Discovery of a Low-mass Companion Embedded in the Disk of the Young Massive Star MWC 297 with VLT/SPHERE^*

Table 1 summarizes the stellar and disk properties. The effective temperature and interstellar extinction were derived following van den Ancker et al. (1998): the observed SED (between 0.3 and 1.2 μm) was fitted using atmospheric models of Kurucz (1991) and the dereddening law from Cardelli et al. (1989) with ${R}_{{\rm{v}}}$ = 3.1. The stellar radius was estimated based on L_bol and T_eff.

We observed MWC 297 on 2015 April 29 and on 2018 June 28 with SPHERE in the IRDIFS-EXT mode, i.e., simultaneous integral field spectroscopy (IFS) in the YJH bands, and dual-band imaging in the K band. The first observation (2015 April 29) was taken in FIELD tracking mode, while for the second set (2018 July 28) we used the PUPIL tracking mode (Table 2). The IFS data are cubes of 39 monochromatic images in the NIR encompassing a field of view of $1\buildrel{\prime\prime}\over{.} 73\times 1\buildrel{\prime\prime}\over{.} 73$. The spectral resolution was R ∼ 30 for the IRDIFS-EXT mode (Y–H, 0.95 < λ < 1.65 μm). The N-ALC-YJH-S coronagraph (inner working angle ∼015) was used.

We obtained "Flux" and "Star Center" calibration images at the beginning and end of both observing sequences. The Flux images were obtained by offsetting the central star from the coronagraphic spot and used to measure the unsaturated peak flux of the star. The Star Center images allowed us to measure the position of the star behind the coronagraph, located at the center of the four replicas produced by the adaptive optics system.

We used the ESO pipeline⁵ to reduce the IFS data. We used a function implemented in the Vortex Imaging Pipeline⁶ (VIP; Gomez Gonzalez et al. 2017) to correct for clumps of bad pixels through an iterative sigma filtering process. For the centering, we increase the signal-to-noise ratio (S/N) of the star replicas using a high-pass filter that subtracts the image itself with a median low-pass filtered version of the image. We fitted the four replicas with a 2D Moffat function (in VIP) to derive the centroid of the star in each frame. We then interpolated the values of the derived center (taken at the beginning and end of the observations) considering the observation time of the science images. Finally, the error was considered to be the discrepancy in the value between two sequential sets of center images (on average σ_x = 0.04 pixels and σ_y = 0.08 pixels).

Calibrated frames are still affected by quasi-static speckles produced by the star (Marois et al. 2006). Speckles move radially with wavelength, while real features remain fixed (spectral information). This is key to the spectral differential imaging (SDI) algorithm (e.g., Sparks & Ford 2002). Also, fixing the pupil of an altitude-azimuth telescope during an observing sequence, most quasi-static speckles remain fixed in the image, while real features rotate (angular information). Angular differential imaging (ADI; e.g., Marois et al. 2006) is based on this idea. The IFS cubes contain the spectral information, while angular information is available when the rotator is moved to maintain the pupil fixed. We used principal component analysis (PCA)-based algorithms in VIP to model and subtract the stellar point-spread function (PSF) and associated speckles. For both sets of observations we applied PCA-SDI, using the spectral information alone, and PCA-SADI, where the PCA library was built using both the angular and spectral information (Pueyo et al. 2012). We also tested the algorithm in two separate steps (PCA-SDI + PCA-ADI; Christiaens et al. 2019), but obtained noisy final images. For the second observational set, we also used another algorithm: PCA-ADI, using only the angular information, performed either in full frames (Soummer et al. 2012) or in concentric 2 FWHM wide annuli on individual spectral channels (Absil et al. 2013).

We detected a bright companion in the outer disk of MWC 297 on 2015 April 29 located ∼246.4 au from the central star. The detection was obtained using the PCA-SADI (Figure 1, top left) and PCA-SDI techniques, with 4σ and 5σ significance. The companion was detected in the averaged H-band image (S/N ≳ 4), but not in J and Y.

Zoom In Zoom Out Reset image size

We performed follow-up observations with longer integration time (Table 2) on 2018 July 28. We redetected the companion in the H and J bands with four different post-processing methods (S/N > 4). We also detected it in the Y band just using ADI. Figure 1 shows that the pointlike source is detected regardless of the post-processing method (SADI, ADI, ANNULI, and SDI—not shown here) and of wavelength (H, J, and Y bands all show the companion).

We used the Markov Chain Monte Carlo (MCMC), a nested sampling algorithm coupled to the negative fake companion technique implemented in VIP, to derive the position and the flux of the companion at each wavelength (e.g., Marois et al. 2010; Wertz et al. 2017). We first estimated the position and flux using the Nelder–Mead simplex-based algorithm (Nelder & Mead 1965), and then fed these first estimates to the MCMC routine. A negative PSF was injected in the original data cube in order to completely delete the signal of the real companion measured in the final PCA-ADI post-processed image. The process produces a posterior distribution of the three parameters and stops upon convergence to minimal absolute residuals in an aperture centered on the location of the companion. Finally, this routine gave us the companion separation, position angle (PA), and flux, with errors (Table 2).

4.3.1. Astrometry

Using MCMC on the second epoch, we derived the position (r and PA) and relative error of the companion for each wavelength and computed the weighted average (Figures 2(a) and (b), red line). For the first epoch, it was not possible to use MCMC. We therefore fitted a 2D Gaussian to derive the position and took the weighted average of the results of the SDI and SADI methods. We considered a pixel scale of 7.46 ± 0.02 mas pixel⁻¹ (Maire et al. 2016). The separation uncertainty was computed as a sum in quadrature of the uncertainties on the stellar and companion position and on the pixel scale for each frame (Figure 2(a); Table 2). We took into account the target distance error to derive the separation in au (Table 2).

Zoom In Zoom Out Reset image size

The PA is affected by the error on the true north angle determination of −102.18 ± 0.13 deg (Maire et al. 2016), used to derive the astrometry. We propagated the errors in the position and true north to get the final error (Figure 2(b); Table 2).

Figure 2(d) suggests a companion comoving with the host star on a trajectory more consistent with Keplerian motion. A background star would move on the trajectory shown by the black line: its proper motion after 3.25 yr exceeds the centroid discrepancy of the two observation sets (orange and blue), albeit within 2σ uncertainty. Assuming a face-on circular orbit (recalling the ∼5° disk inclination), Keplerian motion would account for a shift in PA of 126, inside the error bar of the first detection.

Our two-epoch astrometry alone does not rule out the possibility of a background object. Therefore, we used the TRILEGAL model of the Galaxy to estimate the probability of being a background star (Girardi et al. 2005). TRILEGAL yields 6553 stars with an H-band apparent magnitude brighter or equal to that of our companion candidate (H  ≤ 13.84 mag; Section 4.3.2) within a 30' × 30' patch of sky centered on the star, hence a density of 0.002 arcsec⁻². The probability is thus $1-{ \mathcal P }(n=0| \lambda =0.002,B=4)\approx 0.8 \%$, where ${ \mathcal P }(\lambda ,B)$ is the spatial homogeneous Poisson point process probability with rate λ and area B. Given the separation of ∼07, we conservatively consider a 2'' × 2'' box centered on the star for the area.

4.3.2. Spectrophotometry

The undereddened spectrum of the companion (Figure 2(c), blue points) shows a very red slope, suggesting significant extinction on the companion, not necessarily the same for the star. Each component might be embedded and surrounded by their own disk, in addition to any remnant envelope (e.g., Bowler et al. 2014; Mesa et al. 2019). Therefore, following Christiaens et al. (2018), we considered extinction as a free parameter when fitting BT-SETTL models (Allard et al. 2012). Our grid of BT-SETTL models contains four free parameters: effective temperature, ${T}_{\mathrm{eff}}\,\in \,[1200\,{\rm{K}},5500\,{\rm{K}}]$ in 100 K steps; surface gravity, $\mathrm{log}(g)\in [2.5,5.0]$ in 0.5 dex steps; radius, ${R}_{B}\in \,[0.1\,$R_⊙, 3.5R_⊙] in 0.01 R_⊙ steps; and extinction, ${A}_{V}\,\in \,[0,21]$ mag in 0.1 mag steps. We then considered the same grid, but we fixed the extinction to A_V = 7.72 mag (Figure 2(c), red points), same as for the central star (Section 2).

Next, we considered two libraries of young stellar object (YSO) template spectra: (i) all 76 pre-main-sequence stars spectra compiled in Alcalá et al. (2014) and Manara et al. (2013, 2017), which are members of the TW Hya, σ Ori, Lupus I, III, and IV star-forming regions, spanning G5 to M8.5 spectral types; and (ii) all young dwarfs from the SpeX library (Burgasser 2014), identified based on their gravity class or their membership to young (<10 Myr old) clusters. In either case, we considered two free parameters to account for different A_V and distance between observed and template spectra.

For all spectral fits, we convolved the models and templates with the IFS spectral response before binning them to the same wavelength sampling. We then minimized a goodness-of-fit indicator χ² that accounts for the spectral covariance of the IFS instrument (Greco & Brandt 2016; Delorme et al. 2017).

Figure 2(e) shows the best-fit BT-SETTL and YSOs template spectra (blue and green color) with the undereddened spectrum of the companion candidate (black points). With extinction as a free parameter, the best-fit BT-SETTL model has T_eff = 3500 K, $\mathrm{log}(g)=3.0$, R_B = 1.13R_⊙, and A_V = 11.9 mag (solid line; ${\chi }_{r}^{2}\sim 0.4$), consistent with a young (very low gravity), gravitationally contracting and embedded stellar mass companion surrounded by a lot of dust. By contrast, lower values of extinction (e.g., A_V = 7.72 mag; dotted line), gave significantly worse fits.

The best-fit template spectra correspond to early M-type (M1 to M3.5) YSOs from (i) the 1–3 Myr-old Lupus I cloud (Sz 72 and Sz 74; Alcalá et al. 2014) and (ii) the ∼2 Myr old cluster IC 348 (CXOU J034404.8+315739 and Cl* IC 348 LRL 215; Luhman et al. 2003). Interestingly, both the SpeX targets are located in the youngest part of the IC 348 cluster, where class 0/I objects have been identified (Luhman et al. 2003, 2016). In particular, they could also be class 0/I objects given their significantly lower differential extinction compared to the best-fit extinctions associated with the Lupus I and BT-SETTL spectra, suggesting A_V ≳ 10 mag for the companion.

Based on the empirical relationship between spectral type and effective temperature inferred in Luhman et al. (2003) for IC 348, spectral types M1–M3.5 would correspond to T_eff = 3350–3700 K, consistent with our best-fit BT-SETTL model effective temperature.

Considering an extinction of A_V = 11.9 ± 1.0 mag (Section 4.4), our dereddened J- and H-band absolute magnitudes are 5.3 ± 0.3 and 4.2 ± 0.1 mag (using Cardelli et al. 1989), respectively. We compared the absolute magnitudes and colors with BCAH98, AMES-Cond, and BT-SETTL (Baraffe et al. 1998; Chabrier et al. 2000; Allard et al. 2003) models that suggest a mass of ∼0.10–0.25 M_⊙. Comparing the T_eff and age with stellar isochrones (Baraffe et al. 2015) suggests a mass of 0.25–0.5 M_⊙ (Table 2). Considering that this estimate assumes an age of 1 Myr—the youngest available, but an upper limit for MWC 297 (Vioque et al. 2018 suggest ≈0.02–0.03 Myr)—the companion mass may be lower. This is consistent with the best-fit YSOs and SpeX template spectra with mass ∼0.45–0.50 M_⊙, targets older (1–3 Myr) than MWC 297. For the 2015 epoch, we estimated the mass using only the dereddened absolute H-band magnitude, due to the lack of obvious detection in other bands.