AN INFRARED CENSUS OF DUST IN NEARBY GALAXIES WITH SPITZER (DUSTiNGS). II. DISCOVERY OF METAL-POOR DUSTY AGB STARS

The Surveying the Agents of Galaxy Evolution (SAGE) Spitzer surveys of the Small and Large Magellanic Clouds (SMC/LMC) obtained 3.6–70 μm photometry covering the full spatial extent of each galaxy, thereby detecting the circumstellar dust around the near-complete population of AGB stars in each galaxy (Meixner et al. 2006; Gordon et al. 2011). In both galaxies, a subset of AGB stars with colors J − K ≳ 2 mag show evidence for significantly more dust production than that observed for the average AGB star. Most of these stars are C-rich (Woods et al. 2011), and their mass-loss rates exceed the nuclear-burning mass consumption rate (cf. Boyer et al. 2012). These stars are sometimes referred to as "extreme" (x-)AGB stars, and we use this terminology here to be consistent with the recent work in the MCs (e.g., Blum et al. 2006).

The x-AGB stars comprise <6% of the total AGB population, but they account for more than 75% of the dust produced by cool evolved stars (Matsuura et al. 2009; Srinivasan et al. 2009; Boyer et al. 2012; Riebel et al. 2012). These stars also contribute significantly to the global 8 μm flux of the SMC (Melbourne & Boyer 2013). Other than those in the MCs and the Milky Way, x-AGB stars have been detected and their dust production confirmed in only a few galaxies: M33 (McQuinn et al. 2007; Javadi et al. 2013), M32 (Jones et al. 2015), Sgr dSph (McDonald et al. 2012), Fornax and Sculptor (Matsuura et al. 2007; Groenewegen et al. 2009; Sloan et al. 2009), so it is unclear whether AGB stars in metal-poor galaxies are efficient dust producers.

Boyer et al. (2015, hereafter Paper I) describe the Dust in Nearby Galaxies with Spitzer (DUSTiNGS) survey in detail. DUSTiNGS is a 3.6 and 4.5 μm imaging survey of 50 nearby dwarf galaxies designed to detect all dust-producing evolved stars. Targets include 37 dwarf spheroidal (dSph), eight dwarf irregular (dIrr), and five transition type (dTrans or dSph/dIrr) galaxies (see Table 1 from Paper I). The target galaxies have experienced a variety of different star formation and interaction histories and span a wide range in mass traced by visible light (0  >  M_V  >  15.2 mag) and metallicity (−2.72  <  [Fe/H] < −1.1).

It is difficult to separate AGB stars belonging to a galaxy from unresolved background sources and foreground stars at these wavelengths. In Paper I, we used the large field of view to estimate the size of the thermally pulsing (TP-)AGB population by statistically estimating and removing the foreground and background sources for each galaxy, assuming that the tip of the red giant branch (TRGB) is at M_3.6 = −6 mag. We found that the Andromeda satellites in the sample harbor 40  ±  30 to 506  ±  48 AGB stars each (including dusty and non-dusty stars), with the largest populations in And VII, And XVIII, and And X. The most metal-poor galaxies in the sample have the lowest masses, and therefore are the least likely to harbor AGB stars. In these galaxies, we found only upper limits on the size of the AGB population (<9 each), with the exception of 20 ± 9 AGB stars in the Hercules Dwarf ([Fe/H] = −2.41). The largest AGB populations are in the star-forming dIrr galaxies.

In Paper I, we also found evidence for x-AGB stars in eight of the DUSTiNGS galaxies (And II, Cetus, IC 10, IC 1613, NGC 147, NGC 185, Sextans B, and WLM), though it is impossible to isolate galaxy members from red background sources with only 3.6 and 4.5 μm imaging. In order to identify individual x-AGB stars, the DUSTiNGS images were obtained in two epochs, separated by approximately 6 months. Dust-producing AGB stars are expected to pulsate with periods ranging from 100 to 1000 days (Vassiliadis & Wood 1993), though little is known about how the mid-IR properties of these stars are affected by pulsation. Light curves for pulsating AGB stars in the MCs, the Milky Way, and in nearby dSph galaxies show that pulsation amplitudes decrease from optical to near-infrared (IR) wavelengths (Le Bertre 1992, 1993; Whitelock et al. 2003; Menzies et al. 2010; Battinelli & Demers 2012), but 1–20 μm light curves of Galactic AGB stars from Harvey et al. (1974) and Le Bertre (1992, 1993) suggest that amplitudes of dusty sources may increase in the IR owing to changes in the warm circumstellar dust that is responding to the pulsations of the photosphere. McQuinn et al. (2007) obtained five epochs of imaging of M33 with Spitzer, and were able to identify a large population of dust-producing C star candidates with 3.6 μm amplitudes up to 0.8 mag (amplitudes in that work are the standard deviation over the five epochs). The SAGE observations of the MCs obtained two epochs of imaging, and find that the x-AGB stars show amplitudes up to 1.4 mag at 3.6 μm (Vijh et al. 2009; Polsdofer et al. 2015).

Here, we report the identification of 526 x-AGB star candidates in the DUSTiNGS galaxies that were discovered via their 3.6 and 4.5 μm variability, including 12 in galaxies with [Fe/H] < −2. These are the most metal-poor dust-producing AGB stars known thus far. This paper is organized as follows. In Sections 2 and 3, we describe the data, stellar classification, and variable star identification. In Section 4, we discuss the spatial distribution, colors, and amplitudes of the AGB variables. In Section 5, we compare the DUSTiNGS variables to previously identified variables and C stars. In Section 6, we discuss dust production by these very metal-poor stars. We summarize the results in Section 7.

The DUSTiNGS observations and photometry are described in detail in Paper I. In brief, each galaxy was imaged simultaneously at 3.6 and 4.5 μm to at least the half-light radius (r_h) in two epochs. The separation between epochs was determined by Spitzer's visibility windows for each target. The average epoch separation is 180 days, with a range of 127–240 days (Table 2 from Paper I). All photometry that we use here is from the "good-source" catalog (GSC), which has been culled to include only high-confidence point sources and reliable measurements (Paper I).

Padova stellar evolution models suggest that >90% of the TP-AGB population is brighter than the TRGB at a given time (Marigo et al. 2008, 2013; and G. Bruzual 2012, private communication). The TRGB ranges from M_3.6 = −6.6 to −6 mag, based on Spitzer observations of eight dIrr galaxies (Jackson et al. 2007a, 2007b; Boyer et al. 2009b). To ensure that the majority of the TP-AGB stars are measured, the DUSTiNGS photometry is at least 75% complete down to this limit. However, the Infrared Array Camera (IRAC) colors and magnitudes of AGB stars make it difficult to distinguish them from foreground stars and unresolved background galaxies. This is especially true for target galaxies more distant than ≈250 kpc ((m − M)₀ ≳ 22 mag), for which individual AGB stars can be fainter than unresolved background galaxies. In Paper I, we estimated the total AGB population size by statistically subtracting the background and foreground contamination. Here, we use stellar variability to identify a subset of individual AGB candidates.

Eight of the DUSTiNGS targets were also observed with Spitzer during the cryogenic mission. The observations for these galaxies are described in detail in Jackson et al. (2007a, 2007b) and Boyer et al. (2009b). In brief, the total integration times and dithering strategies are nearly identical to DUSTiNGS, but the fields of view are significantly smaller. For six galaxies (Aquarius, Leo A, LGS 3, Pegasus, Phoenix, and Sextans A), only a single ≈5'×5' IRAC frame was observed. IC 1613 was imaged in a 2×3 IRAC map and WLM in a 3×1 IRAC map (see Figure 15). We downloaded the data from these programs (Programs 128 and 40524, P.Is.: R. D. Gehrz and E. D. Skillman, respectively), processed with the S18.25.0 pipeline, from the Spitzer Heritage Archive and produced photometric catalogs following the same process used for the DUSTiNGS data (Paper I).

The 3.6 μm observations for all eight galaxies and both 3.6 and 4.5 μm observations for IC 1613 and WLM were obtained prior to 2006 (Program 128). The 4.5 μm observations for the remaining six galaxies were obtained in 2007 (Program 40524). See Jackson et al. (2007a, 2007b) and Boyer et al. (2009b) for specific dates for each set of observations. Since this epoch occurs prior to the DUSTiNGS observations, we refer to it as "epoch 0".

In the LMC, Blum et al. (2006) designate stars with colors J − [3.6] > 3.1 mag as x-AGB stars. This color is empirically observed as the point where there is a sudden break, or decrease in source density, in the C star sequence evident in a J − [3.6] versus [3.6] color–magnitude diagram (CMD). This break roughly corresponds to the superwind phase of AGB evolution, wherein the mass-loss rate increases dramatically due to an increase in dust production by a factor of 10 or more (Schröder et al. 1999). At low metallicities, the superwind phase may be triggered by the dredge-up of carbon (Lagadec & Zijlstra 2008). Many AGB stars not classified as x-AGB also show dust emission, though the dust emission from x-AGB stars is significantly stronger.

Most of the DUSTiNGS galaxies do not have available J-band photometry at the necessary depth to use the Blum et al. (2006) classification scheme. Instead, we mimic it based on the positions of AGB stars in the [3.6]–[4.5] CMD (Figure 1(a)) and classify variables as follows.

Zoom In Zoom Out Reset image size

Unknown Variables fainter than the assumed TRGB (M_3.6 = −6 mag; solid line in Figure 1(b)). The TRGB is not known for most of the DUSTiNGS galaxies, but it can be as bright as M_3.6 = −6.6 (Jackson et al. 2007a, 2007b; Boyer et al. 2009b).

Less dusty AGB stars Variables brighter than the assumed TRGB and bluer than the dashed line in Figure 1(b).

x-AGB stars Variables redder than the dashed line in Figure 1(b). The luminosity limit for x-AGB stars becomes fainter at redder colors to allow for the obscuration of the dustiest sources while also avoiding the region dominated by unresolved background sources (see Paper I).

Red supergiants (RSGs) AGB stars can exceed the classical AGB limit when experiencing hot-bottom burning (Boothroyd & Sackmann 1992), so variables brighter than the dot-dashed line in Figure 1 could be either massive RSGs or AGB stars. We classify these stars as AGB or x-AGB depending on their color and also flag them as potential RSGs in Table 3.

Our classification scheme would correctly classify 93%–94% of the Blum et al. (2006) x-AGB stars in the LMC and SMC. Similarly, <2% of other AGB types in the MCs would be mis-classified as x-AGB using our scheme.

We note that the x-AGB stars here are not the same as the very extreme AGB stars in the LMC, as defined by Gruendl et al. (2008). That work discovered 13 LMC C stars that are among the reddest objects in that galaxy ([3.6]–[4.5] up to 3.4 mag). All of the Gruendl et al. (2008) stars are so dust enshrouded that they are faint even at 3.6 μm (M_3.6 > −5 mag), and are beyond the sensitivity limit of DUSTiNGS. Only one star in the DUSTiNGS sample approaches the colors of the Gruendl et al. (2008) sources: a star in IC 10 with [3.6]–[4.5] = 2.4 mag (see Section 4.2 and Figure 1).

Unknown variable stars may be AGB stars in the minimum brightness phase of pulsation, but they could also be non-AGB variable sources such as RR Lyrae, eclipsing binaries, or background active galactic nuclei (AGNs). Cepheid variables are too faint at these wavelengths, so we do not expect them to contaminate the AGB sample.

To identify individual AGB candidates, we use the variability index defined by Vijh et al. (2009):

$$\begin{equation} V_\lambda = \frac{f_{\lambda,i} - f_{\lambda,j}}{\sqrt{\sigma f_{\lambda,i}^2 + \sigma f_{\lambda,j}^2}}, \end{equation} \tag{ 1 }$$

where f_λ is the flux density of a source, σf_λ is the uncertainty in flux, and i and j indicate different epochs. Assuming a two-dimensional Gaussian probability distribution in V_3.6μm and V_4.5μm, we identify high-confidence variable star candidates as those with a joint probability falling outside 3σ. Less confident variables stars are those falling between 2σ–3σ. In both cases, candidates must show variability indices more than 2σ outside the mean at both 3.6 and 4.5 μm, in the same direction. We illustrate this technique for Sextans B in Figure 2.

Zoom In Zoom Out Reset image size

For galaxies with a third epoch of data (Section 2.2), we computed Equation (1) comparing epoch 0 to each of the DUSTiNGS epochs to identify additional variable candidates. Inclusion of epoch 0 results in 14 more 2σ and 19 more 3σ variables for these 8 galaxies, including 16 variable x-AGB candidates (Table 2). These additional variables are located in the central region of each galaxy (Figure 15).

3.2.1. Likelihood of Detection via the Variability Index

Because we have only two to three epochs, the cadence of the observations can be unfavorable for the detection of variables with particular amplitudes and periods. In general, the variability index is less sensitive to small-amplitude variable stars and stars with periods on the order of the separation between the epochs (≈180 days). To determine the "completeness" of the detected variables, we compute the probabilities of detecting variable stars with the DUSTiNGS observations for stars with periods ranging from 50–3000 days and peak-to-peak amplitudes up to 1 mag at 3.6 μm. Figure 3 shows the result of this simulation. For comparison, we also show the SMC and LMC optically detected variable x-AGB stars (from the Optical Gravitational Lensing Experiment III, OGLE III; Udalski et al. 2008a, 2008b). The DUSTiNGS variability index is insensitive to stars with amplitudes less than ≈0.15 mag (especially less-evolved O-rich AGB stars; Figure 14), but it is particularly suited to detect the x-AGB stars with periods ranging from 300 to 600 days.

Zoom In Zoom Out Reset image size

This simulation assumes that the 3.6 μm amplitude is equal to half the I-band amplitude, since it is known that the amplitude tends to decrease at near-IR wavelengths (e.g., Whitelock et al. 2006, 2009; Menzies et al. 2008). Little is known about typical AGB star amplitudes at IRAC wavelengths; for IRC+10216, the L' (3.79 μm) amplitude is ≈0.75A_J (Le Bertre 1992). The most extreme, optically obscured AGB stars ([3.6]–[4.5] ≳ 0.9 mag) may be more easily detected by DUSTiNGS if they follow the general trend for more evolved stars having larger amplitudes (e.g., the C star LI-LMC 1813; van Loon et al. 2003).

3.2.2. Contamination among Variable Candidates

In the entire DUSTiNGS sample, we find 526 variable x-AGB candidates among 710 total variable sources. The number of variable AGB star candidates in each galaxy is listed in Tables 1 and 2. We quote the maximum number of AGB candidates allowing for the uncertainty in the distance moduli (Paper I). Individual variable sources are listed in Table 3 along with mean magnitudes, the lower limits on the 3.6 μm amplitude (Δm_3.6 = $\vert m_{3.6}^{\rm epoch\,1} - m_{3.6}^{\rm epoch\,2} \vert$), and the classification. This table is available to download in the electronic version of this paper.

We have flagged a small subset of the variable sources listed in Table 3 that may be affected by imaging artifacts that cause the star to appear artificially variable. Some x-AGB candidates are located near imaging artifacts caused by "Column Pulldown,"¹⁴ which can be seen as lines of faint pixels emanating from bright point sources. This affects 11 variable x-AGB star candidates in And IX, And XIII, Aquarius, IC 10, NGC 147, Sag DIG, and Sextans A. In addition, some variable sources fall on the edge of the spatial coverage (≲10'' from the edge) where fewer frames contribute to the total depth owing to frame dithering (Paper I). There are fewer frames to assist in the elimination of imaging artifacts in these regions, which may affect the measured fluxes. This affects four variable x-AGB candidates, including two in And IX, 1 in IC 1613, and one in IC 10. We also checked whether proximity to a bright star might affect the measured flux and found one variable x-AGB star in NGC 147 that may be artificially variable. We exclude affected sources from further analysis, with the exception of source #21181 in And IX at α(J2000) = $13^{\rm h}18^{\rm m}54\buildrel{\mathrm{s}}\over{.}04$, δ(J2000) = +43°12'28. This star may be marginally affected by Column Pulldown at 3.6 μm in epoch 2 and 4.5 μm in epoch 1, but since the color is similar between epochs and the amplitude at each wavelength is similar, we include it in the analysis as a high-confidence x-AGB star.

In Paper I, we estimated the total size of the x-AGB population by statistically subtracting the foreground and background sources. For the 6 galaxies with large x-AGB populations and firm (non-upper-limit) estimates from Paper I, the fraction of x-AGB stars detected as variable ranges from 33% to 61% (this includes IC 10, IC 1613, NGC 147, NGC 185, Sextans B, and WLM). Using the AGB star catalogs for the LMC and SMC from Boyer et al. (2011) and the OGLE I-band periods, we find that a DUSTiNGS-like survey would detect ≈62% of the OGLE-detected MC x-AGB stars (probability of a 2σ detection >0.75 with 1000 trials; Section 3.2.1). The difference in detection fractions is more likely a reflection of the differences in the distance to a galaxy and the epoch separation, and not to a real difference in the pulsation amplitudes between galaxies. A more distant galaxy has larger photometric uncertainties resulting in a lower probability of detection, and a longer epoch separation results in a higher probability of detection, in general.

Figure 4 shows the total number of x-AGB variable candidates, normalized to each galaxy's total stellar mass, as a function of metallicity, using the masses and metallicities from McConnachie (2012). There is no clear metallicity trend for the number of x-AGB stars in a galaxy, suggesting that the dominant factor determining the number of x-AGB stars is the star formation history.

Zoom In Zoom Out Reset image size

In Paper I, we showed that the CMDs for most of the DUSTiNGS galaxies are dominated by foreground and/or background sources, making it difficult to see the true color distribution of AGB stars. In Figure 5, we show the location of the variable candidates overplotted on the CMD, separated by the morphological type of the galaxy. The grayscale in panels a–c is the average background-subtracted CMD for all 50 galaxies. For comparison, Figure 5(d) includes a similar diagram with the variables identified in the LMC by Vijh et al. (2009, see Section 3.2.2). We detect few blue AGB stars compared to the LMC, though x-AGB stars are recovered at a similar rate. This is due to dustless stars pulsating with amplitudes that are smaller than the DUSTiNGS photometric uncertainty (Figure 14).

Zoom In Zoom Out Reset image size

Most of the x-AGB star candidates with [3.6]–[4.5] ≳ 1.5 mag are in IC 10, which by far harbors the largest x-AGB population (Table 1) and is thus the most likely to be host to rarer stellar types. Otherwise, only one star each from Sextans B, NGC 185, and NGC 147 is as red as the IC 10 x-AGB stars (Figure 5). With the exception of NGC 147 and NGC 185, the dSph galaxies harbor few examples of x-AGB candidates, as expected based on their star formation histories. Altogether, we find only 10 x-AGB star candidates in the remaining 35 dSph galaxies. Star #21181 in And IX is particularly red (Figure 5(c)), indicating strong dust production (Section 6.2).

The variable x-AGB candidates are most likely C-rich AGB stars, as is indicated in the MCs (e.g., Woods et al. 2011; Riebel et al. 2012). However, a small fraction might be O-rich, especially the brightest examples near the RSG border in Figure 1(a) where hot-bottom burning might disrupt the dredge-up of carbon. It is impossible to distinguish the different chemistries without near-IR photometry or spectra (Section 5.2).

Most of the x-AGB variable stars are confined to within 1–2 half-light radii (r_h; Figure 15). However, in some galaxies the x-AGB stars can be found at large radii (e.g., IC 10). In Sextans B, x-AGB stars fill the full DUSTiNGS coverage (≈13' × 14'). This is much more extended than the small half-light radius listed in McConnachie (2012) (r_h = 09), and agrees better with the new estimate from Bellazzini et al. (2014) of r_h = 19.

As in the LMC and SMC (Blum et al. 2006; Sandstrom et al. 2009; Boyer et al. 2011), the x-AGB star distributions are generally smooth and symmetric, with a few exceptions. In Sextans B, the x-AGB stars beyond the half-light radius are preferentially located to the north, which may be due to the influence of Sextans A and NGC 3109, which are located nearby and toward the south. In Sextans A, x-AGB stars are to the southwest of the center, avoiding the region of star formation on the east side of the galaxy. NGC 185 is lacking x-AGB stars to its north and south, despite the almost circular distribution of other stellar types in the galaxy. In IC 10, the x-AGB candidates avoid a region of strong star formation just to the southeast of the galaxy's center, which may be due in part to crowding (Paper I). Finally, in NGC 147 there is a lack of AGB stars on the southeast side of the galaxy. A detailed examination of the radial distribution of x-AGB stars in each galaxy will be presented in a forthcoming paper.

In general, pulsation amplitudes increase with periods (e.g., Vassiliadis & Wood 1993; Whitelock et al. 2000). Thus, amplitudes will increase as a star evolves, and larger amplitudes may thus be linked with more dust formation. With only two or three epochs, we can present only a lower-limit amplitude for each variable star (Δm = |m^epoch 1 − m^epoch 2|; Table 3). The mean Δm_3.6 detected for x-AGB stars is 0.50 ± 0.23 mag (standard deviation). Using the variable catalog from Vijh et al. (2009) in the LMC, we compute that the same Δm_3.6 for 820 x-AGB stars in the LMC is 0.36 ± 0.19 mag. The lower LMC mean amplitude is probably due to the smaller distance to the LMC. As a result, photometric uncertainties in the LMC data are smaller and enable the detection of smaller amplitudes.

The maximum amplitude is for a star in Sextans B (Δm_3.6 = 1.6 mag) that has a mean color of [3.6]–[4.5] = 0.61 mag and is located on the outskirts of the galaxy (upper right of Figure 15 at α(J2000)$=09^{\rm h}59^{\rm m}46\buildrel{\mathrm{s}}\over{.}07$, δ(J2000)=+05°26'088). The amplitude of this star is similar to the largest Δm_3.6 detected for an x-AGB star in the LMC ($\Delta m_{3.6}^{\rm max} = 1.8$ mag). Since amplitudes tend to decrease with wavelength for long-period variables (LPVs; e.g., Le Bertre 1992), the large amplitudes detected in some DUSTiNGS stars may imply amplitudes of several magnitudes in the optical. An alternate explanation is that the large IR amplitude is the result of a varying temperature of warm circumstellar dust. Since 3.6 μm may be measuring the Wien tail of the dust blackbody, even small changes in the total dust opacity or temperature could result in large changes at 3.6 μm. Changes in the strength of molecular absorption features (possibly from dust veiling) might also account for amplitude changes with wavelength.

In Figure 6, we show Δm_3.6 for each variable x-AGB star as a function of the host galaxy's metallicity. There is no clear trend indicating a change in pulsation properties in metal-poor environments.

Zoom In Zoom Out Reset image size

Figure 6 also shows Δm_3.6 as a function of color. Even with only two epochs, it is evident that the amplitude tends to increase with [3.6]–[4.5]. This trend supports a direct link between the star's pulsation and the dust production, and adds to evidence of the same link seen in the Galaxy, M33, and Sgr dSph variable AGB stars (Whitelock et al. 2006; McQuinn et al. 2007; McDonald et al. 2014).

The difference between the 3.6 and 4.5 μm amplitudes of x-AGB stars is small, with 〈Δm_3.6 − Δm_4.5〉 = 0.02 ± 0.11 mag. There is no trend for a systematic increase or decrease in the amplitudes from 3.6 to 4.5 μm.

All TP-AGB stars are LPVs (20 ≲ P ≲ 1000 days), pulsating regularly or semi-regularly during different phases of their evolution (e.g., Fraser et al. 2008; Soszyński et al. 2009). At the end of their evolution, AGB stars become Miras, which pulsate in the fundamental mode and generally have red colors, large amplitudes, and periods longer than 100 days (Iben & Renzini 1983). Miras and other LPVs have previously been detected in only four of the DUSTiNGS galaxies: NGC 147, NGC 185, IC 1613, and Phoenix.

5.1.1. NGC 147 and NGC 185

Lorenz et al. (2011) (hereafter, L+11) searched for LPVs in NGC 147 and NGC 185 with 30 epochs of photometry in the i-band. They found 182 and 387 LPVs, respectively, that are also detected in the K_S-band. Their survey includes stars within R ∼ 35, or about half the radius probed with DUSTiNGS. We find only six variables in common with NGC 147 and 22 with NGC 185. The i-band amplitudes measured by L+11 are ≈0.2–2 mag. Since we have detected only a small fraction of the L+11 LPVs, it follows that the amplitudes of the stars covered by L+11 are significantly smaller at 3.6 μm than in the i-band, on average. For the variable stars detected both here and by L+11, the mean amplitudes are 0.5 mag larger in the i-band than at 3.6 μm (〈A_i〉 = 0.9 mag and 〈A_3.6〉 = 0.4 mag).

Nowotny et al. (2003) used narrowband photometry centered on the TiO and CN bands to distinguish O-rich from C-rich AGB stars in NGC 147 and NGC 185 (also see Section 5.2). In NGC 185, the 24 L+11 LPVs detected as variable here include 5 O-rich stars (M type), 5 C-rich stars, and 4 stars with C/O ≈ 1 (S type). Most of the M stars that are also LPVs in L+11 are within 2 mag of the assumed TRGB, but of the five M stars that are detected as variable here, four are bright and red enough to be classified as x-AGB variables. This suggests that these four stars may be massive enough to undergo hot-bottom burning. Similarly, in NGC 147, we detect two M type, three C-rich, and one S type star. One of these M-type LPVs is quite bright at M_3.6 = −9.7 mag and may therefore be a massive AGB star or a RSG.

Figure 7 shows the DUSTiNGS CMD for these galaxies with L+11 variables marked. It is clear that optical and near-IR surveys are biased against the reddest sources that are uncovered by DUSTiNGS. This also suggests that red LPVs have larger pulsation amplitudes at 3.6 μm than bluer LPVs (Section 4.4). Within the same coverage surveyed by L+11, we detect 38 and 28 new variable star candidates in NGC 147 and NGC 185. Of these, most are x-AGB candidates.

Zoom In Zoom Out Reset image size

5.1.2. IC 1613

5.1.3. Phoenix

Carbon stars form more easily at low metallicity since less free oxygen is available to tie up newly formed carbon into CO molecules. Analysis of the IR spectra and spectral energy distributions (SEDs) has revealed that most of the x-AGB stars in the LMC and SMC are C-rich (Woods et al. 2011; P. Ruffle et al., in preparation), and it follows that C stars produce the majority of the AGB dust budget. Several of the DUSTiNGS galaxies have been observed using optical narrowband CN/TiO photometry or JHK photometry to identify C stars. The near-IR JHK photometry is less precise than CN/TiO photometry, though at subsolar metallicities, the stars with red J−K colors can be classified as high-confidence C stars (Cioni et al. 2006; Aringer et al. 2009; Boyer et al. 2013).

Neither method is sensitive to the dustiest stars because of circumstellar extinction at optical and near-IR wavelengths. Optical photometry can detect stars as red as [3.6]–[4.5] ≈ 0.5 mag, while the reddest DUSTiNGS star detected via J−K color has [3.6]–[4.5] = 0.9 mag (a star in Sag DIG, see below). Altogether, we confirm 70 variable x-AGB candidates as C stars. These stars are flagged in Table 3.

5.2.1. NGC 185

5.2.2. NGC 147

5.2.3. IC 10

5.2.4. Other Galaxies

Changes in the color of a dusty star over time indicate changes in the temperature and/or the optical depth of the dust. Between the epoch 1 and epoch 2 observations, the colors of the variable stars do not change significantly (Figure 8), suggesting that the pulsations within a cycle do not have a strong effect on the dust on a short timescale. Comparisons in the color between epoch 0 (Section 2.2) and the DUSTiNGS epochs is not possible because the 3.6 and 4.5 μm observations are not simultaneous in epoch 0.

Zoom In Zoom Out Reset image size

A subset of variables do show a significant color change between epochs; two of these stars are marked in Figure 8. We cannot disentangle the effects of temperature and dust mass on the color of these stars with only the DUSTiNGS data. Nonetheless, it is clear that in some of the x-AGB variable stars, the circumstellar environment is changing, sometimes significantly, over a short timescale, implying that dust formation is not continuous and smooth over the dust-producing phase.

A blue [3.6]–[4.5] color is typical for a less dusty C star that has CO+C₃ molecular absorption from 4–6 μm, but this feature becomes veiled as the dust emission increases (van Loon et al. 2008; Boyer et al. 2011). There are a handful of stars that appear blue in epoch 1 and red in epoch 2, suggesting that new dust appeared after the epoch 1 observations. These stars may have undergone an episode of eruptive dust production, perhaps owing to the influence of a binary companion. Further monitoring at IR wavelengths could confirm such phenomena.

Riebel et al. (2012) measured the dust produced by the entire LMC AGB population by fitting the full SEDs from the optical to the IR using the Grid of RSG and AGB modelS (GRAMS; Sargent et al. 2011; Srinivasan et al. 2011). GRAMS assumes a grain mixture of 90% amorphous carbon and 10% SiC with optical constants from Zubko et al. (1996) and Pégourié (1988), respectively, and a standard KMH grain size distribution (Kim et al. 1994). Using the catalog from Riebel et al. (2012), we find that most of stars with [3.6]–[4.5] > 0.1 mag are C-rich and that their dust-production rates ($\dot{D}$) increase with the color as (Figure 9):

$$\begin{equation} \log \dot{D} [M_\odot \,{\rm yr}^{-1}]= -9.5 + [1.4 \times ([3.6]\hbox{--}[4.5])]. \end{equation} \tag{ 2 }$$

We apply this relationship to the x-AGB variable star candidates in the DUSTiNGS galaxies to get a first estimate of their total dust production, using an average between the epoch 1 and 2 colors (Section 6.1). For stars redder than [3.6]–[4.5] ≈ 1.5 mag, the GRAMS dust-production rate appears to saturate at a $\log \dot{D} \sim -7.5\ [M_\odot \ {\rm yr}^{-1}]$. We do not apply this saturation limit to the rates reported in Table 3, but we do apply it in Figure 10 and in the right panel of Figure 11. If real, this saturation limit applies to eight x-AGB variables total in Sextans B, NGC 147, NGC 185, and IC 10 (Figure 5).

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

We caution that the use of different optical constants and/or the assumption of a different wind outflow velocity (GRAMS assumes 10 km s⁻¹) could result in dust-production rates that differ by up to a factor of 10. Therefore, rates reported here are not absolutely accurate and should only be compared to rates measured using similar assumptions. We assume that these parameters do no change from galaxy to galaxy.

Figure 10 shows the total dust production, normalized to the total stellar mass, from the variable x-AGB candidates as a function of metallicity. For the LMC and SMC points, we used the SAGE catalog to select x-AGB stars using the same selection criteria used here for DUSTiNGS galaxies. We note that the LMC and SMC points include the entire x-AGB population, while the DUSTiNGS points include only the x-AGB stars detected as variable (Section 3.2.1). The total dust output shows significant scatter and does not appear to be strongly affected by metallicity. The magenta square in the upper left corner belongs to And IX, and is totally dominated by a single x-AGB variable. This star (#21181) is marginally affected by imaging artifacts (Section 4.1), so its dust-production rate may be inflated. Except for this And IX star, no other x-AGB stars that may be affected by imaging artifacts are included in this plot or in Figure 11.

In Figure 11, each point indicates the mean (or the maximum) dust-production rate among the variable x-AGB stars within one of the galaxies. While the mean $\dot{D}$ is similar at all metallicities, the maximum $\dot{D}$ may show a slight preference for more metal-rich galaxies. However, the five galaxies with the highest maximum $\dot{D}$ (LMC, SMC, IC 10, NGC 147, and NGC 185) all have the largest x-AGB populations, so are less affected by stochastics and are therefore more likely to harbor the rarest, very dusty x-AGB stars.

In globular clusters, dust-producing AGB stars have potentially been detected down to [Fe/H] = −2.4 (e.g., Boyer et al. 2006, 2008, 2009a; McDonald et al. 2009, 2011a, 2011b; Sloan et al. 2010). These stars are low-mass (M ≲ 1 M_☉), oxygen-rich stars that will not ultimately contribute much to the total dust budget of a galaxy (e.g., Zhukovska & Henning 2013; Schneider et al. 2014). Searches for more massive metal-poor dust-producing AGB stars have yielded few examples in Local Group dwarf galaxies (Sloan et al. 2012). The most metal-poor example is Mag 29 in Sculptor dSph ([Fe/H] = −1.68 and $\log \dot{D}\ {[M_\odot \,{\rm yr}^{-1}]} = -8.21$; Sloan et al. 2009), though Sloan et al. (2012) estimate that its metallicity may be as high as [Fe/H] = −1. We find a total of 111 variable x-AGB stars with [Fe/H] < −1.5 and 12 with [Fe/H] < −2, suggesting that carbon formed in and around the stellar core is successfully dredged up and condensed into carbon grains at all metallicities.

We assume here that the wind expansion velocity and the dust properties are the same in all the DUSTiNGS galaxies, but this assumption may not be valid. For example, Wachter et al. (2008) find that the expansion velocity decreases from solar metallicity to SMC metallicity when the ratio C/O is fixed in their models. However, Wachter et al. (2008) also find that when models are given the same absolute abundance of free carbon, the SMC-, LMC-, and solar-metallicity models have similar expansion velocities and dust-condensation efficiency. The DUSTiNGS findings support this scenario.

We show the CMD of x-AGB variable star candidates as a function of metallicity in Figure 12. The sources in Sag DIG, And IX, and LGS 3 ([Fe/H] < −2) are of particular interest, as they indicate that AGB stars can produce large amounts of dust even in the early, metal-poor Universe. The AGB models from Marigo et al. (2013) indicate that C stars can form at least as early as 10⁸ yr at these metallicities (M_initial ∼ 5 M_☉), and perhaps even earlier depending on the details of the interplay between dredge-up and hot-bottom burning. AGB stars must therefore contribute significantly to the large dust reservoirs discovered in distant quasars (Bertoldi et al. 2003; Robson et al. 2004; Beelen et al. 2006).

Zoom In Zoom Out Reset image size

Sag DIG contains the largest very metal-poor x-AGB population ([Fe/H] = −2.1), with 17 stars that are either x-AGB variables or confirmed C stars within the x-AGB region of the CMD (Section 5.2.4). There is one additional unconfirmed source that nonetheless has colors similar to the confirmed x-AGB stars and is within 4' of the galaxy center (Figure 13). We consider this source to be a likely x-AGB star. Similarly, LGS 3 contains one additional possible x-AGB star along with the one that was identified as variable. These likely x-AGB stars are listed in Table 4.

Zoom In Zoom Out Reset image size

No variable x-AGB stars were discovered in the most metal-poor DUSTiNGS galaxies ([Fe/H] < −2.2: UMa II, Segue I, Leo IV, Coma Berenices, Bootes I, or Hercules). These galaxies have very low masses (M_V > −6.6 mag; for comparison And IX has M_V = −8.1 mag) and show no evidence of recent star formation (Paper I and McConnachie 2012), so it is not surprising that they are lacking examples of intermediate-aged AGB stars in general. In Paper I, we estimated the size of the AGB population after subtraction of foreground and background sources. In Hercules, we estimated a total of 20 ± 9 AGB stars, which is the smallest detected AGB population among the DUSTiNGS galaxies. In the other 4 very metal-poor galaxies, we found only upper limits.

If AGB stars can efficiently produce dust at any metallicity as the DUSTiNGS data suggests, it follows that the dust-to-gas ratios (DGRs) in the (C-rich) stellar envelopes may be similar at all metallicities when the stars are in the dust-producing "superwind" phase (Habing 1996; Groenewegen et al. 2007). This implies that the DGR in the interstellar medium (ISM) should also be independent of metallicity if AGB stars are the dominant source of interstellar dust and dust grains are minimally processed after leaving the circumstellar envelope. Observations of nearby galaxies show a clear correlation between metallicity and DGR in the ISM (e.g., Galametz et al. 2011; Sandstrom et al. 2013; Fisher et al. 2014), suggesting that AGB dust may be quickly destroyed or shattered (cf. Temim et al. 2015). The remnants of grain destruction/shattering may then provide the seeds for grain accretion in molecular clouds.