A deep near-infrared spectroscopic survey of the Scutum-Crux arm for Wolf–Rayet stars Free

Abstract

We present a New Technology Telescope/Son-of-Isaac spectroscopic survey of infrared selected Wolf–Rayet (WR) candidates in the Scutum-Crux spiral arm (298° ≤ l ≤ 340°, |b| ≤ 0|$_{.}^{\circ}$|5). We obtained near-IR spectra of 127 candidates, revealing 17 WR stars – a ∼13 per cent success rate – of which 16 are newly identified here. The majority of the new WR stars are classified as narrow-lined WN5–7 stars, with two broad-lined WN4–6 stars and three WC6–8 stars. The new stars, with distances estimated from previous absolute magnitude calibrations, have no obvious association with the Scutum-Crux arm. Refined near-infrared (YHJK) classification criteria based on over a hundred Galactic and Magellanic Cloud WR stars, providing diagnostics for hydrogen in WN stars, plus the identification of WO stars and intermediate WN/C stars. Finally, we find that only a quarter of WR stars in the survey region are associated with star clusters and/or H ii regions, with similar statistics found for luminous blue variables (LBVs) in the Milky Way. The relative isolation of evolved massive stars is discussed, together with the significance of the co-location of LBVs and WR stars in young star clusters.

1 INTRODUCTION

Wolf–Rayet (WR) stars – the progeny of massive O-type stars – are excellent tracers of young stellar populations in galaxies owing to their unique spectroscopic signatures of strong, broad emission lines (Crowther 2007). However, whilst WR surveys of nearby galaxies are nearing completeness (Massey et al. 2014), the WR content of the Milky Way remains woefully incomplete (e.g. Shara et al. 2009) due to high dust obscuration at visual wavelengths. Our detailed knowledge of the evolution of massive stars remains unclear, with inaccuracies in earlier evolutionary phases magnified in the WR phase.

In addition, it is becoming clear that the conventional view of ≥20–25 M_⊙ O stars advancing through the luminous blue variable (LBV) stage en route to the nitrogen (WN) and carbon (WC) sequence WR phase and ultimately a stripped envelope core-collapse supernova (ccSN) is incomplete if not incorrect. First, a high fraction of massive stars are now known to be multiple (Sana et al. 2012), so the major effects of close binary evolution needs to be considered. Secondly, it has been proposed that LBVs are lower mass binary products, from an inspection of their spatial location in the Milky Way and Large Magellanic Cloud (LMC) with respect to WR and O stars (Smith & Tombleson 2015). Thirdly, Sander, Hamann & Todt (2012) argue from a spectroscopic analysis of Milky Way WC stars that the most massive stars do not pass through this phase. Finally, it is not clear whether the most massive stars will undergo a bright SN explosion after core-collapse, since they may collapse directly to a black hole or produce a faint SN and fallback to a black hole (Langer 2012).

Still, our Galaxy contains the largest spatially resolved population of WR stars, predicted to number ∼1200 (Rosslowe & Crowther 2015a,b, hereafter RC15a). The confirmed population has doubled over the previous decade, and currently stands at ∼640.¹ The Galactic disc therefore presents a rich hunting ground for further discoveries. Due to the large foreground interstellar dust extinction towards stars in the galactic disc, near- and mid-infrared surveys are required.

The dense, ionized stellar wind of WR stars facilitates two approaches to infrared surveys. First, their strong, broad emission lines are amenable to near-IR narrow-band imaging (Shara et al. 2009, 2012; Kanarek et al. 2015). Secondly, their dense winds exhibit a free–free excess leading to unusual infrared colours, which have been exploited in the near-IR (Homeier et al. 2003; Hadfield et al. 2007) and mid-IR (Mauerhan, Van Dyk & Morris 2011; Messineo et al. 2012; Faherty et al. 2014). To date, the majority of spectroscopic follow-up has been carried out to an approximate depth of K_S ≲ 11 mag. However, the (coarse) model of the Galactic WR distribution developed by RC15a suggests follow-up spectroscopy is needed to a depth of K_S ∼ 13 mag in order to sample the majority of WR stars.

Here, we exploit prior photometric approaches to spectroscopically survey a region of the Galactic disc to fainter limits than to date (K_S ∼ 13 mag). This has two interrelated goals: (1) the refinement and development of techniques that can be used to classify WR stars using only infrared spectroscopy; (2) comprehensive searches for WR stars in the Milky Way to allow more robust comparisons between their spatial locations and other massive stars; longer term goals involve the second data release (DR2) of Gaia which will provide parallaxes for hundreds of WR stars, permitting their use as tracers of Galactic structure, and will be combined with upcoming large fibre-fed spectroscopic surveys, including WHT/WEAVE (Dalton et al. 2016) and VISTA/4MOST (de Jong et al. 2016).

This paper is structured as follows. In Section 2, we describe our photometric selection criteria and survey region, namely the Scutum-Crux spiral arm, the tangent to which lies at approximately l ≃ 310° (Georgelin & Georgelin 1976). Spectroscopic observations of WR candidates, plus some previously known Galactic WR templates, are presented in Section 3, including a brief description of non-WR stars. Refined near-IR classification criteria for WR stars are presented in Section 4. Results for newly identified WR stars are presented in Section 5, including distance estimates. We consider the spatial location of WR and other massive stars in Section 6, including discussion of prior inferences about the nature of LBVs. Finally, in Section 7 we reflect on the low success rate of the methodology employed, and share some motivating points for future IR surveys targeting WR stars.

2 CANDIDATE SELECTION

Here, we discuss our selection of sightlines towards the Scutum-Crux arm, plus our photometric criteria for the selection of candidate WR stars. Specifically, we focus on l = 298°–340°, which Russeil et al. (2005) have previously highlighted in determining Galactic structure, since this intersects three proposed spiral arm features – Sagittarius-Carina, Scutum-Crux and Norma-Cygnus (Russeil 2003, their fig. 5). The majority of known WR stars in this region lie at distances <6 kpc, consistent with the nearby Sagittarius-Carina arm. However, assuming typical |$M_{K_{\rm S}}\,{=}\,{-}\,5$| and |$A_{K_{\rm S}}\,{=}\,2$| for Galactic WR stars, it is possible to probe heliocentric distances 6–15 kpc by identifying WR stars in the magnitude range K_S = 11–13 mag. Therefore, WR stars may provide a comparable and complimentary tracer of Galactic structure to commonly used non-stellar objects, i.e. H ii regions and atomic H gas. We confined our search to latitudes |b| < 0|$_{.}^{\circ}$|5, ensuring that at the furthest expected distances, candidate WR stars remain within a few scaleheights of the Galactic plane (FWHM = 80 pc for WRs, RC15a).

We selected candidate WR stars for which 298° ≤ l ≤ 340° and |b| ≤ 0|$_{.}^{\circ}$|5 from their near-IR (2MASS; Skrutskie et al. 2006) and mid-IR (GLIMPSE-I; Benjamin et al. 2003) photometry. We limited our survey to GLIMPSE-I point sources with a corresponding 2MASS detection,² requiring a minimum 2MASS quality flag of ‘C’ in the K_S filter, and rejected sources with one or more Source Quality Flags > 10⁵ in the GLIMPSE-I catalogue. We then used the topcat³ tool to apply various cuts in colour and magnitude.

It is necessary to emphasize that not all WR stars occupy this parameter space, although there is no bias towards either WN or WC subtypes. Still, some dusty WC stars are offset from the majority of WR stars in the J − K_S versus K_S − [8.0] colour–colour diagram (Mauerhan et al. 2011, their fig. 1), so our survey criteria are potentially biased against such stars. Approximately 250 sources satisfied these criteria. We subsequently cross-checked the co-ordinates of these with the SIMBAD⁴ data base, to find any with previous identifications. Encouragingly, 14 per cent of these were known WR stars (23 per cent of known WRs in the survey area), 4 per cent had non-WR classifications (mostly Be stars or young stellar objects), leaving ∼200 candidates with no previous spectral classification. Colour–colour diagrams for 191 candidates involving K_S − [8.0] versus (J − K_S) and [3.6] − [8.0] versus [3.6] − [4.5] are presented in Fig. 1 together with reddening-free parameters Q1 and Q2 from Messineo et al. (2012).

3 NTT/SOFI SPECTROSCOPY OF WOLF–RAYET CANDIDATES

We obtained near-IR spectroscopy of 127 WR candidates between 2015 March 29 and 31 (program ID 094.D-0839) using the Son-of-Isaac (SOFI) spectrograph at the New Technology Telescope (NTT). These represent 66 per cent of the IR-selected candidates presented in Fig. 1. Candidates were observed with the red grism (GR) covering the 1.53–2.52 μm spectral region, a dispersion of 10.2 Å pixel⁻¹, and a slit width of 1 arcsec, providing a spectral resolution of R ∼ 600. All sources were observed using a standard ABBA sequence, including a small random offset in the A and B positions between exposures.

Before extracting 1D spectra, we subtracted a median dark frame from each individual frame, then subtracted adjacent AB pairs from one another. The result of this was four dark frame-corrected spectra for each source, free from sky lines. We extracted these four spectra for each object using iraf. Wavelength calibration was performed using strong and isolated sky lines at known wavelengths (Rousselot et al. 2000) present in each raw frame, after which all spectra for each object were co-added.

Throughout each night, we periodically observed bright Vega-type telluric standard stars, at similar airmasses to the WR candidates. The removal of telluric spectral features was achieved using telluric in iraf. We also used these telluric standards, together with Kurucz models of the same spectral types, to perform relative flux calibration, which are subsequently adjusted to match 2MASS photometry.

Of the 127 candidates, 17 stars were identified as WR stars. Of these, one candidate was subsequently matched to the recently discovered WN6 star 1093-1765 (=WR75-30) from Kanarek et al. (2015), such that 16 stars are newly identified as WR stars in this study. Previous surveys of WR stars from IR photometric criteria have achieved similar efficiencies (Mauerhan et al. 2011; Faherty et al. 2014). We briefly discuss the nature of the non-WR stars in Section 3.1 and discuss the newly identified WR stars in Section 4. New WR stars are indicated in Fig. 1, with a subtype dependence apparent in the reddening-free Q1 versus Q2 diagram. Table 1 provides basic observational properties for the new WR stars, for which we obtained NTT/SOFI spectroscopy, while Table B1 (available online) provides a list of all candidates for which we obtained spectroscopy, together with a brief note describing the nature of each source.

In addition to the candidate WR stars, we have also obtained NTT/SOFI spectroscopy of 14 WR stars for which optical classifications have been undertaken, in order to refine near-IR based classification criteria (see Section 4). Finally, we also obtained blue grism (GB) spectroscopy with SOFI for the majority of newly identified WR stars. The GB observations cover the spectral region 0.95–1.64 μm, a dispersion of 7.0 Å pixel⁻¹, and an identical slit width of 1 arcsec, again providing R ∼ 600. Data reduction was undertaken in an identical manner to the GR data sets.

3.1 Non Wolf–Rayet stars

A significant subset of the 110 candidates that were not confirmed to be WR stars exhibited a hydrogen emission line spectrum, with Br γ observed in 60 (55 per cent) cases, plus often higher Brackett series (Br10, 11), and He i 2.058 μm emission present in a quarter of instances. These sources are likely to be massive young stellar objects or Herbig AeBe stars (see e.g. Porter, Drew & Lumsden 1998; Cooper et al. 2013). Br γ emission equivalent widths are typically 10–30 Å, with He i/Br γ ratios of 0.3–1. The majority of Br γ emission lines are unresolved (FWHM ∼ 30 Å) although several stars (e.g. B#127, 147, 149) possess broad emission (FWHM ∼ 50–60 Å). Unusually, B#66 exhibits strong He i 2.058 μm emission, without significant Br γ emission, warranting follow-up observations.

Mid-IR imaging has revealed circumstellar ring nebulae around many evolved stars (e.g. Wachter et al. 2010; Toalá et al. 2015). Such nebulae appear prominently in Spitzer 8.0 μm images, owing to thermal emission from dust swept up by stellar winds. Mindful of this, we inspected 5 arcsec × 5 arcsec Spitzer 8.0 μm images centred on all candidates. One of the Br γ emission line sources, A#9 (SSTGLMC G330.7900-00.4539), is the central star of a striking oval mid-IR ring nebula, S65, which was identified by Churchwell et al. (2006) and studied by Simpson et al. (2012).

Strong absorption lines in the Brackett series are observed in one candidate, B#3, indicating an A or late-B type star, with Br γ absorption observed in another object, B#54, albeit without other prominent features suggesting an early-type star in this instance. Four candidates – B#21, B#100, B#122 and B#153 – exhibit prominent CO 2.3 μm bandhead absorption features, although none of these involve Br γ emission line sources, indicating a late-type star origin. The remaining 44 candidates (40 per cent of the non-WR stars) either have no prominent absorption or emission features, or the S/N achieved was insufficient to identify their nature.

4 NEAR-IR CLASSIFICATION OF WOLF–RAYET STARS

The switch from optical to near-IR spectroscopy for the overwhelming majority of new Galactic WR stars requires a reassessment of spectral classification criteria. Vreux et al. (1990) provided a classification scheme based upon Y-band observations of northern WR stars, while Eenens et al. (1991) devised a near-IR scheme for WC stars from 1–5 μm spectroscopy. More recently, Crowther et al. (2006, hereafter C06) provide near-IR classification diagnostics for WN and WC stars, based on equivalent width ratios. Qualitatively, early-type WC4–6 stars possess broader emission lines than later WC7–9 subtypes, although exceptions do exist (Eenens & Williams 1994).

An updated quantitative near-IR classification of WR stars is made feasible by access to a greatly expanded sample of Galactic and Magellanic Cloud WR stars for which optical classifications have been made, primarily Smith, Shara & Moffat (1996) for WN stars and Smith, Shara & Moffat (1990) for WC stars. The data sets utilized were drawn from various sources, primarily NTT/SOFI and UKIRT/CGS4, as summarized in Table 2. We have also inspected high-resolution, intermediate-resolution IRTF/SpeX spectroscopy of northern Galactic WR stars, provided by W.D. Vacca (private communication), although we focus our criteria on moderate resolution (R = 600–1000), modest signal-to-noise spectroscopy in the 0.95–2.5 μm near-infrared.

To date, no criteria for the identification of WN/C or WO stars from near-IR spectroscopy have been considered, nor has an attempt to distinguish between H-rich and H-deficient WN stars, although C06 did separate broad-lined WN stars (FWHM He ii 1.01μm ≥ 65 Å) from narrow-lined counterparts. In our revised near-IR classification scheme, we attempt to utilize pairs of lines from adjacent ionization stages of helium for WN stars, and adjacent ionization stages of carbon for WC stars. In some instances nitrogen lines are required for WN stars, in common with C06, plus ratios of carbon to helium lines are utilized for WN/C and WC stars, which will also depend upon their relative abundances. We omit from our discussion the near-IR classification of transition Of/WN stars, which has been considered by Bohannan & Crowther (1999) and Crowther & Walborn (2011). WN stars with intrinsic absorption features (WNh abs, WNha) also offer specific challenges that will need to be considered separately.

A detailed description of the updated classification scheme is provided in Appendix A (available online), while we present a summary of Y-, J-, H- and K-band classification diagnostics for WN, WN/C, WC and WO subtypes in Table 3. In Figs A1 and A2 (available online), we present YJ-band and HK-band spectroscopy of template (optically classified) WN, WN/C, WC and WO stars, with line measurements provided in Tables A1– A3, also available online. Overall, the use of solely IR diagnostics provide satisfactory classifications, although confidence in resulting spectral types requires multiband spectral coverage, a minimum spectral resolution of R ∼ 500 and moderate signal to noise. In many instances, observations at a single band prevent a refined classification. For example, WC4–7 stars may not be distinguished using solely K-band spectroscopy, while it is not possible to differentiate between broad-lined WN4–7 stars on the basis of low-S/N spectroscopy in the Y band. Still, reliable, WR subtypes can be obtained from complete 1–2.5 μm spectroscopy, with the exception of broad-lined WN4–6 stars and WC5–6 stars. In addition, WO stars have a very distinctive near-IR spectrum, and WN/C stars possess characteristics in each of Y, J, H and K bands which distinguish them from normal WN stars. In addition, the presence of hydrogen in WN stars can be identified in most subtypes, although very late subtypes are challenging since a low He ii 1.163/P β or He ii 2.189/Br γ ratio may indicate either a high hydrogen content or a low ionization atmosphere.

4.1 Robustness of near-IR classification

In order to assess the robustness of the new scheme, we reclassify several WN and WC stars that have been discovered and classified from red optical spectroscopy. We utilize NTT/SOFI spectroscopy of four WN stars, WR62a, WR68a, WR93a from Homeier et al. (2003) (data set N2 in Table 2), plus WR75-30 from our own observations (data set N5 in Table 2), together with three WC stars, WR107a from Homeier et al. (2003) plus WR75aa and WR75c from our own observations. Near-IR spectra are presented in Figs 2 and 3. Individual line measurements are provided in Tables 4 and 5 for WN and WC stars, respectively, while line ratios are presented in Table 6. Measurements have employed Gaussian fits, using the elf suite of commands within the Starlink dipso package.⁵

4.1.1 WN stars

WR62a was classified as WN5o by Shara et al. (1999, their source #11), and we support its classification as a narrow-lined WN star. Consequently, the primary diagnostics are the He i 1.08/He ii 1.01 μm ratio and the K-band morphology. The former indicates a WN6 subtype, while He i + N iii 2.11 > N v 2.10 favours a WN5–6 subtype. The P β/He ii 1.16 μm ratio indicates WR62a is hydrogen-free, while the Br γ/He ii 2.19 μm suggests a borderline o/(h) classification, so overall we favour WN6o for WR62a. The same arguments and ratios apply to WR68a for which Shara et al. (1999, their source #13) assigned WN6o. We support this classification owing to its morphological similarity to WR62a.

WR93a (Th 3–28), was originally classified as WN2.5–3 by Acker & Stenholm (1990) and revised to WN6 by Miszalski, Mikołajewska & Udalski (2013) from optical spectroscopy. This is also a narrow-lined WN star so again we focus on its He i 1.08/He ii 1.01 μm ratio and the K-band morphology. Both favour a WN6 subtype, with a significant hydrogen content from our multiple diagnostics (darker shaded regions in Fig. A4, available online), so we adopt WN6h for WR93a.

Kanarek et al. (2015, their source 1083-1765) originally classified WR75-30 as a WN6 star from near-IR spectroscopy. The He i 1.70/He ii 1.69 μm ratio favours a WN7 subtype, as does (He i + N iii 2.11)/He i 2.19 μm ∼ 1, while the Br γ/He ii 2.19 μm ratio lies in the hydrogen-free region of Fig. A4 so we favour WN7o for this star.

4.1.2 WC stars

WR75aa and WR75c were identified as WC9 stars by Hopewell et al. (2005) from red optical spectroscopy. All our primary near-IR diagnostics support this assessment, as do the secondary criteria involving helium for WR75c. WR75aa has a borderline WC8–9 classification from the He i-ii 1.7/C iv 1.74 μm ratio, but overall both stars are unambiguous WC9 stars.

Finally, WR107a (#18 from Shara et al. 1999) was originally classified as a WC6 star from red optical spectroscopy. Our primary criteria indicate the follow for WR107a: WC6 ± 1 from both C iii 1.20/C iv 1.19 and C iii 2.11/C iv 2.07, WC5–8 from C ii 0.99/C iii 0.97. Our secondary criteria indicate WC5 from He i 1.08/He ii 1.01, and WC7 from C iii 0.97/He ii 1.01 μm (H-band spectra are unavailable), so although WC6 is plausible we provide a more cautious WC5–7 classification. Indeed, the primary optical diagnostic ratio (C iii 5696/C iv 5808) also favoured WC6–7 according to Shara et al. (1999).

In general, the K band is preferred to shorter wavelengths for classification of highly reddened WR stars, but K-band spectral features of dusty WC stars are often masked by host dust emission. Extremely high S/N is required to identify K-band spectral features of the Quintuplet Q stars. By way of example, Liermann, Hamann & Oskinova (2009) assign a WC8/9d+OB subtype to Q3 (WR102ha) from K-band spectroscopy, whereas WC features are relatively prominent in deep H- and J-band spectroscopy. We confirm a WC9d subtype for Q3 on the basis of high S/N Gemini spectroscopy presented by Najarro et al. (2015), owing to C iii 1.20/C iv 1.19 ≫ 1, C ii 1.78/C iv 1.74 > 1 and C iii 2.11/C iv 2.07 ≫ 1.

5 NEW GALACTIC WOLF–RAYET STARS

We have identified 16 new WR stars, which we have assigned Galactic WR numbers, in accordance with the current IAU convention (see appendix of RC15a). Here, we discuss their spectral types, spatial location and their potential association with Scutum-Crux or other spiral arms. Near-IR spectra of the new WR stars are presented in Figs 2 (IJ) and 3 (HK), together with our NTT/SOFI observations of WR75aa, WR75c, the recently discovered WN star WR75-30 (Kanarek et al. 2015), plus previously unpublished NTT/SOFI spectroscopy of WR62a, WR68a, WR93a and WR107a, as discussed above. Line measurements are provided in Tables 4 and 5 for WN and WC stars, respectively, with diagnostic line ratios presented in Table 6.

5.1 Classification of the new WR stars

5.1.1 Broad-lined WN stars

Only two of the new WN stars, WR64-6 and WR70-14, are identified as broad-lined WN stars, owing to their He ii 1.01 μm line widths (FWHM > 1900 km s⁻¹) and strengths (log W_λ/Å ≥ 2.4), albeit WR64-6 only narrowly complies with the second criterion. A WN6b subtype is favoured for WR64-6 from its He i 1.08/He ii 1.01 μm ratio which is supported by (N iii + He i 2.11) > N v 2.10, while both hydrogen criteria (involving P β and Br γ) indicate no hydrogen, so we adopt WN6b for this star. For WR70-14, the He i 1.08/He ii 1.01 μm ratio is somewhat ambiguous, consistent with WN4–6b, but (N iii + He i 2.11) ∼ N v 2.10 favours WN4b. This is supported by weak N v 1.11 μm in the J  band (Fig. 2). Again, there is no evidence for atmospheric hydrogen from our criteria (Fig. A4, available online), so WN4b is assigned to WR70-14.

5.1.2 Narrow-lined WN stars

The remaining 11 WN stars are a relatively homogeneous group, almost all classified as either WN5o, WN6o or WN7o stars, with only WR47-5 showing evidence of hydrogen so we consider these according to their subtype.

The two highest ionization narrow-lined stars are WR56-1 and WR70-15, according to their He i 1.08/He ii 1.01 ratios (Fig. A3, available online), which indicate WN5 for both stars. This is supported by the He i 1.70/He ii 1.69 ratio for WR56-1, although an earlier WN3–4 subtype is favoured by He i 1.70/He ii 1.69 for WR70-15. We also consider their K-band morphologies, for which (N iii + He i 2.11) > N v 2.10 in both cases, indicating WN5–6. Neither star shows any evidence for atmospheric hydrogen from Fig. A4 so we adopt WN5o for both stars.

The majority of our narrow-lined WN stars are WN6 stars according to their He i 1.08/He ii 1.01 ratios (Fig. A3), with He i 1.70/He ii 1.69 suggesting WN4, 5 or 6. As with the WN5o stars considered above, we also consider the K-band morphology, for which either (N iii + He i 2.11) > N v 2.10, implying WN5–6 or (N iii + He i 2.11) ≫ N v 2.10, implying WN6. Only WR47-5 indicates the presence of (modest) hydrogen from Fig. A4, such that we classify it as WN6(h), but favour WN6o for WR60-8, WR64-2, -3, -4, -5 and WR72-5.

Of the remaining stars, only WR75-31 was observed in the IJ band with SOFI, for which He i 1.08/He ii 1.01 indicates a WN7–8 subtype, with He i 1.70/He ii 1.69 also providing an ambiguous WN7-8 classification. Its K-band morphology strongly favours WN7 since (N iii + He i 2.11) ∼ He ii 2.19, while there is no evidence for atmospheric hydrogen in WR75-31 from Fig. A4, such that we assign WN7o to this star. WR76-11 was observed solely in the H and K bands, but closely resembles WR75-31 such that we classify it as WN7o, as with WR75-30. None of the new WN stars qualify as WN/C stars, since C iii 0.971 μm, C iv 1.19 μm, 1.74 μm and 2.07 μm are weak/absent.

5.1.3 WC stars

Three of the new WR stars are carbon sequence WC stars. Considering the primary diagnostics for WR46-16, WC7 is favoured from the C iv 1.19/C iii 1.20 ratio, WC4–6 from the C iii 0.97/C ii 0.99 ratio and WC5–7 from the C iv 2.07/C iii 2.11 ratio. Secondary indicators suggest WC5–7 from He i 1.08/He ii 1.01, WC5–6 from C iii 0.97/He ii 1.01 and WC6–7 from He i-ii 1.7/C iv 1.74. Overall, we adopt WC6–7 reflecting the tension in primary indicators for WR46-16 (Fig. A7, available online).

WR60-7 is classified as WC8, WC7, WC7–8 from primary diagnostics C iv 1.19/C iii 1.20, C iv 2.07/C iii 2.11 and C iii 0.97/C ii 0.99, respectively. Secondary criteria C iii 0.97/He ii 1.01 and He i-ii 1.7/C iv 1.74 indicate WC6–7, while He i 1.08/He ii 1.01 favours WC7–8. Overall, WC7–8 is selected for WR60-7, reflecting the lack of a consensus amongst primary criteria (Fig. A7).

Finally, primary diagnostics C iv 1.19/C iii 1.20, C iv 2.07/C iii 2.11 and C iii 0.97/C ii 0.99, imply WC8–9, WC8 and WC8–9 for WR70-13, while C iv 1.74 ≥ C ii 1.78 indicates WC8. Consequently, we adopt WC8 for WR70-13, which is supported by our secondary indicator He i-ii 1.7/C iv 1.74, with He i 1.08/He ii 1.01 and C iii 0.97/He ii 1.01 consistent with either WC8 or WC9 (Fig. A7).

5.2 Binarity

Approximately 40 per cent of the Galactic WR population are observed in multiple systems (van der Hucht 2001). This is a lower limit on the true binary fraction, since no systematic survey has been carried out. It is therefore highly likely that some of the newly discovered WR stars are in fact multiple systems. Direct detection of companion stars, usually main-sequence OB stars, is not possible with the current data set since their absorption lines are generally weak with respect to the strong WR emission lines.

It is, however, possible to infer the presence of a companion star by considering the equivalent width of near-IR emission lines which will be diluted by the continuum of a companion star, and/or dust for the case of some WC+OB systems since dust formation is an indicator of binarity in WC stars.

Since a companion star and/or thermal dust emission will not reduce line widths, a weak line compared to single stars at a specific FWHM is suggestive of binarity. In Fig. 4, we compare the FWHM (km s⁻¹) and equivalent widths (in Å) of strong, isolated lines in apparently single Galactic WN stars (He ii 1.012 μm) and WC stars (C iv 1.736 μm) with newly discovered WR stars. We also include weak-lined WN stars with intrinsic absorption lines (WR24, WR87, WR108) which could be mistaken for WN+OB stars, plus dusty WC stars (WR121, WR75aa), whose near-IR emission lines are diluted by hot dust.

Of the newly identified stars, the majority of WR stars possess emission line strengths that are characteristic of single stars. From Fig. 4(a), two exceptions are WR75-31 (WN7(h)) and WR64-4 (WN6o) which possess weak emission for their He ii 1.012 μm FWHM. Both are potential binaries, although WR64-4 is the strongest candidate, such that we revise its spectral type to WN6o+OB. In contrast, WR75-31 has an overall relatively strong emission line spectrum, albeit with an anomalously weak (and low S/N) He ii 1.0124 μm line.

Of the WC stars, none possess unusually weak emission lines based on their C iv 1.736 μm FWHM (Fig. 4b). However, the increased dilution of WC emission lines from 1 to 2.5 μm arising from hot dust in WCd systems also severely modifies equivalent width ratios of C iii-iv lines. For example, W_λ(C iii 2.11)/W_λ(C iii 0.97) = 0.3 for WR88 (WC9) but hot dust in WR121 (WC9d) reduces this ratio to 0.05. A similar reduction in line strength is observed for prominent He ii lines, with W_λ(He ii 2.19)/W_λ(He ii 1.28) = 0.5 for WR88 and 0.17 for WR121. WR135 is a prototypical non-dusty WC8 star with W_λ(C iii 2.11)/W_λ(C iii 0.97) = 0.2, with a ratio of 0.2 for WR60-7 but only 0.1 for WR70-13, suggestive of dust dilution in the latter. Indeed, WR60-7 (WC7–8) and WR70-13 (WC8) possess similar J − H colours, yet the latter has 0.5 mag higher K_S–[8] colour (Table 1), so we amend its spectral type to WC8d. Indeed, WR70-13 is offset from the other WC stars in the reddening free Q1 versus Q2 comparison in Fig. 1. Turning to WR46-18, W_λ(C iii 2.11)/W_λ(C iii 0.97) ∼0.3 for non-dusty WC6–7 stars, with a ratio of 0.4 for WR46-18, arguing against dust emission in this instance.

Discovery of a hard X-ray source associated with any of the WR stars would be highly indicative of stellar wind collision in a massive binary. Indeed, several WR stars towards the Galactic Centre, coincide with hard X-ray sources (e.g. Mauerhan et al. 2010, Nebot Gómez-Morán et al. 2015). From a search of the XMM–Newton science archive and the Chandra source catalogue (1.0), fields including WR60-7 (WC7–8), WR60-8 (WN6o) and WR64-4 (WN6o) had been imaged by Chandra ACIS, although none revealed a source at the location of the WR star.

5.3 Spatial location of the new WR stars

We have estimated distances to the new WR stars by adopting an absolute K_S-band magnitude based on the assigned spectral subtype. To do this, we followed the approach of RC15a, which we briefly summarize here. To calculate the foreground dust extinction to each new WR star, we used subtype-specific intrinsic J−H and H−K_S colours to measure a colour excess. Using the near-IR extinction law of Stead & Hoare (2009), we thereby obtained two measures of extinction in the K_S band, of which an average was taken. Distances were then calculated using the standard relation between absolute and apparent magnitude. We used the uncertainties on calibrated absolute magnitudes, given by RC15a, to calculate upper and lower bounds on the distances calculated. In Table 7, we provide interstellar extinctions, distance moduli/distances, with uncertainties, for every new WR star. Typical extinctions are A_K = 1.2 ± 0.2 mag, with characteristic distances of 9.7 ± 3.8 kpc.

This method inherently assumes the WR star is the sole (or dominant) contributor of near-IR flux to each source. Recalling Section 5.2, this assumption is justified for new WN stars with the exception of WR64-4, the emission line strengths of which suggest a significant contribution from a companion source, implying a larger distance. In addition, we provide a second distance estimate to WR70-13 in Table 7, since there is evidence for a contribution by circumstellar dust to the K band. Adopting the absolute magnitude of a WC8 star that is dominated by hot dust would significantly increase the distance to WR70-13 from 5.3 to 10.7 kpc, though in reality the dust contribution is likely to be modest such that an intermediate distance is more realistic.

5.4 Association of WR stars with the Scutum-Crux and other spiral arms

The new WR stars appear to be evenly distributed throughout the section of Galactic disc observed; neither they, nor the previously mapped WR stars, show any obvious association with the spiral features in the region. Indeed, no pattern resembling a spiral arm can be seen in the distribution of WR stars, as would be expected if they were tightly confined to spiral arms, albeit affected by a systematic offset in distance measurements. The upcoming second data release (DR2) from Gaia should address this general question, although the majority of the newly discovered WR stars are too faint for reliable parallaxes with Gaia. Indeed, only 7 of the 17 stars are included in the Gaia first data release (DR1; Gaia Collaboration 2016), with G = 17.4–20.1 mag.

6 SPATIAL DISTRIBUTION OF WOLF–RAYET STARS AND OTHER MASSIVE STARS IN THE MILKY WAY

6.1 Association of WR stars with star clusters?

If the majority of WR progenitors are born in relatively massive star-forming regions, one might expect them to be in close proximity to their natal star cluster. We have compared the spatial location of the new WR stars with young star clusters from Dutra et al. (2003), Mercer et al. (2005) and Borissova et al. (2011). None of the new WR stars are located within known clusters. At face value, this might be considered to be surprising, but Lundstrom & Stenholm (1984) in the most extensive study of the environment of Galactic WR stars to date, found that only 10–30 per cent of WR stars are located in star clusters.

Since then, the known WR population has quadrupled, so we provide revised statistics for the 298° ≤ l ≤ 340° survey region as a whole, involving 190 + 16 = 206 Galactic WR stars, comprising 119 WN, 1 WN/C and 86 WC stars. Lundstrom & Stenholm (1984) considered a WR star to be associated with a star cluster if its projected distance was within two cluster radii. Here, we soften this requirement, extending the potential association with four-cluster radii, utilizing published cluster centres and radii from Dias et al. (2002), Dutra et al. (2003), Mercer et al. (2005) and Borissova et al. (2011). Overall, 55 WR stars (27 per cent) are associated with a total of 12 star clusters, as shown in Table 8, although it is notable that 44 per cent of all the WR cluster members in our survey range are located within a single open cluster, Westerlund 1 (C06). Consequently, 73 per cent of WR stars in the survey region are not associated with a star cluster.

If the majority of WR progenitors are born in relatively massive star-forming regions, why are so few currently associated with clusters? The lower WR mass limit for single rotating stars at solar metallicity is ∼22 M_⊙ star (Meynet & Maeder 2003). Empirically, such stars are observed in clusters with ≥500 M_⊙ (Larson 2003; Weidner, Kroupa & Bonnell 2010). Stochasticity in the sampling of the initial mass function will result in some massive stars originating in low-mass (10² M_⊙) clusters/star-forming regions, as demonstrated theoretically by Parker & Goodwin (2007). Therefore, some WR stars could be associated with low-mass clusters, with other members of the star-forming region inconspicuous. Indeed, γ Vel (WC8+O), the closest Galactic WR star, is located in a very low mass star-forming region (Jeffries et al. 2014). Other members of such star-forming region would be very difficult to identify at the typical distance of Galactic WR stars.

There is evidence that some dense, young massive clusters rapidly achieve virial equilibrium (e.g. Hénault-Brunet et al. 2012), such that they would retain the bulk of their stellar content over the representative WR ages of 5–10 Myr (Meynet & Maeder 2003). Indeed, the bulk of the WR stars associated with Westerlund 1 lie within the 1.2 arcmin radius – corresponding to 1.4 pc at a distance of 3.8 kpc – such that our four-cluster radius limit is equivalent to ∼5.5 pc. Of course, not all open clusters are dense or bound. For example, Hogg 15 is a sparsely populated cluster with a much larger radius of 3 pc, at its distance of 3 kpc, so will be in the process of dispersing. Consequently, the general lack of an association with star clusters is not wholly unsurprising. Indeed, inspection of the Galactic O Star Catalogue (GOSC v3.0; Maíz Apellániz et al. 2013) reveals 50 (optically visible) O stars in our survey region. Of these, only 18 (36 per cent) are associated with a star cluster, not significantly larger than WR stars since the bulk of these are late-type O stars with comparable ages to many WR stars.

Of course, not all massive stars originate from star clusters. Bressert et al. (2010) demonstrated that only a small fraction of star formation in the solar neighbourhood originate from dense environments. It is probable that the majority of OB stars originate from intermediate density regions, i.e. OB associations and/or extended star-forming regions (see Parker & Dale 2017). Indeed, Wright et al. (2016) have unambiguously demonstrated that the distributed massive stars in Cyg OB2 did not originate from a dense star cluster.

Lundstrom & Stenholm (1984) found that ≥50 per cent of optically visible WR stars were located in OB associations. It is not possible to calculate the fraction of IR-selected WR stars that lie within OB associations since the latter are limited to nearby optical surveys. Instead, it is necessary to compare the location of WR stars with infrared catalogues of star-forming regions. We have compared the locations of all 206 WR stars in our survey region with confirmed H ii regions from the all-sky WISE catalogue of Galactic H ii regions from Anderson et al. (2014). In total, 53 stars are located within ≈1.5 |$R_{\rm H\,\small {ii}}$| of the H ii region, representing 26 per cent of the total WR population, as shown in Table 9. Of course, a subset of these WR stars will be foreground sources, so the quoted statistics represent strict upper limits.

The majority of these WR stars are associated with complexes at G298 (Dragonfish nebula), G305 and G338. By way of example, as many as 12 WR stars are associated with G298 ‘Dragonfish’ complex (Rahman, Moon & Matzner 2011). The VVV CL011 and Mercer 30 clusters are coincident with this region, although radio distances of ∼10.5 kpc significantly exceed spectrophotometric cluster distances. Similar issues affect the association of Mercer 81 with G338.430+0.048, and VVV CL 041 with G317.030+0.028. Finally, stellar winds and/or supernovae associated with Westerlund 1 have sculpted an IR cavity, such that there is no longer an associated H ii region.

Overall, it is apparent that WR stars are rarely located within known H ii regions, albeit with some notable exceptions that include the Carina Nebula in the Milky Way. Typical radii of star-forming regions are 9 arcmin, so for representative radio-derived distances of 7.5 kpc to star-forming regions, we include stars within a projected distance of 30 parsec from the centre of the H ii region. Again, it is possible that stars have migrated away from where they formed. Typical velocity dispersions of stars in such regions are several km s⁻¹, corresponding to several pc Myr⁻¹, such that the WR progenitor could have travelled several tens of parsec. It is important to stress that the parent star-forming region of a WR star would not necessarily display significant radio free–free emission after 5–10 Myr (Meynet & Maeder 2003) since most O stars would have died if there had been a single burst of star formation. Indeed, only 23 (46 per cent) of the 50 GOSC optically visible O stars in our survey region lie in an OB association (Cen OB1, Nor OB1, Ara OB1).

Regardless of whether WR progenitors form in clusters or lower density environments, there are other explanations for their relative isolation. To have travelled more than a few tens of parsec from their birth site, the WR progenitor may have been ejected via dynamical effects or following a supernova kick from a close companion. The former, involving three-body dynamical interactions, is favoured in dense stellar environments, in which the fraction of massive stars ejected is inversely proportional to the stellar mass of the cluster (Fujii & Portegies Zwart 2011). Therefore, the population of WR stars dynamically ejected in this way will be dominated by those from low- to intermediate-mass clusters, explaining why the 10⁵ M_⊙ cluster Westerlund 1 has successfully retained the majority of its WR population. Historically, a supernova origin for runaway WR stars has been considered to be a major factor affecting their spatial distribution (Eldridge, Langer & Tout 2011). This requires a close binary companion, and assumes that all massive stars whose initial masses exceed ∼20 M_⊙ undergo a ccSN. Of course, such stars may collapse directly to a black hole, or form a black hole via fallback, so their companion would not necessarily receive a significant supernova kick (e.g. O'Connor & Ott 2011). Therefore, only a small fraction of isolated WR stars likely originate in this way.

Overall, the most promising scenario for the low observed frequency of WR stars associated with star clusters is via dynamical ejection, but this alone does not explain the very high fraction of WR stars in the field. Instead, it is likely that a majority of Galactic WR progenitors do not form within dense, high-mass star clusters. The observed high fraction of WR stars located in the Galactic field can therefore be attributed to a combination of dynamical ejection from star clusters, and their origins in modest star-forming regions which subsequent dissolve into the field. The latter will not necessarily be recognizable as an OB association during the WR stage since the majority of O stars will have ended their lives after 5–10 Myr, with similar arguments applying to isolated H ii regions. Some distant WR stars will not be recognized as being associated with a star-forming region if other stars in the region possess low masses, as is the case for γ Vel (Jeffries et al. 2014).

6.2 Relative isolation of WR stars and luminous blue variables

The general lack of association between WR stars and O stars – except for the minority located in young, dense star clusters (e.g. NGC 3603, Westerlund 1) – is relevant to the ongoing debate about the nature of LBVs (Humphreys et al. 2016; Smith 2016). Historically, LBVs were considered to be massive stars in transition towards the WR stage. In part, this association arose from the spectroscopic evolution of LBVs to the WN phase (e.g. AG Car; Smith, Crowther & Prinja 1994) and LBV outbursts from WN stars (e.g. R127, HDE 269582; Walborn et al. 2017). Smith & Tombleson (2015) have challenged this view, finding that LBVs in the Milky Way and LMC are more isolated (from O stars) than WR stars. They have argued that they possess lower masses, with their high luminosities arising from being the mass gainers (former secondaries) within close binary systems.

Their conclusions were largely based upon the spatial distribution of O stars, WR stars and LBVs in the LMC, owing to visual studies of Galactic massive stars being severely restricted by dust extinction. Still, the reliance on SIMBAD for catalogues of O stars hinders their conclusions owing to severe incompleteness for both galaxies, while Humphreys et al. (2016) argued for a mass separation between high-luminosity (classical) and low-luminosity LBVs (though see Smith 2016).

Kniazev, Gvaramadze & Berdnikov (2015, 2016) provide the current census of 17 confirmed LBVs in the Milky Way, which is restricted to confirmed spectroscopic variables. Their criteria therefore exclude candidate LBVs possessing ejecta nebulae, which Bohannan (1997) had previously argued should be an additional discriminator. Only three spectroscopically variable LBVs lie within our Galactic survey region – WS1 (Kniazev et al. 2015), Wray 16-137 (Gvaramadze, Kniazev & Berdnikov 2015) and Wd1-W243 (Clark et al. 2005) – so we need to consider the global Milky Way population of LBVs and WR stars.

Of the known WR content in the Milky Way, 27 per cent are members of star clusters (Crowther 2015). Meanwhile, 4 of the 17 spectroscopically variable LBVs are located in star clusters – W243 in Westerlund 1, η Car in Trumpler 16, GCIRS 34W in the Galactic Centre cluster and qF 362 in the Quintuplet cluster (Geballe, Najarro & Figer 2000) – comprising 24 per cent of the total, so their global statistics are comparable. Indeed, a number of widely accepted LBVs known to be associated with star clusters are omitted from the compilation of Kniazev et al. (2015). These include the Pistol star (qF 134), another member of the Quintuplet cluster (Figer et al. 1998), and the LBV in the 1806-20 cluster (Eikenberry et al. 2004; Figer, Najarro & Kudritzki 2004).

5 of the 17 spectroscopically variable LBVs are potentially associated with star-forming regions (Anderson et al. 2014) from a similar exercise to that discussed above for WR stars, namely η Car, Wray 16-137, HD 168607, AFGL 2298 and G24.73+0.69, namely 29 per cent of the total. Consequently, the overwhelming majority of WR stars and LBVs are located in the Galactic field, away from star clusters or star-forming regions. Overall, there is no significant difference in the spatial distribution of WR and LBVs in our Galaxy, with a quarter of both populations associated with young massive star clusters.

Smith & Tombleson (2015) proposed that LBVs generally arise from significantly lower mass progenitors than WR stars, challenging the hitherto scenario that LBVs are transition objects between the O and WR phases. However, those star clusters that host LBVs are relatively young (4–6 Myr; Clark et al. 2005; Bibby et al. 2008; Liermann, Hamann & Oskinova 2012; Schneider et al. 2014), putting aside η Car.⁶ Not only are the statistics of WR and LBV association with star clusters comparable, but crucially four young Milky Way clusters hosting LBVs – Westerlund 1, CL 1806-20, the Galactic Centre and the Quintuplet – also contain (classical) WR stars.

Smith & Tombleson (2015) argued that LBVs arise from mass gainers in close binary systems (Langer 2012; de Mink et al. 2014). Mass gainers will be rejuvenated, yielding an apparently younger star than the rest of the population (Schneider et al. 2014). The presence of spectroscopically variable LBVs (Wd1-W243, GCIRS 34W, qF 362) plus leading LBV candidates (qF 134, LBV 1806-20) in young clusters with progenitor masses in the range 30–40 M_⊙ and coexisting with O stars and WR stars, should permit their scenario to be tested. Of course, LBVs in such systems might be the mass-gaining secondaries whose primaries have already undergone core-collapse. However, LBV 1806-20 is an SB2 system, whose companion is not a WR star since it exhibits He i absorption lines. This appears to rule out the Smith & Tombleson (2015) scenario in this instance.

If LBVs are rejuvenated secondaries in close binaries spanning a wide range of masses, they should also be present in older, massive star clusters such as those hosting large red supergiant (RSG) populations. However, Smith & Tombleson (2015) report that LBVs and RSGs do not share a common parent population, and LBVs are not known within RSG-rich clusters at the base of the Scutum-Crux arm (RSGC 1–3, Figer et al. 2006; Davies et al. 2007; Clark et al. 2009). The presence of ∼50 RSG in these clusters and absence of LBVs either argue against the Smith & Tombleson (2015) scenario, or require a short (∼2 × 10⁴ yr) lifetime for such LBV, in view of the ∼1 Myr RSG lifetime. The only potential LBV within these RSG-rich clusters identified to date is IRAS 18367−0556 in RSGC 2, which possesses a mid-IR ejecta nebula (Davies et al. 2007). Overall, it is apparent that LBVs span a wide range of physical properties (Smith, Vink & de Koter 2004), though the same is true for WR stars, some of which are located in significantly older star clusters (recall table 9 from C06), albeit solely Westerlund 1 hosts RSG and WR stars.

7 CONCLUSIONS

We have undertaken a near-IR spectroscopic search for WR stars along the Scutum-Crux spiral arm, based upon 2MASS and GLIMPSE photometric criteria (Mauerhan et al. 2011; Faherty et al. 2014). Observations of 127 candidate stars (K_S ∼ 10–13 mag) resulted in the confirmation of 17 WR stars (14 WN, 3 WC), representing a success rate of ∼13 per cent, comparable to previous IR-selected studies based on similar criteria (Hadfield et al. 2007; Mauerhan et al. 2011). More sophisticated techniques are clearly required for optimizing future spectroscopic searches. As we have found, large numbers of other stellar types (young stellar objects, Be stars) lie in the same location as WR stars within individual colour–colour diagrams, but may not do so in a multidimensional parameter space. Therefore, future progress might entail a Bayesian approach to optimizing searches for candidate WR stars.

We have extended our earlier near-IR classification scheme (C06) to cover all YJHK bands and all subtypes, including WN/C and WO subtypes. This has been tested on several recently discovered WR stars for which optical classifications have previously been obtained. In general, the near-IR criteria are successful in obtaining reliable subtypes, including the presence of atmospheric hydrogen in WN stars, and identifying transition WN/C stars. However, inferences are weaker if limited wavebands are available, the spectral resolution is modest and/or the signal to noise obtained is low.

The majority of newly discovered WR stars are weak-lined WN5–7 stars, with two broad-lined WN4–6 stars and three WC6–8 stars. Therefore, despite the low success rate, our goal of extending the spectroscopic confirmation of WR stars to the far side of the Milky Way has been successful. Based on the absolute magnitude calibration of C15, inferred distances (∼10 kpc) extend beyond previous spectroscopic surveys, with |$A_{K_{{\rm s}}}{\,\sim \,}1.2$| mag. Spectroscopic searches beyond the Galactic Centre (⁠|$A_{K_{{\rm s}}}{\,\sim }\,3$| mag) will be significantly more challenging.

Only a quarter of WR stars in the selected Galactic longitude range are associated with star clusters and/or H ii regions. We suggest that this arises from a combination of dynamical ejection from (modest stellar mass) clusters and formation within lower density environments (OB associations/extended H ii regions). We also revisit the recent controversy about the association between LBVs and WR stars, or lack thereof. Considering the whole Milky Way, 27 per cent of WR stars are hosted by clusters, comparable to that of spectroscopically variable LBVs (4 from 17 stars). More significantly, several young clusters with main-sequence turn-off masses close to 30–40 M_⊙ host classical WR stars and LBVs, permitting the Smith & Tombleson (2015) suggestion that some LBVs are rejuvenated mass gainers of close binaries to be tested. Specifically, the non-WR companion to LBV 1806-20 and absence of LBVs in RSG-rich star clusters argue against this scenario.

Returning to the main focus of this study, until more robust distances to WR stars – and in turn absolute magnitudes – can be established by Gaia there are legitimate concerns about the completeness of surveys for different subtypes, especially the challenge of identifying faint, weak emission line WN stars with respect to WC stars (Massey, Neugent & Morrell 2015a). Near-IR narrow-band photometric searches suffer from dilution of emission lines by companion stars and hot dust emission from WC binaries, while our broad-band near- to mid-IR photometric approach is limited by the low spatial resolution of Spitzer. Massey, Neugent & Morrell (2015b) have undertaken a deep optical narrow-band survey of the LMC, revealing a population of faint, weak-lined WN3 stars (which they characterize as WN3/O3 stars). Stars with these characteristics – which would usually be classified as WN3ha according to the Smith et al. (1996) scheme – comprise a negligible fraction of Galactic WR stars (e.g. WR3), a minority of the moderately metal-poor LMC WR population, and a majority of the more significantly metal-deficient SMC WR population (Hainich et al. 2015).

It is anticipated that ongoing infrared searches using a combination of continuum methods and emission line characteristics will significantly improve the completeness of WR surveys in a sizeable volume of the Galaxy within the near future.

Acknowledgements

This work is based on observations with ESO telescopes at the La Silla Paranal Observatory under programme 094.D-0839(A).We wish to thank Nicole Homeier, Bill Vacca and Frank Tramper for providing us with near-IR spectroscopic data sets, from NTT/SOFI, IRTF/SpeX and VLT/XSHOOTER, respectively, for several WR stars. We appreciate useful discussions with Simon Goodwin and Richard Parker, and useful comments from the referee which helped improve the focus of the paper. Financial support for CKR was provided by the UK Science and Technology Facilities Council. PAC would like to thank ESO for arranging emergency dental care in La Serena immediately after the 2015 observing run.

REFERENCES

SUPPORTING INFORMATION

Supplementary data are available at MNRAS online.

new_wr_aug17_appendix.pdf

Please note: Oxford University Press is not responsible for the content or functionality of any supporting materials supplied by the authors. Any queries (other than missing material) should be directed to the corresponding author for the article.