Luminous and Variable Stars in M31 and M33. IV. Luminous Blue Variables, Candidate LBVs, B[e] Supergiants, and the Warm Hypergiants: How to Tell Them Apart^*

The upper HR diagram is populated by a diversity of luminous and variable stars of different types, many of which are distinguished by their emission-line spectra and evidence for stellar winds and mass loss. The different groups or types include the luminous blue variables (LBVs), B[e] supergiants, the Wolf–Rayet stars of different types, and more generic hot emission line stars. Many of these stars occupy the same regions of the HR diagram, and they may or may not be related. They could be stars of similar initial mass but in different stages of their evolution or have experienced different mass-loss histories. This diversity is one of the challenges in understanding massive stars, their evolution, and eventual fate.

In this series of papers we have presented the results of a spectroscopic survey of luminous and variable stars in the nearby spirals M31 and M33. In Paper I (Humphreys et al. 2013) we discussed a small group of very luminous intermediate-temperature supergiants, the warm hypergiants, and showed that they were likely post-red supergiants. In Paper II (Humphreys et al. 2014b), we reviewed the spectral characteristics, spectral energy distributions (SEDs), circumstellar ejecta, and mass loss for 82 luminous and variable stars, including the confirmed LBVs, candidate LBVs, and other emission-line stars. We found that many of these stars have warm circumstellar dust, including several of the stars with Fe ii and [Fe ii] emission lines, but concluded that the confirmed LBVs in M31 and M33 do not. Interestingly, the confirmed LBVs have relatively low wind speeds even in their hot, quiescent or visual minimum state compared to the B-type supergiants and Of/WN stars that they spectroscopically resemble.

The final state of the most massive stars as core-collapse supernovae (SNe) is now in question. Smartt et al. (2009) and Smartt (2015) have suggested an upper mass limit of ≈18 M⊙ for the red supergiant progenitors of the SNe II, while Jennings et al. (2014) find a lack of massive progenitors in M31 and M33 and suggest an upper mass of 35–45 M⊙. In our Paper III, Gordon et al. (2016) presented a comprehensive spectroscopic survey of the yellow supergiants (YSGs). Based on spectroscopic evidence for mass loss and the presence of circumstellar dust in their SEDs, we concluded that 30%–40% of the YSGs are likely in a post-red-supergiant state. A comparison with evolutionary tracks shows that these mass-losing post-RSGs have initial masses between 20 and 40 M_⊙, suggesting that red supergiants in this mass range evolve back to warmer temperatures before their terminal state. This is a significant result given the evidence that the most massive stars may not end their lives as supernovae and the recent announcement of a possible failed SN from a red supergiant candidate (Adams et al. 2016).

In this paper (Paper IV), we present spectroscopy of 132 additional luminous stars, variables, and emission-line objects. Most of the stars have emission-line spectra, including LBVs and candidate LBVs, Fe ii emission line stars and the B[e] supergiants, and the warm hypergiants. Many of these objects are spectroscopically similar and are often confused with each other. For example, stars with Fe ii and [Fe ii] emission are sometimes assumed to be LBV candidates. With this large spectroscopic data set including various types of emission-line stars, we can examine their similarities and differences. We briefly describe our new observations in the next section. In the following sections, we discuss their spectroscopic and photometric properties, their SEDs, and the presence or lack of circumstellar dust. In the last section we summarize their characteristics and suggest guidelines that can be used to help distinguish or separate these stars in future work.

Our target selection and observations with the MMT/Hectospec (Fabricant et al. 2005) and LBT/MODS1 were described in Paper I. Our subsequent new spectra are listed here in Table 1. In Paper II, we assigned our targets to six different groups according to their spectroscopic and photometric characteristics described there, and we follow that designation in this paper. All of the hot stars for which we have new spectra are listed in Table 2 in order of Right Ascension with their position⁴ , group type, spectral type where appropriate, and alternate names or designations. The YSGs are listed in Paper III. We use the same star designations, the galaxy name and the R.A. of the star, as in the previous three papers, and reference their earlier spectra described in those papers. The M33C designation in Table 2 comes from an unpublished Hα survey by Kerstin Weis. The "V-" notation is used for stars selected from the emission-line survey of M33 by Valeev et al. (2010). A shortened version of Table 2 is shown here. The full-length table is available in the online edition. The blue and red spectra in FITS format of all of the stars observed with the MMT/Hectospec are available at etacar.umn.edu/LuminousStars. The year of observation is also included in Table 2 to aid in locating the spectra in the online database.

We repeated spectra of several stars discussed in Papers I and II to look for variability, including the confirmed and candidate LBVs, the hypergiants, and several emission-line stars. These stars are listed separately in Table 3 following the same format as Table 2. Any noteworthy spectroscopic changes are noted in the following sections. We also continued to photometrically monitor many of these stars at the Barber Observatory as described in Humphreys et al. (2015). This multicolor photometry is available for most of the stars discussed in this series of papers at http://etacar.umn.edu/LuminousStars/. A paper on the details of the photometry and the variable stars is in preparation (J. C. Martin & R. M. Humphreys 2017, in preparation).

The visual photometry from Massey et al. (2006) was cross-identified with the near- and mid-infrared photometry from 2MASS (Cutri et al. 2003), the Spitzer surveys of M31 (Mould et al. 2008) and M33 (McQuinn et al. 2007; Thompson et al. 2009), and WISE (Wright et al. 2010). The first 10 entries are shown in Table 4, and the full table is available in the online edition. Table 4 also includes information on variability for the M31 stars from the DIRECT survey (see Kaluzny et al. 1998), and for M33 stars from Hartman et al. (2006) and from Burggraf (2014). The Burggraf compilation includes data from several sources from ∼1920 to the present, but most are since 1970. We have also included information on variability from our own monitoring of individual stars since 2010.

The LBV/S Dor variables exhibit extended periods of enhanced mass loss distinguished by a specific spectroscopic and photometric variability (Humphreys & Davidson 1994; Humphreys et al. 2016). During their maximum (visual) light or dense wind stage, their spectra transition from a hot star with mass loss to resemble a cooler A- or F-type supergiant spectrum produced by the optically thick wind. Given the infrequency of these high mass loss episodes, that is, the S Dor variability, few confirmed LBVs are known in M31 and M33.

The confirmed LBVs including the recently identified J004526.62+415006.3 (Humphreys et al. 2015; Sholukhova et al. 2015) are listed in Table 5 with the few candidates discussed here. Their spectra from our observations in 2010 were described in Paper II. These LBVs and candidates have been observed again in 2013–2015 to look for any significant spectroscopic changes. In this section we briefly discuss those LBVs that have shown some spectroscopic variability since 2010. The short-term variability noted here in the spectra for several of the LBVs is very likely typical of their instability and the resulting variations in their winds and mass loss, even when they are in their quiescent state.

The visual spectrum of AE And has always shown strong emission lines of Fe ii and [Fe ii] (Humphreys 1975). In Paper II we noted that the strong H and He i emission lines had weakened compared to an earlier spectrum from 2004. Below 4100 Å these lines were in absorption, and their relative strengths suggested a corresponding B2–B3 spectral type for the underlying star. Thus the wind appeared to be weakening, but its spectrum from 2013 shows that although the absorption lines were still present, the P Cygni emission had reappeared, and in the 2014 spectrum, the absorption has been completely replaced by emission. These variations in the spectrum indicate continued instability and variation in the mass loss even though AE And is in its quiescent state.

Var A-1 in M31 has prominent Fe ii and [Fe ii] emission with H and He i P Cygni profiles and N ii absorption lines. In the past (Paper II), it has shown variation in the strength of the P Cygni profiles, but its spectra from 2010, 2013, and 2015 show no substantial change.

AF And and Var 15 in M31, in quiescence, have spectra like the Of/late-WN stars. Both showed substantial change in the few years from 2010 to 2013 or 2015. The P Cyg absorption minima in the Balmer lines in AF And are much stronger in 2013 and 2015, although there is no similar change in the He i P Cyg profiles.

In Var 15, there was no spectral variability between 2010 and 2013, but between 2013 and 2015, the He i and [Fe ii] emission lines disappeared and the N ii emission lines greatly weakened, as shown in Figure 1. Our photometric monitoring at the Barber Observatory shows that Var 15 brightened by 0.4–0.5 mag in the BV and R bands from 2013 to 2014 September and continued to brighten by another 0.1 mag from 2014 to 2015. During the same time, its B − V color, however, reddened only slightly from 0.4 to 0.5 mag. Thus the spectroscopic and photometric changes are correlated. The disappearance of the He i emission suggests a decrease in the apparent temperature, supported by the somewhat redder color. Previous photometry (Szeifert et al. 1995) from 1992 showed Var 15 significantly brighter but with a similar color, while photometry in Massey et al. (2006) from 2000 shows a fainter star with a much bluer B − V color of −0.01 mag. Clearly, Var 15 has undergone some significant changes in its wind that have largely gone unnoticed. It should be monitored more closely.

Zoom In Zoom Out Reset image size

The new LBV in M31, J004526.62+415006.3, currently in "eruption," developed enhanced emission in the Fe ii lines in the 5000–5600 Å region in the 2015 spectrum relative to 2013, although the absorption-line spectrum is unchanged. Recent photometry shows that it has faded about 0.25 mag, but the color is unchanged.

Var 83 in M33 has a spectrum very much like Var A-1 with N ii absorption lines and Fe ii and [Fe ii] emission lines, and with strong P Cygni profiles in the Balmer lines. Stronger P Cygni emission profiles in the He i lines appeared between 2010 and 2013, but there was no significant change in the 2014 spectrum. However, subsequently Var 83 brightened 0.5 mag in V between 2014 November and 2015 October, but its color did not change. It has remained at ≅15.8 mag since.

Var B and Var 2 in M33 have spectra like the Of/late-WN type stars with nitrogen emission and He i P Cygni profiles. Var 2 also has the He ii λ4686 in emission (Figure 5 in Paper II). No changes are noted in their spectra from 2010 to 2013 and 2014, but the Fe ii and [Fe ii] emission lines in Var B disappeared between 2004 and 2010 (Figure 2 in Paper II).

Var C in M33 is currently in "eruption" (Humphreys et al. 2014a; Burggraf et al. 2015). No additional spectra were obtained after 2013, but the available photometry shows that it is still in its dense wind state, although it has faded about 0.5 mag between 2015 November and 2016 September, and the B − V color has shifted from 0.2 to 0.0, so it is probably transitioning back to its hotter, quiescent state.

This brief survey of the confirmed LBVs in their quiescent state revealed spectroscopic variations that in most cases are small and are due to variability in their winds and mass loss. This is not surprising for stars known for instability.

In Paper II we showed that the SEDs of the confirmed LBV/S Dor variables do not have any evidence for warm dust in the near- and mid-infrared. We also found that the LBVs have relatively slow winds, even in their quiescent state, typically ≈100–200 km s⁻¹, compared to the winds of normal hot supergiants and the Of/WN stars. These characteristics can also be used to identify LBV candidates and separate them from other emission-line stars.

3.1. LBV Candidates

Numerous stars have been suggested as candidate LBVs in the literature (Massey et al. 2007; Clark et al. 2012; Massey et al. 2016; Sholukhova et al. 2015). In Paper II, we also recognized several stars as potential LBVs. Our spectra of several of the stars that have been suggested as LBV candidates are described here.

B526 (Humphreys & Sandage 1980) (=M33C-7292) has been considered a candidate LBV based on its spectral appearance (Clark et al. 2012), but it was also suspected to be more than one star (Monteverde et al. 1996). HST/WFPC2 images⁵ (Figure 2) show that B526 is two stars of comparable brightness about one arcsec apart. For this reason we obtained long-slit spectra with the LBT/MODS1 in good seeing, which easily separated the two stars. The NE star has the spectrum of a hot supergiant with Fe ii emission, weak [Fe ii] emission, H emission with P Cygni profiles, and He i absorption lines, thus resembling many LBVs in their quiescent state. The Ca ii triplet in emission is also present in its red spectrum. The SW component has the normal absorption-line spectrum of a B5-type supergiant. The blue and red spectra of these two stars are shown in Figure 3, and photometry for the two stars measured from the HST images is included in Table 4.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

The infrared photometry from 2MASS and Spitzer/IRAC is for the combined light of the two stars. Although they cannot be separated, their long-wavelength SED clearly shows a lack of warm dust in the system, similar to LBVs. However, its wind velocity of 370 km s⁻¹, measured from the P Cyg minima in the Balmer lines, is significantly higher than in the confirmed LBVs. As noted later, its red spectrum has the [O i] emission at λ6300, a characteristic of the B[e] supergiants, not present in the spectra of LBVs. Thus as we will discuss later, B526NE may be a normal, mass-losing hot supergiant or perhaps a transition star possibly related to the B[e] supergiants; see Section 4.

In Paper II, we suggested that the spectrum of M31-004425.18 had changed from an early A-type spectrum described by Massey et al. (2007), who called it a cool LBV candidate, to a much hotter star, based on its absorption-line spectrum described in Paper II. M31-004425.18 should definitely be listed as an LBV candidate if not a true LBV based on its apparent spectroscopic change like the S Dor variables, although it has a relative low luminosity compared to the known LBVs and lies below the S Dor instability strip; see Figure 12 in Paper II. Our spectra from 2013 and 2015 compared to the 2010 spectrum show small variations in the strength of the He i emission that are indicative of variability in the wind and mass loss typical of LBVs in quiescence.

UIT008 (J013245.41+303858.3) is an Of/late-WN star. In Paper II, we classified it as O7-O8f I and Massey & Johnson (1998) called it Ofpe/WN9.⁶ Hα and Hβ have asymmetric profiles with a broad wing to the red suggestive of electron scattering in the wind. UIT008 also has a very low outflow velocity for a star of this type of only −159 and −138 km s⁻¹ measured respectively from the H and He i emission lines (Paper II). It is on this basis that we call it an LBV candidate. See Section 7 of this paper for a discussion of the Of/late-WN stars as candidate LBVs.

M31-004051.59 has been suggested to be an LBV or candidate LBV by Sholukhova et al. (2015) and Massey et al. (2007, 2016). We discussed its spectrum and SED in Paper III and concluded that its nature is somewhat ambiguous. It is spectroscopically similar to an LBV at maximum light in the dense wind state and to the mass-losing post-RSGs discussed in Paper III. It has the absorption-line spectrum of an early A-type supergiant with strong P Cygni profiles and broad wings at Hα and Hβ. The prominent P Cygni profiles in the multiplet 42 Fe ii lines in the 2013 spectrum weakened in 2015, and Fe ii emission lines appeared at λλ5234,5276,5284. Its SED shows a small excess in the near-infrared due to free–free emission, but no evidence for circumstellar dust like LBVs.

Massey et al. (2016) listed several stars with Fe ii and [Fe ii] emission line spectra as LBV candidates. Fe ii emission lines are common in the confirmed LBVs, but as we discuss in the next section, many of the stars listed in their Tables 3 and 4 are actually B[e] supergiants. The B[e] supergiants also have significant circumstellar dust not observed in the confirmed LBVs.

We emphasized in Paper II that the presence of Fe ii emission lines is ubiquitous in astronomical spectra. Fe ii emission is observed in many different types of stars with a wide range of luminosities and temperatures and in different evolutionary states. We defined the Fe ii emission line group as those stars with a strong blue continuum, with strong hydrogen emission, Fe ii emission, and a lack or dearth of absorption lines such that the nature of the underlying star was not clear.

A significant subset of the Fe ii emission line group known as B[e] stars (Lamers et al. 1998) also show [Fe ii] in emission plus other forbidden lines such as [O i] and [Ca ii]. In a recent paper on a spectroscopic survey of emission-line stars, Aret et al. (2016) designated [O i] λλ6300,6364 emission as one of the characteristics of the B[e] class. They also noted that the [Ca ii] doublet at λλ7291,7324, an indicator of circumstellar gas, while not present in all of the B[e] stars, is more common in those of high mass. The upper level of the transition that produces the [Ca ii] lines is the lower energy state of the near-infrared Ca ii triplet. Thus [Ca ii] emission implies emission in the Ca ii triplet, which we confirmed for the stars with far-red spectra.⁷ The ratios of the Ca ii and [Ca ii] lines can then be used to derive the electron densities in the circumstellar ejecta; see Paper I and II. Furthermore, the Ca ii triplet in emission is produced by overpopulation of the upper level for the Ca ii H and K lines in absorption. This in turn suggests that the underlying star must be relatively cool, perhaps of spectral type late B to A or even later.

The supergiant B[e] stars (hereafter sgB[e]) are of special interest in this study not only because of their luminosities, but also, with their Fe ii and [Fe ii] emission lines, they are often mistaken for LBVs or LBV candidates. In Table 5 we list all of the stars in the Fe ii emission line group from this paper and from Paper II plus two in M31 from the list of candidates in Massey et al. (2016) with a summary of the permitted and forbidden emission lines observed in their spectra: Fe ii, [Fe ii], He i, [O i], [Ca ii], the Ca ii triplet, and the O i triplet at λ7774A. All of these stars have hydrogen emission. We classify the stars with both [Fe ii] and [O i] emission as B[e] stars. All of the B[e] stars in M31 and M33 are luminous stars, that is, supergiants, so we adopt the sgB[e] designation and use it in Tables 2 and 3. Based on this criterion, 18 of the 28 stars in Table 5 are B[e] supergiants. Most also have significant excess infrared radiation due to warm dust, over and above what would be expected from free–free emission from their stellar winds; see Section 6. The remaining 10 stars are Fe ii emission line stars and have neither [Fe ii] nor [O i] emission lines in their spectra. They also lack circumstellar dust like the LBVs. Fifteen of the stars included as LBV candidates by Massey et al. (2016) are B[e] supergiants. Repeat spectra were obtained for five of the sgB[e] stars and one Fe ii emission line star, and no changes were noted.

For comparison, we include the LBVs in Table 5. The notation for the emission lines refers to the LBV when in its quiescent state for which we use our spectra from 2010, unless the same lines showed a significant change in previous or later spectra, such as for M31 Var 15 in 2015 when the [Fe ii] were no longer present and Var B in 2004 (Paper II). This emission-line summary shows a distinct difference with respect to the B[e] supergiants. Most of the LBVs have [Fe ii] or Fe ii emission lines, but none have [O i] in emission or [Ca ii]. The same spectrosopic difference is shown by the candidate LBVs M31-004425.18 and M31-004051.59 discussed in Section 3.1. The spectrum of B526NE, however, does have the [O i] line like the sgB[e] stars. Its immediate neighbor B526SW does not, so we can rule out contamination by emission nebulosity or a residual night sky line as a possible explanation.

In Paper I we identified seven stars that we labeled warm hypergiants based on their high-luminosity A- and F-type absorption-line spectra, the presence of strong Balmer emission with broad wings and P Cygni profiles, and the Ca ii triplet and [Ca ii] lines in emission. The ratios of the Ca ii and [Ca ii] lines provide information on the electron densities in the circumstellar gas, which range from ∼1 to 4 × 10⁷ cm⁻³. They all have significant near- and mid-infrared excess radiation due to free–free emission or thermal emission from dust. Six of the stars have very dusty environments, and on that basis we suggested that they are post-RSG candidates; see also Paper III. B324, the visually most luminous star in M33, does not have circumstellar dust and therefore may be an example of a massive star near the upper luminosity boundary in the HR diagram, evolving toward cooler temperatures. In Paper III, we recognized another warm hypergiant, M31-004621.05, that had previously been considered an LBV candidate (Massey et al. 2007). Its spectrum and SED are in Paper III and its spectrum is also shown here in Figure 4.

Zoom In Zoom Out Reset image size

Repeat spectroscopy was obtained for all of the warm hypergiants. None of them showed any variability except for M31-004322.50. There was no change from 2010 to 2013, but the prominent Fe ii P Cygni emission profiles in multiplet 42 were much weaker in 2015, and the Fe ii emission lines with P Cygni profiles in the red spectrum were replaced by absorption lines, signifying possible changes in its wind and a decrease in the density of its circumstellar gas.

The warm hypergiants are listed in Table 6 with their emission-line summary for comparison with the B[e] supergiants and the LBVs in Table 5. In many respects, their emission-line characteristics plus the presence of circumstellar dust overlap with the properties of the sgB[e] stars. A couple of hypergiants have both [Fe ii] and [O i] emission. We caution, however, about the presence of weak [O i], which could be from surrounding nebulosity or a residual from night sky subtraction, especially if the second, weaker line at λ6364 is not clearly present. But the presence of absorption lines in the warm hypergiants, typical of intermediate-temperature supergiants, clearly separates them from the sgB[e] stars, whose spectra lack apparent absorption lines. Furthermore, the lack of He i emission together with the O i λ7774 triplet in absorption in the hypergiants also indicates a significant temperature difference between them and the sgB[e] stars. The [Ca ii] doublet in emission, and by implication the Ca ii triplet in emission, is another characteristic that some of the B[e] supergiants share with the hypergiants. Aret et al. (2012) noted this spectroscopic similarity in several sgB[e] stars in the LMC and suggested that they may be candidates for blueward evolution from the yellow hypergiant stage. Thus, it is possible that the sgB[e] stars may be the hotter counterparts of a class of stars that are candidates for post-red-supergiant evolution. In Figure 4, we show the spectra of the two warm hypergiants, M31-004522.52 and M31-004621.08, that have [Fe ii] and [O i] emission with one of the B[e] supergiants with the [Ca ii] emission, M33C-15731.

We have emphasized here and in Paper II that the SEDs of the LBVs show free–free emission, but no evidence for warm circumstellar dust in the near- or mid-infrared, in contrast with both the B[e] supergiants and the warm hypergiants. The SEDs for several of these stars belonging to all three groups are shown in Papers I and II. The Fe ii emission line stars, distinct from the B[e] supergiants, also do not have warm dust in their SEDs. To illustrate this difference, we show the SEDs for two of them and two sgB[e] stars from this paper in Figure 5. The presence or not of circumstellar dust is included in Table 5 for each of the stars. One of the B[e] supergiants, M33-013242.26, does not have warm dust but polycyclic aromatic hydrocarbon (PAH) emission from surrounding nebulosity, and M31-004307.1 has free–free emission but weak evidence for circumstellar dust.

Zoom In Zoom Out Reset image size

Oksala et al. (2013) and Kraus et al. (2014) demonstrated that the dusty B[e] stars separate from the LBVs in the $J-H$ versus $H-K$ two-color diagram. Following their example, in Figure 6(a) we show the near-infrared two-color diagram for the LBVs, B[e] supergiants, the Fe ii emission line stars, and the warm hypergiants. This figure shows more scatter than the two-color figures in the Oksala and Kraus papers. The separation between dusty stars and stars without dust at H − K ≅ 0.5 –0.6 mag is not clear-cut. This is partly because many of the sgB[e] stars are below the magnitude limit for the 2MASS survey, lack JHK magnitudes, and therefore are not plotted. Furthermore, stars with strong free–free emission in the near-infrared have larger H − K colors. The dusty stars are more clearly separated using the longer wavelength IRAC magnitudes, [3.6–4.5] versus [5.8–8.0], shown in Figure 6(b). Note that the stars to the lower right in Figure 6(b) have excess emission in the 8 μm band due to PAH emission from surrounding nebulosity.

Zoom In Zoom Out Reset image size

Hot Supergiants and Yellow Supergiants. Many of the stars in Table 2 and most of those in Paper II have emission lines primarily of hydrogen, many with P Cygni profiles, and some also with He i emission. The question is, are these LBVs or LBV candidates? In Humphreys et al. (2014a) and in Paper II we noted that the outflow velocities of the confirmed LBVs in the quiescent state were significantly lower than their hot supergiant counterparts that they spectroscopically resemble. Furthermore, the confirmed LBVs and the candidates that resemble hot, mass-losing supergiants also have Fe ii and [Fe ii] emission lines in their spectra (Table 5). This is of course one of the reasons that the sgB[e] stars are often confused with LBVs. The few LBVs that do not have Fe ii and [Fe ii], such as AF And and Var 2 in M33, are those whose spectra are similar to the Of/late-WN stars. These spectroscopic characteristics, slow winds and Fe ii emission lines, can be used to help identify potential LBV candiates and separate them from the numerous hot supergiants with stellar winds and mass loss in the upper HR diagram.

The same is true for the later-type supergiants, those with A-type spectra and the YSGs. During the LBV "eruption" or dense wind state, their absorption-line spectra resemble cooler A- to F-type supergiants with apparent temperatures of ≈7500°–9000°. They not only have strong hydrogen emission lines with P Cyg profiles, but also Fe ii emission lines, as seen in the maximum light spectra of Var C (Humphreys et al. 2014a) and J004526.62+415006.3 (Humphreys et al. 2015). In Paper III we identified numerous YSGs with hydrogen emission line spectra, many with P Cyg profiles indicative of mass loss. These YSGs occupy the same part of the HR diagram as the LBVs in eruption and like the warm hypergiants and the B[e] supergiants can be mistaken for an LBV. The spectra of these stars lacked the Fe ii emission, and they were not LBVs in eruption. Instead we suggested that they were candidates for post-red-supergiant evolution.

Based on the above characteristics, we did not identify any additional LBV candidates among our new spectra described in Table 2. We also repeated spectroscopy for several hot and intermediate-type supergiants discussed in Paper II to look for potential variability (Table 3). The intermediate-type stars or YSGs did not demonstrate any spectral variability, and none of them have the Fe ii emission lines. The B1Ia + WN supergiant, M33C-9304, did not show any spectral variability. Three of the hot supergiants in M33, however, are listed by Massey et al. (2016) as LBV candidates: M33C-19725(J013339.52+304540.5), B416(J013406.63+304147.8, UIT 301), and J013424.78+303306.6.

We discussed J013406.63+304147.8 at length in Paper II. It is a very luminous O9.5 Ia supergiant. Its spectrum does indeed have weak Fe ii and [Fe ii] emission, but the only P Cyg profiles are the He i λ5876 and λ4471 lines, and the measured outflow velocity of −280 km s⁻¹ is somewhat high for an LBV in quiescence. Its spectrum from 2013 did not show any signficant changes in the emission lines. It is also a possible binary. We concluded that it was not an LBV, but a very luminous star or pair of stars. Its characteristics are a mixture, but with the Fe ii emission lines it is a possible LBV candidate and deserves further observation. We described M33C-19725 (B0.5 I) as peculiar in Paper II due to emission in the core of the He i lines and a second emission component on the red wing of Hα and Hβ. In its spectrum from 2013, the He i lines have changed and now clearly show P Cygni profiles, while the P Cyg absorptions have weakened in the Balmer lines, suggesting changes in the wind and mass-loss rate. The emission feature on the red wing of Hα and Hβ is also gone. Its outflow velocity is normal for an early B-type supergiant. M33C-19725 does not have any Fe ii or [Fe ii] emission lines, and based on the LBV characteristics described above, we do not consider it an LBV candidate. J013424.78+303306.6 is a late B/early A type supergiant with H emission. It spectrum does not have any P Cygni absorption features, but it does show weak Fe ii and [Fe ii] lines that were noticeably weaker in its spectrum from 2014. Thus it may be an LBV candidate or a low-luminosity LBV, possibly in its dense wind state, perhaps similar to M31-004425.18; see Section 3.

Massey et al. (2016) included an additional five and eight supergiants in M31 and M33, respectively, as possible LBV candidates. We have spectra of nine of these stars (Paper II) and obtained spectra of two additional stars from the CfA Optical/Infrared Archive for MMT/Hectospec data: J004221.78+410013.4 and J004444.01+415152.0. They are respectively a B[e] supergiant and an Fe ii emission line star and are included in Table 5. Spectra of the other two candidates were not available in the archive. The spectra of the nine stars⁸ are described in Paper II, Table 1. We examined them again, applying the spectroscopic characteristics for LBV candidates. Except for M33C-4640(J013303.09+303101.8), none qualify as LBV candidates. They have P Cyg profiles in the Balmer lines that sometimes have broad wings due to stellar winds and mass loss, but no Fe ii emission. We discussed M33C-4640 in Paper II and suggested that with its less-luminous A-type spectrum and weak Fe ii emission, it might be a post-red supergiant. Burggraf (2014) concluded that it did not show any significant variability over the preceding 12 years. Although it does not show any evidence for recent mass loss, its spectral characteristics are equally consistent with an LBV candidate of lower luminosity, similar to J013424.78+303306.6 discussed above and M31-004425.18.

Of/late-WN stars. Four of the confirmed LBVs, AF And and Var 15 in M31 and Var B and Var 2 in M33, have spectral characteristics that resemble the Of/late-WN type stars. We listed 11 in Paper II and showed examples of some of their spectra with the spectroscopically similar LBVs. Sixteen more are included here in Table 2. Unfortunately, there is little to spectroscopically distinguish the four confirmed LBVs as a group from the larger set of Of/late-WN stars. This does not mean that all Of/WN stars are candidate LBVs. Of course if they eventually exhibit the characteristic S Dor variability they are LBVs. Short-term spectroscopic changes such as those described in Section 3 for AF And, Var 15, and Var 2 would also be indicators that an Of/late-WN star may be a candidate LBV.

In Paper II, we found that the Of/WN stars have somwhat higher wind velocities of −329 and −313 km s⁻¹ for the H and He i lines, respectvely, compared to −229 and −221 km s⁻¹ for the four LBVs with Of/late-WN type spectra. The spectra of many Of/WN stars have P Cygni profiles in their H and He i emission lines, and although this is not a large difference,⁹  we can use the outflow velocity measured from the absorption minima to identify potential candidates. Nine of the Of/WN stars in Paper II have P Cygni profiles. Three have measured outflow velocities that are consistent with the LBVs: M33C-21386 (−221 km s⁻¹), M33C-15235 (−233 km s⁻¹), and possibly M33-013432.76 (−263 km s⁻¹). In this paper, 10 Of/WN stars have a P Cygni profile, four of which have low outflow velocities: M33C-10788 (−255 km s⁻¹), M33-013334.06 (−224 km s⁻¹), M33C-16364 (−246 km s⁻¹), and maybe M33C-5916 (−278 km s⁻¹), all measured from the He i λ5876 line. Massey et al. (2016) included three stars with Of/late-WN spectra as LBV candidates: J004341.84+411112.0 has wind velocities of −337 and −323 km s⁻¹ from the H and He i lines, J013332.64+304127.2 (M33C-15742) has no P Cyg profiles, and J013351.46+304057.0 is M33C-15235 above. Thus of the 27 Of/late-WN stars for which we have spectra, seven have relatively low outflow velocities and may be considered LBV candidates.

The PSO stars. We also obtained spectra of four LBV candidates in M31 proposed by Lee et al. (2014) based on variability in the Pan-STARRS survey. All four showed variability in the visual on the order of 0.4–0.5 mag and had $H-K$ colors ≤0.5 mag like the confirmed LBVs; see Figure 6(a) in this paper. Two of the stars were fainter than 19 mag and consequently have poor S/N spectra that did not permit a spectral classification, although the red spectrum of PSO-J10.8180+41.6265 shows molecular bands and thus has a blended spectrum. The other two (PSO-J11.0457+41.5548 and PSO-J11.2574+42.0498) have absorption-line spectra typical of B- and A-type supergiants with H and He i emission lines, but no P Cyg profiles and no Fe ii emission lines. The red spectrum of PSO-J11.2574+42.0498 also has molecular bands in the red, and the red spectrum of PSO-J11.0457+41.5548 has [O i] emission, which we attribute to the strong nebular contamination in its red spectrum. Despite their small $H-K$ colors, both PSO-J10.8180+41.6265 and PSO-J11.2574+42.0498 show mid-infrared excess emission in the IRAC bands probably due to their likely M star companions. At this time there is insufficient information to include them as LBV candidates.

The numerous luminous stars with emission lines in the upper HR diagram share spectroscopic properties that at first glance may make them difficult to separate. In this paper we have taken a closer look at their spectra and SEDs, and we conclude that it is relatively straightforward to identify and separate the LBVs, the B[e] supergiants, and the warm hypergiants:

• 1.   
  B[e] supergiants have emission lines of [O i] and [Fe ii], and some also have [Ca ii] in their spectra. Most of the spectroscopically confirmed sgB[e] stars also have warm circumstellar dust in their SEDs.
• 2.   
  Confirmed LBVs do not have the [O i] emission lines in their spectra. Some LBVs have [Fe ii] emission lines, but not all. Their SEDs show free–free emission in the near-infrared but no warm dust. Their most important and defining characteristic is the S Dor-type variability.
• 3.   
  The warm hypergiants spectroscopically resemble the LBVs in their eruption or dense wind state and the B[e] supergiants. However, they are very dusty. Some have [Fe ii] and [O i] emission in their spectra like the sgB[e] stars, but can be distinguished from them by their absorption-line spectra characteristic of A- and F-type supergiants. In contrast, the B[e] supergiants have strong continua and few if any apparent absorption lines.

Candidate LBVs should share the spectral characteristics of the confirmed LBVs plus a lack of warm circumstellar dust. These include P Cygni profiles in the H or He i lines for those with spectra like the hot supergiants. Fe ii emission is present, but not [O i]. [Fe ii] emission may also be present. Some confirmed LBVs resemble the Of/late-WN stars, and although there is no clear-cut spectral difference, a low outflow velocity, ≤300 km s⁻¹, will help to identify potential candidates among the Of/late-WN stars. Based on these criteria, the Fe ii emission line stars (no [O i] and no dust; see Section 4, Table 5) should be considered LBV candidates. Table 7 is a list of candidate LBVs in M31 and M33 in our data based on the guidelines and discussion presented here.

The spectroscopic differences and other characteristics, such as the presence or lack of circumstellar dust, may be clues to other parameters that are more fundamental to the astrophysics of these stars, such as rotation, binarity, and evolutionary state. For example, the warm hypergiants are likely post-RSGs. They may transition to B[e] supergiants or even to the less-luminous LBVs once they shed their dust. And as is often the case in astronomy, there will be some stars that do not quite fit our description or may be transition objects, such as B526NE.

The fifth and final paper in this series will be a discussion of the upper HR diagrams for M31 and M33. We will determine the luminosities of these different emission-line stars to place them on the HR diagram for comparison with the massive star populations and a discussion of their possible evolutionary state.

We thank the anonymous referee for the very positive and constructive review. Research by R. Humphreys on massive stars is supported by the National Science Foundation AST-1109394. J. C. Martin's collaborative work on luminous variables is supported by the National Science Foundation grant AST-1108890. In addition to our own data observed with the MMT/Hectospec, we use spectra obtained from the CFA Optical/Infrared Science Archive. This paper uses data from the MODS1 spectrograph built with funding from NSF grant AST-9987045 and the NSF Telescope System Instrumentation Program (TSIP), with additional funds from the Ohio Board of Regents and the Ohio State University Office of Research. This publication also makes use of data products from the Wide-field Infrared Survey Explorer, which is a joint project of the University of California, Los Angeles, and the Jet Propulsion Laboratory/California Institute of Technology, funded by the National Aeronautics and Space Administration.

Facilities: MMT/Hectospec, LBT/MODS1.