The Origin of B-type Runaway Stars: Non-LTE Abundances as a Diagnostic

Twenty-six targets were observed in 2014 August, using the ESO FEROS instrument (Kaufer et al. 1999), a high-resolution (R ≈ 48,000) prism cross-dispersed echelle spectrograph. FEROS has almost complete spectral coverage from 3500 to 9200 Å and provides a high S/N (≈ 300—see Table 2) in relatively short exposure times for bright stars. All data were reduced using the ESO FEROS pipeline (version 1.57). Multiple exposures were combined using either a median or a weighted average, within iraf.⁶ Both these methods of merging the exposures resulted in very similar spectra, with the weighted average giving a slightly higher signal-to-noise ratio, and so these were adopted. Occasionally cosmic ray events were visible, but these were easily removed from the spectra manually, if they interfered with any analysis.

Four targets (HD 1999, HD 165955, HD 204076, and HD 208213) were deemed unusable for abundance analysis. HD 1999 showed double lines in its spectrum, making quantitative analysis unreliable. HD 165955 has a very high $v\sin i$, so many lines became unobservable. HD 204076 proved to be a double-lined spectroscopic binary when it was observed with UVES. HD 208213 did not have a sufficient S/N to identify the Si iii lines required for our analysis. It was also observed with UVES and discarded for the same reason.

Seventeen targets were observed at the Very Large Telescope using the UVES instrument (Dekker et al. 2000), a high-resolution (R ≈ 80,000), high-efficiency, cross-dispersed echelle spectrograph with a blue arm and a red arm. A standard setting (437+760) was used, yielding a wavelength coverage of 3730–4990 Å in the blue arm and 5650–9460 Å in the red arm. All of the data were taken directly from the ESO archive, having been reduced using the ESO UVES pipeline (Ballester et al. 2000). Multiple exposures were normalized and combined using either a median or a weighted σ-clipping algorithm, within IDL. As the S/N of the individual exposures was high, the final spectra from both methods were effectively indistinguishable. Again, any cosmic ray events that interfered with subsequent analysis were removed manually.

As the UVES data set was obtained to extend our study to a higher $v\sin i$ and fainter targets, some of these proved particularly difficult to analyze and five were discarded, leaving 12 for which atmospheric parameters and abundances have been estimated.

Fourteen runaway candidates were observed at the W. J. McDonald Observatory with the Tull coudé echelle spectrograph at the 2.7 m Harlan Smith telescope (Tull et al. 1995), with a high spectral resolution (R ≈ 60,000), during 2013 May and June. Spectra were obtained covering a wavelength range of 3400–10900 Å. The echelle data were split into orders, each of which was reduced separately using standard iraf procedures.

For stars with significant rotational broadening and/or large surface gravities, the Hδ line profiles extended over a significant fraction of the order. Therefore, the normalization of these data proved more difficult than the normalization of those from UVES and FEROS. To deal with this concern, blaze fits were made to the bracketing orders, where the continuum was obvious, and these were averaged to provide a blaze profile for the order containing Hδ. This was then used to rectify the orders containing Hδ. All data were subsequently normalized and multiple exposures combined using either a median or a weighted σ-clipping algorithm, within IDL, as with the UVES observations discussed above.

Three of the McDonald targets were not analyzed. HD 69686 and HD 118246 did not have a sufficient S/N to identify the rotationally broadened Si iii lines, preventing estimates of the projected rotational velocities, effective temperatures, and microturbulences. HD 203664 had complex and variable spectra (Aerts et al. 2006) and so was removed from the sample.

The three characteristic parameters of a static stellar atmosphere (effective temperature, surface gravity, and microturbulence) are interdependent,  so an iterative process was used to estimate these values (see Fraser et al. 2010; McEvoy et al. 2015 for more details). The parameters are described separately in the subsections below. The final values for these parameters are given in Table 3.

4.1.1. Effective Temperature

Effective temperature (T_eff) estimates were determined using the Si ionization balance. Equivalent widths of the Si iii multiplet (4552, 4567, 4574 Å) were measured, together with those for the Si iv lines at 4089 and 4116 Å in the hotter targets and those of Si ii at 4128, 4130 Å in the cooler stars. For narrow-lined, high S/N targets, a simple Gaussian profile fit was sufficient to obtain a reliable equivalent width measurement (see Figure 1, upper left plot). The uncertainties in these measurements are typically of the order of 10% (Hunter et al. 2007). For stars with higher projected rotational velocities ($v\sin i$ ≥ 50 km s⁻¹), it was more appropriate to fit rotationally broadened profiles, as rotation becomes the dominant broadening mechanism (see Figure 1, upper right plot). In some cases it was not possible to measure the strength of either the Si ii or the Si iv spectrum. For these targets upper limits were set on their equivalent widths, allowing constraints to the effective temperatures to be estimated (see, e.g., Hunter et al. 2007 for more details). This was mostly the case for stars with large projected rotational velocities and midrange temperatures (18,000–26,000 K). The random uncertainty in our effective temperatures is approximately ±1000 K (about 5%), consistent with the high quality of the observational data. In those cases where upper limits were set for the equivalent widths of the Si absorption lines, the values will obviously be more uncertain and so error bars of ≈2000 K are more appropriate.

Zoom In Zoom Out Reset image size

4.1.2. Surface Gravity

4.1.3. Microturbulence

N, Mg, Si, and in some cases C abundances for each star have been estimated using the atmospheric parameters given in Table 3 and measurements of the absorption lines. Abundances are given in Table 3. The atmospheric parameters agree well between the observations with the different spectrographs. Effective temperatures show differences not larger than 1200 K in all cases, while the range of the logarithmic gravity estimates is only greater than 0.2 dex in one instance (HD 121968 with a range of 0.25 dex). Differences in microturbulence are less than 6 km s⁻¹ in all cases. The agreement between atmospheric parameters derived from observations obtained with different instruments is reassuring but not surprising given the high quality of all spectra. This agreement is of course not a measure of any systematic errors.

Extensive appendices in Hunter et al. (2007) show how errors in the atmospheric parameters for B-type stars translate into errors in derived abundances. Hunter et al. (2007) considered the errors to be independent but in reality the situation will be more complicated, as this is not the case. For example, an increase in the effective temperature estimate will lead to an increase in the gravity estimate, and this leads to the theoretical N ii equivalent widths (and hence N abundances) being less sensitive to changes in the atmospheric parameters than if these are considered to vary independently.

For the range in atmospheric parameters found for our sample, the simulations of Hunter et al. (2007; see, e.g., their Figure 6 for the N ii 3995 Å line) imply that our estimated errors in both the gravity and the microturbulence lead to relatively small errors in the N abundance estimate of 0.1 dex or less. Errors in effective temperature estimates are more important and lead to larger uncertainties with an error of ΔT_eff  ± 1000 K translating into an N abundance error of approximately ∓0.2 dex at an effective temperature of 18,000 K but decreasing to ∼0.1 dex at an effective temperature of 25,000 K. In addition, there will be random errors in the N abundance estimates due to uncertainties in the N ii equivalent widths. The latter have been estimated as ±10%, which would imply an uncertainty of ∼0.1 dex. Combining these different sources of error in quadrature would lead to typical uncertainties of 0.2–0.3 dex.

For the atmospheric parameters and abundances in Table 3, the microturbulence was derived from the three Si iii lines with the condition that the three lines return the same Si²⁺ abundance. This method was successful for all but EC 05582-5816, for which lines are greatly rotationally broadened. As we noted above, the alternative method is to assume a standard Si abundance of 7.4 for the determination of the microturbulence and hence other atmospheric parameters that finally define the model atmosphere used to determine the abundances of C, N, and Mg. (This alternative method was used by Hunter et al. 2007 in their analysis of B stars in three Galactic clusters.) The C, N, and Mg abundances are little affected by how the microturbulence is chosen; the preferred and alternative methods give similar results. This is well shown by Figure 2, where we show the N abundances obtained when the microturbulence is set by the condition that the Si abundance is 7.4 plotted versus the abundances obtained when the microturbulence comes from the three Si iii lines. It is seen that the N abundances are not systematically different in the two cases except for two or three outliers. Obvious outliers are HD 18100 and BD +36 2268 with N abundances obtained from the alternative method of 7.97 and 8.20, respectively. A similar correspondence between abundances from the two methods is found for Mg and for the limited number of determinations of the C abundances. Thus, apart from the artificial constraint of a constant Si abundance, the abundance analyses for all other elements are expected to be largely unaffected by the method used to determine the microturbulence and hence the atmospheric parameters.

Zoom In Zoom Out Reset image size

For a subsample of our stars (all stars observed with FEROS) we estimated C abundances. The C ii line at 4267 Å was used to calculate abundances as it is the only measurable line in all of the spectra. This line is known to be susceptible to (subtle) non-LTE effects (Nieva & Przybilla 2006, 2008), which are not fully taken into account in the model ion that was included in the TLUSTY calculations—see discussion by Hunter et al. (2009). Our interpretation of the various abundances is in part referenced to the abundance analysis of Galactic clusters by Hunter et al. (2007), who also exclusively used the 4267 Å line; thus the inability to account fully for the non-LTE effects should be almost canceled by the comparison with the Galactic clusters.

N abundances in Table 3 were estimated from the singlet transition at 3995 Å as it is one of the strongest N ii lines in the optical spectrum and appears unblended. Two examples of spectra of high $v\sin i$ stars around the N line at 3995 Å are shown in Figure 3. Other N ii lines were also present in a significant number of observations, including those between 4601 and 4643 Å and the singlet at 4447 Å. These tended to be more blended and so may lead to less accurate abundance estimates than those from 3995 Å. It is interesting to note that Lyubimkov et al. (2013) found that the line at 3995 Å gave a lower N abundance in their sample compared with the abundance from other transitions and that the differences were a function of effective temperature, ranging from −0.3 dex at 16,000 K to 0.0 dex at 29,000 K. In our case, the difference is about −0.1 dex over the full temperature range with a rise to 0.0 dex below about 18,000 K. These differences with Lyubimkov et al. (2013) likely reflect differences in the adopted model atoms. By referencing our abundances from the 3995 Å line to those from Hunter et al. (2009), we expect to obtain a true measure of abundance differences between the cluster (i.e., the Galactic disk) and the runaway B stars. Mg abundances were estimated for all spectra using the Mg ii transition at 4481 Å, comprising three overlapping lines of a single multiplet.

Zoom In Zoom Out Reset image size

Si abundances are obtained from the lines of Si ii, Si iii, and Si iv referred to in the determination of the atmospheric parameters.

Since the Mg and Si abundance should be unaffected by mixing within evolving B stars but could conceivably be altered in runaway stars created by the BSS, we discuss this pair of elements ahead of our discussion of C and N. Our reference Galactic B stars are the three open clusters analyzed by Hunter et al. (2009): NGC 6611, at a Galactocentric distance R_G of 6.1 kpc; NGC 3293, at ${R}_{G}=7.6\,\mathrm{kpc}$; and NGC 4755, at ${R}_{G}=8.2\,\mathrm{kpc}$, where R_G is taken from Rolleston et al. (2000). (Daflon & Cunha 2004 give somewhat different estimates.) Abundances provided by Hunter et al. (2009) correspond to the atmospheric parameters chosen by assuming an Si abundance of 7.4 (see Section 4 for comments on such a choice). Since the 2.1 kpc difference in Galactocentric distance across the three open clusters should correspond to an abundance difference of only about 0.09 dex, the assumption of constant Si abundance is likely not a source of significant systematic error. The Mg abundances for the three Galactic clusters are just consistent with the anticipated 0.09 dex decline between NGC 6611 and NGC 4755. Of course the Si abundances are not expected to betray the abundance gradient because the condition Si = 7.4 was imposed on the analysis.

For the runaway B stars, the Mg and Si abundances from Table 3 are compared in Figure 4, which shows that Mg and Si abundances are highly correlated and that each range over 1.0 dex. The straight line in Figure 4 corresponds to a constant Mg/Si ratio set at the "standard" ratio of −0.17 dex adopted by Hunter et al. (2009). The spread in Mg abundances among the runaway stars far exceeds the range among stars from an individual open cluster, which is a fair measure of the measurement uncertainties given that stars within a cluster share a common Mg abundance. A major contribution to the Mg and Si abundance spread among the runaway stars surely arises because their birthplaces span quite a range in Galactocentric distances even though present Galactocentric distances may not be that different, a range much greater than the range spanned by the three reference open clusters. Indeed, Galactic orbits calculated by Silva & Napiwotzki (2011) give birthplaces from the Galactic center out to nearly 14 kpc for the sample drawn from their paper. However, plots of the Mg and Si abundances versus the estimated Galactocentric distance of a star's birthplace are too ill defined to confirm the abundance gradients obtained from in situ abundance measurements of young stars and H ii regions. All but three of our stars have estimated birthplaces between 4 and 10 kpc. The failure to reproduce the observed abundance gradients may be attributable to uncertainties in the locations of the birthplaces and to the contamination of the stars in the BSS by Mg and Si from the SN.

Zoom In Zoom Out Reset image size

Scatter in Figure 4 about the linear relation is consistent with Mg and Si abundance uncertainties and with the spread in Mg abundances within each of the open clusters (see Figure 7, for example). Presence of the linear correlation and a constant Mg/Si ratio are qualitatively expected from the similar nucleosynthetic yields of Mg and Si from SNe II, which are surely the controlling influences on the Galactic abundance gradient whatever the influence of the myriad other factors (e.g., the initial mass function, the star formation rate, Galactic infall etc.) entering recipes for Galactic chemical evolution.

Hunter et al. (2009) found that among the B stars in the three reference clusters, the C and N abundances were independent of position on a star's evolutionary track except that half of the few supergiants were enriched in CN-cycled material. Similarly, Lyubimkov et al. (2013) found normal C and N abundances in a sample of 22 slightly evolved B stars (i.e., no supergiants) except for two possibly mixed stars. Thus the expectation is that the majority of the B stars would have preserved their initial C and N abundances before undergoing conversion to a runaway star. The likely corollary of this expectation is that alterations to the C and N abundances may be clues to the process creating the runaway star but that internal CN cycling and mixing to the surface may have occurred independently of formation as a runaway star.

As noted above, the C abundances in Table 3 are provided only for the stars observed with FEROS. Figure 5 shows the C and N abundances for this minority. In Table 3 the C and N abundances for EC 05582-5816, unlike for other stars, are taken from an analysis with the Si abundance fixed at 7.4 but, as noted above, the C and N abundances are expected to be independent of this assumption concerning the Si abundance; the high $v\sin i$ of EC 05582-5816 precluded the use of the Si iii multiplet to determine the microturbulence. The C and N abundances are tightly correlated with all stars showing a similar C/N to within the measurement uncertainties. In addition, the three Galactic clusters analyzed by Hunter et al. (2009) have mean C and N abundances falling within the band defined by the runaway stars; that is, mean abundances of 8.00 ± 0.19 and 7.62 ± 0.12 for C and N, respectively, for all nonsupergiants in the three Galactic clusters with these abundances provided by analyses assuming that Si has its standard abundance.

Zoom In Zoom Out Reset image size

These carbon abundances for the subset of our runaway stars show no evidence for a runaway star enriched in CN-cycled material, i.e., a low C abundance paired with a high N abundance such that the total number of C and N atoms is conserved. Very severe contamination with CN-cycled material would result in an N enrichment of up to 0.5–0.6 dex and a severe depletion of C. No such stars are seen in Figure 5. The presence of CN-cycled material may occur in the B star prior to its conversion to a runaway star and is thus not a determining signature of a process resulting in a runaway star.

To extend the search for abundance anomalies to the complete sample, we examine next the relationship between the N and the Mg abundances. Figure 6, drawing on Table 3, compares the N and the Mg abundances with a distinction made according to surface gravity (i.e., evolutionary status) with stars with log g < 3.2 (i.e., supergiants) represented by open circles and stars of higher surface gravity (i.e., main-sequence and slightly evolved stars) represented by filled circles. Figure 7 shows the N and Mg abundances for B stars belonging to the cluster NGC 3293, the best represented of the three clusters studied by Hunter et al. (2009).

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

With the possible exception of N enrichment from internal mixing or mass loss, one expects the cluster's stars in Figure 7 to share the same N and certainly the same Mg abundance. Thus, the scatter in the N and the Mg abundances in Figure 7 represents the measurement uncertainties that, thanks to the similarity of analytical techniques, will be a very close approximation to the uncertainties affecting the points in Figure 6. (One star—a supergiant—in NGC 3293 appears N-enriched, but this star is not C-depleted.) A comparison of Figures 6 and 7 shows the spread shown by NGC 3293's B stars and hence we conclude that the runaways show an intrinsic star-to-star difference in composition (i.e., C, N, Mg, and Si abundances but with similar abundance ratios), which we attribute to differences in a star's birthplace, with the Galactic abundance gradient being a very likely controlling factor.

EC 05582-5816 and HD 179407 appear as outliers in Figure 6. The two outliers have similar compositions except for their Mg abundances and both appear related to the inner Galaxy. Silva & Napiwotzki (2011) estimate EC 05582-5816's birthplace at 2 pc from the Galactic center. HD 179407 is presently at ${R}_{G}=3.5\,\mathrm{kpc}$ according to Smartt et al. (1997). The pair have very similar C and N abundances (Table 3) but differ substantially in Mg with an abundance of 8.0 for HD 179407 but 7.0 for EC 05582-5816. The Si abundance of HD 179407 is consistent with its Mg abundance. For EC 05582-5816, the present analysis assumes an Si abundance of 7.4. Galactic abundance gradients likely account for the high abundances found for HD 179407. There is a remarkable similarity in composition between EC 05582-5816 and the B9 III secondary of the BH binary V4641 Sgr. This B star has [Fe/H] = 0 and normal abundances of C, O Mg, Al, Si, and Ti but high N and Na overabundances ([N/H] = [Na/H] = +0.8) (Sadakane et al. 2006). If placed in Figure 6, it would provide a third outlier and fall near the two existing outliers. Reobservation and reanalysis of EC 05582-5816, a rapid rotator, is to be encouraged and extended to additional elements.

N, Mg, and Si abundances across the sampled runaway B stars are similar to those seen in B stars in Galactic clusters. The spread in the C, N, Mg, and Si abundances and the nearly uniform abundance ratios C/N/Mg/Si among runaway stars in our sample results from the combination of two facts: (1) the stars' birthplaces sample a wide range in Galactocentric distances, and (2) the abundances in star-forming regions decline with increasing Galactocentric distance at a rate of approximately 0.05 dex kpc⁻¹. Within our sample there are no certain outliers with abundance peculiarities. Thus the conclusion is that the two scenarios—BSS and CES—capable of ejecting B (and other) stars from the Galactic disk into the Galactic halo do not as a rule change the surface chemical composition of the runaway star, as sampled by C, N, Mg, and Si. This conclusion from our non-LTE analysis confirms earlier LTE analyses in which the compositions of a runaway star and a comparable B star in the Galactic disk are compared—see, e.g., Martin (2004). Although changes may occur in rare cases, the general conclusion is both disappointing and challenging: disappointing in that the composition, a readily obtainable quantity, appears not to be a discriminant between competing ejection mechanisms; challenging in that other observable quantities now have to been relied on to identify the principal ejection mechanism across a sample of runaway stars and for individual runaway stars. Possible observables include projected rotational velocities and binarity.

The distribution of projected rotational velocities $v\sin i$ for our sample of runaways is very similar to that provided by the B stars in the three Galactic clusters observed by Hunter et al. (2009): velocities up to 300 km s⁻¹ with a peak near 50 km s⁻¹ and few slow rotators $(v\sin i\leqslant 30$ km s⁻¹). Martin (2006) considered rotational velocities for his sample of runaways and found their frequency distribution to be similar to that assembled by Guthrie (1984) from "young" OB associations whose result is similar to that from the three Galactic clusters. In contrast, B stars in "old" associations and field B stars in the Galactic disk have distributions that rise with a decreasing $v\sin i$ (Wolff et al. 1982; Guthrie 1984). A more recent catalog of projected rotational velocities for 102 northern B stars is provided by Abt et al. (2002).

Tying the $v\sin i$ distribution of the runaways to the ejection mechanisms is unfortunately hampered by the absence of quantitative predictions for the BSS and the CES. In the BSS, if the runaway originates in a close binary system the runaway may be tidally locked, resulting in the rotation velocity being closely related to its orbital velocity. In turn, the ejection velocity is expected to be related to the orbital velocity. These ideas lead to the expectation that the ejection velocity should be positively correlated with the rotational velocity, but as Martin (2006) demonstrated, runaway stars do not show the expected correlation and therefore the BSS is not the leading producer of runaway stars. For the CES, the likelihood of ejection is plausibly higher in denser environments. On this simple premise and the match between the $v\sin i$ frequency distribution for runaways and young associations, Martin (2006) considered that it seems likely that the [CES] is the dominant mechanism behind his runaway sample and by extension ours too, which has overlap with his sample.

In the BSS, the resulting runaway B star will either be a single B star or form a binary system with a low-mass compact companion (e.g., an NS). A runaway that is single will be paired with a distant ejected NS or a BH moving in a generally opposite direction. Portegies Zwart (2000) suggests that 20%–40% of the runaway B stars should be accompanied by an NS and thus the runaway B star should appear as a single-lined spectroscopic binary with a period of several hundred days.

In the CES, simulations (Leonard & Duncan 1990) suggest that 10% of the runaways will be binaries composed of normal main-sequence stars and likely observable as double-lined spectroscopic binaries.

A definitive test of these predictions for the BSS and the CES will require a radial velocity long-term survey of a large sample of runaway B stars to be put up against more precise predictions about the frequency and nature of the binary populations from the two scenarios. Then, it may be possible to assess in a statistical fashion the relative production rates from the BSS and the CES. For the binaries, it should be possible to assign them to either the BSS (i.e., an NS/pulsar companion) or the CES (i.e., a normal main-sequence companion). For the runaway single stars, the attribution to the formation scenario will be difficult unless one can uncover a subtle abundance anomaly or apply a precise determination of space motions (GAIA)⁸ to assist in the identification of the pulsar or the original stellar association.

A definitive observational test is not possible at present. The sole radial velocity survey of high Galactic OB stars was conducted by Gies & Bolton (1986), who concluded that runaway OB stars are deficient in close binaries by a factor of 2–4. Their sample of 15 confirmed runaways provided two binaries (both double-lined systems), and if five probable runways are added, the binary probability becomes 2/20 or 10%. The deficiency of binaries is suggested by reference to the binary fraction of 31% among normal O stars (Garmany et al. 1980) and of 38% among normal B stars (Abt & Levy 1978). Four of our stars (HD 97991, 149363, 214930, and 219188) were in the Gies & Bolton (1986) survey and declared by them to have a constant radial velocity. Martin (2003) found the runaway HD 138503 to be a double-lined spectroscopic binary, a star in Silva & Napiwotzki's (2011) list of runaway stars. HD 1999 and HD 204076, which we observed, are double-lined binaries. Two radio surveys for pulsars reported no detections coincident with runaway OB stars (Philp et al. 1996; Sayer et al. 1996). Unfortunately, the correction for the beaming of pulsar radiation results in an uninteresting limit on the fraction of runaway stars with low-mass compact companions; Sayer et al. (1996) estimate that less than 25%–50% of OB runaways have an NS companion, an (1996) estimate consistent with the prediction by Portegies Zwart (2000).

Our sample can make only a modest contribution to the frequency of binary runaway stars as a test of the BSS and the CES. For the majority of our sample the star was observed once and an assessment of the radial velocity variation must come from velocities reported in the literature. Eight stars were observed at two or more telescopes and thus at different epochs, and the two or three radial velocity measurements may be intercompared and also checked against the literature. (This restricted search for velocity variations is akin to that reported by Martin (2006), who observed many runaway stars at least twice over a few days and also checked velocities against the literature.) We have defined a star with a radial velocity variation between measurements of greater than 20 km s⁻¹ as a possible member of a binary system. This cutoff is set by an inspection of other studies of OB stars (Sana et al. 2013; Dunstall et al. 2015). In some stars, significant velocity variations arise from pulsations, which complicates the search for orbital radial velocity variations, and if neglected as a source of radial velocity variations, the fraction of spectroscopic binaries is overestimated.

Radial velocities are provided in Table 4. Of the eight stars observed twice or thrice by us, seven appear to have a constant velocity. The star with a variable radial velocity is HD 204076, which is certainly a spectroscopic binary; the UVES but not the FEROS spectrum showed double lines. HD 219188 is possibly a variable. There is a 25 km s⁻¹ difference between our two measurements, and online catalogs give a velocity that is either close to the mean of our two or a less positive velocity: 84 km s⁻¹ according to Silva & Napiwotzki (2011) or 64 km s⁻¹ according to Gontcharov (2006) and Kharchenko et al. (2007).

Inspection of Table 4 suggests that four stars observed once by us may be binaries because of a difference with radial velocities reported in the literature. EC 20140-6935 (HD 192273) was noted as a possible binary by Magee et al. (1998), who found a 45 km s⁻¹ velocity difference between their measurement and that reported by Rolleston et al. (1997). Silva & Napiwotzki (2011) gave the velocity as +17 km s⁻¹ from Rolleston et al. (1997). Our velocity is in good agreement with that by Magee et al. (1998) HD 188618 was suspected by Martin (2006) to be a binary from the velocity difference between his measurement of 29 ± 6 km s⁻¹ and a previous measurement of −15 km s⁻¹ by Duflot et al. (1998). Our −46 km s⁻¹ extends the velocity range. HD 179407's present and previous velocity measurements barely satisfy our 20 km s⁻¹ condition. HIP 70205 (LP 857-24) exhibits a 300 km s⁻¹ difference but the only previous measurement is from the RAVE survey (Kordopatis et al. 2013), which is possibly ill suited to velocity measurements of B stars that provide few lines in the RAVE bandpass. Obviously, this star deserves further attention. Finally there is HD 114569, which has no previous velocity measurement.

It seems fair to conclude that our sample contains few spectroscopic binaries and certainly fewer than the approximately 30% provided by a sample of field B stars in the Galactic disk (Abt & Levy 1978). Our sample is very slightly biased by the exclusion of known double-lined or even single-lined spectroscopic binaries. HD 1999 and the double-lined eclipsing star HD 138503 (Martin 2003) are the only binary stars observed, but they are not in Table 4. With the present understanding of the formation of single and double runaway stars by the BSS and the CES, it is not possible to interpret the low fraction of spectroscopic binaries as a pointer to the more important formation mechanism.