The Complex Upper HR Diagram

Several decades of observations of the most massive and most luminous stars have revealed a complex upper HR Diagram, shaped by mass loss, and inhabited by a variety of evolved stars exhibiting the consequences of their mass loss histories. This review presents a brief historical overview of the HR Diagram for massive stars, highlighting some of the primary discoveries and results from their observation in nearby galaxies. The chapters in this volume include reviews of our current understanding of different groups of evolved massive stars, all losing mass and in different stages of their evolution; the Luminous Blue Variables (LBVs), B[e] supergiants, the warm hypergiants, Wolf-Rayet stars, and the population of OB stars and supergiants in the Magellanic Clouds.


Introduction
The reviews and papers in this volume focus on the properties of the most luminous stars in nearby galaxies, those galaxies in which the brightest individual stars are resolved and can be observed. The most luminous stars are also the most massive and because of their intrinsic brightness and relatively short lifetimes they provide our first probes of the progress of stellar evolution in different environments.
The study of massive stars in other galaxies offers many advantages. Foremost of course is distance. Studies of stellar populations in nearby galaxies have the advantage that all the stars are at approximately the same distance, a distance that is relatively well known, especially in comparison with the uncertain distances of individual stars in our own galaxy. In the Milky Way, our observations are also limited to a relatively small volume by interstellar extinction which can be high and uncertain at increasing distances. In external galaxies, extinction by dust is still a problem but the foreground extinction is well determined from maps of the interstellar "cirrus" or dust along the line of sight. Internal extinction within the galaxy can be variable and must still be corrected.
In this introductory chapter, I present a brief historical overview with emphasis on some of the main developments in the study of massive stars in nearby galaxies. Some of the first work on stars in other galaxies was driven by the identification of Cepheids for the extragalactic distance scale, see for example, the comprehensive survey of NGC 2403 by Tammann & Sandage [1]. Other types of variables were also recognized including the first discussion of a class of luminous variables by Hubble & Sandage [2] in the Local Group spirals M31 and M33 that we now call Luminous Blue Variables (LBVs). This work was done on photographic plates and the magnitudes of individual stars were often measured by hand.
The study of individual stars and their placement on the HR Diagram requires several types of data, accurate multicolor photometry for spectral energy distributions (SEDs) and the measurment of interstellar extinction, infrared observations of circumstellar dust and mass loss but most important, spectroscopy, for classification, temperatures, luminosity indicators, emission lines, and evidence for mass loss from line profiles. And when observing stars in other galaxies, we also have to be concerned about foreground contamination; stars in the Milky Way seen projected against the distant galaxy. Photometry is not sufficient to remove these Galactic stars, especially those of intermediate temperature and late spectral type; spectra are required.
The primary galaxies for the chapters in this volume are the Local Group spirals M31 and M33, and the Large and Small Magellanic Clouds due to their relative proximity and the number of surveys of their luminous stellar populations. Other Local Group galaxies are also included as well as results for stars in a few nearby spiral galaxies.

Early Work -The Magellanic Clouds
The early objective prism surveys by the Harvard College Observatory and the resulting Henry Draper Catalog and Extension (HD and HDE) provided the first survey with spectral types of the brightest stars in the Magellanic Clouds [3,4]. The HDE however did not cover the Small Cloud. Feast, Thackeray & Wesselink [5] published the first detailed list of the brightest stars in the Magellanic Clouds with spectral classifications based on slit spectra of individual stars. Their paper included magnitudes, colors, positions, radial velocities and notes on the individual stars relative to emission lines and other features for 50 stars in the SMC and 105 in the LMC. The HDE catalog was the primary source for the LMC stars, plus stars selected as blue based on their colors in the 30 Dor region. The SMC list relied on "B" type stars from the HD plus emission line stars from the survey by Henize [6]. Their work did not include the red or M-type supergiants. The reddest stars in their study were the very luminous F and G-type supergiants. Their HR Diagram, reproduced here in Figure 1, while not complete, shows the visually brightest stars with M v approaching -10 mag.
Additional work on the fainter stellar populations of hot or OB-type stars and red M-type stars was provided by the lower resolution objective prism surveys throughout the 1960's and 1970's. These included surveys for OB type stars in the LMC and SMC by Sanduleak [7,8] and by Fehrenbach & Duflot [9] and for the M type stars by Westerlund [10,11], Blanco, McCarthy & Blanco [12] and Blanco & McCarthy [13]. These necessary photographic surveys provided extensive finding lists for further spectral classification leading to the first comprehensive HR diagrams for these two galaxies and the comparison of their massive and luminous stellar populations with the Milky Way.

The HR Diagram
Fundamental data for the most luminous stars have application for numerous astrophysical questions specifically with respect to stellar evolution, the final stages of the most massive stars as the progenitors of supernovae, and the dependence of their basic parameters on the host galaxy. In the 1970's there was also considerable interest in the luminosity calibration of the brightest stars and their potential as extragalactic distance indicators [14].
The massive star population in the Milky Way, although restricted to a relatively small volume, within ≈ 3 kpc of the Sun, is critical as a reference population and as representative of our local region of the Galaxy, despite uncertain distances and possible incompleteness. Humphreys [15] published an HR Diagram for the Galactic supergiants and O stars with spectral types and photometry in stellar associations and clusters with known distances to derive their luminosities. For a more complete population in the upper HR diagram, the less luminous early B-type main sequence and giant stars were also included. Compared to previous work, numerous surveys, especially of the Southern sky, had added greatly to the number of confirmed supergiants and O stars [16][17][18] and red supergiants [19,20]. For comparison with evolutionary tracks, the derived absolute visual magnitudes and spectral types were transformed to absolute bolometric luminosities and effective temperatures based on the available calibrations.
Humphreys [21] subsequently published HR Diagrams for the LMC. The basic data for the confirmed supergiants and early-type stars came from the extensive catalogs of spectral types and photometry by Ardeberg et al. [22] and Brunet et al. [23], from Feast, Thackeray & Wesselink [5], and from Walborn [24] for the early O-type stars. The data for the M-type supergiants came from the spectroscopic survey in the same paper.
An empirical comparison of these HR Diagrams, for the massive stars in our region of the Milky Way and for the LMC, revealed comparable populations of massive stars based on the distribution of their spectral types and luminosities across the HR Diagrams. The most important result was the recognition by Humphreys & Davidson [25] of an empirical upper luminosity boundary or upper limit in the luminosity vs temperture diagrams.

The Humphreys-Davidson Limit
The original HR diagrams from the Humphreys and Davidson 1979 paper are reproduced here in Figures 2 and 3. with the upper boundary shown as a solid line. The original eyeball fit was first drawn to approximate the upper boundary of the supergiant luminosities in the Milky Way. The same line was then transferred to the LMC diagram which also matched the observed upper envelope to the LMC luminosities. The upper boundary for both galaxies is an envelope of declining luminosity and decreasing temperature for the hottest stars and a relatively tight upper limit to the luminosities of the cooler stars (less than ≈ 10,000K) near M Bol ≈ -9.5 Mag.
This empirical boundary is often referred to in the literature as the "Humphreys-Davidson" Limit. It was not predicted by theory or the stellar structure models and evolutionary tracks at that time. The lack of evolved stars, post main sequence stars, above a certain luminosity implies an upper limit to the masses of stars that can evolve to become red supergiants thus altering the previously expected evolution of the  most massive stars across the HR Diagram. In the original study, this limit corresponded to an initial mass near 60 M ⊙ . Improved models with mass loss and rotation suggest a mass more like 40 -50 M ⊙ today.
Spectroscopy of luminous star candidates in other Local Group galaxies confirmed the upper luminosity boundary in galaxies of different types, but in most cases surveys and population studies were minimal; M31 [26], IC 1613 and NGC 6822 [27]. Humphreys & Sandage [28] completed a major survey for the brightest blue and red stars in M33, that provided the basis for spectroscopy and identification of the most luminous supergiants in that nearby spiral [29], and in subsequent studies. Similar HR Diagrams for the massive stars in the SMC were published a few years later by Humphreys [30] based on a combination of previous work and new observations.
At the time the Humphreys-Davidson paper was published, it was apparent that significant mass loss occurred in both the blue and red supergiants and there was increasing interest in the role that mass loss may have on their evolution. The lack of evolved cooler counterparts to the most massive evolved hot stars in both galaxies suggested that their post main sequence evolution was of special interest. A few high luminosity stars were known for their instabilities, variability and evidence for high mass loss such as eta Car and P Cyg in the Milky Way and S Dor in the LMC. A number of luminous blue variables, spectroscopically similar to eta Car and P Cyg, were now recognized in other galaxies [31,32]. As a group, they were known as S Doradus Variables or as Hubble-Sandage Variables in M31 and M33. Several of these stars were included on the HR Diagrams. They were increasing evidence that the phenomena of high mass loss and instabilities, as observed in eta Car and P Cyg, were more common. Humphreys and Davidson thus suggested that the most massive hot stars could not evolve to cooler temperatures because of their instabilities resulting in high mass loss. The temperature dependence of the luminosity limit for the hottest stars was evidence that the instability was mass dependent. This mass loss could be unsteady and much greater at times resulting in high mass loss events. The relatively tight upper luminosity boundary for the cooler stars represented the upper limit to the initial masses of stars that could evolve across the HR Diagram in a stable way to become red supergiants (RSGs).
Some of the first evolutionary tracks with mass loss were also being published at that time [33][34][35][36]. It was shown that high mass loss (higher than observed) would cause the tracks to reverse and the stars evolve back to warmer temperatures.

Surveys and More Surveys
We emphasize the importance of spectroscopy for the identification and analysis of the most luminous stars, but surveys are essential for identifying candidates. Beginning in the 1980's astronomers began to add multi-wavelength surveys, primarily in the near-infrared, to complement the traditional optical photometry. Elias, Frogel, & Humphreys [37] obtained near IR photometry for known and candidate red supergiants in the LMC and SMC for a comparison with the Galactic population. The all-sky near-infrared 2MASS survey from 1.2 to 2.2µm [38] reached the brightest stars in M31 and M33 as well as the Clouds. The infrared observations allowed astronomers to look for free-free emission from the stellar winds and the presence of circumstellar dust, another indicator of mass loss, and to correct the luminosites for possible additional extinction due to circumstellar dust.
Space-based telescopes such as UIT and later GALEX added FUV and NUV imaging and photometry for more complete SEDS at the shorter wavelengths and more accurate estimates of the total luminosities for the hottest supergiants. The mid-infrared surveys with Spitzer/IRAC of the Magellanic Clouds [39,40] and M31 [41] and M33 [42] added fluxes from 3 to 8µm and even longer wavelengths to search for colder dust thus allowing us to investigate their mass loss histories.
With the advent of the wide-field CCD mosaic cameras, groundbased, multi-wavelength optical surveys, such as the Local Group (LGGS) by Massey et al. [43], added to the fundamental data. These groundbased surveys however are seeing-limited and lack spatial resolution so the images are often multiple. Here again specroscopy or higher resolution imaging with Adaptive Optics on large telescopes can identify and even separate the stars. These groundbased surveys are enhanced by observations with the Hubble Space Telescope such as the PHAT surveys in M31 and M33 [44]. Numerous imaging programs of other galaxies intended for other purposes with HST have provided lists of resolved stars for further observation in, for example, M101 [45] and NGC 2403 and M81 [46,72].

Stellar Population Comparisons
One of the outstanding stellar evolution questions is how stellar populations may depend on their environment; the properties of the host galaxy, its mass and luminosity, the fraction of interstellar gas and dust, and especially on the chemical composition or metallicity. As the most luminous and visually brightest stars, massive stars provide the first indicator or measurement of how star formation, evolution, and the terminal state, depend on these factors. Mass loss is known to alter stellar evolution, and in the standard picture of line-driven winds from hot stars, mass loss is also expected to be metallicity dependent and significantly less in lower metallicity systems.
The first evidence for significant differences in the properties of the massive star populations was the well-established absolute magnitude dependence of the visually "brightest blue star" on the Galaxy type or luminosity in the surveys for extragalactic distance indicators [14]. The smaller, less massive galaxies had fewer of the most massive stars thus their evolved counterparts were statistically less likely to become the most luminous observed blue stars. These results suggested that the star formation rate for these most massive stars was less in the smaller galaxies. In contrast, the luminosities of the brightest red stars showed little or no dependence on the host galaxy. The stars of somewhat lower initial mass thus existed in sufficient numbers in the smaller galaxies to produce evolved descendents in the red supergiant region, and the luminosities of their brightest members reflected the upper luminosity boundary.
The metallicity dependence of the overall characteristics of the massive star poplulations in different galaxies has been less apparent. Comparison of the HR Diagrams of the massive stars in our region of the Milky Way with the LMC revealed very similar populations [25]. The LMC oxygen abundance is lower than Solar but by no more than a factor of two. Consequently, a comparison with the outer regions of the Milky Way, also with reduced metallicity, not surprisedly, showed little variation. Among our Local Group galaxies with comprehensive stellar surveys for luminous stars, the SMC has the lowest metallicity, about 1/10th Solar, and a reduced oxygen abundance by about a factor of five.
Early tell-tale evidence for an observable metallicity affect, was the distribution of the spectral types of the M supergiants [47]. A preliminary survey of red supergiants in the LMC and SMC compared to known Galactic RSGs, revealed a dramatic shift in their spectral types in the SMC to much earlier spectral types, compared to the LMC and Milky Way stars. Except for one M2 -type star in the SMC, all of the others were type M0 or earlier. This apparent shift, attributed to weaker TiO bands, was due to the lower SMC metallicity resulting in lower opacities in the atmospheres. The spectra thus arise in warmer layers. This result was confirmed in later surveys that extended to fainter magnitudes [37,48].
Stellar wind theory predicts a measurable dependence of the mass loss rate on metallicity [49], decreasing with declining heavy element abundances, but the measured rates [50] are somewhat higher than expected. Clumping in the stellar winds is another complication which when included in the mass loss models reduces the mass loss rates [51][52][53]. Measurement of the stellar wind properties and mass loss rates in the luminous, hot OB-type stars in nearby galaxies, especially in the Magellanic Clouds, has progressed with the advent of very large telescopes equipped with high resolution spectrographs. The chapter in this volume by Paul Crowther on the FLAMES survey the luminous, hot stars in the Clouds will discuss their winds and mass loss rates.

The Most Luminous Stars of Different Types
Since the early work of the 1970's and 80's to identify the most luminous and brightest stars in nearby galaxies, numerous surveys and studies of the massive stars, primarily in Local Group galaxies, have greatly expanded the completeness of the population samples. These include, in the LMC and SMC: surveys for the yellow and red supergiants [54,55], and in M31 and M33: studies of the luminous star population [56][57][58][59][60] and surveys for the yellow and red supergiants [61][62][63][64][65]. Surveys and follow-up spectroscopy of the luminous blue and red stars in the Local Group irregulars NGC 6822 and IC 1613 are less complete [27,[66][67][68], but they provide an additional sample of the massive stars in two smaller galaxies and also with reduced metallicity. Table 1 presents a summary of the most luminous stars of different spectral types or temperature ranges in six Local Group Galaxies with their morphological types and integrated visual luminosities. Initial samples of the massive stars in the well-studied nearby spirals, M101, M81 and NGC 2403, outside our Local Group, have also been observed [45,46,69,71,72] and are included here. The bolometric luminosities of the stars are listed for the three highest in each spectral type group for each galaxy. The adopted distance moduli ( Table 2) were used to determine the luminosities. The O-type stars are not included because many are eventually recognized as binary or are in multiple systems. Likewise, those stars in extremely crowded fields, including many OB-type stars, are not listed since they may be blended. Consequently, this Table does not include those stars that will be the intrinsically most luminous, most massive members of their home galaxies. A few stars of special interest are identified by name in the Table. Some of them will be discussed in later chapters in this volume. Note that the Galactic stars are not included. The upper HR Diagram for the Milky Way needs to re-examined when the Gaia survey is complete with improved distances and very likely with a larger volume sample.  RSGs -9.4(2) -9.6 -9.6(MOH-G64) -9.2 -9.2 -9.4 -8.5 -9.5: -9.9: -9.3 -9.5(2) -9.1 -9.1 -9.1 -8.8 -8.4(2) -9.2(2) -9.5 -9.1(2) -9.4(2) -9.0 -8. Some comments with respect to the data in this Table are helpful. It is clear that data for some of the spectral type groups is lacking or incomplete such as for the YSGs (FGK) in NGC 6822 and IC 1613. Those stars are undoubtedly present, but have just not been identified in the published surveys, and they may also be of somewhat lower luminosity in those two galaxies. A survey to identify and classify the hotter supergiants in M101 has not been completed, and although a survey identifying RSGs in NGC 2403 exists [71] confirming spectroscopy and photometry is lacking. In general, the numbers for the hot supergiant group (O9.5 -B5) may not include the most luminous members because many of these stars are in crowded regions.
The six Local Group galaxies in Table 1 present a diverse group of galaxy types with a wide range of luminosities. The well known dependence of the most luminous "blue" stars on the parent galaxy is especially notable for the evolved, post-main sequence A-type supergiants in the two lowest luminosity and lowest mass galaxies, NGC 6822 and IC 1613. Together with the SMC, these are also the galaxies with the lowest metallicity, but the heavy element abundance in NGC 6822 however, is intermediate between the SMC and LMC. Reduced metallicity was also expected to reduce the opacity in the stellar atmospheres and increase their absolute visual magnitudes but the data for these galaxies showed the opposite effect. So although metallicity undoubtedly plays a role in stellar evolution, the luminosity and mass of the parent galaxy is the primary determinant, to the first order, for the luminosities of the most luminous stars. This is a size of sample effect. Assuming a similar slope for the mass function, the most massive galaxies will have a larger progenitor population of massive stars, and consequently, at any given time, we will therefore be more likely to observe their most luminous, evolved counterparts.
We also see another effect with this group of galaxies related to the galaxy type. M31 (Sb I-II) and M81 (Sb I -II), both high luminosity spirals, have lower luminosity evolved A-type blue stars compared with the Sc-type spirals and Magellanic irregulars. This second effect very likely reflects a dependence on the lower star formation rate, not mass, in the Sb spirals. Both of these spirals have above solar metallicites, although M81 has a metallicity gradient in its disk like the Milky Way and the Sc spirals M33 and M101, and solar-type abundances in the outer parts.
This degeneracy, wherein the luminosity of the most luminous blue stars are dependent on the host galaxy, complicated their use as direct distance indicators. The most luminous red supergiants however, exhibit a nearly constant upper luminosity, related to the Humphreys-Davidson limit, over a wide range of galaxy types and luminosities. The presence of extensive circumstellar dust and uncertain reddening in the most luminous RSGs though limits their usefulness as distance indicators in the visual, but their luminosities in the infrared need to be further studied. Kudritzki [73] has introduced a new distance determination method for B and A-type supergiants, the Flux-weighted Gravity-Luminosity Relation (FGLR) that depends on quantitative analysis of the Balmer lines measured in low resolution spectra with good S/N ratio. It has been successfully applied to several nearby galaxies [72,74,75], demonstrating that the visually brightest stars, the A-type supergiants, have potential as distance indicators at very large distances.

The Complex Upper HR Diagram
The upper luminosity boundary to the HR Diagram, or Humphreys-Davidson Limit, complicates our understanding of massive star evolution. Above some initial mass, ≈ 40 -50 M ⊙ , the stars do not evolve across the HR Diagram to become RSGs. Their evolution to cooler temperatures is most likely halted by high mass loss events and their subsequent proximity to the Eddington Limit [76,77]. They then evolve back to warmer temperatures, now increasingly subject to instabilities, before their terminal state. Consequently, they have a different evolutionary path and different mass loss histories than their somewhat lower mass counterparts in the same galaxy or even the same cluster. These lower mass supergiants, ≈ 9 or 10 up to 30 -40 M ⊙ or so, evolve across the HR Diagram , becoming red supergiants which alters their interior structure and with enhanced mass loss. Some may also evolve back to a warmer state, perhaps as warm hypergiants, to become LBVs, or B[e] supergiants, where their final fate is most likely as core-collapse supernovae.
Consequently, the upper HR diagram is populated by a diversity of evolved, luminous and variable stars of different types that challenge our understanding of their physics, evolution and eventual fate. Many of them are distinguished by their emission line spectra, and evidence for stellar winds and mass loss. In addition, some of them exhibit periods of enhanced mass loss, such as the LBVs/S Dor variables and the warm and cool hypergiants with their resolved ejecta. In addition, Wolf-Rayet stars of various types, Oe and Of stars, the B[e] supergiants, and the Fe II emission line stars occupy the same parts of the HR Diagram. 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, and some may be binaries. This diversity is one of the challenges to understanding massive stars, their evolution, and eventual fate.
The HR Diagrams below for M31 and M33 show the distribution in the luminosity-temperature plane of three of these mass losing classes of stars, the LBVs, B[e] supergiants and the warm hypergiants. One of the outstanding questions is the final fate of the most massive stars. It used to be simply assumed that all stars much above initial masses of 10 M ⊙ or so, would end their brief lives as some kind of supernova, but their final stage as core-collapse SNe is now in question. Smartt [79,80] has suggested an upper mass limit of ≈ 18 M ⊙ for the red supergiant progenitors of the Type II-P SNe, while Jennings et al. [81] find a lack of massive supernova progenitors in M31 and M33 and suggest an upper mass of 35-45 M ⊙ . So, do the most massive stars collapse directly to black holes instead, long suspected for extreme, very massive stars like eta Carinae?
In addition, the supernova surveys have identified numerous non-terminal giant eruptions, in which the object greatly increases its total luminosity possibly expelling several solar masses and the star survives. Some of these events are confused with true SNe and thus have been called "supernova impostors". This is a diverse group of objects with a range of luminosities and possible progenitors. A few impostors appear to be normal LBV/S Dor variables in their eruptive or maximum light state. Most are giant eruptions, possibly similar to η Car [82] from evolved massive stars, while some are red transients (ILRTs, [83]) from a lower mass population. The origin of the instability in these giant eruptions is unknown, but proximity to the Eddington Limit is crucial [84]. It isn't known what role these high mass loss events may play in the final stages of massive star evolution or their relation to the evolved massive star population and to other stars with instabilites like the LBVs and the warm hypergiants.
Thus we observe a complex upper HR diagram with different evolutionary paths dependent on initial mass, and several types of evolved stars not only experiencing continuous mass loss, but also high mass loss events. The study of luminous stars in the nearer galaxies provides us with an improved census of these evolved stars, their relative numbers, physical properties and behavior, and clues to their evolutionary state and possible relationship to each other on the HR Diagram. In this volume on luminous stars in nearby galaxies, the reviews focus on different examples of evolved massive stars; the luminous O and B-type stars in the Magellanic Clouds, LBVs, B[e] supergiants, the warm hypergiants, and the Wolf-Rayet stars.
Although the chapters are about different types of evolved massive stars, many will be in the same galaxies. We have therefore adopted the following distance moduli based on the Cepheid scale for these nearby galaxies for consistency and for ease of cross-referencing and comparison.