WorksPhysics and Astronomy
The formation and evolution of Wolf-Rayet stars with temperature-dependent winds
Abstract
Massive stars strongly influence their environments through their radiation, stellar winds and core-collapse supernovae, while their own evolution is strongly shaped by wind mass loss. Wolf-Rayet (WR) stars are characterised by strong, optically thick winds, and the implementation of these winds remains an important uncertainty in stellar evolution models. Recent hydrodynamically consistent atmosphere calculations predict WR mass-loss rates that depend strongly on the temperature at the wind-launching region. We investigate the effects of incorporating these latest theoretical WR wind prescriptions into single and binary stellar evolution models at Milky Way, Large Magellanic Cloud and Small Magellanic Cloud (SMC) metallicities. We find that the temperature dependence of the WR wind can significantly alter the evolution of stars near the threshold for WR formation. In particular, an implementation that transitions from a hydrogen rich WNL wind prescription to a classical WR wind driven by the hot iron opacity bump causes some SMC-metallicity models to spend an extended period at temperatures of approximately 70 to 90 kK. This produces a concentration of models in the temperature range occupied by most observed SMC WR stars, a feature that has previously been difficult to explain with stellar evolution models. Our results demonstrate that incorporating physically motivated, temperature dependent WR winds can improve the agreement between stellar evolution models and observed WR populations.
Introduction
Massive stars, those with an initial mass , play a crucial role in a range of astrophysical processes. Despite their rarity, they dominate the ionising and mechanical feedback of the interstellar medium (ISM) through their high luminosities and strong stellar winds (Smith, Norris & Crowther 2002; Robertson et al. 2010; Krause et al. 2013). They further enrich the ISM chemically and energetically through core-collapse supernovae (CCSNe) at the end of their lives. Neutron stars (NSs) and black holes (BHs) are the remnants of these massive stars (Heger et al. 2003; Ertl et al. 2020) and, in binary systems, they can become sources of gravitational waves originating from their coalescence (Abbott et al. 2016).
Unlike for their low-mass counterparts, wind mass loss from massive stars governs their evolution and affects their spectral appearances throughout their evolution. Notably, classical Wolf–Rayet (cWR) stars form as a late stage in the life of massive stars when they have completely lost their outer hydrogen layers and are fusing helium or heavier elements in their cores. Their strong, optically thick winds give rise to their characteristic prominent, broad emission lines, which are observable even in integrated spectra of entire galaxies (Conti 1991; Schaerer, Contini & Kunth 1999). WR classifications are based on the relative strengths of their spectral lines: WN, WC and WO stars have prominent nitrogen, carbon and oxygen emission lines, respectively (Smith 1968; Sander et al. 2014). This may be viewed as an evolutionary sequence. As the star becomes increasingly stripped and evolved its surface becomes N-rich through CNO-cycle processing and CNO-rich from helium burning (Crowther 2007; Shenar et al. 2019). However, such a sequence is not guaranteed before the end of the life of a star. A precursor to a WN star may fall under the hydrogen-rich WNh classification. Such stars have some remaining surface hydrogen and may still be main-sequence (MS) core-hydrogen burning stars (de Koter, Heap & Hubeny 1997; Hamann, Gräfener & Liermann 2006; Gräfener et al. 2011).
WR winds are understood to be radiatively driven with iron opacities playing a central role in the acceleration of the wind (Nugis & Lamers 2002; Gräfener & Hamann 2005; Vink & de Koter 2005). Although the luminosities and wind driving mechanism of WR stars are similar to those of OB-type stars, their mass-loss rates, inferred from spectra and atmosphere models, are approximately an order of magnitude larger (Nugis & Lamers 2000; Puls, Vink & Najarro 2008). However, the CAK theory of radiative driving by Castor, Abbott & Klein (1975), which successfully explains OB star winds, cannot account for WR winds, primarily owing to its simplifying assumption of single scattering. Indeed, Sander & Vink (2020) (hereinafter S20) showed that WR-type mass-loss begins at the onset of multiple scattering in the wind. Further, proximity to the Eddington limit may be a key cause of WR-type mass loss (Gräfener et al. 2011). It has recently been shown that Eddington-limit-induced mass ejections can reproduce both the Humphreys–Davidson limit and observed WR-star numbers (Pauli et al. 2026). Sharp changes in mass loss associated with iron ionisation and transitions between wind-driving regimes are also found in near-Eddington WNh stars and OBA hypergiants (Lefever et al. 2025; Bernini-Peron et al. 2026).
The velocity profile of a radiatively driven wind has often taken the form of a -law (Pauldrach, Puls & Kudritzki 1986), , where is the distance to the star’s centre, is the terminal wind velocity, is the hydrostatic stellar surface radius, and is a free parameter constrained by spectroscopic observations. Whilst the CAK theory predicts , when accounting for the so-called finite-disk correction3 fits OB star winds best (Friend & Abbott 1986; Puls et al. 1996) and applies to WR winds (Lefever et al. 2023). Such a -law has been used in WR atmosphere models to estimate mass-loss rates, (e.g. Nugis & Lamers 2002). Furthermore, empirical as a function of stellar parameters have been formulated from spectroscopic observations (e.g. Hamann, Koesterke & Wessolowski 1995; Nugis & Lamers 2000; Yoon 2017) and these have repeatedly been implemented in WR evolution models (e.g. [51] (Georgy et al. 2012; Grassitelli et al. 2016; Li et al. 2024)). These previous efforts demonstrate a WR radius problem (Grassitelli et al. 2018): empirical results for the WR radius seem too high, and so the corresponding temperatures are too low, in comparison to where evolutionary models predict them to be. The photosphere of a star, where its effective temperature is defined, is typically where the optical depth . However, for the optically thick winds of WR stars, this is satisfied in the supersonically outflowing material. Hence the location of the hydrostatic surface as derived from model atmospheres relies on the treatment of wind structure (e.g. Groh et al. 2014), for example the implementation of a -law or hydrodynamically consistent models.
Recent years have seen progress in the theoretical modelling of the hydrodynamic structure of WR atmospheres (e.g. with the PoWR code; Gräfener, Koesterke & Hamann 2002; Hamann & Gräfener 2003; Sander et al. 2015). A crucial region in hot WR atmospheres at temperatures kK is the hot iron bump, so called owing to the enhanced bound-bound iron opacities. This opacity bump starts to launch the supersonic wind at deeper layers of the atmosphere (Gräfener & Hamann 2005; Sander et al. 2020; Sander & Vink 2020; Lefever et al. 2026). The surface radii derived from hydrodynamically consistent models usually lie close to these deep wind-launching layers, and are generally significantly lower than those derived empirically. Whilst recent evolutionary studies (e.g. Romagnolo, Gräfener-Matamala & Beczynski 2024; Gormaz-Matamala, Romagnolo & Beczynski 2025) have employed the rates of S204, the latest updates from Sander et al. [85] (hereinafter S23) have not yet been incorporated in stellar models. The inclusion of this updated temperature dependency in the mass-loss rate and its effects on WR evolution is the focus of this study. We investigate the evolution of high-mass stars covering a range of stellar masses, metallicities and the latest WR wind mass-loss prescriptions for both single and binary stars. In § 2, we describe the numerical method and physical assumptions of our stellar evolution models. In § 3, we present the key results of effects on stellar evolution and WR star formation. We discuss the ionising energy output in § 4 and final stellar masses in § 5. We present simulated stellar populations and provide a comparison to observations in § 6. The work is summarised and conclusions are drawn in § 7.
Numerical method
We use version 15140 of the open-source 1D stellar evolution code Modules for Experiments in Stellar Astrophysics (MESA; Paxton et al. 2011, 2013, 2015, 2018, 2019) to evolve non-rotating single and binary stars with three metallicities, , and . These approximate the metallicity of young massive stars in the Milky Way (MW)5, the Large Magellanic Cloud (LMC) and the Small Magellanic Cloud (SMC), respectively. The chosen metallicities are appropriate for comparisons to observations of WR stars in these galaxies. The initial, zero-age main-sequence (ZAMS) masses of the single stars and the primary stars of binaries in our models are , 20, 23, 27, 31, 36, 45, 52, 60, 74, 99 and . The low end of this mass range is motivated by the initial mass limits for WR star formation (for stars in binaries at , the most likely conditions to form a WR star) found by Shenar et al. (2020a) and Gilkis et al. (2025). These were and , respectively. Further, our preliminary results for binary stars at indicate that stars with do not give rise to a WR phase during their evolution. Stars more massive than are not modelled because of the high uncertainty in the mass loss and evolution of these very massive stars (VMSs; Gräfener & Hamann 2008; Crowther et al. 2010; Bestenlehner 2020; Sabhahit et al. 2022).
For binary systems, whose orbits we assume to be circular (a reasonable assumption for close binary stars; Faccioli, Alcock & Cook 2008) the mass ratio is and 0.8, and the initial orbital periods are and 30 d. While the entire parameter space of possible initial conditions has not been covered, we sample contrasting regimes of close binary interaction, and the grid is not intended to provide a comprehensive exploration of binary parameter space. Four different wind mass-loss prescriptions for the WR phase of stellar evolution are used and assessed. The details of these are given in § 2.1.
The physical assumptions underlying the stellar processes being modelled largely follow Gilkis et al. (2025), including a prescription for convective core overshooting following Jermyn et al. (2022). Binary interactions are included through Roche lobe overflow (RLOF) mass transfer, focusing on the effect of envelope stripping of the primary which could subsequently become a WR star. We are not concerned with the evolution of the accreting secondary, which we model as a point mass.
The stars are evolved from the ZAMS until the end of core carbon burning. This captures almost the entirety of the stellar and WR lifetimes without the complexities and uncertainties of modelling CCSNe and/or NS and BH formation at the end of their lives. Because the evolutionary timescales of nuclear burning become increasingly short towards the end of stellar life, the termination point of the simulations should not have a major effect on modelling observed WR stars or estimating their final masses.
Wind mass loss
To account for the wide range of luminosities and temperatures encountered during stellar evolution, we apply several mass-loss recipes with empirical rates supported by observations and theoretical rates in the regimes for which they were developed. The approach is broadly similar to that of Gilkis et al. (2025) and depicted schematically in Figure 1. An optically thin wind mass-loss rate is first calculated as (following Vink, de Koter & Lamers (2001) with the metallicity dependence of Vink & Sander (2021)) for hydrogen-rich stars6 or (following Vink (2017)) for helium-rich stars, and interpolating in between. Four temperature regimes are then considered, as follows.

Figure 1. Wind mass-loss prescription used for massive stars at , and . Details of the mass-loss rates and temperatures are given in the text.
The hottest regime, , is for the classical WR stars which have a temperature greater than the maximum expected temperature for the cooler WNL stars, . Here, follows the theoretical recipes of S20 or S23. Both have been employed here and their results compared. We note that if the star is not luminous enough, or if the predicted is not high enough, is used. S20 provide where, for a star with luminosity and mass ,
is the Eddington factor. The approximation for the number of free electrons per atomic mass unit, , from Gilkis et al. (2025) is used. This allows for surface composition changes and yields for the mostly ionised, H-free atmospheric compositions. The S20 recipe is
with the -dependent coefficients , , and given by them. S23 provide where is the effective temperature at the launching point of the wind in their WR atmosphere models. This is identified with the effective temperature of the hydrostatic surface in our MESA models. Although the MESA outer boundary is formally defined at an optical depth of , this should not be confused with the photosphere of an optically thick WR wind, which can lie far above the hydrostatic stellar surface. Rather, the MESA radius approximately corresponds to the hydrostatic boundary underlying the PoWR atmosphere, making a reasonable approximation for implementing the S23 prescription.
The S23 recipe is
This physically-motivated temperature dependency demonstrates that for stars of equal mass, luminosity and composition (and hence and ), the cooler star, which is larger and has a lower surface gravity, loses its outer layers more easily and thus has a greater wind mass-loss rate.
The next temperature range covers a transition from the cooler WNL stars (which often have significant amounts of surface H) to the hotter cWR stars, which have a minimum classical WR temperature . Here, an interpolation between and ensures a smooth transition in the mass-loss rate. Gilkis et al. (2025) used for the S20 wind but the models of S23 either only explored temperatures as low as or the winds could not be driven by the hot iron bump at any lower temperatures. Hence is more appropriate for the S23 wind. Four different WR wind schemes with varying temperature limits are used and assessed. A summary of the parameters is given in Table 1. Schemes 1 and 4 are the most suitable implementations of the S20 and S23 winds, respectively, whilst schemes 2 and 3 act as reference cases for comparison. We do not apply scheme 3 to the binary star models and so it is not included in the simulated stellar populations in § 6.
| WR wind scheme # | cWR wind | /kK | /kK |
| 1 | S20 | 80 | 70 |
| 2 | S23 | 80 | 70 |
| 3 | S23 | 90 | 90 |
| 4 | S23 | 100 | 90 |
Table 1. Parameters of the different WR wind schemes examined in this work. Definitions of the wind types and temperatures are given in the text.
The next temperature regime is for the WNL stars, with temperatures between and . The so-called bi-stability jump refers to a change in the wind properties around owing to a shift in the ionisation of iron. It is calculated according to Vink et al. (2001). In this temperature range, the WNL mass-loss rate is found as the maximum of and , computed from Sabbah et al. (2022) for and , and from Sabbah et al. (2023) for . The mass-loss properties of the transition from optically thin O-star winds to optically thick WNh winds remain an active area of study. Recent hydrodynamically consistent PoWR calculations have further investigated this transition and provided updated mass-loss predictions as functions of the stellar parameters (Sabhahit, Vink & Sander 2026a). The precise temperature range over which the WNL and cWR prescriptions should be connected is uncertain, since neither prescription is designed to describe the full transition between these wind regimes. We therefore regard the transition temperatures in Table 1 as part of the wind implementation rather than fundamental physical boundaries. Schemes 2–4 allow us to assess the sensitivity of the evolution to this choice, with scheme 4 providing a smooth transition while restricting direct application of the S23 prescription to the temperature range in which hot iron bump is expected to drive winds.
The last temperature range, , applies to cool stars over a range of evolutionary phases. The mass-loss rate is taken as the maximum of a cool star rate (following de Jager, Nieuwenhuijzen & van der Hucht [13]) and .
WR star properties
A star is classified as a WR star when its spectrum displays strong emission lines. This occurs when the mass-loss rate is large relative to the size of the star. The strength of the emission features can be characterised by the transformed radius (Schmutz, Hamann & Wessolowski 1989),
where is a clumping factor which describes an inhomogeneous wind density (Hamann & Koesterke 1998). Physically, the higher density in clumps produces stronger emission than a smooth wind would at the same total mass-loss rate. This term was introduced in spectroscopic analyses to lower mass-loss rates by a factor of between 2 and 3, and thus aligns models more closely with observations (Nugis & Lamers 2000). Modern WR atmosphere codes such as PoWR use a depth-dependent clumping , where it rises from to some (e.g. by S20, and , 10 and 4 by S23).
In our models, WR stars are classified based on the scheme used by Gilkis et al. (2025). The main points are summarised here. A star is provisionally classified and assigned and based on its surface mass fractions of H, He, N, C and O:
WNh if , then , ;
WC if , , , then , ;
WO if , , , then , ;
WN otherwise, then , .
The star is then actually designated as a WR and given its classification if (Shenar et al. 2020a). From its definition in Eq. (4), note that stronger winds correspond to higher and thus lower .
Stellar evolution and WR formation
A sample set of stellar evolution tracks is presented on the Hertzsprung–Russell diagram (HRD) in Figure 2; these are for and use WR wind scheme 4. At , the lower mass stars of this study initially become more luminous and expand as they evolve away from the ZAMS, reaching the cool-supergiant region before subsequently evolving back towards higher effective temperatures. With increasing initial mass, the redward excursion becomes progressively less pronounced: the intermediate-mass models turn back towards the blue at increasingly high effective temperatures, while the most massive stars remain comparatively hot and compact throughout much of their evolution. In particular, the 99 and 135 stars undergo only modest redward evolution before evolving to higher effective temperatures. This behaviour reflects the increasing importance of stellar winds at high luminosities. These remove the outer hydrogen-rich layers before the stars can expand to the cool-supergiant regime.

Figure 2. Stellar evolution tracks for single stars with WR wind scheme 4, and ZAMS masses given in the legend. Only stars with an initial mass high enough for the WR phase to be reached are plotted. When a star is predicted to have a WR spectral appearance, it is emphasised with thick grey shading. Each track starts on the ZAMS, denoted by the black dashed line, and lines of constant stellar radius are denoted by the labelled grey dashed lines.
The evolutionary track of the 31 star in Figure 2 differs from those of more massive stars because it retains some surface hydrogen until the end of their evolution, whereas the higher-mass models become completely hydrogen-free. The sharp increase in luminosity seen towards the end of the higher-mass tracks occurs during core carbon burning and may be associated with structural readjustment during the advanced burning stages, when neutrino losses become increasingly important.
The effect of the different WR wind schemes on stellar evolution is presented in Figure 3. Single star tracks at are shown. Generally, scheme 2 leads to the greatest mass loss in the early WR phase, leading to lower luminosities later on. Because the S23 recipe is applied outside of its suggested temperature range for this scheme, these mass-loss rates may be overestimated and can be seen to influence the evolution substantially. We note that the tracks with scheme 3 resemble those of scheme 2 and provide a jagged evolution across the HRD. Therefore, we conclude that scheme 4, with its interpolation allowing for a smooth transition of wind mass-loss rates, is the most appropriate implementation of the S23 wind. For the highest-mass stars, the general evolutionary trend is not sensitive to the choice of wind prescription, though the luminosities at which the stars spend their WR phase vary by up to 25%. The intermediate mass stars in this study are least affected by the WR wind scheme choice. They show little variation between the tracks.

Figure 3. Stellar evolution tracks for single stars at , comparing the effect of the different WR wind schemes. The black dashed line is the ZAMS.
The effect of metallicity on stellar evolution is presented in Figure 4. Single star tracks with WR wind scheme 4 are shown. As expected, stars with higher have greater mass-loss rates, leading to a more rapid reduction in stellar radius and luminosity. This effect is more pronounced for the higher mass stars because their winds are predominantly radiatively driven and so exhibit a stronger -dependence. Examination of the tracks gives a hint as to how the metallicity affects which stars evolve to show WR-like spectral features: below a certain mass threshold at each metallic-

Figure 4. Stellar evolution tracks for single stars with WR wind scheme 4, comparing the effect of the different metallicities. The black dashed line is the ZAMS.
ity, WR stars are not predicted. These thresholds are presented in Table 2.
Figure 4 also displays an interesting feature for the intermediate-mass stars at with WR wind scheme 4. Rather than evolving rapidly to the highest temperatures, these stars spend an extended period with between 70 and 90 kK. This behaviour occurs near the transition between the WNL and cWR wind regimes, where the adopted mass-loss prescriptions have qualitatively different dependencies on the stellar parameters. In the S23 prescription, the cWR mass-loss rate depends steeply on temperature, approximately as at fixed and . So evolution towards lower temperatures within the regime in which S23 is applied strongly increases the cWR mass-loss rate and promotes further stripping of the envelope. However, below , the WNL prescription of Sabhahit et al. (2023) has no similarly strong explicit temperature dependence, with the enhanced wind rate instead depending steeply on luminosity and mass. The transition between these two wind regimes can modify the response of the star to further expansion or contraction and slow its evolution in effective temperature.
As also found for enhanced WNh winds by Sabhahit et al. (2023), strong mass loss governed primarily by the luminosity-to-mass ratio can lead to extended evolution over a relatively narrow temperature range. Consequently, stars near the WR formation threshold can spend a substantial fraction of their WR lifetime at between 70 and 90 kK, increasing their probability of being observed there. This could help explain the characteristic temperature distribution of WR stars in the SMC (§ 6).
A comparison between single and binary star evolution is presented in Figure 5, with initial conditions similar to those in Figure 2. For the highest masses, the single-star and binary tracks overlap as the stellar wind is sufficiently strong to strip away the envelope without requiring a companion. At lower masses, the tracks diverge when the stars expand enough to interact. The envelope stripping that ensues for the primary star causes it to lose its outer hydrogen and evolve blueward on the HRD, allowing the WR phase to be reached for lower masses.

Figure 5. Stellar evolution tracks at for the primary stars in binaries compared to single stars of the same initial mass, denoted by thin grey lines, with initial period and mass ratio .
The formation of WR stars is a -dependent phenomenon largely owing to the weaker stellar winds at lower . Of interest is the lowest initial mass (denoted by for single stars and for the primary stars in binaries) for which the WR phase is predicted to form. The results for single and binary stars are presented in Table 2. While the overall metallicity dependence is consistent with previous works ((Shenar et al. 2020a)), we find notable differences in the inferred minimum initial masses. In particular, at binary interaction can produce a WR star from a primary with an initial mass as low as for . This is more than below the corresponding threshold inferred by Shenar et al. (2020a). These objects are relatively low-mass WR stars that emerge only during the late stages of their evolution, which likely contributes to the difference between the inferred thresholds.
Binary interaction can help form WR stars through envelope stripping of the primary, leading to a lower initial mass threshold for them to form. Whilst it is commonly assumed that binary interaction dominates the formation of WR stars at low , Shenar et al. (2020a) have demonstrated that this is not necessarily true because both the emergence of WR features and strong wind mass loss depend on metallicity. At the two higher metallicities considered, is insensitive to the initial orbital period. At the lowest metallicity, however, the shorter-period binary has a substantially lower . This likely reflects differences in the mass-transfer history, although the relationship between earlier stripping and WR formation is non-trivial owing to its effects on the primary’s luminosity, effective temperature, and subsequent wind mass loss.
Ionising radiation
Massive stars emit vast amounts of energy into their surroundings owing to their high luminosities. Their high effective temperatures mean that a significant fraction of their emission is in the UV and can thus contribute to the ionisation of their environment. In particular, the key species to consider are neutral hydrogen atoms (H I) with ionisation energy , neutral helium atoms (He I) with , and singly ionised helium ions (He II) with ((Mohr, Newell & Taylor 2016)). This is important for our understanding of early-universe reionisation as well as the formation of ionised bubbles around star-forming regions ((Bianchi, Cristiani & Kim 2001; Krause et al. 2013)).
However, despite their high temperatures, WR stars do not necessarily emit a significant number of He II ionising photons. The transparency of the wind to photons with energies above 54 eV depends on the density of the wind and thus the mass-loss rate. In denser winds, He II recombines to He II, making the atmosphere opaque to these high energy photons ([85] (Schmutz, Leitherer & Gruenwald 1992)). Furthermore, the presence of line blanketing also reduces the emitted ionising fluxes from WR atmospheres ((Smith et al. 2002)).
We have used our stellar evolution simulations to provide estimates of the energy outputs of the models. To calculate the total energy emitted over a stellar lifetime, , we simply multiply the stellar luminosity at each time-step with the time-step duration and summed over all time-steps. For the single stars, is largely independent of for , whilst each decrease in metallicity from to 0.0056 to 0.00224 yields an increase in of about 25% for the higher mass stars (lower models generally have higher luminosities). The choice of WR wind scheme shows almost no effect on the total energy output of the stars at all metallicities.
Estimating the ionising energy outputs of the stars requires knowing both their luminosity and how that energy emission is distributed across the electromagnetic spectrum. To this end, the results of S23 have been applied. For He II ionising photons, we use a -dependent cut-off for their transformed mass-loss rate , estimated from Fig. 17 of S23 to be
If is larger than this limit, the wind is optically thick. Otherwise, the wind is optically thin. Following S23, a temperature is defined as the effective temperature at the point where an optical depth of is reached. This point typically defines the photosphere and thus dictates the emitted and observed spectra. S23 provide a fitting formula for in their equation 3. They derived this for their models so we have shifted down by 45% to be consistent with the assigned to our models (§ 2.2). This correction also has the benefit of avoiding any unphysical results when , because we generally expect the effective temperature to decrease outwards.
We define a radiative emission temperature for He II ionisation. For thin winds and for thick winds , motivated by considering whether the emergent flux originates from the hydrostatic surface or the outflowing wind. For H I ionising photons, no such wind density-transparency relationship is observed (Fig. 21 of Sander & Vink 2020), so at all times. A Planck distribution at is then used to estimate the fraction of energy greater than each ionisation energy limit at each time-step. An estimate of the total ionising energy output over stellar lifetimes, , is then found by summing over each time-step as above, now weighted by the ionising energy fraction. We note that the use of a Planck distribution provides only an approximate estimate of the ionising energy output. In particular, departures from local thermodynamic equilibrium (NLTE), line blanketing and absorption within the stellar wind can substantially modify the emergent spectrum, especially at the highest photon energies. Consequently, the quantitative predictions for the ionising energy output should be treated with caution. Our primary aim here is to illustrate how optically thick winds can shift the emergent spectrum towards lower effective temperatures, thereby reducing the ionising output, rather than to provide precise predictions of ionising fluxes. A more quantitative assessment would require detailed atmosphere calculations.
Because for H I ionising photons, the resulting does not differ significantly from , showing the same dependencies on and the WR wind scheme used. The increase of with initial stellar mass is more exaggerated than for because the higher mass stars are on average hotter throughout their lives and so a greater proportion of their emission is above the Lyman limit. Figure 6 presents the total He II ionising energy predicted to be emitted by the single stars evolved with the different metallicities and WR wind schemes. A much more complicated picture arises for the less massive stars of this study, for which is actually smaller at lower . This effect, noticeable for the single stars with at , arises because these single stars do not reach an evolved WR phase. They die as relatively cool supergiants, having never been hot enough to emit significant amounts of He II ionising radiation. In our approximate treatment, seems to saturate at , reflecting the suppression of the emergent high-energy radiation by the denser winds of more massive stars. Similar results arise for the binary stars, where shows a complicated dependency on the various stellar, binary and wind parameters.

Figure 6. Lifetime-integrated He II ionising energy output over stellar lifetimes for single star models, covering the range of metallicities and WR wind schemes. These were approximated with WR atmosphere results from Sander et al. [85], discussed in detail in the text.
Final masses
We consider the effects of the WR wind prescription on the final stellar mass. The final mass of a WR star significantly affects the appearance of a supernova if the star explodes (e.g. Gal-Yam et al. 2022), or the mass of a BH if there is no supernova (e.g. Gilkis et al. 2026).
Having gone through successive stages of nuclear burning and mass loss in winds and binary interaction, our stars finish at the end of core carbon burning with a final stellar mass . The short evolutionary timescales at the end of the evolution mean that the final masses in our models are approximately equal to the pre-core-collapse mass . Any further wind mass loss is expected to be negligible. However, eruptions can occur in the remaining months or years before core-collapse (Humphreys & Davidson 1994; Jiang et al. 2018; Cheng et al. 2024).
Figure 7 shows as a function of for the primary star in binaries with d and . As discussed, the stronger winds at high cause much greater mass loss, leading to significantly smaller final masses. In particular, for , the final mass saturates at about for . This is a result of the post-MS evolution of these high-mass stars being similar to that of their single-star counterparts (Figure 2), because their strong winds dominate the envelope stripping even in binary systems. This low saturation at disagrees qualitatively with previous works (e.g. Woosley, Sukhbold & Janka 2020; Winch et al. [107]). However the models of Woosley et al. (2020) start as naked He stars, having skipped the WNL phase entirely, so this discrepancy could result from the high uncertainty in the mass loss of VMSs. While Winch et al. [107] show that rapid rotation can substantially modify through rotationally enhanced and mechanical mass loss, their models span to , predominantly below the metallicities considered here, and have initial rotation rates of to , whereas our models are non-rotating. Further, is lowest at all for WR wind scheme 2. Again this is likely due to the misapplication of the S23 recipe for cWR stars. So it is promising to see quantitative agreement between the results of WR wind schemes 1 and 4.

Figure 7. Final mass as a function of initial mass of the stellar evolution models. Shown here are the results for the primary star of binary systems with an initial orbital period of and a mass ratio of . These depend on the choice of metallicity (different colours) and WR wind scheme (different markers and line styles).
Observed BH masses provide an additional constraint on the final stellar masses predicted by our models. In the Milky Way, the mass of the BH in Cygnus X-1 was estimated to be approximately by Miller-Jones et al. [55], exceeding the saturation mass of approximately found in our solar-metallicity models. However, a subsequent analysis by Ramachandran et al. (2025) revised this estimate to –, overlapping the range predicted by our solar-metallicity models. In the LMC, the BH masses inferred for LMC X-1 (; Orosz et al. 2009) and VFTS 243 (; Shenar et al. 2022a) are below the final stellar masses predicted for the more massive progenitors in our models. These comparisons are indicative rather than definitive, because the final stellar mass need not equal the BH mass and the observed systems may have followed different evolutionary pathways.
Probability-weighted stellar populations
We use our models to construct populations of WR stars at MW, LMC and SMC metallicities. A probability is assigned to each model based on estimated star formation statistics. The approach is similar to that of Gilkis et al. (2025). We use an initial mass function (IMF) of suitable for the high-mass stars investigated in this study (Salpeter 1955). Each discrete model mass represents the interval bounded by the midpoints between adjacent grid masses and its weight is therefore
The population is restricted to , which define the outer boundaries of the lowest- and highest-mass bins, respectively. For binaries in the parameter ranges considered here, we assume a flat distribution in the mass ratio, (Shenar et al. 2022b), and an initial-period distribution (Sana et al. 2012). We adopt and as the limits of these distributions. The sampled values and d divide these ranges at the midpoints between adjacent grid points (in and , respectively), with and obtained by integrating the corresponding distributions over these intervals, just as in Eq. (6). We assume a binary fraction (Sana et al. 2012). The weight of each track is then
Finally, each time-step is assigned a weight , after which the weights are normalised over the complete population:
The resulting population thus accounts for both the expected formation probability of each system and the time spent at each stage of its evolution. We have also restricted the weights to only include time-steps when the star is considered to be a WR star as outlined in § 2.2. Separate populations are constructed for each metallicity and WR wind scheme and are examined on the HRD and – plane during the WR phase.
Observations of Galactic single WN stars (Hamann et al. 2019), Galactic single WC and WO stars (Sander et al. [83]), a Galactic binary WN5h star Müller-Horn et al. 2026, LMC WN stars (Hainich et al. 2014), LMC binary WN stars (Shenar et al. 2019; Tehrani et al. 2019; Shenar et al. 2020b), SMC single WN stars (Hainich et al. 2015), SMC binary WN stars (Shenar et al. 2016) and WC and WO stars in all three populations (De Marco et al. 2000; Dessart et al. 2000; Crowther et al. 2002; Tramper et al. 2015; Tramper, Sana & de Koter 2016) have been analysed. Updated values for the SMC WN binary star AB 6 and the LMC WN binary star R 144, from Shenar et al. 2018 and Shenar et al. 2021, respectively, have been used. Characterising observed WR properties requires atmospheric and spectral modelling, for which these studies have used PoWR (Gräfener et al. 2002; Hamann & Gräfener 2003; Sander et al. 2015) or CMFGEN (Hillier & Miller 1998). These typically employ a -law velocity profile for the supersonic part of the wind. Some use a constant clumping factor and some use a depth-dependent .
Figure 8 compares the probability-weighted model populations with observed WR stars in the three metallicity environments. The models reproduce much of the observed WR luminosity range and predict concentrations of WN and WC stars towards the lower end of their expected temperature ranges, although substantial probability remains at higher temperatures than inferred observationally. The model’s probability peak at around to may partly reflect an overestimate of the effective temperatures, because recent work has shown that detailed atmosphere boundary conditions can substantially reduce WR effective temperatures at solar metallicity without significantly altering their internal structure (Voje, Palacios & Martins 2026). Conversely, observationally inferred WR temperatures may be underestimated, particularly for stars with optically thick winds, owing to uncertainties in the atmospheric and wind structure. Further, the hottest and most evolved WO stars with are both theoretically predicted and observed to be a rare but important class of WR stars.
Figure 8. Probability map of our WR stars with WR wind scheme 4, for (top), (middle) and (bottom), on the HRD with observed Local-Group WR stars overlaid. The observations are coloured based on their WR classification. Closed symbols denote single WR stars and open symbols denote WR stars in binary systems.
The LMC () population shows some agreement with our simulations, although the most luminous observed WR stars, reaching , are likely core-hydrogen-burning stars with initial masses beyond the upper limit of our models. At , a slight gap in luminosity in the range is predicted by our models. This gap might be a result of a limited number of simulations and larger model grids could provide a more accurate predicted distribution.
Under WR wind scheme 4 at , the intermediate mass WR stars spend extended periods at to , as noted in Section 3. This is particularly pronounced for stars near the WR star formation threshold, namely those at low and at (relatively) low mass in binary systems. Almost all of the observed WR stars in the SMC appear in this temperature range. This has previously been difficult to reproduce theoretically, and our new models with the most up-to-date WR wind prescription alleviate this discrepancy. We note also that the observed temperatures are more reliable for stars in the SMC because their winds are not as optically thick.
Figure 9 is constructed similarly to Figure 8 but displays the simulated and observed Galactic populations on the log –log plane. While the models span much of the observed range of mass-loss rates and luminosities, they also predict a population extending to lower mass-loss rates than those of most observed WR stars. The recent identification of a low-luminosity WR star by Müller-Horn et al. (2026) provides evidence that such objects exist, although their prevalence remains uncertain. This low-mass-loss population may represent an extension of the observed WR distribution that has largely escaped detection, or it may partly reflect the classification of model stars as WR stars when their winds are insufficiently strong to produce characteristic WR spectra.
Figure 9. Probability map of simulated WR stars, for and WR wind scheme 4, on the log –log plane with observed Galactic WR stars overlaid, similar in construction to Figure 8.
Summary and conclusions
We used the stellar evolution code MESA to model single and binary stars at Milky Way, LMC and SMC metallicities, incorporating four wind prescriptions, including the temperature-dependent mass-loss recipe recently developed by Sander et al. [85]. These models provide an opportunity to link theoretical wind-driving physics to observable WR properties across different environments.
WR formation thresholds depend on metallicity and binary interaction but are largely insensitive to the wind prescription. Binary mass transfer enables WR star formation from lower initial masses than single-star evolution. Stellar evolution differences emerge near the WR formation mass limit. With our wind scheme 4 (including the Sander et al. [85] temperature dependence) some stars stall between 70 and 90 kK, matching observed SMC WR temperatures that earlier models struggled to reproduce. Total and H-ionising fluxes show little dependence on wind prescription. The He II-ionising fluxes saturate at high masses because thick winds absorb high-energy photons. The population synthesis spans much of the observed WR luminosity and mass-loss-rate ranges, while some discrepancies remain in the effective-temperature distribution. Incorporating the updated temperature-dependent WR wind [85] in stellar evolution models improves agreement with observed WR temperature distributions, especially in low-metallicity environments, while preserving consistency in predicted luminosities, ionising fluxes and mass-loss rates. This represents a step forward in linking theoretical wind physics to WR population properties, though some aspects, such as binary interactions and eruptive mass loss, remain key uncertainties. Addressing these open issues will be key to fully understanding how WR stars form, evolve and contribute to the feedback and chemical enrichment of galaxies.
Acknowledgments
AG acknowledges funding from STFC grant UKRI4122. TS acknowledges support from the ERC under the European Union’s Horizon 2020 research and innovation program, grant agreement number 101164755 and from the Israel Science Foundation (ISF) under grant number 2434/24. CAT thanks Churchill College for his fellowship.
Data availability statement
The data underlying this article will be made available upon publication.
References
- [13]de Jager C., Nieuwenhuijzen H., van der Hucht K. A., 1988, A&AS, 72, 259
- [51]Li Z., Li Y., 2023, ApJS, 268, 51
- [55]Miller-Jones J. C. A., et al., 2021, Science, 371, 1046
- [83]Sander A. A. C., Hamann W. R., Todt H., Hainich R., Shenar T., Ramachandran V., Oskinova L. M., 2019, A&A, 621, A92
- [85]Sander A. A. C., Lefever R. R., Poniatowski L. G., Ramachandran V., Sabbahit G. N., Vink J. S., 2023, A&A, 670, A83
- [107]Winch E. R. J., Sabbahit G. N., Vink J. S., Higgins E. R., 2025, MNRAS, 540, 90

