The mass-loss rates of red supergiants at low metallicity: Detection of rotational CO emission from two red supergiants in the Large Magellanic Cloud

Using the PACS and SPIRE spectrometers on-board the Herschel Space Observatory, we obtained spectra of two red supergiants (RSGs) in the Large Magellanic Cloud (LMC). Multiple rotational CO emission lines (J=6-5 to 15-14) and 15 H2O lines were detected from IRAS 05280-6910, and one CO line was detected from WOH G64. This is the first time CO rotational lines have been detected from evolved stars in the LMC. Their CO line intensities are as strong as those of the Galactic RSG, VY CMa. Modelling the CO lines and the spectral energy distribution results in an estimated mass-loss rate for IRAS 05280-6910 of 3x10^-4 Msun per yr. The model assumes a gas-to-dust ratio and a CO-to-H2 abundance ratio is estimated from the Galactic values scaled by the LMC metallicity ([Fe/H]~-0.3), i.e., that the CO-to-dust ratio is constant for Galactic and LMC metallicities within the uncertainties of the model. The key factor determining the CO line intensities and the mass-loss rate found to be the stellar luminosity.


INTRODUCTION
At the end of their lives, stars lose a large fraction of their mass from their surfaces. High-mass (> 8M ) stars evolve into the redsupergiant (RSG) phase, and lose a large amount of mass through stellar winds. RSGs are considered to be the major progenitors of type II-P supernovae (Smartt 2009). High-mass stars lose a large amount of mass during the RSG phase, determining the RSG mass at the time of a SN explosion (Meynet et al. 2015). In order to fully understand post-main sequence stellar evolution, it is important to characterise the mass loss processes of RSGs.
The hydrodynamical simulation of RSG mass-loss is not yet fully modelled theoretically, but the basic mechanisms are believed to be the same as for asymptotic giant branch (AGB) stars, which are their lower-mass evolutionary counterparts. AGB stars are more populous than RSGs and thus better investigated. It is widely accepted that stellar winds from RSGs and AGB stars are dust driven: stellar pulsations elevate some gas from the stellar surface, and from the levitated atmosphere dust grains are condensed (e.g. Höfner & Dorfi 1997). The dust grains experience radiation pressure from the central star, triggering outward movement. The dust motion drags the surrounding gas, driving stellar winds of both gas and dust.
Due to this dust-driven mechanism, the mass-loss rates from RSGs and AGB stars should depend on two major parameters: the luminosity of the central star, and the metallicity of the galaxy.
The luminosity of the central star determines the radiation pressure on the dust grains. Wachter et al. (2008) predicted that the mass-loss rate should correlate with luminosity. While Galactic RSGs have large uncertainties in measuring their luminosities, extragalactic RSGs have well determined luminosities, as the distance to the stars can be adopted as the distance to the galaxy. With its close distance of 50 kpc (Pietrzynski et al. 2013), the Large Magellanic Cloud (LMC) can be used to study the impact on luminosity of mass-loss rates.
Secondly, the mass-loss rate is expected to approximately correlate with metallicity, which can be represented by the metallicity of the parent galaxy. RSG mass loss is driven by radiation pressure on dust grains (e.g. Justtanont & Tielens 1992). Dust grains from oxygen-rich stars are composed of metals, such as Fe, Si, O, Al, hence, a lower metallicity should result in a smaller mass of dust grains. For a lower dust mass, the integrated cross-section for radiation pressure on dust grains should reduce, resulting in a lower mass-loss rate at least for oxygen-rich stars (Bowen & Willson 1991). Indeed, Marshall et al. (2004) measured the expansion velocities of LMC RSGs, using OH masers, and found lower expansion velocities for LMC RSGs compared to their Galactic counterparts, suggesting a metallicity effect on the dust radiation-pressure driven mass-loss rate.
To date, studies of mass-loss rates in LMC RSGs have been based on their IR spectral energy distributions (e.g. van Loon et al. 1999;Groenewegen et al. 2007;Sargent et al. 2010;Riebel et al. 2012), measuring their dust mass-loss rate, whereas historically, Galactic AGB stars and RSGs have been studied more intensively, using CO rotational lines (e.g. Knapp & Morris 1985;Ramstedt et al. 2008). Complementary studies of CO emission in the LMC RSGs are important.
With the aim of investigating the physics and chemistry of the circumstellar envelopes of evolved stars at low metallicity, we have measured molecular emission lines at far-infrared (IR) and submillimeter wavelengths, including CO rotational emission from two IR RSGs in the LMC. Additionally, H 2 O lines have been detected. This is the first detection of CO rotational emission lines from RSGs in the LMC to our knowledge. We report the analysis of these molecular lines.

OBSERVATIONS AND DATA REDUCTION
We observed two RSGs with the Herschel Space Observatory (Pilbratt et al. 2010). The primary target, IRAS 05280−6910 was chosen because it is the brightest LMC RSG at far-infrared wavelengths (Boyer et al. 2010), with a 250 µm flux of 205 mJy. Its mid-infrared spectrum shows 10 µm silicate band in absorption with a peak of the SED at about 25 µm. OH and H 2 O masers have been detected (Wood et al. 1992;van Loon et al. 2001). IRAS 05280−6910 was found towards the cluster NGC 1984(van Loon et al. 2005. The second target is WOH G64, a well-known OH/IR star in the LMC (Elias et al. 1986). This star has a strong silicate absorption in its mid-infrared spectrum, accompanied by OH masers (Elias et al. 1986;Roche et al. 1993;Wood et al. 1986). Based on mid-infrared interferometric observations, Ohnaka et al. (2008) suggested the presence of a face-on torus around WOH G64. This is one of the most luminous mass-losing RSGs in the LMC, with a bolometric luminosity of ∼ −9 mag (van Loon et al. 1999;Levesque et al. 2009) The submillimeter spectrum of IRAS 05280−6910 was obtained, using the SPIRE Fourier Transform spectrometer (Griffin & et al. 2010;Swinyard et al. 2010) on board the Herschel Space Observatory. The data were acquired on 2012 June 16th and 17th (OBSID: 1342247097 and 1342247098) in a Herschel Open Time Program (OT1 − mmatsuur − 1). The total duration of the observation was 13752 s×2. The SPIRE FTS simultaneously covers the short wavelength band (SSW; SPIRE Short Wavelength Spectrometer Array; 190-313 µm; 957-1577 GHz) and long wavelength band (SLW; SPIRE Long Wavelength Spectrometer Array; 303-650 µm; 461-989 GHz), yielding a frequency coverage from 444 to 1540 GHz, with a spectral resolution of 1.2 GHz. The FTS sensitivity varies across the spectra, with the best sensitivity available region at about 700-800 GHz (Swinyard et al. 2014), corresponding to the CO J=6-5 and 7-8 lines. Unfortunately, the SLW spectrum suffered from a temperature drift in the SPIRE detectors, so that the continuum level is unreliable. We discorded the spectrum below ∼700 GHz, and added some uncertainties to the SLW spectra above ∼700 GHz. The spectrum was reduced in HIPE version 11 (Ott 2010).
The PACS (Poglitsch et al. 2010) spectrum of IRAS 05280−6910 was obtained on 19th March 2013 (OB-SID: 1342267866), in a Herschel Open Time Program (OT2 − bsargent − 1). The program targeted the 173 µm CO J=15-14 and 186 µm CO J=14-13 lines. The target CO lines were covered by the first-order grating with a resolving power of R ∼ 1500, while the second-order grating covered the 92 µm H 2 O 6 43 -6 34 line, with a resolving power of R ∼ 2000. The chopped-nodded PACS range spectroscopy mode was used. The total duration of the observation was 2240 s.
PACS obtained the spectrum of WOH G64 on 2013 April 12th (OBSID: 1342269921), as part of the same programme (OT2 − bsargent − 1). The spectral setting was the same as that for IRAS 05280−6910, and the total observing time was 6712 s.
The PACS spectra were reduced with hipe version 13. The absolute flux calibration was performed relative to the telescope background. Both targets were considered point-like, i.e. the fluxes were extracted from the central 'spaxel' only, incling proper beam correction. Calibration set 65 was used for all reduction and calibration steps.  The SPIRE FTS spectra of the LMC red-supergiant, IRAS 05280-6910, compared with the SPIRE FTS spectra of the Galactic red-supergiant, VY CMa. The spectra were scaled for clarity. The spectrum of IRAS 05280−6910 is unapodized, with the detected lines appearing as sharp peaks, but with instrumental line shape of a sinc function included. The spectrum VY CMa was apodized, i.e. instrumental line shape was removed, but the spectral resolution was lowered.

SPECTRA AND DETECTED MOLECULES
The detected CO lines are displayed in Fig. 2. The spectra were fitted using sinc instrumental line shape (Naylor et al. 2014). A systematic velocity of 270 km s −1 for the LMC (Marshall et al. 2004) was adopted.
Our PACS programme targeted two CO transitions (J=14-13 and 15-14) in the first grating order. As a bonus, the 92 µm H 2 O line was detected in the second order spectrum. Figure 3 displays the spectra of IRAS 05280−6910 and WOH G64, showing all three lines from IRAS 05280−6910 while only CO J=15-14 was detected from WOH G64. The lines observed by PACS were fitted by Gaussians with a quadratic underlying continuum, allowing the line strength, line centre, line width, and coefficients of the quadratic continuum all to be free parameters. These gaussians were fit to the lines plus surrounding long-and short-wavelength continuum, except that the wings of the 92 µm H 2 O line of IRAS 05280−6910 were removed from the fitting process, in order for the Gaussian to converge to a sensible fit. The Gaussian fittings are plotted with red dotted lines in the figure, and the line intensities were estimated by integrating the Gaussian profiles. The measured line intensities are summarised in Table 2.

Spectral Line Energy Distributions
In order to examine the excitation temperature and column density of CO, Spectral Line Energy Distributions (SLEDs) were used. Fig.4 shows the SLED of IRAS 05280−6910. The single CO line measurement of WOH G64 has also been plotted. The figure also shows Herschel measurements of SLEDs of the Galactic RSG  Khouri et al. 2014b). The SLEDs of VY CMa and W Hya were scaled to the LMC distance, by adopting their stellar distances of 1.14 kpc and 78 pc, respectively (Choi et al. 2008;Knapp et al. 2003).

Non-LTE line radiative transfer modelling
In order to derive physical parameters from CO line intensities, a non-local thermodynamic equilibrium (non-LTE) line radiative transfer code was used to fit the SLEDs. As only IRAS 05280−6910 has multiple CO transitions detected, we focus our modelling effort on this RSG. The first step was to construct a dust radiative transfer model to fit the spectral energy distribution (SED). This step provided some essential parameters for the CO modelling, such as dust temperature structure, which has an effect on the heating balance between dust and gas collisions. Modelling the SED with dusty provides the inner radius, the dust radial density and the temperature distribution of the dust envelope. These parameters were also converted into a dust mass-loss rate, and, hence, gas mass-loss rate, with an assumption of the gas-to-dust ratio. Boyer et al. (2010) already presented modelling of the SED of IRAS 05280−6910 using 2-dust code (Ueta & Meixner 2003). We started by reproducing the 2-dust results with Dusty (Ivezic & Elitzur 1997), as the Dusty model is one of our default input models to CO modelling. We used the infrared flux measurements of IRAS 05280−6910 collected by Boyer et al. (2010), including 3-24 µm photometric data from Spitzer Magellanic Survey, SAGE (Meixner et al. 2006), the measurements of Herschel Magellanic Survey, HERITAGE (Meixner et al. 2013) covering 100-500 µm, and at 5-35 µm a Spitzer IRS spectrum (SAGE-spec; Kemper et al. 2010). We used K s-band magnitudes from the VISTA Magellanic Cloud survey (VMC; Cioni et al. 2011), prior to their public release of the magnitude of this star. The PSF fitting resulted in a magnitude of 12.604±0.080 in the K s-band. It was converted to a Vega-system magnitude following the procedure described in Rubele et al. (2015), then con-verted to a flux in Jy. The source was not detected in the VMC z and J-bands.
An effective temperature of 3000 K was adopted for IRAS 05280−6910 (Boyer et al. 2010). This is similar to the estimated temperature of WOH G64 and VY CMa which have similar mass-loss rates of ∼ 10 −4 M yr −1 . The effective temperature of VY CMa has been estimated to be in the range of 2700 K (Monnier et al. 2004) to 3650 K (Massey et al. 2006). Elias et al. (1986) estimated a spectral type of M7.5 for WOH G64, corresponding to an effective temperature lower than 3450 K (=M5 ; Levesque et al. 2005), while Ohnaka et al. (2008) suggested to be 3200-3400 K. Davies et al. (2013) argued that the effective temperature of red-supergiants should be higher than previously thought by ∼500 K. Our experimental modelling with the effective temperature increased to 3500 K resulted in a negligible (< 5 %) increase in CO line intensities. We therefore adopted the 3000 K effective temperature from Boyer et al. (2010) but little difference in the CO modelling results for an effective temperature of 3500 K.
Our Dusty fitted results for IRAS 05280−6910 are displayed in Fig.5. The overall shape of the SED was fitted with silicate dust emission. As found by Boyer et al. (2010), our modelling produced silicate absorption stronger than that observed. This could be due to the outflow being slightly asymmetric, as recently found  , with a single CO line measurement for WOH G64. The SLEDs of VY CMa and W Hya were scaled to the LMC distance, by adopting their stellar distances of 1.14 kpc and 78 pc, respectively (Choi et al. 2008;Knapp et al. 2003 Table 3, which are consistent with those obtained from Boyer et al. (2010)'s 2-dust modelling. We evaluated the Dusty input parameters with a χ 2 -analysis. We found that T dust and τ are coupled parameters, and the uncertainty of the dust temperature is ±50 K, and that of the optical depth at 0.55 µm (logτ 0.55 ) is ±0.1, with increasing dust temperature requiring higher dust optical depth. We used dust optical constants for amorphous silicates (Ossenkopf et al. 1992), and the MRN grain size distribution with an index of 3.5, and grain size of 0.005-1 µm (Mathis et al. 1977). Assuming a gas-to-dust ratio of 500, which is expected from the abundance of refractory elements in the LMC ISM (Gordon et al. 2014), the SED modelling resulted in a gas-mass loss rate of 3 × 10 −4 M yr −1 . That carries 30 % uncertainty, according to the Dusty manual 1 .
The second part of our modelling was to reproduce the CO rotational lines using the non-LTE line radiative transfer calculation and level population code, SMMOL (Rawlings & Yates 2001). The SMMOL code adopts the accelerated lambda integration (ALI) scheme (Scharmer & Carlsson 1985;Rybicki & Hummer 1991) with an ability to solve the radiative transfer problems in optically thin and thick circumstellar envelopes. Solving level populations requires cross sections for molecular collision, and we adopted the CO-H 2 cross sections calculated by Yang et al. (2010), with molecular data from Müller et al. (2005). The code has been used for modelling the SLEDs of CO from J =4-3 to 22-21, as well as for H 2 O masers in the Galactic RSG, VY CMa ). The majority of the SMMOL input parameters for IRAS 05280−6910 were constrained from SED modelling work prior to the CO modelling. Integrating over the SED provided luminosity of 2.2 × 10 5 L (Boyer et al. 2010). As discussed above, an effective temperature of 3000 K was adopted for IRAS 05280−6910 (Boyer et al. 2010). These two parameters yielded a radius for the star (R star ) of 1.21 × 10 14 cm (=1738 R ). The Dusty model provided the inner radius, radial density and temperature structure of the dust envelope, as well as the optical depth (τ 0.55 ). The parameters used for the IRAS 05280−6910 non-LTE modelling are listed in Table 3. Another important parameter is the gas terminal velocity, which was adopted to be 17 km s −1 , based on OH maser measurements (Marshall et al. 2004). The majority of our input parameters were already well constrained from previous studies.
The CO-H 2 abundance ratio is affected by the metallicity. A chemical model for the oxygen-rich AGB stars show that essentially all available C atoms are locked up to CO (Willacy & Millar 1997), so that the C abundance with respect to H 2 gives the CO abundance. Scaling with the solar-abundance values for carbon with respect to hydrogen (8.43; Asplund et al. 2009), and taking [Z/H] = −0.3 for the LMC, the estimated CO-H 2 abundance is 2.7 × 10 −4 . Overall, the model predicts that the high-J CO line intensities scaled with the CO abundance. i.e. a metallicity effect is found for the high-J CO line intensities. The adopted CO/H 2 abundance ratio was based on the C and O abundances of the Sun, scaled to the LMC metallicity, however, C and O abundances of high-mass stars can be potentially modified due to nuclear synthesis (Meynet et al. 2015). The current work ignores this effect.
Three parameters were adopted from the VY CMa modelling : the temperature and velocity structure gradient indices, and the turbulent velocity. A turbulent velocity of 1 km s −1 was adopted (Decin et al. 2006;Matsuura et al. 2014). The kinetic temperature gradient index α, where T kin ∝ (r/R inner ) −α , has not yet been determined for circumstellar envelopes at low metallicity. The balance between cooling and heating processes determines the temperature gradient. Two major cooling processes are adiabatic cooling and H 2 O line emission (Decin et al. 2006). The H 2 O abundance has a metallicity dependence, as it is limited by the available oxygen abundance. The H 2 O cooling is expected to be less efficient at low metallicity. On the other hand, key gas heating processes in the circumstellar envelope are associated with collisions with dust grains (Decin et al. 2006). At low metallicity, the dust-to-gas ratio is small; hence the heating rate is expected to be low, too. Depending on the balance between these heating and cooling processes, the temperature gradient can change slightly at low metallicity. However, we found that changing α at R > R dust from 0.5 to 0.6 made less than a few per cent difference to the predicted CO line intensities. Such a small difference in the gas gradient in the circumstellar envelope is not crucial for CO modelling. We adopted of 0.6, the α value found for for VY CMa , and the adopted temperature structure is plotted in Figure 6. Similarly, we examined α at R < R dust , by changing the value of VY CMa (0.15) to 2.5. There was negligible difference in the predicted CO line intensities. The value of α=0.25 was used at R < R dust .
The velocity structure parameter, γ describes the velocity structure of the wind acceleration region. However, the CO emit- ting region (approximately r > 100R * ) is outside of the wind accelerating region, so γ is not important for the CO modelling. Experimentally, when we changed γ up to 5, the obtained CO line intensities remained unchanged, because the velocity structure at r > 100R * does not change. Constraining γ requires measurements of the velocity structure in the inner region. That has been estimated for Galactic red-supergiants , suggesting γ to be less than 1. We adopted γ = 0.2, following our modelling of VY CMa .
A density (ρ) law index β = 2, where ρ ∝ (r/R inner ) −β , is adopted for the gas with a constant expansion velocity. Although this is not the case near the velocity accelerating region, because this inner part does not affect the observed CO line intensities, it is not a major issue.
The outer radius (R out ) was chosen to be large enough that predicted CO line intensities were unaffected by R out . As J=6-5 and higher CO transitions have their line forming regions much closer to the central star than low-J CO lines (1-0 or 2-1) (Khouri et al. 2014b), these high-J line intensities are less sensitive to the choice of R out . The R out is the value corresponding to the dissociation radius of CO due to ISM UV radiation. In the LMC, a higher ISM UV radiation field is expected than for the Galactic ISM, due to low dust extinction. Hence, UV photons might penetrate further into the circumstellar envelope, dissociating CO at smaller radii in the LMC RSGs than for the Galactic counterparts McDonald & Zijlstra (2014). Fortunately, such a low metallicity effect would not be important for the high-J CO line intensities.
In summary, although there are several parameters in SMMOL modelling, key input parameters are the gas mass-loss rate and CO/H 2 abundance ratio, which determine overall height of CO line intensities, and temperature index α that manipulates the tilt of the CO SLED.

Modelling results
Our CO modelling results for IRAS 05280−6910 are displayed in Fig. 4. The measured CO line intensities are well fitted, particularly J=11-10 to 15-14. By modelling CO emission lines, CO massloss rate is determined, which is used to estimate gas mass-loss rate, assuming CO/H 2 abundance ratio. We found that the estimated mass-loss rates of IRAS 05280−6910 from the dust modelling to the SED and from modelling to CO emission are consistent.
The metallicity affects both the CO/H 2 abundance ratio and Silicate (Ossenkopf & Henning 1994) 3 ± 1 × 10 −4 gas-to-dust 500 (Gordon et al. 2014) log τ 0.55 0.56±0.1 SMMOL R * (cm) 1.21 × 10 14 (calculated from the luminosity and T * ) R in,molecules (cm) 1.21 × 10 14 (= R * ) R in,dust (cm) 2.07 × 10 16 (adopted from the Dusty output) R out (cm) 2.07 × 10 18 1.0 number of CO/H 2 2.7 × 10 −4 log τ 0.55 1.56 (adopted from the Dusty output) M g (M yr −1 ) 3 × 10 −4 (consistent to the Dusty output) gas-to-dust 500 (Gordon et al. 2014) v term (km s −1 ) 17 (Marshall et al. 2004) v in (km s −1 ) 4.0 R in,velocity (cm) 1.283 × 10 15 γ (velocity gradient) 0.2 T sub (K) 20 T : temperature, R: radius, τ 0.55 : optical depth at 0.55 µm,Ṁ g : gas mass-loss rate, β: density law index, (r/R inner ) −β , α: kinetic temperature law index, (r/R inner ) −α , v term : wind terminal velocity, v in : wind inner velocity, R in,velocity : radius where the wind velocity of v in starts, γ: velocity law index, v(r) = v inner + (v ∞ − v inner ) (1 − R inner,dust /r) γ , for r R inner,dust , and T sub : sublimation temperature of CO. dust-to-gas ratio in a similar manner. Both ratios are essentially scale with the metallicity, represented by [Fe/H]. The gas-to-dust ratio accounts for the available refractory elemental abundance. Assuming that silicates are the major dust component in oxygen-rich circumstellar environment, such as the RSGs, and that the silicon abundance limits the total dust mass, the estimated gas-to-dust mass ratio would be about 450. We adopted Gordon et al.'s (2014) estimate of 500, which is consistent with the above estimate. Similarly, the CO-H 2 abundance ratio accounts for the number of carbon atoms, and this parameter also scale to [Fe/H]. Essentially, the  Khouri et al. (2014a), Loup et al. (1993) values of 'dust-to-CO ratio' are the same both with solar and LMC metallicities.

CO emission
With the Herschel SPIRE and PACS spectrometers, we have detected CO emission from IRAS 05280−6910 and WOH G64, with multiple CO detections from IRAS 05280−6910. We modelled the IRAS 05280−6910 SED with a dust radiative transfer code, constraining the mass-loss rate and the structure of the circumstellar envelope. These dust model parameters were used to further model the CO rotational lines, resulting in good fits to the lines. Although the LMC has approximately half the solar metallicity (Cole et al. 2005), there is no obvious metallicity effect on CO line intensities between Galactic and LMC red-supergiants. Figure 4 shows the measured and modelled CO SLEDs of the Galactic red-supergiant VY CMa ) and the Galactic AGB star W Hya (Khouri et al. 2014b). The overall shapes of the CO SLEDs are alike for all three stars, showing a sharp increase of CO at J=1-0 to 4-3, and flattening out at higher J. No clear evidence of reduced CO intensity at lower metallicity is found in our limited sample. Lagadec et al. (2010) reported the detection of CO rotational lines from carbon-rich AGB stars, which are likely to be linked with the Sagittarius Dwarf Spheroidal Galaxy (Sgr dSph) stream. The Sgr dSph has a metallicity of [Fe/H]∼ −1.1 (van den Bergh 2000). They concluded that there is no reduction in the gas mass-loss rates from carbon-rich AGB stars at this low metallicity. In the case of carbon-rich AGB stars, the main composition of the dust is carbon, and carbon has been synthesised and dredged up to the surface (e.g. Karakas et al. 2009), and radiation pressure on carbonaceous dust grains can trigger mass loss. Hence, it is not totally surprising that no correlation was found between mass-loss rates and metallicities for carbon-rich AGB stars.
By contrast, the dust mass in oxygen-rich AGB stars and RSGs is expected to be restricted by the number of silicon atoms. The silicon abundance is intrinsic to the star, i.e., limited by the silicon atoms incorporated the stars at the time of stellar formation, although silicon might be synthesised during the very last few hundred years of the RSG phase (Weaver & Woosley 1980). Reduced mass-loss rates are predicted at low metallicities for oxygen-rich AGB stars and RSGs (Bowen & Willson 1991). Therefore, it is more surprising to see no metallicity effects on the mass-loss rates for RSGs than it is for carbon-rich AGB stars.
Instead, we found that the CO line intensities and the massloss rate largely depends on the luminosity/luminosity class. The AGB star, W Hya, which has about a factor of 50-100 lower lumi-nosity than VY CMa and IRAS 05280−6910 (Table 4), has about a factor of 1000 weaker CO ladder than the other two. The luminosity appears to play the dominant role in CO line intensities and mass-loss rates.

Mass-loss rates and luminosities at low metallicity
We found from our small sample that CO line intensities and gas mass-loss rates tend to increase with higher luminosities. In order to further investigate the impact of luminosity, as well as metallicity, on the mass-loss rate of evolved stars, we compare here the massloss rates of these stars with that of a larger sample.
The sample of Galactic stars was taken from De  with 47 AGB stars and RSGs. We uses this sample because they used both infrared SED and CO lines in order to derive the mass-loss rates. The Luminosities of Galactic stars were estimated from a period-luminosity relation. For LMC stars, the sample of the 172 oxygen-rich stars from Groenewegen et al. (2009) was used, where the mass-loss rate estimates were obtained by infrared SED modelling. They assumed an expansion velocity of 10 km s −1 and a gas-to-dust ratio of 200 for all AGB stars and RSGs. These assumptions can provide a factor of a few uncertainty in their mass-loss rates, but do not have a critical impact on our systematic analysis. Figure 7 shows the mass-loss rate vs. luminosity relation for the LMC and Galactic evolved stars. The Galactic evolved stars follow an increasing trend of mass-loss rate with higher luminosity. This trend was fitted with a 'robust' least deviation method by removing outliners (Press et al. 1992), yielding log(Ṁ g ) = −10.79 + 1.29 log(L * ). (1) The fitted line is indicated in Fig. 7. Both VY CMa and W Hya, which have Herschel CO SLEDs, follow this line. In contrast, the LMC sample scatters across Fig. 7. Although the majority of LMC stars are placed lower than the line fitted to the Galactic objects, two stars we studied (IRAS 05280−6910 and WOH G64) follow this Galactic trend. Further, these two stars have among the highest mass-loss rates in the LMC. Actually, the Galactic line approximately captures the upper range of mass-loss rates at a given luminosity amongst the LMC sample.
The fitted mass-loss rate vs. luminosity relation for Galactic stars is similar to that found by van Loon et al. (2005) for LMC stars. Their analysis of 23 stars in the LMC produced the relation plotted as a dashed line in Fig.7. The line plotted in the figure is for T e f f =3500 K. For a T e f f =3000 K star, the mass-loss rate is a factor of 2.6 higher at a given luminosity, which is small compared with scatter in the sample. van Loon et al. (2005)'s LMC fit follows a similar trend to that found here for the Galactic stars, but this fit indicates only the upper range of mass-loss rate at a given luminosity for the larger LMC sample (Groenewegen et al. 2009).
The question is why LMC evolved stars show a much larger scatter of mass-loss rate at a given luminosity than Galactic objects. We argue that this is due to biases in the samples. Both the LMC and Galactic stars could have a large scatter, but Galactic sample tend to have picked up bright CO objects. De 's study was optimised for verifying models for circumstellar envelopes with firm CO detections, rather than optimising for a volume-limited CO line survey. The majority of the CO sample of De  were taken at JCMT or APEX (Kemper et al. 2003). As Kemper et al. (2003) selected bright samples from Loup et al. (1993)'s CO catalogue, it is possible that mass-loss stars with weak CO lines have not been included in their target list. Hence, the Galactic sample could appear at the an upper end of the mass-loss vs. luminosity relation.
As there are no systematic LMC studies of CO emission beyond our 2 objects, studies of mass-loss rates in LMC evolved stars have relied on infrared SEDs. The LMC sample here is from Groenewegen et al. (2009), who assembled Spitzer IRS spectra from six different observing programmes. The targets of these programmes were selected mostly from the 2MASS near-infrared survey (Skrutskie et al. 2006), or from a combination of 2MASS and MSX (Egan et al. 2001), with the exception of a few objects with optical light curves (Sloan et al. 2008). The near-infrared and optical surveys are efficient for detecting evolved stars with relatively low massloss rates, but inefficient of detecting dust-embedded AGB stars and RSGs. For example, IRAS 05280−6910 does not have a 2MASS counterpart. The surveys detected a range of stars with mass-loss rate below > 10 −5 M yr −1 but only a few stars with L > 10 5 L , and mass-loss rate > 10 −6 M yr −1 was found in Groenewegen et al.'s (2009) sample. We argue that due to the selection methods, Groenewegen et al.'s (2009) LMC sample does not contain RSGs with high optical depth dust shells.
The reason for the wide scatter in mass-loss rate and luminosity could be very interesting. One of the possibilities is that the AGB stars that experience the third-dredge up have an enhanced mass-loss rate (e.g. Vassiliadis & Wood 1993), reaching the upper range of the mass-loss rate and luminosity relation. Mauron & Josselin (2011) reported a metallicity dependence for IR-SED mass loss rates, using RSGs in the Large and Small Magellanic Clouds (Bonanos et al. 2010). The mass-loss rates would be scaled by the metallicity (Z/Z ) 0.7 . Their definition of mass-loss rate was based on K s − [12], a good indicator of dust mass-loss rate. Such a metallicity dependence was not detected from a comparison on three Herschel CO SLEDs nor from a comparison between the results by Groenewegen et al. (2009) andDe Beck et al. (2010). Detecting metallicity effects could require a much larger CO sample.

The contribution of high mass-loss red-supergiants to the global dust and gas source in the LMC
Evolved stars, both AGB stars and RSGs are considered to be important sources of dust and gas of the ISM (Gehrz 1989;Dwek 1998;Tielens 2010) After Spitzer's launch, infrared colours and magnitudes were used to estimate individual mass-loss rates of evolved stars, and global gas and dust inputs from evolved stars into the ISM estimated for studied in the Large and Small Magellanic Clouds (e.g. Matsuura et al. 2009;Srinivasan et al. 2009). These studies found that a few high mass-loss rate stars are dominate the global inputs of dust and gas from evolved stars. The mass loss rate of IRAS 05280−6910 is at the high-end range even amongst RSGs, and it must be an important source of gas and dust. Matsuura et al. (2013) used 2MASS K s magnitudes for object classifications for high mass-loss rate RSGs, but IRAS 05280−6910 does not have 2MASS measurements. Similarly, Boyer et al. (2012) and Riebel et al. (2012) also used 2MASS photometry for their target selection. So IRAS 05280−6910 was excluded from the accounting of the global gas and dust inputs into the ISM in these two studies.
High mass-loss rate stars, such as IRAS 05280−6910 can contribute a significant fraction of the total dust and gas input from evolved stars into the ISM. Matsuura et al. (2013) estimated the total gas input rate from AGB stars and RSGs to be 1.5×10 −2 M yr −1 . With a gas mass-loss rate of 2 × 10 −4 M yr −1 , IRAS 05280−6910 contribute 2 % of the total gas input from LMC evolved stars. Boyer et al. (2012) estimated the total 'dust' input rate from AGB stars and RSGs to be to be 1.4 × 10 −5 M yr −1 . The dust mass-loss rate of IRAS 05280−6910 is estimated to be 5 × 10 −7 M yr −1 , contributing about 4 % of the total dust input from evolved stars. If reasonable fractions of dust grains from RSGs may survive subsequent supernova explosions, high mass-loss rate RSGs can account for a substantial fraction of input to the global gas and dust budget of the LMC.
The dust contributions from RSGs to the ISM dust are subject to subsequent dust destruction by supernova (SN) shock waves, and that is largely uncertain. Theoretical estimated values of the dust survival rate depends on models, from 20-62 % (Silvia et al. 2010 (Barlow et al. 2010;Indebetouw et al. 2014). In contrast, the dominant energy in the Crab Nebula is currently from its pulsar wind nebula. In the older supernova remnant, the Cygnus Loop, an enhanced C abundance in shocked region suggests the destruction of carbonaceous dust Raymond et al. (2013), but this could be swept-up ISM dust. Temim et al. (2015) and Lakićević et al. (2015) studied supernova remnants (SNRs) in the Large Magellanic Clouds, and estimated that dust overall destroyed by supernovae. Lakićević et al. (2015) found that the temperature of SNR dust emission to be higher than that of ISM dust, which affect their dust mass analysis. In the optically thin case, the thermal emission from warmer SN dust can dominate the total emission, hiding weaker emission from colder ISM dust. This effect distort the measured ISM dust mass towards the SNRs rather than the ISM dust mass being actually lower towards the SNRs due to SN shocks destroying surrounding ISM dust grains. A challenge remains in measuring dust mass destroyed by supernovae.

SUMMARY
We have detected far-infrared and sub-millimetre CO and H 2 O lines from two red-supergiants in the LMC. These are the first detections of these lines in LMC evolved stars.
Modelling with radiative transfer codes of the dust and line emission resulted in good fits to the infrared-SED and CO transitions of IRAS 05280−6910. A mass-loss rate of 3 × 10 −4 M yr −1 was obtained, and a value is at the high end of mass-loss rates found for red-supergiants. A gas-to-dust ratio of 500 and a CO/H 2 abundance of 2.7×10 −4 were adopted for the modelling. The gas-to-dust ratio was adopted from the value estimated for the LMC ISM, but is consistent with the solar neighbourhood value scaled by the LMC metallicity. The CO/H 2 abundance was adopted from the Galactic value scaled by the LMC metallicity. Our good modelling results support CO/dust ratio appears to be constant in the Galactic and LMC RSGs.
There is a general increasing trend of mass-loss rate with higher luminosity. The LMC red-supergiants with CO detections appear at the an upper range of mass-loss rate at a given luminosity. Among CO-detected AGB stars and red-supergiants, there is no obvious metallicity effect found between Galactic and LMC stars and object-to-object variation overwhelms the metallicity dependence, if any.