RED SUPERGIANT STARS AS COSMIC ABUNDANCE PROBES: KMOS OBSERVATIONS IN NGC 6822

Our targets were selected from optical photometry (Massey et al. 2007), combined with near-IR (JHK_s) photometry (for details see Sibbons et al. 2012) from the Wide-Field Camera (WFCAM) on the UKIRT. The two catalogs were cross-matched and only sources classified as stellar in the photometry for all filters were considered.

Our spectroscopic targets were selected principally based on their optical colors, as defined by Massey (1998) and Levesque & Massey (2012). Figure 1 shows cross-matched stars, with the dividing line at (B − V) = 1.25 × (V − R) + 0.45. All stars redder than this line, above a given magnitude threshold, and with V − R > 0.6 are potential RSGs.

Zoom In Zoom Out Reset image size

Distinguishing between RSGs and the most luminous stars on the asymptotic giant branch (AGB) is difficult owing to their similar temperatures and overlapping luminosities. Near-IR photometry can help to delineate these populations, with RSGs located in a relatively well-defined region in the near-IR (J − K) color–magnitude diagram (CMD), as discussed by Nikolaev & Weinberg (2000). The near-IR CMD from the WFCAM data is shown in Figure 2 and was used to further inform our target selection. Employing the updated CMD criteria from Cioni et al. (2014)—modified for the distance and reddening to NGC 6822—all of our potential targets from the combined optical and near-IR criteria are (notionally) RSGs.Near-IR photometry for our targets was also given by Whitelock et al. (2013) from observations with the Infrared Survey Facility; we note that from their analysis of the multi-epoch data, not one of our targets was identified as an AGB star.

Zoom In Zoom Out Reset image size

The combined selection methods yielded 58 candidate RSGs, from which 18 stars were observed with KMOS, as shown in Figure 3. The selection of the final targets was defined by the KMOS arm allocation software karma (Wegner & Muschielok 2008), where the field center was selected to maximise the number of allocated arms, with priority given to the brightest targets. Optical spectroscopy of eight of our observed stars, confirming them as RSGs, was presented by Levesque & Massey (2012).

Zoom In Zoom Out Reset image size

The observations were obtained as part of the KMOS Science Verification program on 2013 June 30 (PI: Evans, 60.A-9452(A)), with a total exposure time of 2400 s (comprising 8 × 300 s detector integrations). KMOS has 24 deployable integral-field units (IFUs), each of which covers an area of 28 × 28 within a 72 field of view. The 24 IFUs are split into three groups of eight, with the light from each group relayed to different spectrographs.

Offset sky frames (05 to the east) were interleaved between the science observations in an object (O), sky (S) sequence of: O, S, O, O. The observations were performed with the YJ grating (giving coverage from 1.02–1.36 μm); estimates of the mean delivered resolving power for each spectrograph (obtained from the KMOS/esorex pipeline for two arc lines) are listed in Table 1.

In addition to the science observations, a standard set of KMOS calibration frames was obtained consisting of dark, flat, and arc-lamp calibrations (with flats and arcs taken at six different rotator angles). A telluric standard star was observed with the arms configured in the science positions, i.e., using the KMOS_spec_cal_stdstarscipatt template in which the standard star is observed sequentially through all IFUs. The observed standard was HIP 97618, with a spectral type of B6 III (Houk & Smith-Moore 1988).

A summary of the observed targets is given in Table 2. A signal-to-noise ratio (S/N) of ≳100 per resolution element is required for satisfactory results from this analysis method (see Gazak et al. 2014a). We estimated the S/N of the spectra by comparing the counts in the brightest spatial pixels (within the 1.15–1.22 μm region) of each source with the counts in equivalent spatial pixels in the corresponding sky exposures (between the sky lines). The S/N estimated is knowingly an underestimate of the true S/N achieved.

The default template for telluric observations with KMOS is to observe a standard star in one IFU in each of the three spectrographs. However, there is an alternative template which allows users to observe a standard star in each of the 24 IFUs. This strategy should provide an optimum telluric correction for the KMOS IFUs but reduces observing efficiency.

A comparison between the two methods in the H-band was given by Davies et al. (2013b), who concluded that using the more efficient three-arm method was suitable for most science purposes. However, an equivalent analysis in the YJ-band was not available. To determine if the more rigorous telluric approach is required for our analysis, we observed a telluric standard star (HIP 97618) in each of the 24 IFUs. This gave us the data to investigate both telluric correction methods and to directly compare the results.

We first compared the standard-star spectrum in each IFU with that used by the pipeline routines for the three-arm template in each of the spectrographs. Figure 4 shows the differences between the standard-star spectra across the IFUs, where the differences in the YJ-band are comparable to those in the H-band (cf. Figure 7 from Davies et al. 2013b). The qualitative agreement between the IFUs in our region of interest (1.15–1.22 μm) is generally very good.

Zoom In Zoom Out Reset image size

To quantify the difference the two telluric methods would make to our analysis, we performed the steps described in Section 3.2 for both templates. We then used the two sets of reduced science data (reduced with both methods of the telluric correction) to compute stellar parameters for our targets. The results of this comparison are detailed in Section 4.1.

To improve the accuracy of the telluric correction, for both methods mentioned above, we implemented additional recipes beyond those of the KMOS/esorex pipeline. These recipes were employed to account for two effects which could potentially degrade the quality of the telluric correction. The first corrects for any potential shift in wavelength between each science spectrum and its associated telluric standard. The most effective way to implement this is to cross-correlate each pair of science and telluric-standard spectra. Any shift between the two is then applied to the telluric standard using a cubic-spline interpolation routine.

The second correction applied is a simple spectral scaling algorithm. This routine corrects for differences in line intensity of the most prominent features common to both the telluric and science spectra. To find the optimal scaling parameter, the following formula is used:

$$\begin{eqnarray}&&{{T}_{2}}=({{T}_{1}}+c)/({{T}_{1}}+c),\end{eqnarray} \tag{ 1 }$$

where T₂ is the corrected telluric-standard spectrum, T₁ is the initial telluric standard spectrum and c is the scaling parameter.

To determine the required scaling, telluric spectra are computed for −0.5 < c < 0.5, in increments of 0.02 (where a perfect value, i.e., no difference in line strength, would be c = 0). Each telluric spectrum is used to correct the science data and the standard deviation of the counts across the spectral region is computed for each corrected spectrum. The minimum value of the standard-deviation matrix defines the optimum scaling. For this algorithm, only the region of interest for our analysis is considered (i.e., 1.15–1.22 μm).

The final set of telluric-standard spectra from the KMOS/esorex reductions were modified using these additional routines and were then used to correct the science observations for the effects of the Earth's atmosphere.

As an alternative to observing telluric standard stars, a new telluric correction package, molecfit, allows one to calculate a telluric spectrum based on atmospheric modeling. Briefly, the software uses a reference atmospheric profile to estimate the true profile for the time and location of the science observation. This model is then used to create a telluric spectrum which can be used to correct the observations.

This software has been shown to work well on a variety of VLT instruments (Smette et al. 2015) and has been rigorously tested using X-shooter spectra (Kausch et al. 2015). However, the package has yet to be tested thoroughly on lower resolution observations such as those from KMOS. Our first tests appear encouraging; however, pending further characterization of the KMOS data cubes (e.g., small variations in spectral resolving power leading to sky residuals, see Section 3.3), we will investigate the potential of the molecfit package in future papers of this series.

Initial inspection of the extracted stellar spectra revealed minor residuals from the sky subtraction process. Reducing these cases with the "sky_tweak" option within the KMOS/esorex reduction pipeline was ineffective to improve the subtraction of these features. Any residual sky features could potentially influence our results by perturbing the continuum placement within the model fits, which is an important aspect of the fitting process (see Gazak et al. 2014a; Davies et al. 2015, for more discussion). Thus, pending a more rigorous treatment of the data (e.g., to take into account the changing spectral resolution across the array), we exclude objects showing sky residuals from our analysis. Of the 18 observed targets, 11 were used to derive stellar parameters (as indicated in Table 2).

Radial velocities for each target are listed in Table 2. To check the accuracy of the wavelength solution provided by the data reduction pipeline, this solution is cross-correlated with the wavelength solution of a spectrum of the Earth's telluric features.⁸ In general, this is a small correction. The science spectra are then telluric corrected in the manner described above.

Radial velocities are estimated by measuring the peak of the cross-correlation function between the telluric-corrected science spectra (taking the spectrum from the brightest spatial pixel) and an appropriate synthetic RSG spectrum. The errors are calculated by taking the dispersion of each line used to derive the radial velocities scaled by the number of lines used, following Lapenna et al. (2015). A robust radial velocity could not be derived for one of our candidates (NGC 6822-RSG06) owing to strong residual sky subtraction features, therefore, it is not considered further.

Radial velocity estimates are shown as a function of distance to the center of NGC 6822 in Figure 5. The average radial velocity for our targets is −68 ± 12 km s⁻¹, in good agreement with the systemic radial velocity of the H i disk (−57 ± 2 km s⁻¹; Koribalski et al. 2004). Our radial velocities also agree with estimates for the two A-type supergiants from Venn et al. (2001). This result confirms that our candidates are NGC 6822 members.

We used these Science Verification data to determine which of the two telluric standard methods is most appropriate for our analysis. Table 4 details the stellar parameters derived for each target using both telluric methods, and these parameters are compared in Figure 7. The mean difference in metallicity from the two methods is ${\Delta }[Z]=\;0.04\;\pm \;0.07$. Therefore, for our analysis, there is no significant difference between the two telluric approaches.

Zoom In Zoom Out Reset image size

Table 4 summarizes the derived stellar parameters. For the remainder of this paper, when discussing stellar parameters, we adopt those derived using the 24 arm telluric method (i.e., the results in the left-hand part of Table 4.) The average metallicity for our sample of 11 RSGs in NGC 6822 is $[\mathop{Z}\limits^{{}}]$ = −0.52 ± 0.21. This result is in good agreement with the average metallicity derived in NGC 6822 from blue supergiant stars (BSGs; Muschielok et al. 1999; Venn et al. 2001).

A direct comparison with metallicities from BSGs is legitimate as the results derived here yield a global metallicity ([Z]) that closely resembles the Fe/H ratio derived in Venn et al. (2001). While our [Z] measurements are also affected by Si and Ti, we assume [Z] = [Fe/H] for the purposes of our discussion. Likewise, we can compare oxygen abundances (relative to solar) obtained from H ii regions as a proxy for [Z] by introducing the solar oxygen abundance 12 + log(O/H)_⊙ = 8.69 (Asplund et al. 2009) through the relation [Z] = 12 + log(O/H) − 8.69.

The RSG and BSG stages are different evolutionary phases within the life cycle of a massive star, while H ii regions are the birth clouds which give rise to the youngest stellar population. As the lifetimes of RSGs and BSGs are <50 Myr, their metallicity estimates are also expected to be representative of their birth clouds.

To investigate the spatial distribution of chemical abundances in NGC 6822, in Figure 8 we show the metallicities of our RSGs as a function of radial distance from the center of the galaxy, as well as the results from Venn et al. (2001) and the indicative estimates from Muschielok et al. (1999).

Zoom In Zoom Out Reset image size

A least-squares fit to the KMOS results reveals a low-significance abundance gradient within the central 1 kpc of NGC 6822 of −0.5 ± 0.4 dex kpc⁻¹. The extrapolated central metallicity from the fit (i.e., at R = 0) of [Z] = −0.30 ± 0.15 derived remains consistent with the average metallicity assuming no gradient.

Figure 9 shows the location of our sample in the Hertzsprung–Russell (H–R) diagram. Bolometric corrections were computed using the calibration in Davies et al. (2013a) to calculate luminosities. This figure shows that the temperatures derived using the J-band method are systematically cooler than the end of the evolutionary models (which terminate at the end of the carbon-burning phase for massive stars) for Z = 0.002 from Georgy et al. (2013). This is discussed in Section 5.2.

Zoom In Zoom Out Reset image size

We find an average metallicity for our sample of $[\mathop{Z}\limits^{{}}]$ = −0.52 ± 0.21, which agrees well with the results derived from BSGs (Muschielok et al. 1999; Venn et al. 2001; Przybilla 2002) and H ii regions (Lee et al. 2006).

We also find evidence for a low-significance metallicity gradient within the central 1 kpc of NGC 6822 (−0.5 ± 0.4 dex kpc⁻¹ with a $\chi _{{\rm red}}^{2}$ = 1.16; see Figure 8). The gradient derived is consistent with the trend reported in Venn et al. (2001) from their results for the two BSGs compared with H ii regions from Pagel et al. (1980) and two planetary nebulae from Richer & McCall (1995) at larger galactocentric distances. Our result is also consistent with the gradient derived from a sample of 49 local star-forming galaxies (Ho et al. 2015). Including the results for BSGs from Venn et al. (2001) in our analysis gives a consistent gradient (−0.48 ± 0.33 dex kpc⁻¹) with a $\chi _{{\rm red}}^{2}$ = 1.06. Results from Muschielok et al. (1999) are not included in the fit as these measurements were qualitative estimates of metallicity.

In contrast, Lee et al. (2006) used the oxygen abundances from 19 H ii regions and found no clear evidence for a metallicity gradient. Using a subset of the highest quality H ii region data available, these authors found a gradient of −0.16 ± 0.05 dex kpc⁻¹. Including these results into our analysis degrades the fit and changes the derived gradient significantly (−0.18 ± 0.05 dex kpc⁻¹; $\chi _{{\rm red}}^{2}$ = 1.78). At this point it is not clear whether the indication of a gradient obtained from the RSGs and BSGs is an artifact of the small sample size or indicates a difference with respect to the H ii region study.

There have been a number of studies of the metallicity of the older stellar population in NGC 6822. From spectroscopy of red giant branch stars, Tolstoy et al. (2001) found a mean metallicity of [Fe/H] = −0.9 with a reasonably large spread (see their Figure 19). More recently, Sibbons et al. (2012) derived metallicities using a population of AGB stars within the central 4 kpc of NGC 6822. They found an average metallicity of [Fe/H] = −1.29 ± 0.07 dex. Likewise, Kirby et al. (2013) used spectra of red giant stars within the central 2 kpc and found an average metallicity of [Fe/H] = −1.05 ± 0.49 dex. We note that none of these studies found any compelling evidence for spatial variations in the stellar metallicities, which is not surprising given that, in disk galaxies, radial migration is thought to smooth out any abundance gradients over time. The stellar populations used for these studies are known to be significantly older than our sample; therefore, owing to the chemical evolution in the time since the birth of the older populations, we expect the measured metallicities to be significantly lower.

The low metallicity of the young stellar population and the interstellar medium (ISM) in NGC 6822 can be understood as a consequence of the fact that it is a relatively gas-rich galaxy with a mass M_{H i} = 1.45 × 10⁸ M_⊙ (Koribalski et al. 2004) and a total stellar mass of M_* = (0.83−1.70) × 10⁸ M_⊙ (Woo et al. 2008; Kirby et al. 2013; Weisz et al. 2014).

The simple closed-box chemical-evolution model relates the metallicity mass fraction Z(t) at any time to the ratio of stellar to gas mass ${{M}_{*}}/{{M}_{g}}$ through

$$\begin{eqnarray}&&Z(t)=\frac{y}{1-R}{\rm ln} \left[ 1+\frac{{{M}_{*}}(t)}{{{M}_{g}}(t)} \right],\end{eqnarray} \tag{ 2 }$$

where y is the fraction of metals per stellar mass produced through stellar nucleosynthesis (the so-called yield) and R is the fraction of stellar mass returned to the ISM through stellar mass loss.

According to Kudritzki et al. (2015), the ratio y/(1 − R) can be empirically determined from the fact that the metallicity of the young stellar population in the solar neighborhood is solar, with a mass fraction Z_⊙ = 0.014 (Nieva & Przybilla 2012). With a stellar-to-gas mass column density of 4.48 in the solar neighborhood (Wolfire et al. 2003; Bovy & Rix 2013) one then obtains y/(1 − R) = 0.0082 = 0.59 Z_⊙ with an uncertainty of 15% dominated by the 0.05 dex uncertainty of the metallicity determination of the young population.

Accounting for the presence of helium and metals in the neutral interstellar gas, we can turn the observed H i mass in NGC 6822 into a gas mass via M_g = 1.36 M_{H i} and use the simple closed-box model to predict a metallicity of [Z] = −0.44 to −0.69, in good agreement with our value obtained from RSG spectroscopy.

As discussed above, the older stellar population of AGB stars has a metallicity roughly 0.8 dex lower than the RSGs. In the framework of the simple closed-box model this would correspond to a period in time where the ratio of stellar to gas mass was ∼0.1 (much lower than the present value of 0.42–0.86). The present star-formation rate of NGC 6822 is ∼0.02 M_⊙ yr⁻¹ (Gratier et al. 2010; Efremova et al. 2011). At such a level of star formation it would have taken 5 Gyr to produce the currently observed stellar mass and to arrive at the average metallicity of the young stellar population from that of the AGB stars (of course, again relying on the simple closed-box model). Evidence suggests that the star-formation rate was substantially lower in the past (Efremova et al. 2011; Weisz et al. 2014); therefore, the buildup of the observed stellar mass would have taken correspondingly longer.

Given the irregularities present in the stellar and gaseous morphology of NGC 6822, this galaxy may not be a good example of a closed-box system; however, it is remarkable that the closed-box model reproduces the observed metallicity so closely.

Effective temperatures have been derived for 11 RSGs from our observed sample in NGC 6822. To date, this represents the fourth data set used to derive stellar parameters in this way and the first with KMOS. The previous three data sets which have been analyzed are those of 11 RSGs in Perseus OB-1, a Galactic star cluster (Gazak et al. 2014a), nine RSGs in the LMC and 10 RSGs in the SMC (both from Davies et al. 2015). These results range from Z = Z_⊙ in Perseus OB-1 to Z = 0.3 Z_⊙ in the SMC, around 0.5 dex in metallicity.

We compare the effective temperatures derived in this study with those of the previous results in different environments. Their distribution is shown as a function of metallicity in Figure 10. Additionally, Figure 11 shows the H–R diagram for the four sets of results. Bolometric corrections to calculate the luminosities for each sample were computed using the calibration in Davies et al. (2013a).

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

From these figures, we see no evidence for significant variations in the average temperatures of RSGs with respect to metallicity. This is in contrast to current evolutionary models which display a change of ∼450 K, for a M = 15 M_⊙ model, over a range of solar to SMC-like metallicities (Ekström et al. 2012; Georgy et al. 2013).

For solar metallicity, observations in Perseus OB-1 are in good agreement with the models (see Figure 9 in Gazak et al. 2014a). However, at SMC-like metallicity, the end-points of the models are systematically warmer than the observations. The temperature of the end-points of the evolutionary models of massive stars could depend on the choice of the convective mixing-length parameter (Schaller et al. 1992). That the models produce a higher temperature than observed could imply that the choice of a solar-like mixing-length parameter does not hold for higher mass stars at lower metallicity.

Lastly, we note that the average spectral type of RSGs tends toward an earlier spectral type with decreasing metallicity over this range (Humphreys 1979; Levesque & Massey 2012). We argue that this is not in contradiction to the above results. Spectral types are derived for RSGs using the optical TiO band-heads at ∼0.65 μm, whereas in this study temperatures are derived using near-IR atomic features (as well as the line-free pseudo-continuum). The strengths of TiO bands are dependent upon metallicity which means that the spectral classification for RSGs at a constant temperature will differ (Davies et al. 2013b). Therefore, although historically spectral type has been used as a proxy for temperature, this assumption does not provide an accurate picture for RSGs.

As mentioned in Section 2.1, massive AGB stars are potential contaminants to our sample. These stars have similar properties to RSGs and can occupy similar mass ranges as lower mass RSGs (Herwig 2005), with lifetimes of around ∼25 Myr (Doherty et al. 2010). Wood et al. (1983) argued for an AGB luminosity limit (owing to the limit on the mass of the degenerate core) of M_bol ∼ −7.1. Using this maximum luminosity, corrected for the distance to NGC 6822, yields K = 14.0. Four of our analyzed stars have K-band magnitudes fainter than this limit, but excluding the results for these does not significantly alter any of our results. However, given the near-IR luminosity and colors of these stars, if indeed they are AGB stars, they are probably younger massive AGB stars (e.g., Siess 2008) and therefore would still trace the relatively young stellar population.