ON THE NATURE OF THE HERBIG B[e] STAR BINARY SYSTEM V921 SCORPII: DISCOVERY OF A CLOSE COMPANION AND RELATION TO THE LARGE-SCALE BIPOLAR NEBULA^*

Our observations on V921 Sco were carried out using the AMBER instrument (Petrov et al. 2007), which allowed us to coherently combine the light from three of the VLTI 1.8 m auxiliary telescopes. The observations were conducted using AMBER's low spectral resolution mode, which covers the near-infrared H and K bands (1.44–2.50 μm) with a spectral resolution R ∼ 35. The observations were carried out between 2008 April 3 and 2009 March 19 on four different array configurations (Table 1), providing a good uv-coverage with baseline lengths between 10 and 127 m. For all observations, we used a detector integration time (DIT) of 100 ms.

For a significant fraction of our data we find that the closure phases (CPs) vary significantly between subsequent exposures (i.e., on timescales of several minutes). As discussed in Section 3, these rapid variations are very likely due to the presence of a companion star. Therefore, in contrast to the standard AMBER data reduction procedure, we decided not to average the individual exposures, but to fit the quantities derived from the individual data exposures separately.

Each observation on V921 Sco was accompanied by observations on interferometric calibrator stars of known intrinsic diameter, allowing us to monitor the instrumental and atmospheric transfer function. Both for the science and calibrator star observations, we extract raw visibilities and CPs using the amdlib (V3.0) data reduction software (Tatulli et al. 2007; Chelli et al. 2009). In order to associate the CP sign with the on-sky orientation, we use a reference data set³ on the binary star θ¹ Orionis C (Kraus et al. 2009b).

In order to investigate the large-scale environment around V921 Sco, we employed the IMACS instrument on the Magellan/Baade 6.5 m telescope. The images were recorded on 2011 March 13 under exceptional atmospheric conditions (seeing FWHM ∼ 04) and cover a field of 154 with a pixel size of 011 pixel⁻¹. We employed the Bessell B, V, and R filters and two narrowband filters centered around the [S ii] and Hα line (filters "676circular" and "Halpha656") using DITs of 240 s, 240 s, 180 s, 300 s, and 360 s, respectively. The images were bias-subtracted and flat-fielded using standard IRAF data reduction routines and are shown in Figures 1(A)–(F).

Zoom In Zoom Out Reset image size

Using the acquisition camera of the FIRE spectrograph (Simcoe et al. 2008) at Magellan/Clay we also recorded on 2011 March 12 a J-band image of V921 Sco (394 × 394 pixels with a pixel size 0147 pixel⁻¹), which is shown in Figure 1(H).

In order to derive the basic source structure of V921 Sco, we reconstructed model-independent aperture-synthesis images from our AMBER data. Since the object structure might change with wavelength, we subdivide our data set in three wavelength bins, which we denote with H (1.4 μm ⩽ λ < 1.9 μm), K₁ (1.9 μm ⩽ λ < 2.15 μm), and K₂ (2.15 μm ⩽ λ < 2.5 μm). For each wavelength bin, we reconstruct a separate image (Figure 2, top) and then combined the independent images in a color composite (Figure 2, bottom).

Zoom In Zoom Out Reset image size

For the image reconstruction, we employed the building block mapping algorithm (Hofmann & Weigelt 1993), which was already used in several of our earlier long-baseline interferometric imaging projects (Kraus et al. 2007, 2009b, 2010b). The presented images were obtained without (Figure 2, left) and with (Figure 2, right) a regularization function. Image reconstruction with regularization means the minimization of the cost function

$$\begin{equation} J[o_k({\bf x})] := Q[o_k({\bf x})] + \mu \cdot H[o_k({\bf x})], \end{equation} \tag{ 1 }$$

where Q[o_k(x)] describes the χ² function of the measured bispectrum data O⁽³⁾(f_u, f_v) and the bispectrum of the actual iterated image o_k(x). H[o_k(x)] is a regularization term, and μ is a weighting factor called the Lagrange multiplicator. For our reconstructions we used the maximum entropy regularization function

$$\begin{equation} H[o_k({\bf x})] := \int \left[o_k({\bf x})\cdot \log \left({\frac{o_k({\bf x})}{p({\bf x})}}\right) - o_k({\bf x}) + p({\bf x})\right] d{\bf x}. \end{equation} \tag{ 2 }$$

As prior function p(x) we used a smooth version of a building block reconstruction obtained without a regularization function (μ = 0). The start image was a circumsymmetric Gaussian with a size obtained by fitting the measured visibilities. Reconstructions for different Lagrange multiplicators (μ = 10⁻⁶ to 10⁻³) and for different reconstruction windows (radii: 60–100 mas) were obtained. The presented images are the best-fit reconstructions, i.e., those reconstructions with minimum reduced χ² values of the squared visibilities and CPs.

Each of the reconstructed images clearly reveals a close companion, which is located at a separation of ∼25 mas north of the primary star⁴ (Θ ∼ 353°) and which we denote in the following V921 Sco B. In our images, V921 Sco B appears point-like, while the primary star is clearly associated with extended emission that appears elongated along position angle (P.A.) ϕ ∼ 145°. In the images, it is also evident that the contributions of V921 Sco B to the total flux decrease with wavelength.

In order to better characterize the geometry and physical conditions of the disk-like structure in our image and to derive the relative astrometry of V921 Sco A–B, we fitted geometric models to our interferometric data.

In all models, the primary star is included as a point source where the photospheric emission at a given wavelength is given by the flux ratio of a B0 (T_eff = 14, 000 K, g = 4.04) Kurucz model atmosphere (Kurucz 1970) to the measured total SED flux F_tot, with contributions F_A/F_tot ranging from 0.27 (at 1.5 μm), 0.13 (at 2.0 μm) to 0.07 (at 2.5 μm), where F_A denotes the photospheric flux contribution of the primary star. We adopt A_V = 4.8 ± 0.2, a stellar radius of R_⋆ = 17.3 ± 0.6 R_☉, and a distance of d = 1150 ± 150 pc (Borges Fernandes et al. 2007). Given that the spectral type of the primary was determined in earlier studies without knowledge of the companion star, we note that these spectral-type estimates are potentially biased, which could affect the photospheric flux of the primary star in our models. However, the absolute level and wavelength dependence of the photospheric emission in the near-infrared wavelength regime depend only weakly on the precise spectral type and will therefore not significantly affect our modeling results.

The companion star is described by five free parameters, namely, the P.A. (Θ) and separation (ρ) measured from the primary star, the angular extent of the circumstellar material (parameterized as Gaussians with FWHM θ_B), and two parameters ((F_B/F_tot)(λ_ref), s) to describe the photospheric flux contributions of V921 Sco B to the total flux. The flux contributions might change with wavelength due to differences in the stellar effective temperatures, local extinction effects, or circumstellar emission. Given the still relatively short wavelength coverage, we approximate these effects with the linear relation (F_B/F_tot)(λ) = (F_B/F_tot)(λ_ref) + s · (λ − λ_ref), where we choose λ_ref := 2 μm as arbitrary reference wavelength.

Free parameters of the circumprimary disk are the P.A. (ϕ), angular extent along the major axis (Gaussian FWHM θ_A), and inclination angle (i, measured from the polar axis). Due to temperature gradients in the circumstellar material, it is possible that the source brightness distribution might change with wavelength. To test this hypothesis, we performed our geometric fits both to the complete data set and to the aforementioned data subsets (H, K₁, K₂ wavelength bins).

In order to find the best fit, we employ a Levenberg–Marquardt least-square-fitting procedure and minimize the likelihood estimator χ²_r = χ_{r, V}² + χ²_{r, Φ}, where χ²_{r, V} and χ²_{r, Φ} are the reduced least squares between the measured and model visibilities and CPs, respectively (Equations (1) and (2) in Kraus et al. 2009a). For our best-fit solution (Table 2), the uncertainties on the individual parameters have been estimated using the bootstrapping technique. The fit provides a slightly better representation of the CP than the visibility data, as indicated by the χ²_{r, V} and χ²_{r, Φ} values. This likely indicates that the circumprimary disk geometry is not well represented by a Gaussian-brightness distribution and we will consider more physical models in an upcoming study (S. Kraus et al. 2012, in preparation).

Table 2. Model-fitting Results for the VLTI/AMBER Continuum Observations (Section 3.2)

Epoch	Spectral	Secondary Star					Circumprimary Disk
	Band	Θ	ρ	θB	$\frac{F_{\rm B}}{F_{\rm tot}}$	s	θA	i	ϕ	χ2r, V	χ2r, Φ	χ2r
		(deg)	(mas)	(mas)	(at 2 μm)	(μm−1)	(mas)	(deg)	(deg)
2008	H	353.0	25.0	⩽0.5	0.074	0a	6.12	51.1	143.2	2.46	1.40	1.73
2008	K1	353.4	25.0	⩽0.5	0.070	0a	6.85	48.6	143.5	2.50	2.14	2.25
2008	K2	353.1	25.0	⩽0.5	0.064	0a	7.55	48.5	149.0	3.91	1.97	3.27
2008	All	353.8	25.0	⩽0.2	0.054	−0.0056	7.5	50.3	147.8	4.30	3.09	4.88
		±1.6	±0.8		±0.018	±0.019	±0.2	±1.9	±4.3
2009	All	347.3	25.5	⩽0.2a	0.054a	−0.0056a	7.5a	50.3a	147.8a	4.17	1.78	3.39
		±1.0	±1.2

Note. ^aIn our fitting procedure, this parameter was kept fixed.

Download table as:  ASCIITypeset image

Given that the position of the companion might change notably between 2008 and 2009, we first fitted only the 2008 data in order to characterize the properties of the stellar components and the circumstellar material.

Following the detailed characterization of the V921 Sco system for epoch 2008 (Section 3.2), we then investigated whether we find evidence for orbital motion between 2008 and 2009. Compared to 2008, a significantly smaller number of AMBER observations was recorded in 2009 (Table 1), which leads us to fix the geometry of the circumprimary disk using the best-fit parameters from epoch 2008 and treat only the two astrometric parameters (Θ, ρ) and the flux ratio (F_B/F_tot) as free fitting parameters.

The resulting best-fit parameters can be found in Table 2 and indicate that the position of the secondary has moved significantly in P.A. (∼7°) during the covered ∼8 months, while only marginal changes in separation (ρ ≲ 0.5 mas) or in the flux ratio could be detected over this time period.

Assuming a face-on circular orbit, we can estimate from the detected signs of orbital motion (Table 2) the period of the orbit to P ∼ 35 yr. Of course, for a precise mass determination, long-term follow-up observations will be necessary in order to derive the full dynamical orbit.

Both in our images and our model fits, the companion appears spatially unresolved (Gaussian FWHM <0.3 mas, corresponding to <0.35 AU at 1.15 kpc), which suggests that V921 Sco B is not associated with circumstellar material. Based on this result, we can compute the flux ratio of V921 Sco A and B for the H band ((F_B/F_A)_H = 0.83 ± 0.15) and K band ((F_B/F_A)_K = 1.18 ± 0.12). These values suggest that V921 Sco B is of cooler temperature and later spectral type than the primary V921 Sco A, although the current uncertainties are still too large for a quantitative spectral classification.

Our IMACS narrowband and broadband images (Figure 1) reveal a remarkably complex environment around V921 Sco. The overall shape of the nebula is bipolar, where the southwestern part appears much fainter and less extended, perhaps indicating that this part is facing away from the observer and therefore suffers from a larger amount of obscuration from material in the ambient cloud. Obscuration from foreground material might also be responsible for the dark filaments that appear in the southeastern part of the nebula (shaded area in Figure 1(G)).

Remarkably, the symmetry axis of the bipolar nebula (50° ± 5°) appears well aligned with the polar axis (578 ± 43) of the AU-scale disk detected with interferometry, suggesting that the nebula might have been shaped through outflow-activity from V921 Sco. From our kinematical modeling (S. Kraus et al. 2012, in preparation), we conclude that the disk is notably inclined, with the northeastern disk axis facing toward Earth. Therefore, based on the measured disk inclination and orientation, one would expect the northeastern lobe of the bipolar nebula to appear brighter, as is observed. This scenario provides a natural explanation for the large number of arc- and cone-shaped structures which can be seen in our IMACS images (Figure 1) and that appear centered on V921 Sco. We have identified the most significant of these arc- and cone-shaped features in our deepest image (R band, Figures 1(E) and (F)). After confirming the features in images taken with other filters, we marked them in Figure 1(G).

The distribution of the individual arc fragments, in particular in the northeastern lobe, suggests that we might observe up to five layers of ejecta, which have been created during episodic events of extreme mass loss. In order to obtain a rough estimate for the period of the mass-loss events, we measured the typical separation between the different layers (4''–5'') and apply a projection factor in order to correct for the system inclination angle of 503 ± 19 (Section 2). Assuming the maximum expansion velocity of 1400 km s⁻¹ measured by Borges Fernandes et al. (2009) and a distance of 1.15 kpc, we find that a mass-loss period of ∼25 yr would be required, which seems consistent with the period of the companion, as estimated for a circular orbit (Section 4.1). Compared to the bow-shock structures observed around other high-mass young stellar objects (YSOs; e.g., Kraus et al. 2010b), the arcs around V921 Sco have a clumpy and sometimes truncated structure, possibly indicating a lower degree of collimation in the V921 Sco outflow.

In a field of about 1 arcmin, we detect in our IMACS and FIRE images at least 26 stellar sources. Habart et al. (2003) obtained spectra for a subset of these sources (see identification in Figure 1(I)) and classified them as embedded low- and intermediate-mass YSOs. Increasing the number density in the cluster, our observations strengthen the argument that V921 Sco is likely in a young evolutionary stage and, like many massive YSOs, embedded in a star-forming region of low- and intermediate-mass YSOs.

Our detection of a companion around V921 Sco is interesting in the context of earlier suggestions that the B[e] phenomenon might be related to dynamical interaction in a close binary system, either as a result of a recent stellar merger (e.g., Podsiadlowski et al. 2006) or through material which might have been ejected during phases of binary interaction (e.g., Sheikina et al. 2000; Miroshnichenko 2007; Kraus et al. 2010a). Unfortunately, only very few B[e] stars have been studied so far with sufficient angular resolution to make definite statements about multiplicity. One of the best-studied B[e] systems is HD 87643, where Millour et al. (2009) detected a close companion around this supergiant star. Both the projected separation (∼50 AU) and orbital period (∼50 yr) of this system show similarities with the characteristics we deduce for V921 Sco (∼29 AU, P ∼ 35 yr). Also, both systems are embedded in an extended nebulosity with a shell-like substructure, indicating episodic mass loss. Therefore, it is plausible that the mass loss in both systems might be triggered by dynamical interaction of the companions with the circumprimary disks. The B[e]-star-characteristic permitted and forbidden line emission might then originate in low-density material which is periodically stripped away from the circumprimary disk. This scenario works independently of the evolutionary status and physics of the circumprimary disks (accretion or excretion disks) and might explain the diversity of stellar systems (Herbig B[e], supergiant, symbiotic star, compact planetary nebulae) in which the B[e] phenomenon is observed.

Follow-up interferometric studies on V921 Sco and HD 87643 might confirm this scenario, for instance by measuring the orbital eccentricity and by imaging the expected star–disk interaction effects during periastron passage.