A Long‐Period Spectroscopic Binary in the O‐Star Multiple System HD 193322

The O‐type star HD 193322A (HR 7767 = STF 2666 = ADS 13672 = HIP 100069) is the central object and brightest member of the open cluster, Collinder 419 (Cr 419 = C2016+405 = OCL 177), which has a distance of roughly 1.4 kpc from the Sun (Hron 1987). The star is also the A component of a multiple system with three visual companions and one speckle companion (see the Washington Double Star Catalog; Worley & Douglass 1997); the closest visual companion, the B component, has a separation of 27 and a magnitude difference of ΔH_p = 2.3 according to the Hipparcos Catalogue (ESA 1997). The spectral classification of the AB pair is O9 V:((n)) according to Walborn (1972), where the suffix indicates "slightly broadened lines." However, the measured projected rotational velocity is moderately low (Vsin i = 67 km s⁻¹ according to Howarth et al. 1997, and 86 km s⁻¹ according to Penny 1996), so the classification suffix "((n))" suggests that the spectrum is a composite of broad‐ and narrow‐lined components. In fact, Hutchings, Nemec, & Cassidy (1979) interpreted the spectrum in terms of a very broad absorption line component (Vsin i = 455 ± 20 km s⁻¹) and a narrow‐lined shell component. Early radial velocity measurements by Adams, Joy, & Sanford (1924) suggested variability (see Abt & Biggs 1972), but the target has not received much attention in the intervening years. Garmany, Conti, & Massey (1980) concluded that HD 193322A was probably radial velocity constant based on just a few measurements.

Our interest in the multiplicity properties of this star was sparked by two discoveries made by our group. First, McAlister et al. (1987) found that the A component is a speckle binary (designated CHARA 96 Aa) that has since shown considerable motion (suggesting an orbital period of approximately 31 yr; Mason et al. 1998). Second, Fullerton (1990) obtained a high‐quality series of spectra of HD 193322A that showed a systematic linear radial velocity variation over an interval of 6 days. We have subsequently obtained new radial velocity measurements from a variety of ground‐based telescopes and the International Ultraviolet Explorer Satellite (IUE), and we show here that the A component is one of only a few known O‐type spectroscopic binaries with an orbital period of about 1 yr. The A component is thus a close triple system, and such triples may have important dynamical consequences for young clusters of massive stars. We plan to present an analysis of the longer period speckle orbit in a future paper.

Radial velocities were determined from recent and archival IUE spectra, literature sources, and recent optical spectra. The resultant mean velocities are presented in Table 1. We caution the reader that our compilation represents a very diverse set of measurements of many different spectral lines made using different techniques, but any resulting systematic errors appear to be smaller than the amplitude of the orbital motion.

IUE high‐dispersion spectra obtained over the period from 1978 to 1995 with the Short Wavelength Prime (SWP) camera in echelle mode (resolution, R = λ/Δλ = 10,000) were manipulated by using the techniques described in Gies et al. (1993) to produce a matrix of pseudo‐rectified spectra sampled with a common wavelength grid. Radial velocities corresponding to these spectra were then determined by cross‐correlating each spectrum with that of HD 34078, a narrow‐lined star with a similar classification (O9.5 V; Walborn 1972). We set to unity those regions in the neighborhood of Lyα and the strong stellar wind resonance lines prior to performing the cross‐correlation. The relative velocity was determined by fitting a parabola to five velocity points (spanning 50 km s⁻¹) surrounding the minimum position in the resulting cross‐correlation functions. The spectral lines and resulting cross‐correlation functions are relatively narrow and approximately Gaussian in shape, and a parabolic fitting of the central core of the profile provides a reasonable match of that part of the profile and an adequate measurement of its position. The transformation from relative to absolute velocity was made by adding 54.4 km s⁻¹, the radial velocity of HD 34078 (Gies & Bolton 1986), to each of the relative velocities obtained. The radial velocities corresponding to the IUE spectra given in Table 1 are the transformed values. The standard deviation of the nine velocities from IUE spectra obtained over 3 days in 1987, σ = 1.3 km s⁻¹, was used to estimate the error (col. [5]) for all the IUE‐derived radial velocities.

We performed a literature search for radial velocity data for HD 193322. Four sources, Plaskett & Pearce (1931), Conti, Leep, & Lorre (1977), Garmany et al. (1980), and Fehrenbach et al. (1997), yielded averaged radial velocity data that are displayed and referenced in Table 1. The number of lines used in computing the average velocities (col. [4]) and the standard errors (col. [5]) are those given by the authors. Fullerton (1990) gives individual radial velocities for the C iv λλ5801, 5812 and He i λ5876 lines, and we give in Table 1 the average and standard deviation of each set.

We made several additional observations using the 0.9 m Coudé Feed Telescope at Kitt Peak National Observatory and the 1.9 m telescope at the David Dunlap Observatory. The KPNO spectra of HD 193322 made in 1993 (by M. E. H.) were obtained using the large collimator, grating A (632 grooves mm⁻¹, blazed at 12000 Å) in second order, and camera 5 (focal length 108.0 cm), and the detector was the T2KB CCD. These spectra cover the range 6430–6769 Å with a reciprocal dispersion of 0.167 Å pixel⁻¹ (with a resolution of 0.33 Å FWHM or R = 20,000). We measured radial velocities in these spectra by parabolic fits of the lower half of the He i λ6678 profile and of the core of Hα. The KPNO spectra obtained in 1994 and 1995 (by M. L. T.) used a similar arrangement but used the F3KB CCD as a detector. The spectrum made in 1994 covers the range 6456–6773 Å with a reciprocal dispersion of 0.104 Å pixel⁻¹ (0.31 Å FWHM, R = 21,000), and we measured only the radial velocity of He i λ6678. The spectrum made in 1995 covers the range 5572–5894 Å with a reciprocal dispersion of 0.106 Å pixel⁻¹ (0.32 Å FWHM, R = 18,000), and here we measured the same three lines observed by Fullerton (1990) plus O iii λ5592. The DDO spectra (obtained by C. T. B.) were obtained with the Cassegrain spectrograph, grating 600C, and a Thomson 1024 × 1024 CCD. They have lower resolution (reciprocal dispersion of 0.61 Å pixel⁻¹ and resolution of 1.83 Å FWHM or R = 2200) but cover a much larger range of the spectrum (3500–4120 Å for the first DDO spectrum and 3990–4600 Å for the rest). We measured radial velocities for these spectra by parabolic fitting of the lower halves of the lines (and using wavelengths from Bolton & Rogers 1978).

There is clear evidence of variability in the radial velocities presented in Table 1 (see, for example, the large range relative to the errors in the uniform set of measurements from IUE), so we searched for a periodic signal using the discrete Fourier transform and CLEAN deconvolution algorithm presented by Roberts, Lehar, & Dreher (1987). We used a realization of the algorithm written (by A. W. F.) in IDL.⁵ We found only one clear peak in the deconvolved periodogram, corresponding to a period of 3112. We used this value as a starting point for a nonlinear least‐squares fit of the orbital elements using the program described by Morbey & Brosterhus (1974). We found that a fully elliptical solution did not significantly improve the fit over a circular solution, and so we adopted a circular orbit for simplicity. We also found that the earliest measurements (by Plaskett & Pearce 1931 and by Fehrenbach et al. 1997) showed the largest residuals from the fit, so in the final solution these were assigned zero weight. Our derived orbital elements are listed in Table 2 (where T₀ is the epoch of radial velocity maximum and the zero point of orbital phase), and the radial velocity curve is illustrated in Figure 1. The residuals from the fit (col. [6] of Table 1) are comparable to the expected observational errors and possible systematic measurement errors, but the residuals may also reflect motion in the wider orbit associated with the speckle companion. We plan to make a combined radial velocity and astrometric solution for both orbits once the orbital motion of the wider system is better established.

Zoom In Zoom Out Reset image size

There is no clear indication of the spectral lines of the close companion in the spectra we have examined (see below), but we can estimate the probable mass ratio using the statistical method of Mazeh & Goldberg (1992). Mazeh & Goldberg describe how the expectation value of q = M₂/M₁ for a given single‐lined system is derived from the observed mass function, f(m), the probable mass of the primary, M₁ (which we estimated from the mass–spectral classification calibration of Howarth & Prinja 1989), and the normalized distribution of mass ratio among the parent population of binaries. We calculated the expectation value of q assuming M₁ = 24 ± 3 M_⊙ and two kinds of mass ratio distributions (flat and observed; see Mason et al. 1998), and we find q = 0.54 ± 0.02, which suggests that the close companion is an early B‐type star (M₂ = 13 ± 2 M_⊙). If both stars in the close binary are main‐sequence objects, then we expect a magnitude difference of approximately Δm = 1.3 (see Fig. 2 of Schönberner & Harmanec 1995). In the following section we discuss the composite spectrum and the lack of any immediately apparent spectral features associated with the secondary.

The detection of a spectroscopic binary with a 311 day period in addition to the wider speckle binary indicates that the flux from three stars is present in the visible spectrum (and possibly more in the IUE spectra since its 3 ^'' circular entrance aperture may also include the B component 27 away). The speckle observations suggest a magnitude difference of Δm = 0.5 ± 0.3 between the inner binary and the outer companion, and taken together with the magnitude difference estimate for the inner binary, this suggests that the three stars, inner primary, inner secondary, and outer speckle companion, contribute to the visual flux by approximately 47%, 14%, and 39%, respectively. Thus, the individual spectral components might be visible in the composite spectrum under favorable circumstances. Here we show, however, that the two companions are probably broad‐lined objects that will require both high signal‐to‐noise ratio (S/N) spectra and a good separation algorithm (such as Doppler tomography; see Liu et al. 1997) for a successful resolution of components.

The highest S/N ratio spectra available at the moment are those made by Fullerton (1990) at the CFHT in the yellow part of the spectrum. We show in Figure 2 the profiles of the He i λ5876 line in HD 193322A at the beginning and ending of the run, which clearly demonstrate the redward motion of the narrow‐lined component. There is also an extremely broad‐lined, shallow component (with a width equivalent to a projected rotational velocity of Vsin i ≈ 350 km s⁻¹) that shows little if any motion over the same interval. The equivalent widths of these two components are comparable, and it is the presence of this broad component (in some of the lines) that led Walborn (1972) to append the "((n))" suffix to the classification. We see similar broad components in the KPNO spectra of He i λ6678 (blended with He ii λ6683 in the red wing) and in the DDO spectra of lines such as He ii λ4541, Si iii λλ4552, 4567, 4574, and O ii λλ4590, 4596. However, the broad component is absent in the CFHT spectra of C iv λλ5801, 5812 and in the KPNO spectrum of O iii λ5592, and these characteristics suggest that the broad component arises in the spectrum of an early B‐type star (Conti 1974). Because this broad component appears essentially stationary in our spectra (and, in particular, does not show motion opposite to that of the primary), we tentatively assign it to the slow‐moving speckle companion. The lack of any readily visible spectral component from the remaining, close secondary star is surprising considering our rough estimate for its flux contribution (14%). The simplest explanation is that it too has a broad‐lined spectrum, and so we caution the reader that the apparent broad‐lined component may result from the spectra of both the nearby and distant companions.

Zoom In Zoom Out Reset image size

The excellent seeing conditions at CFHT allowed Fullerton (1990) to obtain a single, uncontaminated spectrum of the visual companion, HD 193322B, and the He i λ5876 profile in the spectrum of this star is shown as the upper plot in Figure 2. The equivalent width of this line is greater in the B component (indicative of an early B‐type classification), and the profile shows an extended blue wing that is suggestive of a second component. The radial velocity obtained by fitting a single parabola to the core of the entire composite profile is -16.7 ± 3.0 km s⁻¹, well below the systemic velocity of the A component (Table 2). We did not attempt to make a two‐component fit to this profile because of the unknown broadening of each component and because we have only one profile available. Taken together, these results suggest that the B component is an unresolved double‐lined spectroscopic binary. If so, then the central HD 193322AB system contains at least five stars (possibly seven if the C and D components are physical).

The HD 193322A system is one of some 20 O‐type speckle binaries known to be triples with an inner spectroscopic binary and distant companion (Mason et al. 1998). This system holds the claim for the longest period for the inner binary among such triples (and long for any spectroscopic binary containing O stars; only HD 15558, with a period of 439 days, and 15 Mon, with a period of 23.6 yr, have longer periods in the list of Mason et al. 1998). The resulting ratio of semimajor axes of the outer and inner binaries, a_outer/a_inner ≈ 60, is small, and the configuration may be prone to dynamical instability (Eggleton & Kiseleva 1996). The presence of such close triples in young clusters may provide the mechanism to eject runaway OB stars through close gravitational encounters with binaries (Gies & Bolton 1986; Leonard 1995; Leonard & Duncan 1990; Clarke & Pringle 1992; Iben & Tutukov 1997).