ON THE MASS-LOSS RATE OF MASSIVE STARS IN THE LOW-METALLICITY GALAXIES IC 1613, WLM, AND NGC 3109^*

The evolution of massive stars is greatly affected by the amount of mass and angular momentum lost during their lifetime. Understanding the mechanisms responsible for these losses is a fundamental goal in stellar astrophysics. For instance, mass loss influences the characteristics of the supernova explosion with which a massive star ends its life, as well as the number of potential single-star progenitors of long-duration gamma-ray bursts (e.g., Yoon & Langer 2005; Woosley & Bloom 2006).

An important mass-loss mechanism, at least at galactic and Magellanic Cloud metallicities, is the transfer of momentum from photons to the atmospheric gas through line interactions, initiating and driving an outflow (e.g., Castor et al. 1975; Kudritzki & Puls 2000; Vink et al. 2001). The strength of these radiation-driven winds therefore depends on the effective number of absorption lines, in particular near the photospheric flux maximum in the ultraviolet. It is dominated by the absorption of light through a copious amount of metallic ion lines that are present in this wavelength region. As a consequence, the mass-loss rate ($\dot{M}$) of stars hotter than 25,000 K is predicted to scale with metallicity (Z) as $\dot{M} \propto Z^{0.69\pm 0.10}$ (Vink et al. 2001). Mokiem et al. (2007) showed that this prediction holds for early-type stars in the Galaxy (Z = Z_☉), Large Magellanic Cloud (LMC; Z = 0.5 Z_☉), and Small Magellanic Cloud (SMC; Z = 0.2 Z_☉), yielding the empirical relation $\dot{M} \propto Z^{0.78 \pm 0.17}$. To date, however, no observational constraints exist at sub-SMC metallicities.

For galactic stars more luminous than 10^5.8 L_☉ (i.e., above the Humphreys–Davidson limit), episodic mass loss is expected to occur during the luminous blue variable and/or Wolf–Rayet phase (e.g., Humphreys & Davidson 1994; Smith et al. 2004). Also, dust/pulsation-driven mass loss takes place in the red supergiant phase of lower luminosity objects (e.g., Heger et al. 1997; Yoon & Cantiello 2010). However, if the relative contribution of line driving remains sizable at all Z, an important consequence of the theory of radiation-driven winds would be that low-metallicity massive stars lose less mass and angular momentum over their lifetime than their high-metallicity counterparts. At sub-SMC metallicities, single stars with a large initial rotational velocity may avoid envelope expansion and the associated efficient transfer of angular momentum from the core to the envelope, thus keeping a rapidly spinning core. This makes them potential progenitors of long-duration gamma-ray bursts (e.g., Woosley & Bloom 2006).

In order to observationally constrain stellar and wind parameters at sub-SMC metallicities, we have used the X-Shooter spectrograph mounted on ESO's Very Large Telescope (VLT) UT2 (D'Odorico et al. 2006; Vernet et al. 2011) to secure intermediate-resolution spectra of some of the most massive stars in three nearby dwarf galaxies, IC 1613, WLM, and NCG 3109, each having a metallicity of Z ≈ 1/7 Z_☉. The throughput of the instrument, combined with the collecting area of an 8.2 m telescope, allows us, for the first time, to perform a detailed quantitative spectral analysis at a resolution R ∼ 6000–9000 and test the theory of line driving in O-type stars at a sub-SMC metallicity (see also Herrero et al. 2011).

With the aid of previous, low-resolution studies (Bresolin et al. 2006, 2007; Evans et al. 2007), we selected the six visually brightest O-type stars in these galaxies (four in IC 1613, one each in WLM and NGC 3109). The low line-of-sight extinction toward these galaxies and their similar metallicity allow us, for this particular study, to treat these stars as a group.

In Section 2, the observations and data reduction are described. Section 3 discusses the method of analysis and Section 4 presents the results. Finally, in Section 5, we discuss the implications of our findings.

The spectrum of NGC 3109-20 has been obtained during a X-Shooter Guaranteed Time Observations (GTO) run on 2010 April 23, with an exposure time of 5×900 s. In a successive GTO run from 2010 September 12 to 16, the stars in IC 1613 and WLM were observed, with an exposure time of 6×900 s for all stars. All observations were obtained in nodding mode with a nod throw of 5'', using a slit width of 08 for the UVB arm (300–550 nm), 09 for the VIS arm (550–1020 nm), and 09 for the NIR arm (1000–2500 nm), yielding a spectral resolving power of R = 6200, 8800, and 5600, respectively. Conditions were clear or photometric with an average seeing below 08 in the R band. The moon illumination fraction was below 0.2. Fundamental properties of the observed stars are listed in Table 1.

2.1. Data Reduction

The data have been reduced using the X-Shooter pipeline v1.2.2 in physical model mode (Goldoni et al. 2006; Modigliani et al. 2010). The nodding mode usually allows for easy nebular subtraction. However, all observations except those of WLM-A11 show signs of variable strength of the nebular emission lines over the length of the nod throw, preventing the use of this method. For these stars the sky subtraction was done by using suitable regions of sky close to the object, resulting in a good nebular correction for all stars except IC 1613-A15 (see below). The signal-to-noise ratio (S/N) per resolution element ranges from 32 to 67 in the UVB arm and from 21 to 29 in the VIS arm. The detected flux in the NIR part of the spectrum was low (S/N < 5), so we could not use this part of the spectrum in our analysis.

The nebular emission in IC 1613-A15 varies in radial velocity (RV) along the slit, causing residuals in the final spectrum. The regions affected by these residuals were removed from the final spectrum and not used in the atmosphere fit.

The resulting one-dimensional spectra were normalized to the continuum. Figure 1 presents the blue spectrum and Hα region of all six stars. Figure 2 shows the profiles of the mass-loss sensitive He ii 4686 and Hα lines in more detail. The spectral type of the stars was determined by comparing with a spectral atlas of O-type stars (H. Sana et al. 2011, in preparation), and for all but one agree with the type listed in the literature. The only change is that the spectral type of IC1613-A13 is refined from O3-4 V((f)) (Bresolin et al. 2007) to O3 V((f)).

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

2.2. Oxygen Abundance Determination

The stellar properties and wind characteristics of the stars have been determined using an automated fitting method developed by Mokiem et al. (2005). This method combines the non-LTE stellar atmosphere model FASTWIND (Puls et al. 2005) with the generic fitting algorithm PIKAIA (Charbonneau 1995). It allows for a fast and homogeneous analysis of our sample of stars using a selection of hydrogen, He i, and He ii lines.

The absolute visual magnitude (M_V) of the stars is needed as an input parameter. They have been determined using the V magnitudes from Table 1 and a distance and reddening of d = 721 kpc (Pietrzyński et al. 2006) and E(B − V) = 0.025 (Schlegel et al. 1998) for IC 1613, d = 995 kpc and E(B − V) = 0.08 for WLM (Urbaneja et al. 2008), and d = 1300 kpc (Soszyński et al. 2006) and E(B − V) = 0.14 (Davidge 1993) for NGC 3109. The resulting values of M_V agree well with the values for their spectral type (Martins et al. 2005). The RV has been measured by fitting a Gaussian profile to the hydrogen lines and the obtained values are in agreement with the radial velocities of the host galaxies.

The fitting algorithm covers a large parameter space, fitting line profiles of the same 11 spectral lines as described by Mokiem et al. (2005) to determine the effective temperature (T_eff), surface gravity (g), mass-loss rate ($\dot{M}$), surface helium abundance (Y_He), depth-independent microturbulent velocity (v_tur), and projected rotational velocity (v_rotsin i). The bolometric luminosity L is derived by applying the bolometric correction to the absolute visual magnitude used as input. This luminosity, together with the obtained temperature, is then used to determine the radius (R) of the star. The terminal wind velocity (v_∞) cannot be constrained from the optical, but is related to the surface escape velocity v_esc: v_∞ = 2.6 v_esc (at Z = Z_☉; Lamers et al. 1995; Kudritzki & Puls 2000). The coefficient of the wind velocity structure β has been fixed to 0.8 for the dwarfs, 0.9 for the giants, and 0.95 for the supergiants, conforming with theoretical predictions (L. Muijres et al. 2011, in preparation).

Table 2 presents the best-fit values and derived properties for the six stars. The values for T_eff are higher than those of their Galactic counterparts of the same spectral type (Martins et al. 2005), in agreement with their low metallicity (Mokiem et al. 2004). IC 1613-A13 shows an unusually high value for Y_He. This is likely caused by the degeneracy between effective temperature and surface helium abundance at high temperatures (resulting from the loss of He i as a diagnostic). The large uncertainties in the determination of the micro-turbulent velocity show that this parameter cannot be constrained well with our data.

Table 2. Best-fit Parameters and Derived Properties of the Observed Stars

ID	Teff	logg	log$\dot{M}$	YHe	vtur	vrotsin i	v∞	logL	R
	(kK)	(cm s−2)	(M☉ yr−1)		(km s−1)	(km s−1)	(km s−1)	(L☉)	(R☉)
IC 1613
–A13	47.6+4.73 − 4.95	3.73+0.13 − 0.22	−6.26+0.45 − 0.50	0.32+0.04 − 0.17	4+25↓	94+40 − 28	1869	5.78	11.4
–A15	33.7+3.10 − 2.65	3.76↑ − 0.39	−6.36+0.40↓	0.16+0.14 − 0.10	9+17↓	34+42 − 24	1971	5.24	12.2
–B11	31.3+2.00 − 2.00	3.41+0.24 − 0.21	−6.16+0.30 − 1.30	0.13+0.11 − 0.07	12+13↓	88+36 − 28	1601	5.45	18.1
–C9	35.7+1.70 − 1.85	3.58+0.19 − 0.24	−6.26+0.25 − 0.60	0.12+0.10 − 0.04	15+8↓	72+28 − 28	1697	5.43	13.6
WLM
–A11	29.7+2.45 − 2.75	3.25+0.29 − 0.19	−5.56+0.20 − 0.30	0.110.11↓	8+14↓	70+40 − 36	1711	5.79	29.8
NGC 3109
–20	34.2+4.70 − 3.05	3.48+0.37 − 0.40	−5.41+0.25 − 0.35	0.11↑↓	16↑↓	98+86 − 72	2049	5.88	24.7

Note. Arrows indicate upper or lower limits.

Download table as:  ASCIITypeset image

The errors on the parameters have been analyzed by calculating the probability (P = 1 − Γ(χ²/2, ν/2), where Γ is the incomplete gamma function and ν is the degrees of freedom) for all calculated models. Because P is very sensitive to the value of χ², we normalize all χ² values such that the best χ²_red is equal to 1, i.e., we assume that deviations of the original best χ²_red from unity are induced by under- or overestimated error bars on the normalized flux. This approach is similar in spirit to using relative weighting in the χ² merit function and to propagating the root mean square of the fit to scale the error bars (e.g., Press et al. 1986). The uncertainties are obtained by considering the range of models which satisfy P > 5%. These errors do not take into account uncertainties in the luminosity. However, as the mass-loss rate approximately scales with luminosity as $\dot{M} \propto L^{5/4}$, typical uncertainties in L do not have a significant impact on our conclusions.

The derived mass-loss rates are represented in the modified wind momentum versus luminosity diagram (WLD; Figure 3). The modified wind momentum is defined as $D_{\mathrm{mom}} = \dot{M} v_{\infty } \sqrt{R/R_{\odot }}$. This quantity is ideally suited to study the $\dot{M}(Z)$ relation because it is almost independent of mass, and v_∞ and R are usually relatively well constrained. As the Hα recombination line is essentially sensitive to the invariant wind-strength parameter $Q=\dot{M} / (R^{3/2}v_{\infty })$ that is inferred from the spectral analysis, D (for given T_eff) scales with L, making it less sensitive to uncertainties in the luminosity. As v_∞ is expected to scale with metallicity (v_∞∝Z^0.13; Leitherer et al. 1992), and the wind-strength parameter Q is invariant, the derived mass-loss rate is subject to a similar scaling. This scaling of $\dot{M}$ and v_∞ has been applied to the values given in Table 2.

Zoom In Zoom Out Reset image size

Compared to theoretical expectations (Vink et al. 2001; Z = 0.14 Z_☉ dashed line in Figure 3), all stars except IC 1613-A13 tend toward a higher than predicted mass-loss rate for their metallicity. Compared to the empirical relations found by Mokiem et al. (2007), our values are reminiscent of the values measured for the LMC. For IC 1613-A15 we could only obtain an upper limit due to the nebular contamination, and IC 1613-B11 and C9 could have a low enough mass-loss rate within the errors. WLM-A11 and NGC 3109-20 have a well-defined mass-loss rate which is almost an order of magnitude too large for their metallicity.

4.1. Systematic Uncertainties in the Mass-loss Determination

4.2. Possible Intrinsic Causes for the High Mass-loss Rates

In this Letter, we have pointed out a discrepancy between the observed and predicted mass-loss rates from massive stars in low-metallicity environments, and discussed possible explanations.

A potential violation of the expected metallicity scaling of the radiation-driven mass-loss rate at Z ∼ 1/7 Z_☉, resulting in higher than expected mass-loss rates, would have far-reaching implications. First, one expects low-Z O-type stars to suffer more from spin-down through angular momentum loss in the stellar wind. Consequently, the rotational mixing efficiency may be reduced, leading, for instance, to a more modest nitrogen enrichment than currently thought (Brott et al. 2011).

Second, one anticipates the single O-star population in low-metallicity environments to produce more observationally identifiable Wolf–Rayet stars, being the successors of O-type stars having their outer envelope stripped by mass loss, and therefore increased number of Ib and, potentially, Ic supernovae. The single-star channel would, however, produce less progenitors of long-duration gamma-ray bursts, as the stars lose more angular momentum by their outflow.

Finally, if the larger than expected wind strength at Z ∼ 1/7 Z_☉ persists to extremely low metallicities (though at present the driving mechanism is unknown), stellar winds will impact the evolution of massive Population III stars and the chemical enrichment of the intergalactic medium of the early universe.

We thank the referee, Dr. Kudritzki, for his useful comments, Dr. Martayan for his support during the observations, and O. Hartoog for the interesting discussions.

Facility: VLT:Kueyen (X-Shooter) - Very Large Telescope (Kueyen)