel companion more precisely and independent of evolutionary models we obtained follow-up spectroscopy. Here we present our results from these observations, targeted to identify the nature of the sub-stellar companion.

We used the adaptive optics integral-field spectrograph SINFONI, mounted at UT4 of the ESO VLT. The observations of the PZ Tel companion were carried out in H+K band (resolution 1500) using the best resolving 0.0125 mas/spaxel scale of the instrument with a FoV of 0.8″ $\times$ 0.8″. Further details are summarized in Table 1.

We used the SINFONI data reduction pipeline version 2.0.5 offered by ESO (Jung et al. 2006) with reduction routines developed by the SINFONI consortium (Abuter et al. 2006). After standard reduction, including dark subtraction, flat fielding, distortion correction and wavelength calibration, all nodding cycles were combined to a final data cube.

We used the Starfinder package of IDL (Diolaiti et al. 2000) and an updated version of an algorithm to remove the halo of PZ Tel by 2 dimensional polynomial fit of the primary PSF, both described in detail in Seifahrt et al. (2007) and Schmidt et al. (2008). The data quality of July 2011 was too bad for the data to be used in general, however for the data of August 2011 we get the result presented as the second image in each row in Fig. 1. The technique was developed for wider companions and the primary being off the FoV, thus an overshooting polynomial removal is present after application.

Since the primary is in the FoV this additional information can be used for the removal of the primary halo. We use a technique called spectral deconvolution introduced by Sparks & Ford (2002) for HST data and used for the first time for ground-based data by Thatte et al. (2007). The idea is to remove speckles and other remnants, which move with wavelength across the cube and can be distinguished from the non-moving companion, by rescaling the cube according to the wavelength of each plane in the cube. Then the companion is at different positions at different wavelength/spectral channels in the new cube, while wavelength dependent speckles and remnants remain at the same positions and can be identified using a median and removed from the original cube. Please see Thatte et al. (2007) for details of the technique. It, however, only works if the wavelength coverage and the projected separation of the two components have an appropriate ratio to each other, ensuring that the companion moves enough after rescaling to remove its PSF by median combination of a small fraction of wavelength channels at either end of the data cube. Such a clean collapsed image devoid of flux from the companion can be constructed in our case as the $\epsilon$ value defined in Thatte et al. (2007) easily exceeds unity, being 1.83 for the projected separation of PZ Tel & its sub-stellar companion (see next section).

In a first step we have to remove the strongest part of the influence of the halo of the primary, as the halo full width strongly changes with wavelength because of the difference in adaptive optics (AO) performance at different wavelength. We achieve that by determination and removal of the radial symmetric part of the primary PSF, as can be seen in the third image of each row in Fig. 1. The center determination and reshifting of the PSF was improved even further to the results of Starfinder (Diolaiti et al. 2000) by fitting a polynomial of 2nd degree to the x shift values of the primary PSF from spectral channel to spectral channel in the cube to suppress any rapid changes on short wavelength differences. This can be neglected for y shift values as the cube was aligned in y according to the position angle of $\sim$ 60^∘ of the sub-stellar companion with respect to the primary for the observations.

We then perform the spectral deconvolution technique using a self-written version in IDL. To fully optimize the technique we also remove the PSF of the secondary within an iterative spectral deconvolution implementation in 10 iteration steps, as similarly done in Thatte et al. (2007) and arrive at the final images of each row in Fig. 1. As can be seen, not all remnants could be removed. We note that this is expected as the reproducibility of speckles and similar remnants and hence the efficiency of the technique decreases with increasing Seeing and decreasing coherence time, both not ideal in our observations (Table 1).

As telluric standard HIP 94378, a B5V star was used for H+K band. In order to correct for features of this standard star, the Bracket-$\gamma$ line in the K band as well as the $\zeta$, $\theta$, $\eta$ and $\iota$ lines of the Bracket series in the H band were removed before usage of the standard. A correction temperature of 15400 K was assumed for the star, as given for a B5V in Kenyon & Hartmann (1995). We use the optimal extraction technique by Horne (1986) to improve the quality of the standard, as it was observed at a slightly worse Seeing (Table 1).

We cannot use PSF fitting of the sub-stellar companion as the S/N is quite low in parts of the H band, hence aperture spectroscopy is our choice. However, we realized that the spectral shape changes with different aperture sizes, a natural outcome of different Strehl ratios achieved at the different wavelength used. Hence, we apply a correction fuction to the extracted spectrum: We extract the spectrum of the primary at the same aperture size as the companion and at the maximum possible aperture size and multiply the companion spectrum by the quotient of these spectra to correct for the differences of AO performance achieved at different wavelengths. The optimal extraction is not used in order not to emphasize artificial deviations.

Without the correction function we find temperatures for the companion about 700 K lower than our final result. Nevertheless different aperature sizes still give different results, which can be attributed to different AO performances achieved for primary and companion and/or remnants in the final cubes not removed by reduction and spectral deconvolution. Because of the proximity of the companion the AO performance at primary and companion position should be very comparable. Since we see in addition strong and spatially close positive remnants in the K band, artificially increasing the flux for increasing aperture sizes, we conclude that these remnants are the dominant source of changes in the spectral shape in accordance with aperture size. Hence, the accuracy of the spectrum should increase with decreasing aperture size, while its precision gets worse.

At the beginning of the H band at 1.45 $\mu$m, the sparrow resolution limit (Sparrow 1916) is at 34.6 mas or 2.8 spaxels, hence we choose an aperture size of 2.5 spaxels. This choice guarantees the best accuracy (without further inclusion of remnants), while a lower value would be below the resolution and comes into conflict with the precision of spatial PSF alignment at different wavelengths.

Using the photometry of PZ Tel A from the Two Micron All Sky Survey (2MASS) catalog (Cutri et al. 2003; Skrutskie et al. 2006) of $K$ = 6.366 $\pm$ 0.024 mag we can estimate all further parameters of the sub-stellar companion using all results from the spectroscopic analysis described in the previous section. As no absolute photometric calibration is possible with the spectroscopic standard, we preliminary estimate the parameters of the PZ Tel companion assuming negligible photometric variability of both sources, most likely not correct according to variability indications presented in the previous sections.

We derive a luminosity of $\log(L_{bol}/L_{\odot})$= -2.66${}^{+0.06}_{-0.08}$ for the PZ Tel companion from the extinction corrected apparent brightness Ks₀= 11.86${}^{+0.07}_{-0.10}$ mag (from the 2MASS brightness, the magnitude difference (Table 2), $A_{V}$= 0.53 ${}^{+0.84}_{-0.53}$ mag and extinction law by Rieke & Lebofsky 1985) using a bolometric correction of B.C._K = 3.1 $\pm$ 0.1 mag from Golimowski et al. (2004, for spectral type M6–L0) at a distance of 51.49${}^{+2.74}_{-2.47}$ pc (van Leeuwen 2007). From the luminosity and temperature T_eff= 2500 ${}^{+138}_{-115}$ K, we calculate the radius to be R= 0.25${}^{+0.03}_{-0.04}$ R_⊙ or 2.42${}^{+0.28}_{-0.34}$ R_Jup. From radius and surface gravity $\log{g}$= 3.50${}^{+0.51}_{-0.30}$ dex, we find a mass of the PZ Tel companion of M= 0.0071${}^{+0.0161}_{-0.0041}$ M_⊙ or 7.5${}^{+16.9}_{-4.3}$ M_Jup.

For the bolometric correction we used the spectral type corresponding to the temperature range found here. Using the H₂O index defined in Allers et al. (2007) we find a possible spectral range of L1.5 – L4 for the sub-stellar companion, using the modified version in Bonnefoy et al. (2014) we find a possible spectral range of M7 – L9, roughly consistent with our findings of M6 – L0. Bonnefoy et al. (2014) modified the index to avoid noisy regions, definitely necessary in the present case, as the blue part of the H-band has the lowest signal to noise. Our derived surface gravity uncertainties are about equal to the values of 0.5 dex, as computed by Bonnefoy et al. (2014) for similar objects.

The derived values, diagramed in Figs. 4–6, agree within 1 $\sigma$ with the extinction $A_{V}$ and metallicity [M/H] for PZ Tel A by Bailer-Jones (2011) and Jenkins et al. (2012), respectively as well as with the temperature by Mugrauer et al. (2010). Temperature and surface gravity deviate in regard to literature estimates by 1.2 & 1.3 $\sigma$ (Biller et al. 2010) and 2.9 & 2.5 $\sigma$ (Jenkins et al. 2012), respectively. Finally our mass result deviates by 1.2 $\sigma$ (28${}^{+12}_{-4}$ M_Jup, Mugrauer et al. 2010), by 1.6 $\sigma$ (36 $\pm$ 6 M_Jup, Biller et al. 2010) and by 3.2 $\sigma$ (62 $\pm$ 9 M_Jup, Jenkins et al. 2012).

Although our mass estimate is independent of evolutionary models, we can use them as comparison to put our results into context and to check which age is indicated by the models for our spectroscopic results. According to Chabrier et al. (2000) DUSTY models best fits are achieved between 5 – 10 Myr isochrones, fitting our spectral results for temperature, surface gravity and luminosity within 1 $\sigma$ errors. This age range is consistent with the age of the $\beta$ Pic moving group (12${}^{+8}_{-4}$ Myr), deviant from the age determined by Jenkins et al. (2012) (24 $\pm$ 3 Myr), while very well consistent with the lithium depletion age of PZ Tel A of 7${}^{+4}_{-2}$ Myr by the same authors. Very recently Binks & Jeffries (2014) combined data of eight low-mass candidates with literature data of $\beta$ Pic moving group members and find a lithium depletion age of 21 $\pm$ 4 Myr, being also inconsistent with the age found here.

Finally we arrive at a possible mass range of 3.2 – 24.4 M_Jup. According to these estimates the PZ Tel companion is most likely a brown dwarf of about 21 Jupiter masses, as the evolutionary models reject a $\log{g}$ = 3.5 dex to be only valid for objects younger than 1 Myr and predict a surface gravity $\log{g}$ $\sim$ 3.95 dex for the given lithium depletion age range, being within the 1 $\sigma$ uncertainty of our results. However, spectra at improved observing conditions and with longer spectral range, especially including alkali metal lines for a more precise surface gravity determination should be able to narrow down the parameters of the PZ Tel companion in the future. Fortunately such spectra are increasingly easy to acquire because of the strong orbital separation increase, probably persistent for the upcoming years to decades (Mugrauer et al. 2012).