May 272017

By Jeffrey Cummings (STSCI and Johns Hopkins University).

Stars evolve through many phases in their lifetime. Towards the end, when they have fused all of the Hydrogen (H) in their cores to Helium (He), they expand, cool down, and become giants. During these final stages, stars also begin to lose a significant amount of mass from their surfaces. A star like our Sun will lose approximately 50% of their mass, while more massive stars (> 8 Msun) will lose as much as 80% to 85%. Nearly all stars will eventually shed their outer layers and expose their hot and ultrahigh density cores. These remnants are called white dwarfs and, unless they experience mass transfer from a companion, their fate is to slowly cool with time.

However, a small percentage (approximately less than 2%) of high-mass stars can reach the necessary high densities and pressures in their cores to undergo a different fate: iron (Fe) will be created in their central core, but because it cannot undergo fusion to continue to generate energy, the star eventually becomes unstable and gravitationally collapses on itself creating a core-collapse supernova (CCSN). While it is commonly adopted in astronomy that stars with initial masses greater than 8 Msun will end their lives as a CCSN, we still do not fully understand at which stellar mass this transition occurs.

Knowledge of this transition mass is important because it sheds light on stellar evolution processes like mass-loss and convection/dredge up, and on the CCSN rate and how large of an effect these supernovae have on the production of elements (important to understand chemical evolution), on the energetics/feedback in galaxies, and on star formation.

Determination of the CCSN Mass Transition

There are three main methods to determine this mass. They all have their challenges and limitations, but jointly they may be able to begin to precisely constrain this transition mass:

1) Supernova studies. With the ability of the Hubble Space Telescope to resolve stars out to nearby galaxies, when a nearby CCSN occurs there is likely deep and resolved archival images of the progenitor star. Additionally, the stellar population of lower-mass stars that formed with the progenitor can subsequently be observed. Using stellar evolutionary models and the photometry of the progenitor and/or of the nearby lower-mass stars that formed with it, it is possible to infer the progenitor’s mass (e.g., Smartt 2009, Smartt 2015, Williams et al. 2014, Jennings et al. 2014). Within the past several years, the sample of nearby supernova has increased enough to provide meaningful statistics, indicating that the lowest-mass stars that undergo supernova have masses of ~9.5 Msun (Smartt 2015; Figure 1). These data also suggest that stars more massive than 18 Msun likely collapse but do not undergo an observable supernova.

BlogFig1Figure 1:  Nearby supernovae are displayed with their inferred initial masses and errors.  These masses are based on STARS and Geneva models (Eldridge & Tout 2004; Hirschi et al. 2004).  In solid black, a trend is fit to the data suggesting a lower limit of ~9.5 Msun.  In dashed black a second trend illustrates the expected number of stars at each initial mass based on a Salpeter initial mass function.  This shows that such a large number of 10 to 17 Msun supernovae progenitors should be accompanied by a smaller but observable number of higher mass supernovae progenitors.  This suggests that stars with mass greater than ~18 Msun may not undergo standard observable supernovae.

2) Stellar evolution models. A second method to estimate the CCSN transition mass is to use stellar evolution models to determine at which mass the core will reach the necessary conditions to create an Fe core and induce core-collapse (or produce an electron-capture CCSN). While a variety of models currently exist to account for the complex physical processes involved, many of them are beginning to converge around 9.3–9.8 Msun (Eldridge & Tout 2004, Poelarends 2007, Ibeling & Heger 2013, Doherty et al. 2015). Figure 2 illustrates the models from Doherty et al. (2015) and the transition masses at differing composition, with solar being represented by Z = 0.02.


Figure 2: This diagram illustrates the evolutionary result of a star with an initial mass at a given composition (Z = total mass fraction of a star that is not H or He), where solar composition is Z = 0.02.  This model distinguishes between the different types of white dwarfs (WDs) and illustrates where white dwarf formation ends and supernovae begin.  These models suggest a narrow region where electron-capture supernovae (ECSN) occur followed by the traditional Fe-core CCSN and shows that composition is an important secondary variable in these mass transitions.

3) White dwarf studies. Another way to constrain the transition mass is to study the remnants of the stars that do not undergo CCSN, i.e. white dwarfs. Spectroscopic analysis of high-mass white dwarfs in star clusters, together with evolutionary models, allow us to both directly determine the mass of the white dwarf and to infer the mass of the white dwarf’s progenitor. This is known as the initial-final mass relation (IFMR). Figure 3 illustrates the IFMR derived from white dwarfs across multiple clusters and a broad range of masses (Cummings et al. 2015; 2016a; 2016b; in prep.), and with evolutionary timescales based on PARSEC stellar models (Bressan et al. 2012). The IFMR provides a direct constraint on the mass loss that occurs during stellar evolution and how it varies with initial mass. In terms of constraining the CCSN transition mass, there currently are very few ultramassive white dwarfs that have been discovered in star clusters, but our current project is to survey for more white dwarfs approaching the highest mass a white dwarf can stably have (the Chandrasekhar mass limit at 1.375 Msun). The initial mass of a star that will create a Chandrasekhar mass white dwarf will also define the CCSN transition mass, which will provide a critical check of the other methods. However, based on cautious extrapolation, we note that our current relation suggests a consistency with the CCSN mass transition occurring near 9.5 Msun.


Figure 3: The current IFMR based on the spectroscopic analysis of white dwarfs that are members of star clusters.  Comparison of the spectroscopically determined mass and age of the white dwarf can in comparison to the age of the cluster be traced back to the mass of the star that would have formed a white dwarf at that time.  The solid line represents a fit to the data while the dashed line represents the theoretical IFMR of Choi et al. (2016).  This shows a relatively clean relation with the clear formation of larger and larger white dwarfs from larger and larger stars.  The highest-masses, however, remain poorly constrained, but precise measurement of where this relation reaches the Chandrasekhar mass (1.375 Msun; the upper limit of the y-axis), will define the progenitor mass at which stars begin to undergo CCSN.

The three methods described above, mostly independent from each other, are beginning to suggest a convergence around a CCSN transition mass of ~9.5 Msun. This implies that the commonly adopted value of 8 Msun, which is based on older and more limited data and theoretical models, has led us in the past to overestimate the number of CCSN by ~30%. Such an overestimation would greatly affect our understanding of the chemical evolution, energetics, and feedback in galaxies. Furthermore, the maximum mass for CCSN inferred by Smartt (2015), 18 Msun, further decreases the number of expected CCSN in a stellar population. This may solve the supernova rate problem that resulted from assuming that all stars with a mass greater than 8 Msun should undergo a CCSN. Under that assumption, the number of predicted CCSN is twice the observed rate (Horiuchi et al. 2011). But if only stars in the 9.5–18 Msun mass range undergo CCSN, based on the standard Salpeter stellar initial mass function, the new estimated CCSN rate is almost exactly half of the original estimate. This would bring observations and models into remarkable agreement.

Future work

Significant work remains to be done. More nearby supernovae need to be observed, with subsequent study of their progenitors, to further constrain their lower and upper mass limits. More ultramassive white dwarfs need to be discovered in star clusters, a focus of our research group, to refine our understanding of higher-mass stars and their mass-loss processes. Increasing the white dwarf sample size will provide an independent measurement of the CCSN transition mass, and it may also begin to provide an important observational test for the effects of differing stellar composition on mass-loss, evolution, and this mass transition. For stellar evolution models, a more coordinated approach between these observations and theory will improve the models, which also affects the inference of progenitor masses in the supernova and white dwarf studies. An iterative process, aiming for self-consistency across all steps, will ideally bring a more precise convergence of the mass transition at which supernova begin to occur. In any case, the assumption that all stars with mass greater than 8 Msun will undergo a CCSN is appearing more and more inaccurate.


  • Choi, J., Dotter, A., Conroy, C., et al. 2016, ApJ, 823, 102
  • Cummings, J. D., Kalirai, J. S., Tremblay, P.-E., & Ramirez-Ruiz, E. 2015, ApJ, 807, 90
  • Cummings, J. D., Kalirai, J. S., Tremblay, P.-E., & Ramirez-Ruiz, E. 2016a, ApJ, 818, 84
  • Cummings, J.D., Kalirai, J.S., Tremblay, P.E., Ramirez-Ruiz, E., & Bergeron, P. 2016b, ApJ, 820, L18
  • Doherty, C.L., Gil-Pons, P., Siess, L., Lattanzio, J.C., & Lau, H.H.B. 2015, MNRAS, 446, 2599
  • Eldridge, J. J., & Tout, C. A. 2004, MNRAS, 353, 87
  • Hirschi, R., Meynet, G., & Maeder, A. 2004, A&A, 425, 649
  • Horiuchi, S., Beacom, J. F., Kochanek, C. S., Prieto, J. L., Stanek, K. Z., & Thompson, T. A. 2011, ApJ, 738, 154
  • Ibeling D., Heger A., 2013, ApJ, 765, L43
  • Jennings, Z. G., Williams, B. F., Murphy, J. W., Dalcanton, J. J., Gilbert, K. M., Dolphin, A. E., Weisz, D. R., & Fouesneau, M. 2014, ApJ, 795, 170
  • Poelarends A. J. T., 2007, PhD thesis, Astronomical Institute Utrecht (P07)
  • Smartt, S. J. 2009, ARA&A, 47, 63
  • Smartt, S.J. 2015, PASA, 32, 16
  • Williams, B. F., Peterson, S., Murphy, J., Gilbert, K., Dalcanton, J. J., Dolphin, A. E., & Jennings, Z. G. 2014, ApJ, 791, 105
Oct 022016

By Ori Fox (STScI)

Core-collapse supernovae (SNe) are the explosions of massive stars (>8 Msun) that reach the end of their lifetime. No longer able to radiatively support themselves by nuclear core burning after depleting their fuel, the stars collapse and release gravitational energy that rips apart the star entirely. The resulting explosions exhibit differences in their spectra and light curves that can be grouped into one of several subclasses.

From a theoretical perspective, these differences once seemed straightforward. Single star models indicate that the strength of a stellar wind increases as a function of the star’s initial mass and metallicity (Heger et al. 2003). In turn, stronger winds can remove more of a star’s outer envelope, resulting in the distribution of observed SN subclasses. Accordingly, a Type II SN has hydrogen in the spectrum, suggesting a lower mass (~8-25 Msun) red supergiant (RSG) progenitor. In contrast, a Type Ic SN has neither hydrogen nor helium in its spectrum, suggesting a higher mass (>40 Msun) progenitor.

Direct images of the individual stars before they explode provide the strongest observational constraints, but are difficult to obtain because they require deep, high-resolution, multi-color, pre-explosion imaging. Before the Hubble Space Telescope (HST) was launched, one of the few progenitors directly observed was the progenitor to SN 1987A in the Large Magellanic Cloud (LMC) at just 0.05 Mpc. The Cerro Tololo Inter-American 4-meter telescope obtained several images of the LMC between 1974 and 1983 (Walborn et al. 1987). The direct observations showed a progenitor consistent with a blue supergiant, which contradicted most stellar evolution theory and set the field on the course it is still on today.

AAT 50. The field of supernova 1987A, before and after (March, 1

Figure 1: The famous SN 1987 both before (right) and during (left) the explosion. The exploding star, named Sanduleak -69deg 202, was a blue supergiant.

Ground-based imaging is only sufficient for detecting progenitors out to 1-2 Mpc. HST extended this range out to about 20 Mpc. Cost and time, however, prohibit HST from obtaining pre-explosion imaging of the thousands of galaxies within this volume. Instead, these data must be obtained serendipitously via other science programs. The number of galaxies with pre-explosion imaging has grown steadily since HST was launched in 1990. With only a few SNe within this volume each year, a statistically significant sample of SNe with corresponding HST pre-explosion images was not accumulated until the mid-2000s (Smartt 2009). As predicted by the theory, Type II SNe had RSG progenitors. The most mystifying result, however, was the fact that the Type I SNe (i.e., those without hydrogen) had no confirmed massive (and thereby luminous) star progenitors, even to very deep limits.

The solution to this mystery is still not solved but may involve binary star progenitor systems, which are now known to account for ~75% of massive star systems (Sana et al. 2012). As opposed to single stars systems, where stars lose their envelopes in their winds, a binary companion star can remove the outer envelope of the primary via tidal stripping. This process allows for increased mass-loss from lower mass, less luminous stars that may evade detection in pre-explosion imaging. This scenario has long been preferred for a specific subclass referred to as the Type IIb (i.e., a hybrid of the Type II and Ib subclass) since most, but not all, of the outer Hydrogen envelope is removed.


Figure 2: This illustration shows the key steps in the evolution of a Type IIb supernova. Panel 1: Two very hot stars orbit about each other in a binary system. Panel 2: The slightly more massive member of the pair evolves into a bloated red giant and spills the hydrogen in its outer envelope onto the companion star. Panel 3: The more massive star explodes as a supernova. Panel 4: The companion star survives the explosion. Because it has locked up most of the hydrogen in the system, it is a larger and hotter star than when it was born. The fireball of the supernova fades. (Credit: NASA, ESA, and A. Feild (STScI))

While the primary (i.e., exploding) star in the binary system may be too faint to be detected in the pre explosion imaging, the companion star may be bright enough to test the binary hypothesis. As the primary star loses mass, the companion will gain mass and become more luminous and blue. Despite these changes, detecting the companion star in a binary system is not straightforward. The stellar spectrum of the companion will peak towards the ultraviolet (UV). Since most serendipitous pre-explosion imaging does not consist of UV observations, a UV search for the companion must occur only once the SN has faded. To date, a companion star has only been observed in a single instance for the Type IIb SN 1993J in M81 at just 3.5 Mpc (Maund et al. 2004, Fox et al. 2014).


Figure 3: This is an artist’s rendition of supernova 1993J, an exploding star in the galaxy M81 whose light reached us 21 years ago. The supernova originated in a binary system where one member was a massive star that exploded after siphoning most of its hydrogen envelope to its companion star. After two decades, astronomers have at last identified the blue helium-burning companion star, seen at the center of the expanding nebula of debris from the supernova. HST identified the UV glow of the surviving companion embedded in the fading glow of the supernova. (Credit: NASA, ESA, and G. Bacon (STScI))

The future of progenitor detections lies with HST and the James Webb Space Telescope (JWST). HST offers UV-sensitive instruments that allow us to search for the binary companions to these stripped envelope SNe. JWST will offer more than 7 times the light collecting area than HST. While JWST lacks UV capabilities necessary for companion star searches, it will increase the sensitivity to primary stars that peak at redder wavelengths. This increased sensitivity will not only provide stronger constraints on the progenitors, but it will allow progenitor searches to extend out to larger distances, thereby increasing the search volume and sample size. These new progenitors discoveries will have direct implications on our understanding of star formation, stellar evolution models, and mass loss processes.


  • Heger, A., et al. 2003, ApJ, 591, 288
  • Walborn, N. R., et al. 1987, ApJ, 321, L41
  • Smartt, S. 2009, ARA&A, 47, 63
  • Smith, N., et al. 2011, MNRAS, 412, 1522
  • Sana, H., et al. 2012, Science, 337, 444
  • Maund, J., et al. 2004, Nature, 427, 129
  • Fox, O. D., et al. 2014, ApJ, 790, 17
Sep 072016

By Linda Smith, European Space Agency/STScI

The upper mass limit for stars is not known with any certainty. The best means of observationally determining this parameter is to study the content of young, massive star clusters. The clusters need to be young (< 2 Myr) because of the short lifetime of the most massive stars, and they need to be massive enough (> 105 Msun) to sample the full extent of the initial mass function (IMF).

In 2005, Don Figer derived an upper mass limit for stars of 150 Msunusing the Arches cluster near the center of our Galaxy. However, the Arches cluster is too old at 4 Myr to sample the true initial mass function (IMF) because stars more massive than 150 Msun will have already exploded.

In a star-forming region, the most massive stars will dominate the ionization and stellar wind feedback for the first few million years. The amount of feedback will be severely underestimated from models if the upper stellar mass cut-off of the IMF is too low. Most stellar population synthesis models, which are used to infer the stellar content and feedback of unresolved star-forming regions, adopt cut-off values of 100 or 120 Msun (e.g. Starburst99; Leitherer et al. 1999).

The massive star cluster R136 in the 30 Doradus region of the Large Magellanic Cloud (LMC) is the only nearby resolved cluster which is young and massive enough to measure the IMF, and thus empirically determine the stellar upper mass cutoff. In a series of papers, Crowther et al. (2010, 2016) used far-ultraviolet (FUV) spectra obtained with spectrographs on HST to determine the masses of the massive stars using modeling techniques. They found that the R136 cluster is only 1.5 ± 0.5 Myr old and contains eight stars more massive than 100 Msun with the most massive star (called R136a1) having a current mass of 315±50 Msun. The four most massive stars account for one-quarter of the total ionizing flux from the star cluster. These very massive stars (VMS, M > 100 Msun) have very dense, optically thick winds and their emission-line spectra resemble Wolf-Rayet (W-R) stars but they are hydrogen-rich (see the recent blog article by Tony Marston on W-R stars).

Beyond R136, the best means of finding VMS is to look for their spectral signatures in the integrated FUV light of young massive star clusters in star-forming galaxies. NGC 5253 is a blue compact galaxy with a young central starburst at a distance of 3.15 Mpc. The galaxy is part of the Legacy Extragalactic UV Survey (LEGUS; see, a Cycle 21 HST large program. In a paper by Calzetti et al. (2015), we combined the LEGUS imaging with HST archive images and derived the masses and ages of the bright, young star cluster population of NGC 5253 using 13 band photometry. In Fig. 1, the LEGUS image of NGC 5253 is shown. Fig 2 shows the two clusters (numbered #5 and #11) at the center of the galaxy. Cluster #5 coincides with the peak of the Hα emission in the galaxy and cluster #11 with a massive ultracompact H II region.


Figure 1: Three color composite of the central 300 x 250 pc of NGC 5253 from Calzetti et al. (2015). The 11 brightest clusters are identified and numbered.


Figure 2: Detailed view of the two nuclear clusters #5 and #11 shown in Fig. 1, which are separated by a projected distance of 5 pc.

We found that the two nuclear clusters have ages of only 1±1 Myr and masses of 7.5 x 104 and 2.5 x 105 Msun.   Interestingly, the very young ages we derive contradict the age of 3-5 Myr, inferred from the presence of W-R emission-line features in the optical spectrum of cluster #5. Could these W-R features arise from very massive stars instead? To answer this, we examined archival FUV STIS and FOS spectra and optical spectra from the Very Large Telescope (VLT) of cluster #5 to search for the spectral features of VMS. This study is described in Smith et al. (2016).

The FUV spectra show that cluster #5 does indeed have the signature of very massive stars rather than much older classical W-R stars. The FUV spectrum of cluster #5 is shown in Fig. 3 and compared to the integrated FUV STIS spectrum of R136a (Crowther et al. 2016), which has been scaled to the distance of NGC 5253. The similarity between the two spectra is striking. The crucial VMS spectral features are the presence of blue-shifted O V λ1371 wind absorption, broad He II λ1640 emission, and the absence of a Si IV λ1400 P Cygni profile (expected in W-R stars). Crowther et al. (2016) find that 95% of the broad He II emission shown in the R136a spectrum in Fig. 3 originates solely from VMS. Thus the presence of this feature in emission together with the O V wind absorption indicates a very young age (< 2 Myr) and a mass function that extends well beyond 100 Msun.


Figure 3: The HST FUV spectrum of NGC 5253 cluster #5 compared to the integrated HST/STIS spectrum of R136a (Crowther et al. 2016). The R136a spectrum has been scaled to the distance of NGC 5253. The flux is in units of 1015  erg s-1 cm-2 Å-1

The presence of very massive stars in cluster #5 (and also probably cluster #11) can also explain the very high observed ionizing flux. Previous studies have assumed an age of 3-5 Myr and find that standard stellar population synthesis codes significantly under-predict the ionizing flux. For an age of 1 Myr, the predicted ionizing flux is still too low by a factor of 2 for a standard IMF with an upper mass cut-off of 100 Msun. However, only 12 VMS with M > 150 Msunare needed to make up the deficit.

The UV spectrum of cluster #5 shows many similarities with the rest frame spectra of metal-poor, high-redshift galaxies with broad He II emission and strong  O III] and C III] nebular emission lines. If VMS exist in young star-forming regions at high redshift, their presence should be revealed in the UV rest-frame spectra to be obtained by the James Webb Space Telescope. For all studies near and far, it is crucial to extend stellar population synthesis models into the VMS regime to correctly model the spectra, and account for the radiative and stellar wind feedback, which will be dominated by VMS for the first 1–3 Myr in massive star-forming regions.


  • Calzetti, D. et al., 2015, ApJ, 811, 75
  • Crowther, P.A. et al., 2010, MNRAS, 408, 731
  • Crowther, P.A. et al., 2016, MNRAS, 458, 624
  • Figer, D.F., 2005, Nature, 434, 192
  • Leitherer, C. et al., 1999, ApJS, 123, 3
  • Smith, L.J. et al., 2016, ApJ, 823:38
Aug 012016

By Anthony Marston, European Space Agency/STScI

What are Wolf-Rayet stars?

Wolf-Rayet (WR) stars are believed to be evolved massive stars that initially started their lives with masses of  > 20 Msun. With such high masses, they evolve very quickly to the WR state from high-mass hydrogen burning O stars in 1-2Myr. Currently, evidence suggests that the majority of WR stars are either in or affected by having been in relative close binaries, that can affect their evolution.

There are several evolutionary paths and theories as to the evolutionary direction of WR stars. It is postulated that different evolutionary paths exist depending on the how much initial mass exists above 20 Msun, as well as whether they are single or binary stars. For most WR stars, a mass-loss phase of a few tens of thousands of years probably occurs. Evidence for this is seen in the nitrogen-enriched ejecta nebulae that are seen around many WR stars. Ejecta are believed to be associated with a slow wind phase following the fast wind of the main sequence O star phase. Once evolved to a WR star there is again a fast wind phase which can quickly interact with a slow moving ejecta nebula. But not all WR stars are seen to have ejecta.

There are three subtypes of WR stars: WN subtypes show prominent nitrogen emission lines in their spectra, WC subtypes show prominent carbon emission lines, and WO subtypes show strong high excitation oxygen emission lines. These form an apparent evolutionary sequence with the spectra showing the products of Hydrogen burning for WN stars, Helium burning for WC star spectra and higher level burnings for WO stars. WO subtype stars in the Galaxy are very rare (three are known) and probably represent a final WR evolutionary phase before becoming a supernova (probably of type Ib).

How common are they and how are they distributed in the Galaxy?

By the end of 2000 just over 200 WR stars were known in the Galaxy. Most of these were discovered in studies of clusters or serendipitously. They were shown to follow the spiral pattern of the Galaxy and showed a distribution that mimicked other Galactic star formation site indicators. Indeed, WR stars have in various ways been used as markers of very recent high-mass star formation and star formation bursts since they only live a few million years.

In his review of WR stars, Karel van der Hucht (2001) indicated that the likely population of WR stars in the Galaxy could be several thousand rather than the few hundred known. This was in part due to obvious observational restrictions, such as unseen populations on the opposite side of the Galaxy. With the advent of sensitive infrared detectors the possibility of finding distant and/or obscured WR stars became more realistic. Two approaches have been developed for finding Galactic WR stars in recent years.

The “narrow-band” approach (Shara et al. 2009) uses narrow-band images centered on strong emission lines seen in WR stars (e.g. HeII) and subtracts from them narrow-band images covering only the continuum (or broad-band infrared images). The candidates revealed are followed up with infrared and/or optical spectroscopy to confirm their nature.

The “broad-band” approach is based on the near- to mid-infrared colors which are peculiar to stars with strong winds – and in particular WR stars. Figure 1 shows how the free-free emission from the fast WR wind of the nearby WR star WR11 (g Vel) has a distinct spectral index which is substantially different from stellar photospheres leading to WR stars being overabundant in certain areas of broadband infrared (2MASS, Spitzer/IRAC, WISE) point source color-color space (see Figure 2). Even though the this approach is slightly more prone to confusion issues than the “narrow-band” method, it has a couple of advantages over the latter: the potential for picking up weak-lined WR stars or ones where lines are diluted relative to the continuum due to a massive companion or local hot dust emission. It also uses already existing infrared point source catalogs (e.g. the GLIMPSE catalog of source within | b < 1 | in the Galactic plane). As of July 2016, the total number of known Galactic WR stars is 634 (

figure1Figure 1: Spectral energy distribution of g Velorum (Williams et al. 1990) showing the excess free-free emission from the stellar wind in the infrared wavelengths as compared to photospheric emission (straight black line). The GLIMPSE catalog which used Spitzer/IRAC data will show WR stars with colors distinct from the vast majority of stars.

Our group, with core members Schuyler Van Dyk (Caltech), Pat Morris (Caltech), Jon Mauerhan (UC Berkeley) and Anthony Marston (ESA-STScI), uses the broad-band method. It was first developed by Marston in 2004 to identify candidates in ESO/SOFI infrared spectroscopic observations and it helped identify 60 new WR stars by Mauerhan et al (2011). The color selection uses data from the GLIMPSE catalog, consisting of several 10’s of million sources detected in the Galactic plane using broadband Spitzer/IRAC 3.4 – 8 mm measurements combined with band-merged flux data from 2MASS (broadband near-infrared JHKs). In certain studies, X-ray emission sources and, more lately, WISE point source colors have been used in identifying WR candidates.  Spectroscopic follow-up has concentrated on obtaining K-band spectra, as WR stars are typically identified by strong HeII emission lines such as the 2.189mm line. For the less reddened candidates, optical spectroscopic follow-up has also been possible.

Historically we have found that 10-15% of candidates turn out to be bona fide WR, stars while~ 85% of all candidates are emission-line stars, most often Be stars. Small numbers of O/Of stars B[e] supergiants and stars exhibiting infrared CO bandhead absorption lines have been picked up where combinations of photosphere, dust emission and free-free emission has brought objects into our infrared color space. Improvements to our color-space selection have increased the success rate of WR detections out of the candidates, notably for more reddened/distant objects where the candidate confirmation rate can go as high as 25% (see Figure 2). We are currently looking into a machine-learning capability for assessing the likelihood of an object being a WR star from color-space criteria. The ultimate goal is to be able make accurate predictions of WR numbers of different subtypes in the Galaxy.


Figure 2: Infrared color-color plots showing the candidate objects observed by Mauerhan et al (2011). The green symbols were newly discovered WN subtypes and red WC subtype stars. Blue points represent candidates that follow up spectroscopy showed were not WR stars. Grey shaded areas indicate the part of the color-color plots where 50% or more candidates were found to be WR stars.

What have WR stars taught us about high-mass star formation?

The number ratios of WR to O stars and Red Supergiant (RSG) or Luminous Blue Variable stars are key values for constraining stellar evolution theories of massive star evolution. In a simple way, ratios provide an indication of relative timescales for lifetimes. Another indication of timescales, and possibly different evolutionary links between subtypes, mass-loss phases and initial stellar masses, is the number distribution of WR subtypes, both WN and WC (plus the rare WO stars).

The distribution of WR stars (studied e.g. using the Spitzer’s GLIMPSE survey across the Galactic plane) marks star formation sites across the Galaxy and indicate likely sites of future supernovae. However, it has become clear over time that many WR stars, that are no more than a few Myr in age, appear to be found well away from the centers of star-forming clusters in the Galaxy. A projection of most of the known WR stars with secure distance shows that some WR stars also appear to be more than 100 pc above/below the plane of the Galaxy (Rosslowe & Crowther, 2015). A possible explanation of why some WR stars appear to be located away from their birth site could be the presence of fast transverse motions caused by expulsion from their cluster formation site. Another possibility could be that these stars were part of small clusters but, being much more luminous than other cluster members, they appear to be isolated. But in recent years, in the study of star-forming regions like the Cygnus OB2 cluster, we have learned of an unexpected third possible explanation.

Various studies suggest that the Cygnus OB2 cluster, being 1 Myr of age, has not evolved significantly from its original distribution. This means that the massive stars, and WR stars in particular, are near the sites where they were born. However, none of the WR stars are in the massive star cluster at the center of the Cyg OB2 association (see Figure 3), and not only that, none show evidence of bow shocks from significant transverse velocities, suggesting these stars were born in situ. We now know, from studies with Herschel, that filaments of high-density gas can extend through star-forming regions with “strings” of star-forming cores being found along them. And in fact, filaments pervade the Cyg OB2 area leading to the possibility of forming high-mass stars outside of major stellar clusters, possibly instigated to form high-mass stars through a triggering event, such as expanding gas shell collisions.


Figure 3: The Cygnus OB2 association as seen by Herschel PACS/SPIRE (colored background from Schneider et al, 2016). WR stars and Luminous Blue Variable stars (likely precursors of WR stars in stellar evolution) are found well away from the major cluster of O stars shown as white points a bit to the right of center of the field (Comeron et al 2008, Wright et al 2015).

There are therefore two possible scenarios:

  • WR stars are born in situ and away from stellar clusters (but likely within stellar associations) – which means distributed high-mass star formation occurs for some of the most massive stars probably from filaments.
  • WR stars are kicked out of stellar clusters due to dynamics of the early cluster of stars or through binary/supernova interactions, apparently affecting a large fraction of the very massive stars in the stellar cluster.

As we have seen, the study of Wolf Rayet stars has shed new light on unexpected physical processes associated to high-mass star formation. In the future, we will advance their study by: (1) Using machine-learning and improved color-selection techniques to find new WR stars and assess their distributions in the Galaxy, including in high-mass star-forming regions. (2) Pinning down number ratios of WR subtypes and other massive star types. (3) Using the GAIA catalog to get proper motions of WR stars to identify runaway stars. (4) Searching for bow shocks, in particular in the mid-infrared with WISE, as it has been found that they are particularly prominent at IR wavelengths.



  • Blaauw,  A.,1993, ASPC, 35, 207
  • Comeron et al, 2008, A&A, 486, 453
  • Crowther et al, 2006, MNRAS, 372, 1407
  • van der Hucht, K., 2001, New AR, 45, 135
  • Mauerhan, J., et al, 2011, AJ, 142, 40
  • Rosslowe, C., & Crowther, P., 2015, MNRAS, 447, 2322
  • Schneider, N., et al, 2016, A&A, 591, A40
  • Shara et al, 2009, AJ, 138, 402
  • Williams, P., et al, 1990, MNRAS, 244, 101
  • Wright, N., et al, 2015, MNRAS, 449, 741