LUMINOUS AND VARIABLE STARS IN M31 AND M33. III. THE YELLOW AND RED SUPERGIANTS AND POST-RED SUPERGIANT EVOLUTION^*

Our targets were primarily selected from the published surveys of M31 and M33 for yellow and red supergiants (Drout et al. 2009, 2012; Massey et al. 2009). Their red and yellow candidates were all chosen from the Local Group Galaxies Survey (LGGS; Massey et al. 2007b). Although their adopted magnitude limit and color range for the YSGs (V < 18.5 and $0.4\leqslant {\text{}}B\quad -\quad {\text{}}V\leqslant 1.4$) correspond to those of F- and G-supergiants, the same color range will include a large fraction of foreground contamination from Galactic dwarfs and halo giants (Massey et al. 2006; Drout et al. 2012).

To establish membership for the yellow candidates, the M31 survey of Drout et al. (2009) relied on radial-velocity measurements. However, as their Figure 10 illustrates, even restricting the candidates to the velocity range expected for M31 includes substantial foreground contamination. Drout et al. (2009) therefore used a relative velocity—the measured velocity compared to the expected velocity of the star at its position in M31—to establish probable membership. They identified 54 rank-1 (highly probable) and 66 rank-2 (likely) YSGs in M31 from a sample of 2901 targets. The 96% foreground contamination clearly demonstrates the difficulties of color and magnitude criteria for determining membership.

For the candidate YSGs in M33 (Drout et al. 2012), the authors again relied on the relative radial velocities, but also added a measurement of the strong luminosity-sensitive O i λ7774 blend in A- and F-type supergiants, which greatly increased the probability that the stars were supergiant members. With these criteria, they identified 121 rank-1 YSGs and 14 rank-2 in M33.

Fortunately for the RSG candidates, the two-color B − V versus V − R diagram has been demonstrated to be an effective metric for distinguishing red dwarfs and supergiants (Massey et al. 2009; Drout et al. 2012), from which the authors identify 437 RSG candidates in M31 and 408 in M33. For the M31 candidates, 124 had additional radial-velocity information for membership determination, and 16 were spectroscopically confirmed as M-type supergiants. For M33, the 408 candidate RSGs from Drout et al. (2012) were reduced to 204 (189 rank-1, 15 rank-2) likely RSGs using radial-velocity criteria.

In addition to the 120 and 135 YSG candidates from the Drout/Massey catalogs of M31 and M33, respectively, we include 18 confirmed YSGs from Humphreys et al. (2014, hereafter Paper II), seven warm hypergiants from Paper I, 39 Hα emission stars with intermediate colors from the survey by Valeev et al. (2010), and seven Hα emission sources from an unpublished survey by K. Weis (see Paper II). With these catalogs, we assembled a final target list of 124 and 165 candidate YSGs (after cross-identification among the listed works) for spectroscopy from M31 and M33. We did not obtain follow-up spectroscopy of the RSG candidates; our discussion of them instead relies on published photometry and analysis of their SEDs for CS dust (Section 4).

Our spectra of the YSG candidates were obtained with the Hectospec Multi-Object Spectrograph (Fabricant et al. 1998, 2005) on the MMT at Mount Hopkins over several observing sessions in 2013, 2014, and 2015. The Hectospec has a 1° field of view with 300 fibers each subtending 15 on the sky. We used the 600 line mm⁻¹ grating with a tilt of 4800 Å, yielding ≈2500 Å coverage with 0.54 Å pixel⁻¹ resolution and R of ∼2000. The same grating with a tilt of 6800 Å was used for the red spectra with similar coverage and resolution and R of ∼3600. The total integration time for each field was 90 minutes for the red and 120 minutes for the blue. The journal of observations is in Table 1.

Due to the large angular size of M31 on the sky, observations of this galaxy were split across two fields, labeled A and B in Table 1, centered at 00:43:36.5 + 41:32:54.6 and 00:41:15.9 + 40:40:31.2, respectively. Weather conditions at Mount Hopkins during the 2015 season prevented observations on the last set of supergiant candidates in the red filter setting.

The spectra were reduced using an exportable version of the CfA/SAO SPECROAD package for Hectospec data.¹ The spectra were bias-subtracted, flat-fielded, and wavelength-calibrated. Due to crowding, sky subtraction was performed using preselected sky fibers off the field of each galaxy. These sky fiber positions were chosen from Hα maps in regions where nebular contamination would be minimized. Flux calibration was done in IRAF using standard stars Feige 34 and 66 from Hectospec observations during the 2013–2015 seasons.

In M31, we obtained spectra for 113 of the 120 YSG candidates from Drout et al. (2009) plus follow-up spectra for the 10 previously confirmed warm supergiant and hypergiant stars from Papers I and II (six of which were cross-listed in the Drout catalog) for a total of 117 spectra. In M33, 71 of the 135 YSG candidates from Drout et al. (2012) were observed, plus 14 confirmed supergiants from Papers I and II (four cross-listed in the Drout catalog), and 37 Hα-emission sources from Valeev et al. (2010) (15 cross-listed in the Drout and Humphreys catalogs) for a total of 103 spectra. For all of these sources, as well as the remaining YSG candidates from the Drout catalogs for which we did not observe spectra, we obtain photometry from published catalogs, discussed in Section 3.2.

Our primary goal for the spectroscopy of the YSG candidates is to search for evidence of mass loss and winds from emission lines and P Cygni profiles when present. Since foreground contamination is an obstacle for identifying yellow and RSG members in external galaxies, the same spectra can be used for spectral and luminosity classification. We refine the classification of the YSG candidates with established luminosity and spectral type indicators in the blue and red spectra.

The blends of Ti ii and Fe ii at λλ4172–4178 and λλ4395–4400 are strong luminosity criteria in the blue when compared against Fe i lines that show little luminosity sensitivity such as λ4046 and λ4271. The O iλ7774 triplet in the red spectra is also a particularly strong luminosity indicator in A- to F-type supergiants. This feature is present in all of the candidate YSGs from M33, since Drout et al. (2012) identified all probable members based on this criterion.

The Sr iiλλ4077, 4216 lines are especially useful for temperature classifications of the YSGs. Comparing the relative strength of Sr iiλ4077 to the nearby Hδ feature, and the relative strengths of the Fe iiλ4233 and Ca iλ4226 lines, for example, provides a quick diagnostic for all F-type supergiants.

Later type supergiants (G-type) are identified by the growth of the G-band, a wide absorption feature around λ4300 Å due to CH. Luminosity criteria for G-type stars are similar to those in F supergiants. The Mg i triplet λλ5167, 5172, 5183 is strong in the later type dwarfs and allows for filtering foreground contaminants from our sample.

In M31, we identify 75 YSGs, including the previously confirmed stars from Papers I and II. Seventy of the 113 observed stars from Drout et al. (2009) are confirmed YSGs, and 42 are foreground dwarfs or subgiants. Therefore, the M31 catalog of Drout et al. (2009) was ∼35% contaminated by foreground stars. The remaining eight rank-1/rank-2 candidates of Drout et al. (2009) for which we did not obtain spectra are analyzed in Section 3.3 for evidence of mass loss in their SEDs along with the confirmed YSGs. We identify 86 YSGs in M33, which also includes the warm supergiants and hypergiants discussed in Papers I and II. Sixty-two of the 71 observed candidates from Drout et al. (2012) are spectroscopically confirmed as YSGs. The remaining nine observed sources were identified as foreground dwarfs. Thus, the M33 catalog was only ∼7% contaminated by foreground stars. Since their M33 survey used the luminosity-sensitive O iλ7774 line in addition to relative velocities, the cleaner sample is not surprising.

Twelve YSGs in M31—including the hypergiants M31-004322.50, M31-004444.52, M31-004522.58, and hypergiant candidate J004621.08+421308.2 (see Section 3.4)—and 18 in M33 (including hypergiants B324, Var A, N093351, and N125093) exhibit spectroscopic evidence for stellar winds. The notable spectral features include P Cygni profiles in the hydrogen emission lines, broad wings in Hα or Hβ emission indicative of Thomson scattering, and [Ca ii]/Ca ii triplet emission. Example spectra are shown in Figure 1, highlighting two stars with Hα emission indicative of stellar winds and CS outflows.

Zoom In Zoom Out Reset image size

We find that approximately 17% of the observed YSGs in M31 and 21% in M33 demonstrate evidence for mass loss in their spectra. We discuss the evidence for CS dust ejecta in their SEDs in Section 3.3. Representative A- and F-type supergiants from both galaxies are illustrated in Figure 2.

Zoom In Zoom Out Reset image size

Table 2 is a list of the confirmed YSGs in both galaxies, 75 in M31 and 86 in M33, with their spectral types. Notes to the table include comments on the evidence for winds and mass loss in their spectra and references to cross-identified objects. The rank is included for stars from the Drout surveys. Table 9 lists all of the foreground stars—confirmed dwarfs and subgiant stars.²

Reduced spectra for the confirmed YSGs and foreground stars observed from 2013 to 2015 can be found at http://etacar.umn.edu/LuminousStars/M31M33/.

For each source in our target list, we cross-identify the visual photometry from the LGGS (Massey et al. 2006) with the near- and mid-infrared photomery from 2MASS (Skrutskie et al. 2006) at J, H, and K_s, the Spitzer/IRAC surveys of M31 (Mould et al. 2008) and M33 (McQuinn et al. 2007; Thompson et al. 2009) at 3.6, 4.5, 5.8, and 8 μm, and WISE (Wright et al. 2010) at 3.4 μm (W1), 4.6 μm (W2), 12 μm (W3), and 22 μm (W4). For cross-identification between the 2MASS and IRAC coordinates, we use a search radius of 05.

The WISE satellite has angular resolutions of 61, 64, 65, and 120 in the four bands, which presents some issues for cross-identification in the crowded M31 and M33 fields. We selected a 6'' search radius for matching the LGGS/2MASS coordinates to WISE, which is consistent with the FWHM of the WISE point-spread function at 3.4 μm (Wright et al. 2010). Since the longer-wavelength photometry has such a large beamsize, we recognize that some of our matched candidates may contain multiple sources or be contaminated by polycyclic aromatic hydrocarbon (PAH) emission. To mitigate this, the prime candidates for CS dust (as characterized by infrared excess in the WISE bands, Section 3.3) were each checked visually in the 2MASS K_s-band images, and the photometry was rejected if the sources were likely composites. The resulting multi-wavelength photometry for the spectroscopically confirmed YSGs, as well as the YSG candidates, is summarized in Table 3.

Mould et al. (2008) and McQuinn et al. (2007) additionally provide catalogs of infrared variable sources in M31 and M33, respectively. We checked each source for variability against those catalogs, as well as the DIRECT survey (Kaluzny et al. 1998; Macri et al. 2001), and find low-amplitude fluctuations (<0.1 mag) in the IRAC bands, most likely associated with Alpha Cygni variability. For M33 sources, we also checked against the optical survey of Hartman et al. (2006) and find similarly low-level flux variability in g', r', and i' band observations. YSGs and YSG candidates that show variability in either the optical or infrared are indicated in Table 3.

To determine whether the YSGs have excess free–free emission in the near-infrared (1–2 μm) due to stellar winds and/or an excess at longer wavelengths due to CS dust, we must first correct the SEDs for interstellar extinction. Many of these targets are likely embedded in their own CS ejecta or warm CS dust. Additionally, we have noticed in our previous work that the extinction can vary considerably across the face of these galaxies, especially in M31. Drout et al. (2009) assumed a fixed $E(B-V)=0.13$ reddening law for all YSGs in M31, and Drout et al. (2012) similarly adopted $E(B-V)=0.12$ for sources in M33. We instead proceed more conservatively and calculate the extinction for each source individually.

For those stars with spectral types, we compare the observed B − V color to the intrinsic colors of supergiants from Flower (1977) and calculate A_V from the standard extinction curves (Cardelli et al. 1989) with R = 3.2. This procedure is uncertain for stars with strong emission lines in their spectra, so we also estimate the visual extinction using two other methods: the reddening-free Q-method (Hiltner & Johnson 1956; Johnson 1958) for nearby OB-type stars in the LGGS within 2–3'' of each target, assuming that their UBV colors are normal, and the relation between the neutral hydrogen column density (N_H) and the color excess, ${E}_{B-V}$ (Savage & Jenkins 1972; Knapp et al. 1973).³

We measure N_H from the recent H i surveys of M31 (Braun et al. 2009) and M33 (Gratier et al. 2010). Since we do not know the exact location of the stars along the line of sight with respect to the neutral hydrogen, we follow Paper II and define the total A_V as the foreground A_V (≈0.3 mag) plus half of the measured N_H. Since the H i surveys have spatial resolutions of 30'' and 17'' for M31 and M33, respectively, we favor the extinction estimates from the two other methods when available. We use the Q-method and N_H measurements for the supergiant candidates for which we did not obtain spectra. The results from these different methods, the adopted A_V, and the resulting extinction-corrected M_V are summarized in Table 4.

For the spectroscopically confirmed YSGs with known spectral types, we calculate the bolometric luminosities by applying bolometric corrections from Flower (1996) to M_V. Bolometric corrections for stars in this temperature range are small, typically $\leqslant 0.2$ mag. For the candidate supergiants without spectra, we integrate the SED from the optical to the 2MASS K_s band (2.2 μm). If the SEDs show an infrared excess, and thus evidence of CS dust, we integrate the SED out to the IRAC 8 μm band and/or the 22 μm WISE band if available and if not obviously contaminated by nebulosity in the beam. Those sources are indicated with an asterisk in Table 4.

Figures 3 and 4 are example SEDs from supergiants in M31 and M33. The observed visual, 2MASS, and IRAC magnitudes are shown as filled circles, and the WISE data as open circles. The optical and 2MASS extinction-corrected photometry is shown as gray squares. We fit a blackbody curve to the extinction-corrected optical photometry to model the contribution from the central star. If the flux in the near-infrared 2MASS, and IRAC bands exceeds the expected Rayleigh–Jeans tail of the stellar component, we identify this as an infrared excess. Many of the supergiants in our sample show the characteristic upturn redward of 8 μm due to PAH emission (Draine & Li 2007). Due to the large beamsize of WISE, it is likely that some sources are contaminated by PAH emission from H ii regions in the mid-infrared. However, if a source also has an apparent excess in the near-infrared 2MASS and IRAC bands, the infrared photometry provides evidence of mass loss. An infrared excess in the 1–2 μm 2MASS bands is characteristic of free–free emission in stellar winds, with the 3.6–8 μm IRAC data providing evidence for warm CS dust. Free–free emission is generally identified as constant F_ν in the near-infrared, often extending out to 5 μm (see Figure 6). The IRAC photometry can be used to estimate the mass of the dusty CS material (see Section 5.1). We note that the data provided in the broadband visual (LGGS), near-infrared (2MASS), and mid-infrared (IRAC and WISE) photometry were not all observed simultaneously. The resulting SEDs, then, do not represent a single snapshot in time.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Twenty-six YSG candidates have indicators for free–free emission in the near-IR 2MASS photometry and/or CS dust emission in the mid-IR IRAC or WISE bands. D-004009.13, as its SED in Figure 3 shows, likely has both nebular contamination and dust. Combining both the spectroscopic and photometric data, we find a total of 32 sources in M31 with evidence for mass loss either from the stellar wind features in their spectra or free–free/CS dust emission in their SEDs. Six show evidence for both: the warm hypergiants M31-004322.50, M31-004444.52, M31-004522.58, the new hypergiant J004621.08+421308.2 (see Section 3.4), and two F-type supergiants, M31-004424.21 and M31-004518.76. We do not have spectra for two of the sources with evidence of free–free emission, D-003745.26⁴ and D-003936.96, so we cannot confirm membership in M31. Therefore, of the 75 confirmed YSGs in M31, 30 (or 40%) are likely post-RSG candidates, plus two sources that require follow-up spectroscopy to confirm supergiant status. Five sources (D-003711.98, D-003725.57, D-003907.59, D-004102.78, D-004118.69) are likely contaminated with nebular PAH emission from nearby H ii regions.

In M33, 22 stars show evidence for free–free emission and/or CS dust emission in their SEDs. Combining the spectroscopic and photometric data indicators, we find a total of 30 sources in M33 with evidence for mass loss. Nine have both the spectroscopic stellar wind and CS dust features: the warm hypergiants Var A, M33-013442.14, N093351, N125093, and the supergiants V002627, D-013233.85, V021266, V130270, V104958. Three of the sources have not been spectroscopically confirmed as members of M33 (D-013345.50, D-013349.85, D-013358.05).⁵ Therefore, of the 86 confirmed YSGs in M33, we identify 27 (or ∼31%) as likely post-RSG candidates, plus three sources requiring follow-up spectroscopy. Thirty-three sources show evidence for nebular PAH contamination.

In Table 5 we list all the stars, both confirmed and unconfirmed YSGs, that show at least one indicator of CS ejecta: spectroscopic evidence of a stellar wind, free–free emission, and/or thermal dust emission in the SED. If the stellar spectrum contains nebular emission markers such as [O iii], we indicate in the table that the source is likely contaminated with nebular emission. Estimates of total mass lost are discussed in Section 5.1.

J004621.05+421308.06 was considered a candidate LBV by Massey et al. (2007a) and King et al. (1998). Due to its position in the northern arm of M31, it was outside both of our two Hectospec fields for M31. Consequently, we obtained a long-slit spectrum of this target on 2014 November 22 with the MODS1 spectrograph on the Large Binocular Telescope. The instrument setup and observing procedure are described in Paper I. Its blue and red spectra are shown in Figure 5. J004621.05+421308.06 has the absorption-line spectrum of a late A-type supergiant with strong Balmer emission lines with deep P Cygni profiles and the broad wings characteristic of Thomson scattering. The outflow velocity is −223 km s⁻¹, measured from three P Cygni absorption minima. Numerous Fe ii and [Fe ii] emission lines are present. Like the warm hypergiants discussed in Paper I, its red spectrum shows the Ca ii and [Ca ii] line emission indicative of a low-density CS nebula. The O iλ8846 line is also in emission. The SED shown in Figure 6 reveals a prominent CS dust envelope in the near- and mid-infrared not observed in the LBVs (Paper II). These properties are shared with the warm hypergiants and probable post-RSGs experiencing high mass loss. Their photospheres are not due to the cool dense winds formed by an LBV in eruption, but represent the stellar surface. We therefore suggest that J004621.05+421308.06 belongs with the class of warm hypergiants.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

J004051.59+403303.00 has been described by Massey et al. (2007a) and by Sholukhova et al. (2015) as a candidate LBV. Our blue and red spectra from 2013 shown in Figure 5 are similar to the blue spectrum published by Massey et al. (2006) and the blue and red spectra in Sholukhova et al. (2015), suggesting little spectroscopic change in the last 10 years. The spectra show prominent P Cygni profiles in the Balmer lines with broad wings and in the Fe ii multiplet 42 lines. The mean outflow velocity measured from the absorption minimum in six P Cygni profiles is −152 km s⁻¹, similar to the LBVs and hypergiants (Papers I and II). There are no other Fe ii or [Fe ii] emission lines. The relative strengths of the Mg ii λ4481 and He i λ4471 lines suggest an early A-type supergiant. Its spectrum is similar to the warm hypergiant M31-004444.52 in Paper I, but it does not have the [Ca ii] emission lines in the red. It also resembles J004526.62+415006.3 in 2010, which was later shown to be an LBV entering its maximum-light or dense-wind stage (Humphreys et al. 2015; Sholukhova et al. 2015). Therefore, based on this spectrum, its nature is somewhat ambiguous. Its SED in Figure 6 shows an excess in the near-infrared due to free–free emission, as evidenced by constant flux (F_ν) out to 5 μm. The WISE photometry at 12 and 22 μm may be due to CS dust from silicate emission, but is more likely contaminated by PAH emission from a nearby H ii region and nebulosity.

Thus, J004051.59+403303.00 may be a mass-losing post-RSG like several of the stars discussed in this paper or a candidate LBV. Future spectroscopic and photometric variability will be necessary to confirm that it is an LBV, but even so, given its luminosity, it is very likely in a post-RSG state similar to the less-luminous LBVs.

Thermal emission from dust appears in the mid-infrared IRAC data from 3.6 to 8 μm and is also present in the WISE photometry at longer wavelengths. With some assumptions about the dust grain parameters, we can estimate the mass of the CS material from the mid-infrared flux (see Paper I):

where D is the distance to the source (here, the average distance to M31/M33), F_λ is the mid-infrared flux, a is the grain radius, ρ is the grain density, Q_λ is the absorption efficiency factor for silicate dust grains, and ${B}_{\lambda }(T)$ is the blackbody emission at temperature T.

For many of the YSG and RSG candidates with an infrared excess, the flux is fairly constant across the mid-infrared, which implies that the dust is emitting over a range of temperatures and distances around the central star. Using the prescription of Suh (1999) for silicate dust grains and mass loss around AGB stars, we assume a dust grain size of a = 0.1 μm at a density ρ = 3 g cm⁻³ and an average temperature of 350 K. The absorption (and emission) efficiency factor Q_λ is maximized at the 9.8 μm Si–O vibrational mode (Woolf & Ney 1969) and the 18 μm O–Si–O bending mode (Treffers & Cohen 1974), so the fluxes at these wavelengths would be the ideal tracers of thermal dust emission. Since we lack photometry precisely centered on the silicate features, we calculate the dust mass using the flux at 8 μm, or at 12 μm (W3) if no IRAC data exist for the sources. We assume a nominal gas-to-dust ratio of 100, which allows for the calculation of the total mass lost in each source.

The results are summarized in Table 5 for the YSGs and Table 8 for the RSGs, where the error is calculated as the standard error propagation on the average distance to M31/M33 (<5%) and the photometric errors for measured flux by IRAC/WISE (<3%/<9%, Hora et al. 2004; Wright et al. 2010). For both YSGs and RSGs in M31 and M33, we find a range of at least a factor of 10 for the mass of the CS material. Most of the supergiants have shed ~10^–3–${10}^{-2}\;{M}_{\odot }$, which is consistent with Paper I. Mauron & Josselin (2011) apply the mass-loss prescription of de Jager et al. (1988) to Galactic RSGs to calculate an average mass-loss rate of ∼10⁻⁶ M_⊙ yr⁻¹ from IRAS 60 μm flux. For our dusty RSGs, we can approximate a timescale probed by the IRAC photometry and thus compare our total integrated mass loss to typical RSG mass-loss rates. If we assume an average dust condensation distance of ∼250 AU and an outflow velocity of 20 km s⁻¹, we estimate ∼100 years for the dust condensation time—a rough timescale for the dust we observe at 8 μm. Considering that the CS ejecta most likely contains dust over a range of temperatures (∼150–400 K), as well as the possibility of episodic mass loss in the more massive RSGs, an average mass-loss rate of 10^–5–${10}^{-4}\;{M}_{\odot }\;{\mathrm{yr}}^{-1}$ is consistent with the total mass-lost estimates of 10^–3–${10}^{-2}\;{M}_{\odot }$ over the dust condensation timescale.

For the RSG populations in both galaxies, we plot bolometric luminosity versus total mass lost in Figure 9. The formulation of de Jager et al. (1988) predicts an increasing mass-loss rate with luminosity, and we find a similar trend with total mass lost as traced by dust. We separate the rank-1 RSGs—those candidates with a clear indication of mass loss in their SEDs—from the rank-2 RSG candidates. The rank-2 sources yield ejecta masses at the lower end of the RSG sample. These RSG candidates likely have CS dust, but the infrared excess was not as obvious as in the rank-1 SEDs, thus the derived mass-loss estimate is lower. Since the RSGs with the highest luminosity also have the highest mass loss, these dusty RSGs may evolve back to higher temperatures to become the intermediate-type post-RSGs discussed in this paper.

Zoom In Zoom Out Reset image size

The HRDs for the populations of yellow and red supergiants are shown in Figures 10 and 11. The temperatures for the YSGs are derived from the ${(B-V)}_{0}$ colors using the transformations in Flower (1996) for intermediate-type supergiants. As described in Section 3.3, their luminosities are calculated based on the bolometric corrections given in Flower (1996) or by integrating the SED for those stars with emission-line spectra or with CS dust.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

There are several temperature scales in the literature for RSGs. Since we do not have spectral types for the RSGs, we adopt the temperatures from Massey et al. (2009) for the M31 RSGs and from Drout et al. (2012) for M33 simply for the purpose of placing them on the HRD to compare with the YSG population. Their M31 temperature scale is based on a color–temperature relationship from ${(V-K)}_{0}$ from MARCS atmosphere models (Gustafsson et al. 2008), while their temperatures for the M33 stars are based on ${(V-R)}_{0}$ color transformations from Levesque et al. (2006) in the LMC, which has a metallicity similar to M33. The bolometric luminosities for the RSGs are determined from integrating their SEDs (Section 4).

Stellar evolution tracks from non-rotating models from Ekström et al. (2012) are shown on the HRDs for zero-age main-sequence masses of 15, 25, and 40 M_⊙. In both galaxies, the post-RSG candidates are preferentially more abundant at higher luminosities. Comparison with the evolutionary tracks suggests that most of the progenitor main-sequence stars have masses ≳20 M_⊙. Likewise, the dusty RSGs dominate the higher luminosities. This is most obvious for the M33 population with a smaller sample. This is not surprising, because we know from Figure 9 that the mass lost in the RSGs correlates with luminosity.

We note the presence of several "warm" RSGs in both galaxies. These RSG candidates, with temperatures upwards of 4000 K, fall in the temperature range of the YSGs. The temperature scales are somewhat uncertain, and without spectra of these objects we cannot confirm that some of them may actually be YSGs.

Labeled sources in Figures 10 and 11 are the warm hypergiants from Paper I, as well as the two new hypergiant candidates, J004051.59+403303.00 and J004621.05+421308.06, discussed in Section 3.4.