© ESO, 2015

1. Introduction

The VLT-FLAMES Tarantula Survey (VFTS) is a European Southern Observatory (ESO) Large Programme that has obtained multi-epoch optical spectroscopy of over 800 massive stars in the 30 Doradus region of the Large Magellanic Cloud (LMC). An overview of the survey, including details of the stellar photometry, spectroscopic observations and data reduction, was given by Evans et al. (2011; hereafter Paper I).

The VFTS observations are the first optical spectroscopy for many of the targets, and provide a wealth of new morphological information on the high-mass population of 30 Dor, which will underpin quantitative analyses of the spectra in future papers. Detailed classifications for the 352 O-type spectra (with discussion of notable morphological groups) were presented by Walborn et al. (2014), while classifications for the smaller samples of Wolf–Rayet and late-type stars were given in Paper I. Here we present spectral classifications for 438 B-type objects from the survey. Of these, only 74 (17%) have previously published classifications based on digital data, many of which were obtained at a lower spectral resolving power.

To complement the spectral classifications, we also present estimates of radial velocities for the B-type sample. These estimates provide insights into the different populations within 30 Dor, and enable identification of candidate runaway stars that have been ejected from their birth sites. In the context of runaway stars, the size of the VFTS sample is unprecedented for a single, young star-forming region. The survey has revealed ~20 O-type runaway candidates (Evans et al. 2010, Sana et al., in prep.), with an apparent connection with rapid rotation (Walborn et al. 2014, Sana et al., in prep.). There are also a number of stars with moderate (projected) rotational velocities in the peripheral regions of 30 Dor, which could perhaps be very rapid rotators (but with low inclinations; see Walborn et al. 2014, for a discussion of their spectra). For completeness, we note that two additional O2-type stars near 30 Dor, Sk−68° 137 and BI 253 (=VFTS 072), were suggested as runaways by Walborn et al. (2002a) on the basis of their remote locations for their apparent youth.

In his pioneering study, Blaauw (1961) noted that the fractional incidence of Galactic runaways is larger among O stars than among B stars, a result substantiated by subsequent authors, and which also emerged in the simulations from Portegies Zwart (2000). In this paper we therefore complement the studies of the O-type runaways in the VFTS with analysis of the B-type stars. In particular, the number and mass spectrum of the ejected runaways from 30 Dor will provide important constraints on models of cluster formation and evolution (e.g. Fujii & Portegies Zwart 2011). The estimated radial velocities presented here, combined with those from the O-type stars (Sana et al. 2013), will also provide an important input for comparisons with theoretical predictions, e.g. in the context of simulations to investigate the initial conditions leading to the distinct clump identified to the north-east of R136 (Sabbi et al. 2012).

In Sect. 2 we summarise the relevant features of the VFTS data and their post-processing. Section 3 sets out the framework within which the spectra were classified and discusses notable objects. Section 4 details the methods we used to estimate radial velocities for each target, followed by a discussion of members of the two oldest clusters in the survey region (Hodge 301 and SL 639) in Sect. 5. Finally, in Sect. 6, we use our velocity estimates to identify candidate single-star and binary runaways and discuss their nature and potential origins. In parallel to this study, Dunstall et al. (in prep.) have exploited the multi-epoch VFTS observations to investigate the multiplicity properties of the B-type sample.

2. Observations and data processing

All of the VFTS data discussed here were obtained using the Medusa–Giraffe mode of the Fibre Large Array Multi-Element Spectrograph (FLAMES) instrument on the Very Large Telescope (VLT). The Medusa fibres (which subtend 1.̋2 on the sky) were used to relay light from up to 130 targets simultaneously to the Giraffe spectrograph (see Pasquini et al. 2002).

Full details of the observational strategy and reduction of the data are given in Paper I. In brief, the ESO Common Pipeline Library FLAMES recipes were used to undertake bias subtraction, fibre location, summed extractions of each object, division by a normalised flat-field, and wavelength calibration. Subsequent processing included correction of the spectra to the heliocentric frame, sky subtraction, and rejection of significant cosmic rays.

The Medusa observations used three of the standard Giraffe settings (LR02, LR03, and HR15N), providing spectral coverage of λ3960−5071 Å (at resolving powers, R, of 7000 to 8500), and λ6442−6817 Å (at R = 16 000). These resolutions are greater than for most published standards so, for the purposes of classification, the data were degraded to an effective resolving power of R = 4000, ensuring consistency with the approach to the classification of the O-type stars by Walborn et al. (2014).

For stars that appear to be single (Dunstall et al., in prep.) the LR02 and LR03 spectra were normalised then merged, with the σ-clip levels for cosmic rejection and relative weightings based on the signal-to-noise (S/N) ratios of the individual spectra. Finally, for classification, the merged LR02 and LR03 spectra were combined, then smoothed and rebinned to R = 4000. For stars identified as having significant but relatively modest velocity variations (Δ v_r ≤ 40 kms^-1), the individual spectra were similarly co-added and degraded (as the variations were within the effective velocity resolution of the rebinned data). For single-lined binaries with Δ v_r> 40 kms^-1, the single-epoch LR02 and LR03 spectra with the best S/N ratio were combined and then smoothed/rebinned. We were careful in these cases to ensure that any velocity offsets between features in the small overlap region between the LR02 and LR03 data were not misinterpreted. The small number of double-lined binaries were classified from inspection of the individual spectra.

3. Spectral classification

The spectra were classified by visual comparison to the B-type Galactic standards from Sota et al. (2011) and Sana et al. (in prep.), while also taking into account the effects of the reduced metallicity of the LMC compared to the Galaxy (which can affect the appearance of both the metal and helium absorption lines; e.g. Markova et al. 2009). This was achieved for the luminous, low-gravity objects by interpolating between the available Galactic standards and those for supergiants in the Small Magellanic Cloud (Lennon 1997), and with reference to the framework developed for the LMC by Fitzpatrick (1988, 1991). Classifications of the dwarfs and giants were also informed by comparisons with the standards from Walborn & Fitzpatrick (1990) and the LMC stars classified by Evans et al. (2006).

The primary diagnostic lines are the ionisation ratios of silicon, while also taking into account the appearance of the helium and magnesium (λ4481) lines. For reference, the primary temperature-sequence criteria used to classify the VFTS sample are summarised in Tables 1 and 2 (for supergiants and dwarfs, respectively). Example temperature sequences for supergiants, giants and dwarfs are shown in Figs. A.1−A.3, respectively. At a given spectral type, luminosity classes were assigned from the width of the Balmer lines and (at the earliest types) from the intensity of the silicon absorption lines, while also taking into account the possible effects of rotational broadening on the spectra. Example luminosity sequences for B0.2, B1, and B2.5 types are shown in Figs. A.4−A.6, respectively.

The supergiants and bright giants (i.e. class I and II objects) have rich absorption-line spectra. Given the importance of spectral peculiarities in the context of CNO abundances, and the influence of metallicity effects, these luminous objects were classified independently by three of the authors (CJE, NM, and NRW) with a typical consistency of 0.5 subtypes/luminosity class from the first pass; final types were adopted following discussion of individual objects as required. The larger sample of dwarfs and giants, which exhibited a considerable range in data quality, was classified by CJE. In many instances these classifications are less precise than those possible for the luminous objects, but they should provide sufficient morphological information to enable robust subdivisions of the sample in future analyses.

Each merged spectrum was classified in the framework described above. After initial classification of all the spectra, careful checks were undertaken to ensure self-consistency within the sample. An overlapping sample of ten B0-type stars was classified independently (by NRW) for the study by Walborn et al. (2014), with excellent agreement (to within one luminosity class) in all instances.

The catalogue of classifications and radial-velocity estimates for the B-type objects from the VFTS is presented in Table 7 (available online). For convenience, supplementary information is also provided for each star on its binary status (from Dunstall et al., in prep.), Hα morphology (see Sect. 3.1), optical photometry (from Paper I), alternative identifications used in the literature, and previous spectral classifications.

The S/N ratios of the supergiant spectra were particularly good (typically >100 for both the LR02 and LR03 regions; see Fig. A.1). This allowed us to provide consistent indications of the nitrogen enhancement/deficiency compared to morphologically-normal stars (via Nstr and Nwk qualifiers), as diagnosed from the CNO absorption features in the λλ4640−4650 region (Walborn 1976). The significant differences in CNO morphology are illustrated by the two B1.5 Ia spectra in Fig. 1. Atmospheric analysis of the supergiants, including estimates of nitrogen abundances and a thorough study of the line-broadening in their spectra, will be presented by McEvoy et al. (2014) and Simón-Díaz et al. (in prep.), respectively.

Given the wide range in quality of the dwarf and giant spectra, we do not comment on CNO morphologies in their classifications. Equally, while we employ the “n” and “(n)” indications of broadening (see e.g. Table 3 from Sota et al. 2011) at all luminosity classes, there are likely to be some lacunae (cf. the quantitative v_esini estimates from Dufton et al. 2013).

3.1. Emission-line objects

Where appropriate, spectra were initially classified as “Be+” on the basis of Fe II emission lines in the LR02 and LR03 spectra (see discussion by Evans et al. 2006); the Fe II lines were preferred over the blue-violet Balmer lines owing to the significant (and variable) contamination of the latter by nebular emission.

Independently of the blue-region classifications, we inspected the Hα profiles of every target to investigate their morphologies. The greater resolution of the Hα observations and the extended wings from disc emission made it easier to distinguish contributions from Be-type Balmer emission and nebular contamination than in the blue, but these classifications should still be treated with some caution. The Hα morphologies are encoded in the fourth column of Table 7 (with the notation explained in the corresponding footnote).

Objects classified as “Be+” from the LR02/LR03 spectra were seen to display Hα emission in all cases. Approximately 30 additional stars display Be-like Hα emission without Fe II emission. The Hα emission in luminous supergiants most likely arises from their stellar winds, but for less luminous objects such emission is the primary Be diagnostic. To preserve the distinction of the lines seen in emission, the standard “e” suffix is used in the classifications for the (non-supergiant) stars where only Hα is seen in emission.

The complex star-formation history of 30 Dor means that our sample will span a range of stellar ages, so we are unable to comment on the incidence of the Be-phenomenon for a single, co-eval population. Nonetheless, we note that the incidence of the Be-type classifications for the dwarf and giant stars is 18% (70/388 objects). This is in good agreement with the Galactic fraction (≥17%, Zorec & Briot 1997), and that reported for the field population near NGC 2004 in the LMC (of 16 and 17.5%, from Evans et al. 2006 and Martayan et al. 2006, respectively).

3.1.1. B[e]-type spectra

Two targets in the Medusa–Giraffe sample show forbidden-line emission characteristic of the B[e] category. VFTS 698 is a double-lined binary system with a B[e]-like spectrum, comprising what appears to be an early B-type secondary in orbit around a veiled, more massive companion (see Dunstall et al. 2012). The second B[e] object is VFTS 822, shown in Fig. 2 together with VFTS 1003 (the B[e]-like object from Paper I). Given the intensity of its Si II λλ4128-32 and Mg II λ4481 absorption, the spectrum of VFTS 822 was classified as mid-late B[e] (~B5-8); further discussion of its possible pre-main-sequence nature was given by Kalari et al. (2014).

Although the nature of VFTS 1003 remains uncertain at present, the number of evolved B[e] stars (at most two) in the VFTS is clearly small compared to the number of normal supergiants. This is in keeping with the small number of sgB[e] stars in Galactic clusters (see the appendix of Clark et al. 2013), arguing that the B[e] phenomenon in evolved supergiants must be a relatively brief phase or originate from a rare process.

3.1.2. VFTS 766: A peculiar emission-line star

VFTS 766 displays strong Hα emission and inspection of its blue-region spectrum (see Fig. 2) revealed it as a peculiar object. At first glance its spectrum appears to be that of a mid-late B-type supergiant (B5-B8, as traced by e.g. Si II, Mg II), but with superimposed metallic absorption lines, which are consistent with a cooler component. This seemingly composite appearance can indicate a circumstellar shell or disc observed edge-on, and this hypothesis is supported by the twin-peaked Hβ shell-like emission profile. Its near- and mid-IR colours¹ place it in the region in the J, J − [3.6] colour−magnitude diagram which is occupied by early-type stars associated with free-free/bound-free emission, and we find no evidence of hot dust (cf. Bonanos et al. 2009).

Similar emission profiles are observed in some low-luminosity supergiant B[e] stars (e.g. S59, Gummersbach et al. 1995), with examples that also appear to lack dusty discs (Graus et al. 2012, although none have yet been discovered in the edge-on orientation required to yield shell profiles). However, we speculate that VFTS 766 is a classical Be star, in which the twin-peaked emission in the wings of the Balmer profiles gives rise to the supergiant-like appearance. An example of this effect is the development of a circumstellar disc in the classical edge-on Be star o And, which conspires to yield a narrower absorption feature than the underlying photospheric profile (Clark et al. 2003). Another example is provided by the Be star 48 Lib (see Rivinius et al. 2013, and references therein), which has previously been classified as a supergiant (Houk & Smith-Moore 1988).

This possible explanation for the nature of VFTS 766 is supported by its magnitude relative to other stars in our sample. Spectroscopic monitoring to identify an episode of disc loss, during which the (uncontaminated) photospheric Balmer line profiles might be observed, would clarify its luminosity type and true evolutionary status.

3.1.3. Blue hypergiants

We identifed five blue hypergiants (BHGs) in our sample, namely: VFTS 003 (B1 Ia⁺), 424 (B9 Ia⁺p), 430 (B0.5 Ia⁺((n)) Nwk), 458 (B5 Ia⁺p) and 533 (B1.5 Ia⁺p Nwk). Recent theoretical and observational studies suggest that early-type BHGs might be massive stars which have passed through a blue supergiant phase and are about to enter a luminous blue variable (LBV) phase (Clark et al. 2012; Groh et al. 2014). This suggests that VFTS 003, 430, and 533 might be on the verge of LBV-like behaviour, although we note that VFTS 430 appears somewhat subluminous compared to the blue supergiants in 30 Dor, and is significantly fainter than VFTS 533 (even accounting for its apparently greater line-of-sight extinction, see photometry in Table 7). Later-type BHGs have been observed to undergo LBV-like photometric and spectroscopic behaviour (e.g. Smith et al. 2004; Clark et al. 2012, with potential implications for VFTS 424 and 458 here), suggesting that these two evolutionary phases are essentially synonymous for such stars.

On further inspection of their spectra, broad emission wings were identified in the Hβ profiles of each of the BHGs, except for VFTS 430. This Hβ morphology is seen in some LBVs and other luminous supergiants (e.g. Walborn & Fitzpatrick 2000), and is also present in VFTS 739 in our sample (classified as A0 Ip but, given its omission from Paper I, included here in the B-type sample). The spectra and evolutionary status of these potential LBV precursors/candidates will be discussed in more depth by Walborn et al. (in prep.).

3.1.4. Comparisons with X-ray observations

We compared our sample with the point-source detections from observations in 30 Dor with the Chandra X-ray Observatory (Townsley et al. 2006, 2014); none of the VFTS B-type stars is detected in the extant X-ray data. A new X-ray Visionary Project is now underway, “The Tarantula Revealed by X-rays (T-ReX)”, which will observe 30 Dor with Chandra for 2 Ms. This will push far deeper than the existing data, and we will revisit the VFTS sample once the T-ReX observations and analysis are complete.

4. Stellar radial velocities

4.1. Single stars

Radial-velocity estimates for the stars identified as apparently single by Dunstall et al. (in prep.) were obtained from Gaussian fits to selected absorption lines in the merged LR02 and LR03 spectra of each object (at the native resolution of the data, without the degrading applied for classification purposes). This approach is consistent with that used by Sana et al. (2013) for analysis of the O-type stars, albeit (necessarily) using a different subset of absorption lines. We note that the choice of Gaussian profiles may not be appropriate for fitting the wings of the lines in the presence of different broadening mechanisms, but they should suffice for robust estimates of the centres of our chosen lines.

We also considered a cross-correlation method to estimate the radial velocities. However, given the range in temperature and luminosity of the sample, as well as the varying data quality and problems of nebular contamination, the selection of suitable template spectra for cross-correlation was not trivial and would have introduced additional uncertainties (relating to e.g. the projected rotational velocities, v_esini). Thus, we did not pursue this approach further.

4.1.1. Line selection

The large range in v_esini and effective temperature for the sample meant that it was not possible to use a single set of diagnostic lines for all stars. Moreover, the large intrinsic width and nebular contamination of the Balmer lines precluded their use; similarly, nebular emission affects some of the stronger He I lines. Following the approach taken by Dufton et al. (2013), we used different sets of diagnostic lines (as summarised in Table 3) depending on the v_esini of the star.

For narrow-lined spectra (defined as v_esini ≲ 150 kms^-1), we used three non-diffuse He I lines and four isolated metal lines listed as “Set 1” in Table 3. Where possible, radial velocities were estimated from Si III λ4553 in both the merged LR02 and LR03 spectra, giving up to eight individual estimates for the narrow-lined objects.

We note that He I λ4121 is potentially affected by blending with O II λλ4120.28/54. In many instances these features could be resolved and a double Gaussian profile was fitted to estimate the He I and O II line centres; in cases where the O II lines were weaker or unresolved, they do not appear to have unduly influenced the results (see Table 3, in which the residual for He I λ4121 is consistent with the other lines).

Radial-velocity estimates for stars with broader lines were more difficult, particularly for spectra with significant nebular contamination. In these cases, velocities were estimated by fitting the five He I lines listed as “Set 2” in Table 3, for which the profiles are expected to be effectively symmetric².

Representative examples are shown in Fig. 3. The lower panel shows fits to the Si III triplet for two narrow-lined stars, VFTS 053 and 159, which have v_esini ≤ 40 (i.e. less than the velocity resolution of the data) and 96 ± 9 kms^-1, respectively (from Dufton et al. 2013). As an example of a spectrum of a rapidly-rotating star which also suffers nebular contamination, the upper panel of Fig. 3 shows the Gaussian fit to He I λ4388 for VFTS 636 (v_esini = 371 ± 35 kms^-1; Dufton et al. 2013).

For three stars with later/peculiar types – VFTS 272 (possible shell star), 739 (A0 Ip), and 766 (B5-8e, see Sect. 3.1.2) – we employed weak Fe II and Ti II absorption lines to estimate velocities, although we note that in VFTS 272 and 766 they might not be representative of the true (stellar) radial velocity, as these lines could trace cooler features associated with a disc of material (see e.g. the discussion of VFTS 698 by Dunstall et al. 2012).

4.1.2. Mean velocities

Radial-velocity estimates for the single stars were obtained by calculating an average of the n individual measurements (v_i), weighted by the estimated central depth (d_i) from the Gaussian fit of each line, i.e. ${\mathit{v}}_{\mathrm{r}}\mathrm{=}\frac{{\sum }_{\mathit{i}\hspace{0.17em}\mathrm{=}\hspace{0.17em}\mathrm{1}}^{\mathit{n}}{\mathit{v}}_{\mathit{i}}\hspace{0.17em}{\mathit{d}}_{\mathit{i}}}{{\sum }_{\mathit{i}\hspace{0.17em}\mathrm{=}\hspace{0.17em}\mathrm{1}}^{\mathit{n}}{\mathit{d}}_{\mathit{i}}}\mathrm{·}$(1)The sample standard deviation (σ) for each star was similarly calculated as a weighted average: ${\mathit{\sigma }}^{\mathrm{2}}\mathrm{=}\frac{{\sum }_{\mathit{i}\hspace{0.17em}\mathrm{=}\hspace{0.17em}\mathrm{1}}^{\mathit{n}}\mathrm{(}{\mathit{v}}_{\mathit{i}}\mathrm{-}{\mathit{v}}_{\mathrm{r}}{\mathrm{)}}^{\mathrm{2}}\hspace{0.17em}{\mathit{d}}_{\mathit{i}}}{{\sum }_{\mathit{i}\hspace{0.17em}\mathrm{=}\hspace{0.17em}\mathrm{1}}^{\mathit{n}}{\mathit{d}}_{\mathit{i}}}\mathrm{·}$(2)In cases where the central depths of the absorption lines were less than twice the inverse of the continuum S/N ratio, the individual estimates were excluded from the above calculations (to avoid their larger uncertainties degrading the results from stronger lines). For stars with estimates from three (or more) absorption lines, we present their estimated mean velocities and standard deviations in Cols. 7 and 8 of Table 7. For completeness, the line-by-line estimates from Sets 1 and 2 for the single stars are detailed in Tables A.1 and A.2, respectively; the mean uncertainty of the estimates for the 307 (presumed) single stars is 7.8 kms^-1 (and with a median uncertainty of 6.2 kms^-1).

As a check on our results, we used the methods described above to analyse a sample of the late O-type stars from the VFTS, finding excellent agreement between our velocity estimates and those from Sana et al. (2013, see discussion in Sect. A.2.1. We also investigated the internal consistency of our results from the selected lines by calculating the average difference (Δ) of the individual measurements for each line compared to the adopted mean velocity for each target. The means and standard deviations of these differences are given in Col. 3 of Table 3, with no significant shifts seen for any of the diagnostic lines compared to the adopted radial velocities.

4.2. Radial velocities of the binaries

To complement the analysis of the single stars, we also investigated the radial velocities of the single-lined binaries in the sample. We obtained single-epoch velocity estimates (v_single) for each system using measurements from the best LR02 spectra (in terms of S/N ratio of the co-added exposures from a given night, which span three hours at most).

Given the reduced spectral coverage of the LR02 data compared to the full LR02 + LR03 range, we used all of the available lines from Table 3 to estimate v_single (except for He I λ4121, given the nearby O II feature and the lower S/N ratio of the single-epoch spectra compared to the quality of the co-added spectra of the single stars). Line-by-line estimates and the date for which they were measured are listed in Table A.3. Mean velocities and standard deviations for the selected epoch were then estimated using the same methods as for the single stars.

The mean single-epoch velocities serve as an absolute reference point for the inter-epoch differential velocities calculated by Dunstall et al. (in prep.) yielding absolute estimates for each epoch. We then calculated the mean and standard deviation of the multi-epoch estimates to obtain a velocity estimate for each single-lined binary. These values are presented in Table 7, and are flagged as “B” (binary) in the ninth column. We note that the typical uncertainty on the cross-correlation results from Dunstall et al. was less than 5 kms^-1, while the typical error on the absolute estimates determined here from the single-epoch data was 9 kms^-1; i.e. the uncertainties in our estimates will generally be dominated by the analysis presented here.

Full orbital solutions are presently unavailable for the B-type binaries, so we strongly caution the reader that the velocity estimates in Table 7 are not their centre-of-mass, systemic velocities (although, as discussed in the Appendix, in many instances they may provide reasonable estimates).

4.3. Spatial distribution

In addition to NGC 2070 (the main 30 Dor cluster, which includes R136 at its core) and NGC 2060 (the association to the south-west), there are two smaller clusters in the VFTS survey: Hodge 301 (Hodge 1988) and SL 693 (Shapley & Lindsay 1963); their relative locations are shown in Fig. 4. Both clusters appear older than NGC 2060 and 2070 as they contain evolved supergiants, from B- through to M-types.

We calculated weighted mean velocities ($\overline{{\mathit{v}}_{\mathrm{r}}}$, and their standard deviations) for the full sample of single stars, for the subsamples associated with the four distinct stellar assocations in the survey region (NGC 2060, NGC 2070, Hodge 301, and SL 639, as defined in Table 4), and for the field population of stars outside the four clusters. These results, which exclude candidate runaway stars (see Sect. 6), are summarised in Table 4.

The results for the two younger star-forming regions (i.e. NGC 2060 and 2070) are in excellent agreement with the field population of B-type stars in the region, but the two older clusters (Hodge 301 and SL 639) appear at slightly lower recession velocities, suggesting that they are kinematically distinct from the rest of the sample. The weighted mean velocity for the 273 single stars excluding these two clusters (and potential runaways) is $\overline{{\mathit{v}}_{\mathrm{r}}}$ = 271.6 ± 12.2 kms^-1.

5. Spectral content of the older clusters

The VFTS data comprise the first comprehensive spectroscopy in Hodge 301 and SL 639, so we briefly discuss these clusters in turn.

5.1. Hodge 301

There are 20 VFTS targets within a 20′′ radius (chosen to conservatively identify likely members, and corresponding to ~4.9 pc, assuming a distance modulus of 18.5 mag). Fifteen of these were classified as B-type and their likely membership of the cluster is indicated in the final column of Table 7³.

From analysis of Hubble Space Telescope (HST) imaging, Grebel & Chu (2000) estimated Hodge 301 to be 20−25 Myr old. From narrow-band imaging they also identified 19 Be-type stars, eight of which (VFTS 272, 276, 279, 282, 283, 287, 293, and 301) have spectroscopy from the VFTS. The Hα spectra of each of these contain significant nebular emission (with multiple components in all sightlines except towards VFTS 293). Nonetheless, broad Be-like Hα emission is present in each object, except for VFTS 276 (GC00-Be2).

Grebel & Chu (2000) noted two potential blue stragglers in the cluster, which they argued were too luminous for the main-sequence population (but less luminous than expected for supergiants). The first, VFTS 270 (WB97-2), is a seemingly unremarkable B3 Ib star. The second is VFTS 293 (WB97-9, GC00-Be1), classified here as B2 III-II(n)e. From two previous observations, Walborn & Blades (1997) noted small shifts of its absorption features (relative to superimposed nebular emission) and possible weak He II λ4686 emission at one epoch. If the He II emission were real, Walborn & Blades speculated that it might be an X-ray binary. There is no evidence of λ4686 emission in the LR03 FLAMES spectra (albeit all taken on the same night, see Paper I), but small line-profile variations are seen in the absorption features in the multi-epoch LR02 spectra.

More intriguing is VFTS 310 on the periphery of the cluster and classified as O9.7 V: (Walborn et al. 2014). With such a classification, VFTS 310 would be expected to be significantly younger than the other cluster members, as shown by its location in the Hertzsprung-Russell (H-R) diagram of the cluster (Fig. 5, see Sect. 5.3 for details). The star appears otherwise unremarkable, with a low projected rotational velocity (v_esini≤ 40 kms^-1, Dufton et al. 2013) and a radial velocity (v_r = 272.3 ± 2.8 kms^-1, Sana et al. 2013) which is consistent with those in NGC 2070 and the local field population. Moreover, it does not appear to be a runaway from Hodge 301 (see Sect. 6). Whether this is a true blue straggler or simply a line-of-sight coincidence remains uncertain at this time.

5.2. SL 639

Adopting a 20′′ radius (~4.9 pc), sixteen of the VFTS targets are within SL 693. Fourteen are in the B-type sample, with their membership of SL 639 indicated in the final column of Table 7⁴. Ten of these stars are early-B dwarfs/giants, with six of them displaying Be-like Hα emission. The remaining four are comprised of two supergiants (VFTS 827, B1.5 Ib; VFTS 831, B5 Ia) and two bright giants (VFTS 826, B1 IIn; VFTS 829, B1.5-2 II).

There are previous age estimates for SL 693. From consideration of its main-sequence turn-off, Hodge (1983) estimated its age to be 18 Myr (although he noted a large uncertainty compared to the theoretical models of the time). A larger estimate of 30 Myr was given by Santos et al. (1995) from its reddening-corrected, integrated (U − B)₀ colour and adopting an empirical calibration of colour vs. age from Bica et al. (1990).

5.3. H-R diagram

Motivated by the apparent similarities of the two clusters in terms of their spectral content, we constructed an H-R diagram in order to compare the properties of their likely members. Ahead of detailed quantitative analysis of the VFTS spectra and pending extinction analysis of the whole sample, we adopted effective temperatures (T_eff) on the basis of the spectral types in Table 3 using the calibrations of Trundle et al. (2007), and interpolating between the Galactic and SMC temperatures for A-type supergiants from Evans & Howarth (2003); cooler objects were omitted given the larger uncertainties in their spectral classifications. Stellar luminosities were estimated using the optical photometry from Paper I, intrinsic colours from Fitzgerald (1970), bolometric corrections from Balona (1994), and assuming R_V = 3.5 (see Appendix C of Doran et al. 2013).

The H-R diagram is shown in Fig. 5, with SL 639 and Hodge 301 members indicated by the filled black and open red circles, respectively. Evolutionary tracks and isochrones (for non-rotating models) from Brott et al. (2011) are also plotted for comparison. As noted earlier, the late O-type star in Hodge 301, VFTS 310, stands out as a blue straggler with T_eff > 30 kK.

Tailored atmospheric and extinction analyses of these objects should help to provide further insight in due course. Nonetheless, the stars in SL 639 appear marginally younger than those in Hodge 301. From qualitative comparison with the isochrones, we estimate an age in the range of 10−15 Myr for SL 639, significantly younger than the estimate of 30 Myr from Santos et al. (1995). There is a larger spread in the inferred ages for the stars in Hodge 301 of 15 ± 5 Myr, overlapping with the lower estimate of 20 Myr from the photometric study of Grebel & Chu (2000). Adopting isochrones calculated for the rotating models from Brott et al. had little impact on these results; e.g. assuming an intial rotational velocity of 301 kms^-1, the inferred ages were only marginally younger (by ~1 Myr).

6. Candidate runaways

6.1. Single stars

To identify candidate runaways in the B-type sample we compared the radial-velocity estimates for each star with the initial weighted mean and standard deviation calculated for the stars in NGC 2060, NGC 2070, and the surrounding field population (i.e. excluding Hodge 301 and SL 639). Employing a threshold of | δv_r |  > 3σ we identified eight candidate runaways and then recalculated the mean velocities (and dispersions) excluding these objects. Reapplying a 3σ threshold provided one additional candidate, while a second iteration did not reduce the velocity dispersion further. The mean velocities and dispersions for the different spatial samples (excluding potential runaways) are presented in Table 4. The cumulative distribution of the velocity estimates for the 307 single stars is shown in Fig. 6, with the ±3σ threshold (i.e. ±36.6 kms^-1) for identification of runaway stars indicated by the vertical dotted lines.

Our adopted velocity threshold is comparable in magnitude to the limit used by Blaauw (1961), which was that candidate runaways have a space velocity which differs by more than 40 kms^-1 from the systemic velocity of the region in which they formed. (In this context, our radial-velocity estimates should serve as lower limits to the 3D space velocities.) In contrast, Portegies Zwart (2000) adopted a lower threshold of 25 kms^-1 in his theoretical study of the origins and characteristics of runaways. Given the limitations of the B-type data and the relatively small number of lines available for some objects, we did not consider a comparably low threshold.

Details of the nine candidate runaways are summarised in Table 5, and their locations are shown in Fig. 4. Four of the candidates have differential velocities in the range of 3.2−3.6σ, thus their status as candidates will depend strongly on the adopted velocity criteria. Moreover, our adopted velocity threshold (the dispersion of the weighted mean velocities in the sample) was employed without consideration of the uncertainty on the radial-velocity estimate for a given star. Thus, VFTS 467, although formally with a radial velocity of 4.2σ from the systemic value, has a large uncertainty on its estimated velocity. Indeed, only three He I lines were available for this star and the λ4144 estimate (277.5 kms^-1) is comparable to the systemic value.

The choice to use weighted means to investigate the spatially-different samples in Sect. 4.3 was motivated by the fact that approximately 10% of our velocity estimates have uncertainties of ≥15 kms^-1 (each of which is a broad-lined star, i.e. used the Set 2 lines from Table 3). To investigate the dependence of the number and identity of the candidate runaways on these adopted means, we repeated the above steps for both the (unweighted) mean and median velocities of the sample. The mean velocity for the 281 stars (not including those in Hodge 301 and SL 639) was 269.8 = 16.4 kms^-1, with a median value of 270.7 kms^-1. Adopting either of these systemic values and a 3σ threshold (taking the dispersion as 16.4 kms^-1) led to identification of the five most significant runaways from Table 5 (i.e. VFTS 242, 298, 358, 467, and 831), with a second iteration providing one additional candidate (VFTS 616). The final mean velocity from these calculations is 270.1 ± 13.9, in good agreement with the equivalent subsample in the final row of Table 4. From these checks we conclude that the four most significant candidates (VFTS 242, 298, 358, and 831) are secure as potential runaways, with five more tenative candidates as discussed above.

For the members of Hodge 301, VFTS 309 is a potential outlier (at 3.6σ) compared to the mean velocity for the cluster in Table 4. Indeed, excluding its v_r estimate from the calculated mean and dispersion for Hodge 301 gives $\overline{{\mathit{v}}_{\mathrm{r}}}$ = 261.5 ± 5.0 kms^-1, such that VFTS 309 is then a 4.1σ outlier. However, its velocity is also in reasonable agreement with those for stars in NGC 2060 and NGC 2070, so we are unable to confirm it as a potential runaway from Hodge 301. Finally, we note that the velocity estimate for VFTS 310 is not significant as a potential runaway from Hodge 301.

With our adopted velocity criteria, the nine candidate runaways correspond to 2.9 % of the sample of 307 single stars with radial-velocity estimates. This lower fractional share of runaways compared to the O-type stars (Sana et al., in prep.) is in qualitative agreement with recent N-body predictions for the mass-spectrum of runaways (Banerjee et al. 2012; Banerjee & Kroupa 2012), although quantitative comparison with the predictions is complicated by the different criteria employed (e.g. Banerjee et al. used a radial distance threshold for a star to be considered a runaway). We note that the distribution in Fig. 6 is slightly asymmetric, with two thirds of the candidates having negative differential velocities. This could point to veiling of more distant runaways by the nebula, although this is not seen in the results for the O-type stars (in which a greater fraction are seen to have positive differential velocities, Sana et al., in prep.), suggesting that small-number statistics likely explain the asymmetry.

6.2. Rotational velocities of candidate runaways

Estimated projected rotational velocities (from Dufton et al. 2013; McEvoy et al. 2014) for our candidate B-type runaways are included in Table 5. There appears to be a dichotomy in their rotation rates, as highlighted by Fig. 7. Five stars have small rotational velocities, with the estimates for four stars effectively limited by the velocity resolution of the data (i.e. ~40 kms^-1) and the fifth with v_esini = 64 ± 12 kms^-1. Meanwhile, the remaining four stars each have estimates of v_esini ≥ 345 kms^-1. To investigate the significance of these results, we ran Monte Carlo simulations using the v_esini distributions from Dufton et al. (2013). We note that the results from Dufton et al. were for the unevolved B-type sample (i.e. dwarf and giant objects), so we exclude VFTS 831 from the following discussion.

We generated random rotational velocities for eight stars drawn from the distribution of Dufton et al., and repeated this 10⁶ times to estimate the probability of recovering rotational velocities consistent with our eight (non-supergiant) runaway candidates (i.e. four stars with v_esini ≤ 65, and four with v_esini ≥ 345 kms^-1). We then repeated these runs 10⁴ times to estimate the dispersion on the probabilities. In a second test we limited the low velocity subsample to the three stars with v_esini ≤ 40 kms^-1 (i.e. excluding VFTS 625, which has the least significant runaway velocity), together with the four rapid rotators. We found low probabilities for recovering the observed v_esini values from both tests (as summarised in Table 6), with significance levels of 3.0 and 3.2σ, respectively.

We also investigated the probabilities compared to the cumulative distribution function of unprojected rotational velocities (v_e) from Dufton et al. (their Table 6). In this case we are limited by the unknown inclination angles of our targets but, assuming some relatively conservative lower-limits, we repeated tests similar to those above (in which we adopt the nearest available value from Dufton et al. for our upper limit, i.e. 340 kms^-1). These tests are formally less significant (Table 6), but provide further support that the rotational velocities of our candidate runaways do not appear to be randomly drawn from the distribution for the larger B-type sample. We note that this agrees with similar conclusions regarding the rotational velocities of the O-type runaways (Walborn et al. 2014; Sana et al., in prep.).

6.3. Atmospheric analysis of two candidate runaways

Atmospheric analysis of the B-type spectra is now underway within the VFTS project. In the context of the current study, we were particularly motivated to investigate the properties of the two most noteworthy runaway candidates, VFTS 358 and 831; we now comment briefly on our findings for these two objects.

6.3.1. VFTS 358

This star has the largest differential velocity from the estimated systemic value (δv_r = −106.9 ± 16.2 kms^-1) and also has a relatively large projected rotational velocity (v_esini =345 ± 22 kms^-1; Dufton et al. 2013). This prompted us to investigate its physical properties, in part to see if there was any evidence of past binary interaction (see e.g. de Mink et al. 2014).

Detailed analysis was complicated by the rapid rotation, which made estimation of the atmospheric parameters and element abundances difficult. However, using similar methods as those used by Hunter et al. (2007), we undertook a preliminary analysis using a grid of model atmospheres calculated using the tlusty and synspec codes (Hubeny 1988; Hubeny & Lanz 1995; Hubeny et al. 1998; Lanz & Hubeny 2007); further details of the grids are given by Ryans et al. (2003) and Dufton et al. (2005)⁵.

We estimated the effective temperature (T_eff) and gravity (log g) of VFTS 358 by fitting rotationally-broadened profiles to the Hδ, Hγ and He II λ4686 lines, assuming solar helium abundances. This gave estimates of T_eff ≃ 29 kK and log g ≃ 3.5 dex, although we were unable to estimate the microturbulence because of the paucity of observable metal lines (which also precluded use of the silicon ionisation balance to estimate the temperature).

Absorption from N II λ3995 appears to be present in the spectrum (with an equivalent width of ~120 mÅ), although its identification is marginal, as shown in Fig. 8. Rotationally-broadened model spectra for our adopted physical parameters (and an assumed microturbulence of 5 kms^-1) are shown for three nitrogen abundances: 12 + log (N/H) = 6.9 (the approximate baseline N-abundance for the LMC, see discussion by Hunter et al. 2007), 7.7, and 8.5 dex plotted in red, green, and blue, respectively. This comparison suggests significant nitrogen enrichment, with an abundance of ≳8.5 dex (the maximum abundance considered in our grid). Tests for theoretical profiles with a microturbulence of 20 kms^-1 (which is larger than normally found for stars with this gravity, e.g. Hunter et al. 2007) implied a similarly large nitrogen abundance. Fitting a rotationally-broadened profile to the N II λ3995 line leads to v_esini = 342 kms^-1 (cf. 345 kms^-1 from the helium lines) and a central wavelength of 3995.06 Å (cf. the laboratory wavelength of 3995.00 Å). This agreement provides some support that the feature is indeed from N II absorption.

The neutral helium lines in the observed spectrum appear somewhat stronger than those in the best-fitting model (e.g. the He I λλ4009, 4026 lines in Fig. 8) suggesting that the helium abundance might also be enhanced. In turn, this would tend to decrease the T_eff estimated from the He II line (and the surface gravity). Tests using cooler models indicated that the change in nitrogen abundance would be modest. For example, a change in effective temperature of −2 kK leads to a change in 12 + log (N/H) of approximately −0.2 dex. Hence, the nitrogen abundance would still be significantly greater than the baseline value for the LMC.

At present we are limited by the quality of the VFTS spectrum but from this preliminary analysis there is evidence of significant nitrogen enrichment (and maybe also helium) in the photosphere of VFTS 358. Indeed, the estimated nitrogen abundance is the largest obtained so far from the VFTS survey (including results from the B-type supergiants from McEvoy et al. 2014). The large velocity offset, high rotation rate, enhanced nitrogen (and possibly helium) all fit, at least qualitatively, with a scenario in which VFTS 358 is the former secondary of an interacting binary that became unbound at the explosion of the primary star.

6.3.2. VFTS 831

VFTS 831 is remarkable in two regards. First as a luminous B-type supergiant, likely to be the evolved descendent of a formerly more massive O-type star, and secondly because, from the definition used in Table 4, it is located within SL 639 (≈12.̋5 from the visual centre). With v_r = 210.4 ± 6.2 kms^-1 it also appears to be runaway compared to the other targets in SL639. A model-atmosphere analysis of the spectrum of VFTS 831 has been presented by McEvoy et al. (2014), who classified it as an apparently single star. It is one of the cooler objects in the supergiant sample, with T_eff = 14 000 K and log g = 2.1.

Following the discussion in Sect. 6.2, VFTS 831 does not appear particularly remarkable in terms of stellar rotation. Indeed, it has a relatively low projected rotational velocity and a substantial contribution from macroturbulent broadening (v_esini, Θ_RT) ~ (40, 65) kms^-1, as found in most B-type supergiants (see e.g. Markova et al. 2014; Simón-Díaz & Herrero 2014). From the intrinsically strong N II λ3995 singlet transition, McEvoy et al. (2014) estimated a nitrogen abundance of 12 + log (N/H) = 7.94. Although large, this is consistent with the other single supergiants in the sample, which have abundances spanning approximately 6.8 to 8.1 dex.

6.4. Single-lined binaries

For completeness we also investigated the velocities of single-lined binaries for potential runaway systems. In the final column of Table A.3 we include the observed range of velocities, δv_{r max}, from Dunstall et al. (in prep.). Following the arguments laid out in Sect. A.3 of the Appendix, we adopt an estimated error on the true systemic velocity compared to the single-epoch estimate of 0.15 × δv_{r max}. For stars with v_r above (or below) the systemic velocity, we then subtract (or add) this estimated error, to give a corrected velocity estimate, v_corr. Using the same 3σ criteria discussed above we then looked for potential runaways.

Only one single-lined system, VFTS 730, was found to be a potential runaway, with v_corr = 324.0 kms^-1 (a 4.3σ outlier cf. the adopted systemic velocity). Its range of observed velocities is relatively small (δv_{r max} = 16.6 kms^-1, thus only just above the detection threshold employed by Dunstall et al.) but it could be, for example, a long-period system with an eccentric orbit. Indeed, inspection of its velocity estimates as a function of epoch show them to be monotonically decreasing from 332 to 316 kms^-1, suggesting it as a long-period system. Further investigation of this (and other) binary systems will require comprehensive spectroscopic monitoring to determine their centre-of-mass velocities and other orbital parameters.

6.5. Origins of candidate runaways

The production of runaways is thought to be from one of two channels, with evidence in support of both mechanisms (Hoogerwerf et al. 2000). In the dynamical-ejection scenario the runaway is ejected via gravitational interaction between single-binary or binary-binary systems in a dense star cluster (Poveda et al. 1967). In the binary-ejection scenario, the primary star explodes as a supernova, leaving behind a neutron star or black hole. In most cases the compact object leaves the system depending on the magnitude and direction of the birth kick, and the companion is ejected with a velocity approximately equal to its orbital velocity (Blaauw 1961; Eldridge et al. 2011).

The complex star-formation history of 30 Dor makes it hard to cleanly attribute our candidate runaways to either of the two scenarios. R136 is sufficiently young (1–2 Myr; de Koter et al. 1998; Massey & Hunter 1998) that none of its members is thought to have undergone a supernova explosion. Thus, any runaways from R136 would have to be ejected via dynamical interactions. However, as noted above, the VFTS region contains several older subclusters and associations, including NGC 2060, which contains a supernova remnant (e.g. Chu et al. 1992), and Hodge 301 which was argued by Grebel & Chu (2000) as a likely site of tens of past supernovae from consideration of its mass function.

One of the important diagnostics might prove to be the connection with the observed rotational velocities (Fig. 7), in which the apparently rapid rotators could have a binary origin. For example, the theoretical discussions by de Mink et al. (2013), and the suggestion that the high-v_esini tail of the distribution of the rotational velocities for the VFTS O-type stars could be accounted for by past binary interactions (Ramírez-Agudelo et al. 2013). Detailed dynamical analysis of the O-type stars in the survey has been underway in parallel to the current work, and the broader question of the origins of runaway stars from 30 Dor is discussed in more depth by Sana et al. (in prep.).

7. Summary

We have presented comprehensive spectral classifications and radial-velocity estimates for the B-type stars observed in the VFTS. We now briefly summarise our main findings:

• The members of the two older clusters,Hodge 301 and SL 639,appear kinematically distinct from those in the younger clustersand general field population.

• The preliminary H-R diagram for stars in Hodge 301 and SL 639 (Fig. 5) suggests an age of ~10−15 Myr for the population of SL 639, slightly younger than that of Hodge 301 (15 ± 5 Myr).

• The systemic velocity for 273 single stars, excluding those in the two older clusters and candidate runaways, is $\overline{{\mathit{v}}_{\mathrm{r}}}$ = 271.6 ± 12.2 (s.d.) kms^-1.

• Employing a 3σ velocity threshold, we identify nine single stars as candidate runaways (2.9 % of the single stars with radial-velocity estimates). These are mostly unevolved, early B-type objects, but include a B5-type supergiant (VFTS 831).

• There appears to a bimodal distribution of rotation rates in the candidate runaways. Five stars have v_esini ≤ 65 kms^-1, while the others have v_esini ≥ 345 kms^-1.

• Excluding VFTS 831 (because of its evolved nature), there is a remarkably small probability (~0.001%) that the projected rotational velocities for our other eight candidate runaways could be randomly drawn from those for the full (unevolved) B-type sample.

• VFTS 358 has the largest velocity offset from the systemic velocity of 30 Dor (δv_r = −106.9 ± 16.2 kms^-1) and appears to be rotating rapidly (v_esini = 345 kms^-1). Preliminary spectral analysis suggests evidence of significant nitrogen enrichment, perhaps indicative of past binary interaction.

The classifications presented here will underpin the quantitative analyses in progress on this substantial part of the VFTS data. For example, they will enable investigation of the mass/temperature dependence (if any) of the estimated v_esini results from Dufton et al. (2013), particularly when also combined with the results for the O-type stars from Ramírez-Agudelo et al. (2013). In the longer-term, the ongoing proper-motion studies using HST imaging (programme GO12499, P.I. Lennon) will add another dimension to our studies of the complex dynamics in this region, by providing tangential velocities (see Sabbi et al. 2013).

Our radial-velocity estimates should also lend themselves to detailed investigation of the structural properties of the VFTS sample (e.g. using the techniques discussed by Parker et al. 2014; Wright et al. 2014), particularly when combined with the results for the O-type stars from Sana et al. (2013); this should provide further insights into the formation and evolutionary history of the massive-star population in the region.

Finally, there is an ongoing monitoring programme of FLAMES spectroscopy to characterise a subset of the O-type binaries (PI: Sana); a similar monitoring programme for the ~100 binaries discussed here and by Dunstall et al. (in prep.) would be valuable to determine the full orbital parameters of the lower-mass population of binaries, while also providing robust estimates of centre-of-mass velocities to better explore their dynamics in the context of ejected runaways.

Online material

Appendix A: Ancillary material

Appendix A.1: Spectral montages

To illustrate the spectral sequences as a function of luminosity class, examples of VFTS B-type spectra are shown in Figs. A.1−A.3. The principal lines used in classification, as summarised in Tables 1 and 2 are indicated in each figure. Luminosity sequences at spectral types B0.2, B1 and B2.5 are shown in Figs. A.4−A.6, respectively. For clarity, the spectra in all six figures have been smoothed and rebinned to an effective resolving power of R = 4000.

Appendix A.2: Detailed velocity estimates

For completeness, we list the individual velocity estimates from each line, calculated using the rest wavelengths given in Table 3. Tables A.1 and A.2 present the results for stars using the Set 1 and Set 2 diagnostic lines, respectively; results for the binary stars (following the discussion outlined in Sect. 4.2) are given in Table A.3.

Appendix A.2.1: Comparison with published velocity estimates

The methods and line sets adopted to estimate radial velocities (v_r) for the B-type stars differ to those employed by Sana et al. (2013) for the O-type objects from the survey. As a consistency check between the two studies, we analysed 45 of the apparently-single, late O-type stars (with O9.5 and O9.7 types) and compared our velocity estimates with those from Sana et al. We find a mean and standard deviation between the two methods of Δv = 0.5 ± 6.2 kms^-1 (where Δv = v_r − v_Sana).

The differential results (Δv) for the 45 stars are shown in Fig. A.7 (with the standard deviations added in quadrature). For the purposes of global analysis of the sample and identification of potential runaways, these results are in excellent agreement. As a further check of any systematic trends arising from the different methods, the lower panel of Fig. A.7 shows the same differential velocities, but now as a function of estimated v_esini (from Dufton et al. 2013); no obvious systematic difference is present.

Appendix A.3: Centre-of-mass velocities

We have investigated the expected differences between our mean values and centre-of-mass velocities using a Monte Carlo technique. We assumed a sinusoidal velocity curve (i.e. circular orbit) and, adopting the observational sampling of Field A of the survey (Table A.1 of Paper I), we calculated ranges of observed velocities (i.e. δv_{r max}) and their resulting mean velocity. We assumed a random initial phase, and considered periods ranging from one day up to several years.

These simulated mean velocities provide us with an estimate of the expected error compared to the genuine centre-of-mass velocity. For a given period, we can then average these differences over the range of assumed phases to find a typical error. The ratio of this typical error to δv_{r max} is shown in Fig. A.8. For binaries with periods of ≲250 d, the expected error is approximately 12−15 % of δv_{r max}, with this increasing for longer periods, and then tending to a value of ~25% for periods ≳350 d (greater than the maximum time-sampling of the real data).

These simplistic simulations (e.g. not taking into account potential eccentric orbits) have probably underestimated the uncertainties but, given the observational cadence of the survey, they illustrate that the mean multi-epoch v_r values should provide reasonable estimates of the centre-of-mass velocities for short-period systems (≲250 d), which display relatively small peak-to-peak variations.

---

Acknowledgments

Based on observations at the European Southern Observatory Very Large Telescope in programme 182.D-0222. We thank Mark Gieles for helpful discussions regarding the candidate runaways, and the referee for their constructive comments. STScI is operated by AURA Inc., under NASA contract NAS 5-26555. S.d.M. acknowledges support for this work by NASA through an Einstein Fellowship grant, PF3-140105. J.M.A. acknowledges support by grants AYA2010-17631 and AYA2010-15081 of the Spanish Ministry of Economy and Competitiveness. S.S.-D. acknowledges funding from the Spanish Government Ministerio de Economía y Competitividad (MINECO) through grants AYA2010-21697-C05-04, AYA2012-39364-C02-01 and Severo Ochoa SEV-2011-0187, and the Canary Islands Government under grant PID2010119.

References

All Tables

All Figures