DISCOVERY OF A PROBABLE 4–5 JUPITER-MASS EXOPLANET TO HD 95086 BY DIRECT IMAGING^*

HD 95086 was identified as an early-type member of the Lower Centaurus Crux (LCC) association by de Zeeuw et al. (1999) and also by Madsen et al. (2002). The membership was established on the grounds that the star shares a similar velocity vector in the galactic framework as other LCC members. HD 95086 has a distance of 90.4 ± 3.4 pc (van Leeuwen 2007), which is approximately the mean distance of the LCC association. About the age of the association, Mamajek et al. (2002), followed by Pecaut et al. (2012), derived 17 ± 2 Myr (based on isochrone fitting), while Song et al. (2012) derived ≃ 10 Myr (by comparison with lithium equivalent width of members of other nearby associations). Systematic differences between these two fully independent methods may be responsible for the discrepancy. Nonetheless, the full assessment of the age is beyond the scope of this Letter and adopting 10 or 17 Myr only has a limited impact in the following results.

Houk & Cowley (1975) proposed that HD 95086 has a class III luminosity and an A8 spectral type with a mass of ≃ 1.6 M_☉. However, its good-quality trigonometric parallax and thus derived luminosity and effective temperature undoubtedly place it close to zero-age main-sequence stars of LCC and thus reject the supergiant phase.

Another interesting property of HD 95086 is that observations in the mm (Nilsson et al. 2010) and in the mid- to far-infrared (Rizzuto et al. 2012; Chen et al. 2012) revealed a large dust-to-star luminosity ratio (L_d/L_⋆ = 10⁻³), indicating the presence of a thus far unresolved debris disk.

Finally, Kouwenhoven et al. (2005) observed HD 95086 with adaptive optics and identified a background star with a separation of 487 and a position angle of 316° to the star (based on its K magnitude).

HD 95086 was observed with VLT/NaCo (Lenzen et al. 2003; Rousset et al. 2003) in thermal infrared and angular differential imaging (ADI; Marois et al. 2006) mode as part of our direct-imaging survey of young, dusty, early-type stars (Rameau et al. 2013). NaCo setups were the L' filter (λ₀ = 3.8 μm, Δλ = 0.62 μm) with the L27 camera in service mode for observations in 2012 January. The source was dithered within the instrument field of view (≃ 14'' × 14'') in order to properly estimate and remove the background contribution. An observing sequence was made up of a first short set of unsaturated exposures to serve as calibrations for the point-spread function (PSF) and for the relative photometry and astrometry. The sequence was followed by a one-hour set of deep science observations.

In 2013, follow-up observations were performed and included two runs: one in February (back-up star, visitor mode, very bad conditions) with the Ks filter (λ₀ = 2.18 μm, Δλ = 0.35 μm) and the S13 camera (plate scale ≃ 13.25 mas pixel⁻¹), and one in March at L' with the L27 camera in service mode. Table 1 summarizes the observing log for each run. We recall that HD 95086 has V = 7.36 ± 0.01 mag, Ks = 6.79 ± 0.02 mag, and 6.70 ± 0.04 mag at 3.4 μm from the WISE database.

Finally, an astrometric calibrator, the θ₁ Ori C field, was observed for each observing run.

Data reduction (flat-fielding, bad pixel and sky removal, registration,⁷ and image selection) was performed using the IPAG-ADI pipeline (e.g., Lagrange et al. 2010; Chauvin et al. 2012; Rameau et al. 2013 and references therein). Stellar-halo subtraction was also done using all the ADI algorithms implemented in the pipeline: cADI, sADI (Marois et al. 2006), and LOCI (Lafrenière et al. 2007). The frames were finally de-rotated and mean-combined. The astrometry and photometry of any detected point source as well as their error estimates were done similarly to Chauvin et al. (2012) and Lagrange et al. (2012) by injecting fake planets using the unsaturated PSF reduced images. The noise per pixel was derived from the standard deviation calculated in a ring of 1.5 FWHM width, centered on the star, with a radius of the separation of the source and masking the point-source itself. The flux of the point source was integrated over an aperture of 1.5 FWHM in diameter. The final signal-to-noise ratio (S/N) was then calculated on the same aperture size considering the noise per pixel and the aperture size in pixels.

The θ₁ Ori C field data were reduced for detector calibrations by comparison with Hubble Space Telescope observations by McCaughrean & Stauffer (1994; using the same set of stars TCC0051, 034, 029, and 026 at each epoch). We found, for the 2012 and 2013 data, a true north of −0.37 deg ± 0.02 deg, −0.38 deg ± 0.03 deg, and −0.45 deg ± 0.09 deg, respectively, and a plate scale of 27.11 ± 0.06 mas, 27.10 ± 0.03 mas, and 27.10 ± 0.04 mas, respectively.

Finally, the detection performance was derived by measuring the 5σ level noise in a sliding box of 5 × 5 pixels toward the direction of the point source and corrected for flux loss. The contrast was converted to mass with the "hot-start" COND models of Baraffe et al. (2003).

Two independent pipelines (Boccaletti et al. 2012; Amara & Quanz 2012) were also used for consistency and error estimates.

The data in 2012 January showed the background star detected by Kouwenhoven et al. (2005) at a projected separation of 4540 ± 0015 from the central star and a position angle of 319.03 deg ± 0.25 deg (see Figure 1, right panel). We also detected an additional, fainter signal southeast of the star with a S/N of 9. The robustness of this detection was strengthened due to its systematic confirmation via the mean of a series of tests using three independent pipelines, all our published flavors of ADI algorithms, and an extensive parameter space exploration. The source successfully passed all tests. The top and bottom left panels of Figure 1 represent the companion candidate (CC) located at a separation of 623.9 ± 7.4 mas and a position angle of 151.8 deg ± 0.8 deg from the central star, using the sADI algorithm (with 20 frames combined for N_δ = 1 (FWHM) at r = 540'') and LOCI (with N_δ = 0.75 (FWHM), dr = 3 (FWHM), g = 1, and N_A = 300 (FWHM)), respectively.

Zoom In Zoom Out Reset image size

With HD 95086 being at very low galactic latitude (b ≃ −8°), the contamination by background objects is relatively high, even at L'. A time lapse long enough with another data set was mandatory to prove the companionship of each object (see Section 4.1).

In the 2013 L' data, both objects were detected at similar locations as in 2012. However, the weather conditions strongly varied over the sequence, thereby revealing the CC with a lower S/N of 3. We carried out the same tests as for the 2012 data for this data set and the CC was detected. The signal is thus consistent with the object seen in 2012, even at low S/N. The residual map with the CC southeast from the star is displayed in the top and bottom middle panels of Figure 1 using sADI with the 2012 parameters and LOCI with N_δ = 0.75 (FWHM), dr = 1 (FWHM), g = 0.5, and N_A = 600 (FWHM). The positions are 626.11 ± 12.8 mas and 150.7 deg ± 1.3 deg for the separation and position angle of the CC, and 4505 ± 0016 and 319.42 deg ± 0.26 deg for the background star.

At Ks, we did not detect the CC (Figure 2, bottom left panel). Conversely, the background star was revealed, as well as seven other point sources not seen at L'. This may be consistent with background objects.

Zoom In Zoom Out Reset image size

In the 2012 L' data, we derived the star-to-CC contrast to be 9.79 ± 0.40 mag (L' = 16.49 ± 0.50 mag) and 6.2 ± 0.2 mag for the background star. The error budget includes, from high-to-low significance, photometry of the CC, neutral density, PSF flux estimate, and variability. Similar results have been obtained with other algorithms and pipelines. In the 2013 L' data, the weakness of the CC signal impacted to the estimation of the photometry and higher uncertainties than in 2012 data, but the L' contrast was consistent and of 9.71 ± 0.56 mag. For the background star, we found ΔL' = 6.1 ± 0.2 mag.

Finally, at Ks, we estimated ΔKs = 5.84 ± 0.1 mag for the background star. The non-detection of the CC directly also provided an upper limit to the Ks–L' color of 1.2 mag.

HD 95086 has a proper motion of [−41.41 ± 0.42, 12.47 ± 0.36] mas yr⁻¹ and a parallax of 11.06  ±  0.41 mas (van Leeuwen 2007), hence translating into an amplitude of 58.886 ± 0.002 mas (more than 2 NaCo/L27 pixels) between the two epochs. Figure 2 (top left) shows that the northwestern star is unambiguously of background, based on astrometric measurements rather than on photometric ones (Kouwenhoven et al. 2005). This analysis strengthened our capability to discriminate between background behavior and common proper motion with the parent star, despite the relatively low amplitude between the two epochs.

Figure 2 (top right) presents the sky relative positions for the CC. Its background nature may be excluded with a χ² probability of 3 × 10⁻³ (3σ confidence level). For further investigations of the background hypothesis, we ran simulations with the Besançon galactic model (Robin et al. 2003) to identify the probability of contamination by stars with L' ≃ 17 mag and their properties. We found that in a field of view of radius 1'' around HD 95086, this probability is about 0.11% and dominated by M dwarfs (peak at Ks = 19 mag). Making the assumption of an M-type background star, the resulting Ks–L' ≃ 0.4 color would imply Ks ≃ 16.9 mag, which would easily be detected in our observations. Therefore, we injected a PSF scaled to the flux into the Ks data at the separation of the CC and reduced within the pipeline. The signal was unveiled with a S/N of 15 (see Figure 2, bottom right). As a result, the Ks data set should have exhibited the CC if it had been the reddest contaminant even though the observing conditions were bad. For these reasons, background contamination by a late-K to M dwarf seems improbable and only a very red object (planet, brown dwarf) may match the Ks–L' constraint. Finally, according to Delorme et al. (2010), the probability of finding a fore/background field L or T dwarf around HD 95086 (1'' radius) down to L' ⩽ 17 mag is about 10⁻⁵.

From both astrometry and photometry points of view, we conclude that a contamination appears very unlikely.

In the following, we assume that the CC is a bound companion hereafter named HD 95086 b.

First, the measured separation of 623.9 ± 7.4 mas of HD 95086 b (from the 2012 data with the highest S/N) translates into a projected distance of 56.4 ± 0.7 AU.

Given the observed contrast (ΔL' = 9.79  ±  0.40 mag), distance (90.4  ±  3.4 pc), and HD 95086 magnitude (6.70  ±  0.09), we derived ${M}_{L^{\prime }}=11.71\pm 0.53$ mag for HD 95086 b. This translates, according to the COND evolutionary model (Baraffe et al. 2003), into a mass of 5 ± 1 M_Jup at 17 ± 2 Myr and 4 ± 1 M_Jup at 10 Myr. We checked that, given such a mass, HD 95086 b cannot be detected at Ks. Figure 3 displays L' 5σ-detection performances toward the direction of HD 95086 b. Our sensitivity ruled out any additional companion as light as 4 M_Jup from 48 AU and 1200 AU. The presence of any planet more massive than 8 M_Jup and 12 M_Jup can be excluded beyond 38 AU and 34 AU, respectively, in projected separation. Moreover, we attempted a comparison of the L'-band magnitude of HD 95086 b to "warm-start" evolutionary models' predictions (Spiegel & Burrows 2012) with different initial conditions (see Figure 11 of Bonnefoy et al. 2013). We found a mass greater than 3 M_Jup using the youngest age estimate of the system and three-times solar metallicity hybrid cloud models. The lack of prediction for M ⩾ 15 M_Jup prevented us from giving an upper limit of the mass.

Zoom In Zoom Out Reset image size

Lastly, Figure 4 compares the HD 95086 b magnitude and color upper limit with those of other companions, field dwarfs, and tracks from COND and DUSTY evolutionary models (Baraffe et al. 2003; Chabrier et al. 2000). HD 95086 b, HR 8799 cde, and 2M 1207 b appear to be similar in the sense that they all lie at the L–T transition and are less luminous than all other companions except HR 8799 b. The Ks–L' limit suggests that HD 95086 b is at least as cool as HR 8799 cde and 2M 1207 b (Chauvin et al. 2004). Moreover, with a predicted temperature estimate of 1000 ± 200 K and log g of 3.85 ± 0.5 derived from the L' magnitude, HD 95086 b would enable us to further explore the impact of reduced surface gravity on the strength of methane bands in the near-infrared.

Zoom In Zoom Out Reset image size

We reported the probable discovery of the exoplanet HD 95086 b, which may be the planet with the lowest mass ever imaged around a star.

In a nutshell, our L' observations revealed the probable planet in 2012 with a S/N of 10 and the likely re-detection of it in 2013 at a S/N of 3. It is separated from its host star by 56.4  ±  0.7 AU in projection, has L' = 16.49 ± 0.50 mag, and an upper limit for the Ks–L' color of 1.2 mag from the non-detection at Ks. These Ks observations also allowed us to reject the background hypothesis by the most probable contaminants that would have been detected. In addition, we emphasized the comoving status of the planet with a 3σ confidence level based on our astrometric measurements in 2012 and 2013. Another data set such as the one in 2012 with S/N ⩾5 would significantly improve the astrometric precision and thus ascertain the bound status. Finally, we derived a mass of 4 ± 1 to 5 ± 1 M_Jup for HD 95086 b using the COND models and an age of ≃ 10 or 17 ± 2 Myr for the system.

HD 95086, having a large infrared excess, and its probable planet, likely having q ≃ 0.002, lend support to the assumption that HD 95086 b formed within the circumstellar disk like β Pic b or HR 8799 bcde. Regarding the separation, HD 95086 b has a projected physical separation about 56 AU, which is very similar to the brown dwarf κ And b (Carson et al. 2013) which is smaller than that of HR 8799 b but larger than those of c and d, and much larger than β Pictoris and HR 8799 e. Therefore, this giant planet may also be challenging for the classical formation mechanisms, specifically for core accretion. The timescale to reach the critical core mass of 10 M_⊕ for gas accretion is far longer than the gas dispersal one (from 10⁶ to 10⁷ Myr) and thus inhibits the in situ formation of a giant planet beyond very few tens of AU (Rafikov 2011). Particular circumstances for core accretion (Kenyon & Bromley 2009) or an alternative scenario such as pebble accretion (Lambrechts & Johansen 2012) may occur instead. Another possible mechanism is gravitational instability. At 56 AU, the fragmentation of the protostellar disk of HD 95086 (1.6 M_☉) can occur (Dodson-Robinson et al. 2009), but such a low mass planet might not be formed through direct collapse. It may result from subsequent clump fragmentation or from gravitational instability with peculiar disk properties (Kratter et al. 2010). Finally, there is also the possibility that HD 95086 b was formed closer to the star by core accretion and migrated outward due to interactions with the disk or planet–planet scattering (Crida et al. 2009) to its current position. Orbital monitoring showing eccentricity would likely ascertain the presence of an unseen close-in, higher mass planet.

Future observations with NaCo and next planet imagers will be important for providing new photometric points at predicted H ≃ 18.9 mag and K ≃ 18.5 mag to further explore its atmosphere properties and to search for additional close-in planets.