An Infrared Census of DUST in Nearby Galaxies with Spitzer (DUSTiNGS). IV. Discovery of High-redshift AGB Analogs^*

Photometric surveys typically separate C- and M-type stars using broadband near-IR or narrow-band optical filters. The broadband near-IR filters (JHK) are influenced by VO, TiO, and H₂O molecular features in M-type stars and CN and C₂ in C stars, while the narrow-band filters target TiO and CN molecular features at $\lambda \lt 7000\,\mathring{\rm A}$. The optical surveys are severely photon-limited and fail to detect the stars with even moderate circumstellar dust extinction. The near-IR JHK surveys capture more of the dusty stars, but classification is imprecise (Boyer et al. 2013, 2015a; Ruffle et al. 2015; Jones et al. 2017) and the dustiest stars, which can be faint even in the near-IR, generally remain undetected because of source confusion and insufficient sensitivity from the ground, especially in the K band. These impediments are overcome here with the IR channel of HST's Wide-Field Camera 3 (WFC3; Kimble et al. 2008), which has ample sensitivity and angular resolution in the near-IR for detecting AGB stars out to the edge of the Local Group.

Most HST surveys use the wide WFC3 filters (especially F110W and F160W; Dalcanton et al. 2012a, 2012b; Sabbi et al. 2013) to maximize the imaging depth, but these filters are too wide to be influenced by molecular absorption features in AGB star spectra, resulting in significant color overlap between these spectral types. Boyer et al. (2013) demonstrated the use of HST WFC3/IR medium-band filters to successfully separate C- and M-type stars in a field in the inner disk of M31. These filters fall within the H₂O and CN+C₂ features at 1.2–1.5 μm (filters: F127M, F139M, and F153M, Figure 1), thus firmly dividing the two spectral types with minimal cross-contamination. Furthermore, even the dustiest stars remain separated by spectral type in the WFC3/IR colors, unlike in J − K colors which overlap significantly once dust is included.

Zoom In Zoom Out Reset image size

We observed six DUSTiNGS galaxies with HST's WFC3/IR in Cycle 23 as part of GO-14073 from 2016 October to 2016 August. Table 1 lists the observed galaxies and their properties. We chose these six galaxies because they have large populations of dust-producing AGB candidates and span the entire metallicity range of the DUSTiNGS sample. Four of the galaxies are gas-rich dwarf irregular (dIrr) systems with H ii regions that point to sites of current massive star formation. One galaxy (Pegasus/DDO 216) is a transition-type galaxy (dTrans). It is gas-rich, but there are no H ii regions. One galaxy (NGC 147) is a dwarf spheroidal (dSph). It is a gas-poor galaxy with no current star formation, but a sizeable intermediate-aged AGB population is evident (e.g., Lorenz et al. 2011; Hamedani Golshan et al. 2017). The star formation histories of all six galaxies are described by Weisz et al. (2014).

Table 1.  Target Information

Galaxy	Alt. Name	Morph. Type	d	[Fe/H]	$12+\mathrm{log}({\rm{O}}/{\rm{H}})$	${M}_{{\rm{V}}}$	${A}_{{\rm{V}}}$
		(Mpc)			(mag)	(mag)
NGC 147	DDO 3	dE/dSph	0.76	−1.11	...	−14.6 ± 0.1	0.47
IC 10	UGC 192	dIrr	0.77	−1.28	8.19 ± 0.15	−15.0 ± 0.2	2.33
Pegasus	DDO 216	dTrans/dIrr	0.98	−1.4 ± 0.20	7.93 ± 0.13	−12.2 ± 0.2	0.19
Sextans B	DDO 70	dIrr	1.43	−1.6 ± 0.10	7.53 ± 0.05	−14.5 ± 0.2	0.09
Sextans A	DDO 75	dIrr	1.46	−1.85	7.54 ± 0.06	−14.3 ± 0.1	0.12
Sag DIG	Sgr dIG	dIrr	1.09	−2.1 ± 0.20	7.26–7.50	−11.5 ± 0.3	0.34

Note. Distances are derived from F814W (I-band) TRGB estimates from McQuinn et al. (2017, Paper III). ${A}_{{\rm{V}}}$ is from Weisz et al. (2014). ${M}_{{\rm{V}}}$ is from McConnachie (2012) and references therein. The values of $[\mathrm{Fe}/{\rm{H}}]$ adopted here are derived from RGB colors from Nowotny et al. (2003), Bellazzini et al. (2014), Tikhonov & Galazutdinova (2009), McConnachie et al. (2005), Momany et al. (2002), and Sakai et al. (1996). Stellar metallicities are derived from spectroscopy of RGB stars in NGC 147 agree with our adopted value (Geha et al. 2010). However, Kirby et al. (2017) measured $[\mathrm{Fe}/{\rm{H}}]=-{1.88}_{-0.09}^{+0.13}$ from RGB star spectroscopy in Sag DIG. ISM gas-phase oxygen abundances ($12+\mathrm{log}({\rm{O}}/{\rm{H}})$) are from from Mateo (1998), Lee et al. (2006), and Saviane et al. (2002).

Download table as:  ASCIITypeset image

The $2\buildrel{\,\prime}\over{.} 1\times 2\buildrel{\,\prime}\over{.} 3$ WFC3/IR field of view is very small compared to the DUSTiNGS footprint ($\gtrsim 10^{\prime} \times 10^{\prime}$), so we observed galaxies with two or three fields, placed to maximize the coverage of DUSTiNGS AGB candidates while also maximizing the inclusion of those with [3.6]–[4.5] > 0.5 mag, i.e., the dustiest stars. Figure 2 shows the placement of the WFC3/IR fields. In total, we have covered 99 of the 375 original DUSTiNGS x-AGB variables reported for these six galaxies in Paper II.

Zoom In Zoom Out Reset image size

Observations are summarized in Table 2. We imaged each field with the F127M, F139M, and F153M filters, employing four dithers with the WFC3-IR-DITHER-BOX-MIN pattern to minimize image defects and to maximize the spatial resolution (the resulting mosaics are Nyquist sampled). The total exposure times are 796.9–896.9 s in each filter, depending on the detector sampling sequence used. WFC3/IR is non-destructively read out multiple times during an exposure using sequences combining long and short reads that provide uniform sampling over a wide range of stellar magnitudes (MULTIACCUM mode¹⁶ ). For the first three exposures in each filter, we adopted sample sequence STEP100 with NSAMP = 8. The fourth and final exposure for each filter was set to fit within the remaining orbit visibility, which differed for each galaxy: STEP100,NSAMP = 8; STEP100,NSAMP = 9; STEP50,NSAMP = 10; or SPARS25,NSAMP = 12.

Table 2.  Observations

					F127M	F139M	F153M
Galaxy	Field	R.A.	Decl.	Start Date	${t}_{\exp }$	${t}_{\exp }$	${t}_{\exp }$	Orient.
		(J2000)	(J2000)	(UT)	(s)	(s)	(s)	(E of N)
IC 10	A	00h20m0822	+59d20m2568	2015 Oct 18 04:56:34	875.6	896.9	896.9	229
IC 10	B	00h20m0883	+59d16m3724	2015 Oct 19 04:48:44	875.6	896.9	896.9	216
IC 10	C	00h20m3695	+59d16m4882	2015 Oct 19 16:38:31	875.6	896.9	896.9	211
NGC 147	A	00h32m4189	+48d32m4421	2015 Oct 09 22:28:15	846.9	875.6	846.9	352
NGC 147	B	00h33m2492	+48d32m4026	2015 Oct 07 02:57:33	846.9	875.6	846.9	395
NGC 147	C	00h33m0814	+48d29m1266	2015 Oct 09 02:59:39	846.9	875.6	846.9	365
Pegasus dIrr	A	23h28m3315	+14d43m5462	2015 Oct 02 06:57:09	796.9	825.6	796.9	−101
Pegasus dIrr	B	23h28m4142	+14d44m4116	2015 Oct 04 07:07:30	796.9	825.6	796.9	−289
Sag DIG	A	19h29m4807	−17d39m5921	2016 Aug 07 13:47:10	796.9	846.9	796.9	−490
Sag DIG	B	19h29m5804	−17d41m0773	2016 Aug 07 14:46:50	796.9	846.9	796.9	−446
Sextans A	A	10h10m5520	−04d40m5664	2016 Mar 06 04:03:59	796.9	825.6	796.9	−811
Sextans A	B	10h10m5891	−04d42m5165	2016 Apr 04 01:21:55	796.9	825.6	796.9	−458
Sextans B	A	10h00m0249	+05d19m3836	2016 Mar 14 02:54:20	796.9	825.6	796.9	−241
Sextans B	B	09h59m5300	+05d20m0084	2016 Mar 17 17:08:40	796.9	825.6	796.9	−372

Note. This data is from HST program GO-14073 (P.I. Boyer; doi:10.17909/T9HM3M).

Download table as:  ASCIITypeset image

We used HST Drizzlepac v2.0 to create mosaics in each filter. We combined the calibrated, flat-fielded exposures (flt.fits files) using Astrodrizzle to create a stacked, drizzled image with pixel scale 00642/pix. Stellar positions were measured on the drizzled F127M image, allowing for forced photometry of faint stars in the individual images.

We performed PSF photometry on the flt.fits images using DOLPHOT's WFC3-specific module (Dolphin 2000). We retain only stars with error flag <8 and signal-to-noise >4 and require low sharpness and crowding values. Restricting the sharpness parameter minimizes contamination from extended sources and cosmic rays, and restricting the crowding parameter ensures that a star's flux measurement is not substantially affected by nearby stars. We adopt ${({\rm{\Sigma }}{\mathrm{Sharp}}_{\lambda })}^{2}\lt 0.1$ and $({\rm{\Sigma }}\,{\mathrm{Crowd}}_{\lambda })\lt 1.5$ mag. The crowd parameter is a measure of how much brighter a star would be if nearby stars were not subtracted simultaneously. The sharp parameter is positive for sharp sources (cosmic rays) and negative for extended sources. These initial quality cuts eliminate most contamination and poor measurements from the catalog, but inspection of the images suggests that some extended and/or blended sources remain. We therefore flag sources with ${({\rm{\Sigma }}{\mathrm{Sharp}}_{\lambda })}^{2}\gt 0.02$ as those that require further scrutiny.

Artificial star tests indicate that photometry is >90% complete at ${\rm{S}}/{\rm{N}}\gt 10$. Since our AGB analysis is restricted to sources brighter than this limit, our AGB samples are near-complete. To identify dust-producing stars, we use Spitzer photometry from Paper I. We use the magnitudes measured after coadding the two observing epochs to mitigate the effect of pulsation in the IR data.

The HST catalog was matched to the Spitzer catalog using the DAOMaster routine (Stetson 1987), which iteratively solves for the transformation coefficients and assigns matches, starting with a 3'' radius and decreasing to a 06 radius. The HST and Spitzer data are poorly matched in both spatial resolution and sensitivity; to minimize mismatches between these two data sets, we restrict the HST input catalog to stars with signal-to-noise (S/N) >10 in all three filters thereby excluding the faintest stars that are unlikely to have Spitzer counterparts. IC 10 is the most densely populated galaxy in our sample, and the lower-resolution Spitzer data are therefore strongly affected by crowding (Paper I). We thus restrict the IC 10 HST catalog to ${\rm{S}}/{\rm{N}}\gt 15$ to achieve a good match to the Spitzer sources. We manually checked all DUSTiNGS x-AGB variables from Paper II that were not matched to HST sources using these criteria. Six DUSTiNGS variables match HST sources with low S/N, and these were added back into the catalogs.

All photometry presented in this paper is corrected for extinction using the ${A}_{{\rm{V}}}$ values listed by Weisz et al. (2014). That work takes values from Schlafly & Finkbeiner (2011) if ${A}_{{\rm{V}}}\lt 0.2$. Higher extinction values were estimated by comparing observed and simulated optical color–magnitude diagrams. We assume ${A}_{{\rm{F}}127{\rm{M}}}/{A}_{{\rm{V}}}=0.27391$, ${A}_{{\rm{F}}139{\rm{M}}}/{A}_{{\rm{V}}}\,=0.23979$, ${A}_{{\rm{F}}153{\rm{M}}}/{A}_{{\rm{V}}}=0.20366$, ${A}_{3.6}/{A}_{{\rm{V}}}=0.06706$, and ${A}_{4.5}/{A}_{{\rm{V}}}=0.05591$, from the Padova simulations.¹⁷

Figure 3 shows the HST color–magnitude diagrams, and Table 3 lists the columns included in the final catalog, available to download in the electronic version of this paper and on VizieR. We have made no corrections to the HST astrometry, which has ≈200 mas absolute accuracy and ≈10 mas relative accuracy.¹⁸ The Spitzer positions are aligned to 2MASS by the Spitzer pipeline and thus have an absolute astrometric accuracy ≈150 mas.

Zoom In Zoom Out Reset image size

To select C- and M-type stars, we first measure the tip of the red giant branch (TRGB). Once sub-TRGB stars are eliminated, we use the Aringer et al. (2009, 2016) models to define color cuts that separate AGB spectral types. In all six galaxies, we find a total of 908 C stars and 2120 M stars; Table 4 lists the final adopted star counts.

Table 4.  Source Counts

		C				M
Galaxy	${N}_{\mathrm{TRGB}}$	All	Dustya	Paper II b	Cont.c	All	Dustya	Paper II b	Cont.c
IC 10	2928	531	49	26	5	1766	15	6	2
NGC 147	370	65	21	14	1	265	2	1	3
Pegasus	154	44	8	4	1	32	0	0	3
Sag DIG	149	16	4	3	6	3	2	0	3
Sextans A	386	65	14	10	3	9	3	1	0
Sextans B	545	187	24	11	5	45	4	1	0

Notes. ${N}_{\mathrm{TRGB}}$ indicates the number of stars brighter than the TRGB in F153M. C- and M-type stars are identified by their HST near-IR colors (Section 2.4.2).

^aThe subset of C and M stars with [3.6]–[4.5] color >4σ that are not classified as contaminants. (Section 2.4.3). ^bThe subset of C and M stars identified as x-AGB variables in Paper II not classified as contaminants. ^cThe subset of C and M stars that show >4σ dust excess, but are suspected contaminants based on their location in Figure 8.

Download table as:  ASCIITypeset image

2.4.1. TRGB

To find the TRGB in each filter, we follow the commonly used strategy described by Méndez et al. (2002). First, we select stars with ${\rm{F}}129{\rm{M}}\mbox{--}{\rm{F}}153{\rm{M}}\gt 0.1$ mag to eliminate main-sequence stars. Next, we pass the Gaussian-smoothed luminosity function through a Sobel edge-detection filter. We perform 500 Monte Carlo Bootstrap resampling trials, each time adding 4σ Gaussian random photometric errors and random variations on the bin size and starting magnitude of the luminosity function. The final TRGB and its uncertainty are the mean and standard deviation of a Gaussian function fit to the TRGB estimates from the 500 trials (Table 5). Other techniques to find the TRGB can result in more precise measurements than the edge-detector method we employ, such as Bayesian maximum likelihood (e.g., Makarov et al. 2006; Conn et al. 2012; McQuinn et al. 2016). However, since our red giant branches are deep and well populated, Bayesian techniques result in only a marginal gain in accuracy and reliability.

Table 5.  Near-IR TRGB

TRGB
Galaxy	F814W	F127M	F139M	F153M	${(m-M)}_{0}$
	(mag)	(mag)	(mag)	(mag)	(mag)
N 147	24.39 ± 0.06	19.03 ± 0.04	18.82 ± 0.04	18.45 ± 0.04	24.39 ± 0.06
IC 10	24.43 ± 0.03	19.27 ± 0.04	19.06 ± 0.04	18.68 ± 0.04	24.43 ± 0.03
Peg	24.96 ± 0.04	19.76 ± 0.05	19.55 ± 0.05	19.20 ± 0.05	24.96 ± 0.04
Sex B	25.77 ± 0.03	20.71 ± 0.05	20.52 ± 0.04	20.23 ± 0.05	25.77 ± 0.03
Sex A	25.82 ± 0.03	20.77 ± 0.09	20.58 ± 0.09	20.32 ± 0.09	25.82 ± 0.03
Sag	25.18 ± 0.04	20.18 ± 0.06	20.03 ± 0.06	19.77 ± 0.07	25.18 ± 0.04

Note. Distance modulii are derived from the F814W TRGB (Paper III), using the relationship from Rizzi et al. (2007). These are within 1σ of distance modulii listed by McConnachie (2012), except Pegasus dIrr (2σ) and NGC 147 (3σ).

Download table as:  ASCIITypeset image

The TRGBs are fairly stable against metallicity in these filters, similar to the F814W TRGB (e.g., Rizzi et al. 2007). Using the distance modulii listed in Table 5, all three filters show brighter TRGBs at higher metallicity, with the F153M filter showing the largest metallicity effect (${\rm{\Delta }}\mathrm{TRGB}({\rm{F}}127{\rm{M}}/{\rm{F}}139{\rm{M}}/{\rm{F}}153{\rm{M}})\,=0.35/0.42/0.53$ mag).

2.4.2. Color–Color Selection

To select C- and M-type stars, we start with all sources brighter than the F153M TRGB. This restriction results in the loss of the dustiest AGB stars, which are made faint in F153M by circumstellar extinction. We recover these stars in Section 2.4.3.

Next, we classify the resulting subset of stars based on their location in the F127M–F139M versus F139M–F153M color–color diagram (CCD; Figure 4). When sub-TRGB stars are excluded (as they are here), this diagram has three main branches: an M-type branch, a C star branch, and a branch that includes a mixture of K-type stars, main-sequence stars, and foreground stars. We place the divisions in Figure 4 following the C, M, and K star models from Aringer et al. (2009, 2016). The C star models included in Figure 4 (plus symbols) are those with $-0.8\leqslant \mathrm{log}g\ [\mathrm{cm}\,{{\rm{s}}}^{-2}]\leqslant 0$ and C/O = 1.4 (light orange), 2 (orange), and 5 (red). For the M/K stars (crosses), we include ${T}_{\mathrm{eff}}\leqslant 3700$ K and $-1\leqslant \mathrm{log}g\ [\mathrm{cm}\,{{\rm{s}}}^{-2}]\leqslant -0.5;$ darker blue colors represent lower effective temperatures. Only models with $[\mathrm{Fe}/{\rm{H}}]\ \leqslant \ -1$ are included. The solid lines mark the adopted C- and M-type color divisions. Red and blue arrows illustrate the direction and magnitude of circumstellar extinction for $E(J-{K}_{S})=1$ mag for M-type stars with 60% silicate + 40% AlOx and C stars with 70% amorphous carbon +30% SiC (Groenewegen 2006).

Zoom In Zoom Out Reset image size

Both the C- and M-type models are slightly bluer than the data in both colors. This discrepancy (approximately 0.05 mag) has no effect on the C star definition since C stars are fairly well isolated from both M- and K-type stars. However, the discrepancy is large enough to substantially affect the number of M-type stars selected because of strong contamination from warmer K-type stars. The reason for the slight shifts between the models and data is unclear—one possibility is a bias in our adopted ${A}_{{\rm{V}}}$ values (Table 1). Another culprit might be the adopted water opacity in the models from Aringer et al. (2016), who note that other opacities shift the near-IR colors by about 0.05 mag. Furthermore, deviations from hydrostatic, spherically symmetric atmospheres and non-LTE conditions can affect the water opacity. Because of this slight color mismatch between the data and the models, we cannot use the exact colors of the models to define the transition from K-type stars to M-type stars (at around 3600 K) in the data. Instead, we note that ${\rm{M}}0+$ models occupy the M/K-star sequence beginning just blueward of the knee in the CCD at ${\rm{F}}127{\rm{M}}\mbox{--}{\rm{F}}139{\rm{M}}\approx 0.2$ mag and ${\rm{F}}139{\rm{M}}\mbox{--}{\rm{F}}153{\rm{M}}\approx 0.4$ mag. We therefore select M-type stars based on the location of this knee in the data (Figure 5).

Zoom In Zoom Out Reset image size

The adopted (F127M–F139M, F139M–F153M) positions of the C and M divisions in the CCD are as follows:

$$\begin{eqnarray}&&{\rm{M}}\ \mathrm{stars}:\qquad [(-0.6,0.35),(0.35,0.35),(1.0,1.1)]\end{eqnarray} \tag{ 1 }$$

$$\begin{eqnarray}&&{\rm{C}}\ \mathrm{stars}:\qquad [(0.35,-0.75),(0.25,0.00),\\ &&(0.25,0.23),(1.0,1.1)].\end{eqnarray} \tag{ 2 }$$

The line between C- and M-type stars on the right side of the CCD follows the direction of circumstellar extinction. Groenewegen & Sloan (2017; GS2017) fit ∼500 dusty Magellanic Cloud stars using a modified version of the DUSTY radiative transfer code (Groenewegen 2012), and we passed the resulting spectral energy distributions (SEDs) through the medium-band HST filters used here. GS2017 used high-quality, multi-epoch data in the near-IR, so the SEDs used to derive the magnitudes in Figure 4 are well constrained around 1 μm. The C- or M-type classification in GS2017 is based on features in the stars' mid-IR spectra (e.g., Jones et al. 2017). We include the dustiest LMC stars from the GS2017 sample in Figure 4, with C-type stars as red open diamonds and M-type stars as blue open circles. With the exception of one star in Figure 4, the dusty C- and M-type stars stay separated on the HST CCD, indicating that contamination across the dividing line is low. For reference, we also show the COLIBRI tracks (Marigo et al. 2017), with Nanni et al. (2016) dust growth models.

We note that the GS2017 stars represent the dustiest examples of AGB stars. Less dusty examples that have not completely veiled their molecular features can fall along the entire H₂0 and C/O sequences, with dust affecting the color as indicated by the blue and red extinction arrows in Figure 4.

2.4.3. Dusty Stars

Dusty AGB stars can be fainter than the TRGB both due to variability and to circumstellar extinction. To recover these sources, we turn to the Spitzer data. For both O-rich and C-rich stars in the Magellanic Clouds, the [3.6]–[4.5] color is approximately proportional to the dust-production rates (Riebel et al. 2012; Srinivasan et al. 2016; Sloan et al. 2016), especially at color >0.1 mag (Figure 6). We therefore use the [3.6]–[4.5] color as a proxy for dust excess, which we measure by comparing a source's color to the mean color of stars within 1 mag bins at 4.5 μm. We flag sources as dusty if the excess exceeds 4σ. Dusty objects that are fainter than the F153M TRGB are included in our sample if they are brighter than the 3.6 μm TRGB (Paper III). There are examples of dusty carbon stars that are fainter than the 3.6 μm TRGB (e.g., in the LMC; Gruendl et al. 2008), but given the small stellar masses of our galaxies relative to the LMC, we expect to miss only a few such stars, if any, which does not impact our star counts. Once the dusty stars are recovered, we classify them as C- or M-type using the criteria outlined in Section 2.4.2.

Zoom In Zoom Out Reset image size

Paper II adopted the definition set by Blum et al. (2006), who defined the dustiest AGB stars (dubbed extreme AGB or x-AGB stars) in the LMC as those with $J-[3.6]\gt 3.1$ mag. IR spectroscopy has confirmed that x-AGB stars are dominated by carbon stars, though there are also O-rich examples (Trams et al. 1999; van Loon et al. 2008; Ruffle et al. 2015; Boyer et al. 2015a; Jones et al. 2017). Most of the variable stars detected in the DUSTiNGS survey occupy the same space in [3.6] versus [3.6]–[4.5] as the Magellanic Cloud x-AGB stars, and were classified as such. The HST coverage includes 99 of the DUSTiNGS x-AGB variables. Of these, 90 are included in our HST catalog and 77 are confidently identified as C- or M-type (Table 4). Seven did not fall within the C or M regions of the HST CCD and six are identified as possible contaminants (see Section 2.5).

Visual inspection verifies that nine DUSTiNGS x-AGB variables do not have HST counterparts (Table 6). These nine sources have an average IR color of [3.6]–[4.5] = 1.1 mag (Figure 7), and include five of the six dustiest sources identified in Paper II; those with slightly bluer Spitzer colors in Figure 7 may be somewhat affected by CO absorption in the 4.5 μm band. All nine sources are well isolated within $r\gtrsim 3$ pix in the HST images, which eliminates crowding as a factor in their non-detection and suggests that they are truly fainter than the sensitivity limit of our HST observations. Their Spitzer colors closely follow COLIBRI AGB tracks, and we therefore propose that these are AGB stars with very strong circumstellar dust extinction. In fact, their IR colors and absolute magnitudes are similar to the progenitor of the intermediate-luminosity optical transient SN2008S, which may have been an electron-capture SN from a massive AGB star (Prieto et al. 2008; Khan et al. 2010). Alternatively, they could be other red objects such as active galactic nuclei.

Zoom In Zoom Out Reset image size

Table 6.  DUSTiNGS x-AGB Variables without HST Counterparts

Galaxy	ID	[3.6]	[4.5]	Amp.
		(mag)	(mag)	(mag)
IC 10	115785	17.42 ± 0.06	15.20 ± 0.03	0.62
IC 10	109882	15.61 ± 0.03	14.39 ± 0.03	0.45
IC 10	110204	15.91 ± 0.05	14.58 ± 0.03	0.85
IC 10	111624	15.71 ± 0.07	15.13 ± 0.03	0.74
Sex A	90428	17.01 ± 0.05	15.84 ± 0.05	0.50
Sex A	94477	16.89 ± 0.05	15.45 ± 0.03	0.67
Sex B	85647	15.93 ± 0.02	15.43 ± 0.03	0.21
Sex B	93730	16.38 ± 0.03	15.46 ± 0.03	0.40
Sex B	96433	15.81 ± 0.02	15.10 ± 0.03	0.33

Note. The ID is the DUSTiNGS ID from Papers I and II. The amplitude is the change in 3.6 μm magnitude between the two DUSTiNGS epochs (larger amplitudes are generally indicative of more dust).

Download table as:  ASCIITypeset image

By combining the HST and Spitzer data, we find additional examples of both M- and C-type dusty stars that were not identified as variable x-AGB stars in Paper II, most likely because they were observed at an unfavorable pulsation phase. Altogether, we increase the number of known dusty AGB stars by approximately 50% (Table 4). Most of the x-AGB stars that we identify as AGB stars here are C stars, similar to what is seen in the Magellanic Clouds. However, there are examples of M-type x-AGB stars, most notably in IC 10 (see Section 3.2).

2.4.4. Contamination From K-type Giants

We do not expect significant contamination from other source types among our sample. Possible contaminants include young stellar objects (YSOs), planetary nebulae (PNe), and post-AGB stars, but the comparatively short lifetimes of these objects make them even more rare than AGB stars.

To minimize contamination, we use data from the SMC as a guide. The Surveying the Agents of Galaxy Evolution (SAGE) program targeted hundreds of sources in both the LMC and the SMC with the InfraRed Spectrograph (IRS) on board Spitzer (Kemper et al. 2010) and Ruffle et al. (2015) compared the spectroscopically classified sources to photometric classifications in the SMC. While their near-IR filters are different from ours (they use J and ${K}_{{\rm{S}}}$), it is clear from Figure 13 in Ruffle et al. (2015) that dusty AGB stars follow a branch that extends from the TRGB to red colors and faint magnitudes. A similar branch is evident in our near-IR CMDs (Figure 8). YSOs, PNe, and post-AGB stars, on the other hand, tend to be faint and blue in the near-IR. We therefore flag sources to the left of the solid line in Figure 8 as possible contaminants. The slope of the contaminant line was determined using the expected direction of circumstellar extinction from Groenewegen (2012).

Zoom In Zoom Out Reset image size

Another source of contamination is from red supergiants (RSGs), which have similar infrared properties to AGB stars. RSGs are not easily distinguished with this data set, though they do tend to be warmer than AGB stars, and thus fall toward the lower right end of the M-star sequence in Figure 4. A sample of six confirmed RSGs in IC 10 from Britavskiy et al. (2015) fall below the knee in Figure 4, with some even falling within the foreground sequence. We expect the number of RSGs in our AGB sample to be small, given their comparative rarity. This is discussed more in Section 3.2.

We identify 120 dust-producing carbon stars, almost twice the number detected via two-epoch variability in Paper II. Figure 9 shows the Spitzer color–magnitude diagrams for sources detected in the HST images. These CMDs are significantly cleaner than those presented by Paper I because the contamination from extended background sources is substantially reduced by including the high-resolution HST data. The relative positions of dusty C and M stars in the Spitzer CMD are similar, though the dusty C stars tend to be more tightly concentrated than M stars.

Zoom In Zoom Out Reset image size

Sag DIG and Sextans A are the two most metal-poor galaxies in our sample, with gas-phase ISM metallicities more than an order of magnitude below solar, suggesting that even the youngest stars are very metal-poor. However, both galaxies show a sizeable population of dust-producing C stars, with very similar colors to those seen in more metal-rich galaxies. For C stars, the [3.6]–[4.5] color is approximately proportional to the dust-production rates (Figure 6; Riebel et al. 2012; Sloan et al. 2016; Srinivasan et al. 2016). The similar colors of the C stars across our sample therefore suggest that dust masses are similar at all metallicities. The high efficiency of the third dredge-up at low metallicity (e.g., Karakas et al. 2002) appears to provide metal-poor C stars with plenty of material for dust condensation.

Several dusty sources identified as C stars by the HST color definitions from Section 2.4.2 are flagged as possible contaminants based on their positions in Figure 8. This includes some of the reddest sources identified in the survey (see Figure 9). There is a strong possibility that these sources are C-rich PNe or post-AGB stars. In the Magellanic Clouds, the post-AGB stars tend to be brighter and redder than PNe at 4.5 μm (Ruffle et al. 2015; Jones et al. 2017). For this reason, we favor the possibility that the contaminants around ${M}_{[4.5]}\sim -10$ mag are post-AGB stars, including the reddest object in IC 10.

The reddest (non-potential-contaminant) carbon stars in our sample have [3.6]–[4.5] ∼ 1 mag, which corresponds to a dust-production rate of $\mathrm{log}{\dot{M}}_{\mathrm{dust}}=-9$ to −8 $[{M}_{\odot }\,{\mathrm{yr}}^{-1}$], according to the SMC relationship in Figure 6.

We find a total of 26 dust-producing M-type candidates, comprising 1.2% of the M star population. We list them in Table 7 along with their photometry and classification confidence. The reddest among these have [3.6]–[4.5] ∼ 1 mag, which corresponds to a dust-production rate of $\mathrm{log}{\dot{M}}_{\mathrm{dust}}=-7.5$ to −6.5 $[{M}_{\odot }{\mathrm{yr}}^{-1}]$ using the SMC relationship in Figure 6.

Table 7.  Dusty M Stars

Galaxy	DUSTiNGS	R.A.	F127M	F139M	F153M	[3.6]	[4.5]	Note
	ID	Decl.	(mag)	(mag)	(mag)	(mag)	(mag)
N 147	103322	00h33m0979	18.849 ± 0.005	18.718 ± 0.006	18.293 ± 0.005	16.59 ± 0.05	16.19 ± 0.06	H
		+48d29m0928
N 147	112780	00h33m0561	21.006 ± 0.021	20.524 ± 0.018	19.992 ± 0.015	15.28 ± 0.03	14.70 ± 0.03	Lx
		+48d28m4743
IC 10	102032	00h20m1498	18.883 ± 0.006	18.768 ± 0.006	18.181 ± 0.005	17.10 ± 0.05	16.80 ± 0.05	H
		+59d21m0116
IC 10	105880	00h20m1238	22.437 ± 0.068	21.889 ± 0.052	21.221 ± 0.037	15.36 ± 0.05	14.88 ± 0.06	L
		+59d19m4195
IC 10	105975	00h20m1230	15.666 ± 0.001	15.518 ± 0.001	14.955 ± 0.001	13.16 ± 0.03	12.84 ± 0.03	H
		+59d20m4209
IC 10	107349	00h20m1139	19.530 ± 0.015	19.278 ± 0.017	18.821 ± 0.017	16.68 ± 0.10	16.06 ± 0.07	L
		+59d19m0387
IC 10	112431	00h20m0799	17.275 ± 0.002	17.727 ± 0.003	16.856 ± 0.002	14.91 ± 0.05a	14.73 ± 0.04	Hx
		+59d19m3174
IC 10	118138	00h20m0407	17.420 ± 0.003	17.764 ± 0.004	17.029 ± 0.003	14.30 ± 0.03	14.08 ± 0.03	Hx
		+59d19m3050
IC 10	121876	00h20m0150	18.724 ± 0.005	18.516 ± 0.005	18.010 ± 0.004	15.18 ± 0.03	14.32 ± 0.03	Hx
		+59d20m0205
IC 10	122923	00h20m0081	18.816 ± 0.005	19.590 ± 0.010	18.596 ± 0.006	16.40 ± 0.03	16.04 ± 0.05	H
		+59d20m3204
IC 10	96092	00h20m1900	20.783 ± 0.019	20.413 ± 0.020	19.955 ± 0.018	14.35 ± 0.02	13.51 ± 0.03	Lx
		+59d16m3947
IC 10	98013	00h20m1775	20.568 ± 0.016	19.930 ± 0.012	19.235 ± 0.009	14.18 ± 0.02	13.62 ± 0.03	Lx
		+59d16m1822
IC 10	107616	00h20m1130	19.040 ± 0.006	18.798 ± 0.006	18.259 ± 0.005	16.36 ± 0.05	15.82 ± 0.06	H
		+59d17m1795
IC 10	108360	00h20m1081	20.113 ± 0.012	19.591 ± 0.010	19.030 ± 0.008	15.48 ± 0.03	14.85 ± 0.03	L
		+59d16m1222
IC 10	117402	00h20m0472	20.173 ± 0.012	19.937 ± 0.012	19.315 ± 0.009	17.17 ± 0.07	16.75 ± 0.07	H
		+59d17m1620
IC 10	120247	00h20m0274	22.315 ± 0.063	21.526 ± 0.042	20.642 ± 0.024	15.74 ± 0.03	14.78 ± 0.03	H
		+59d17m1808
IC 10	120468	00h20m0258	21.510 ± 0.033	20.860 ± 0.025	20.074 ± 0.017	15.20 ± 0.02	14.69 ± 0.03	Hx
		+59d16m5246
Sex B	84519	10h00m0361	22.688 ± 0.078	22.249 ± 0.068	21.659 ± 0.053	18.42 ± 0.07	17.50 ± 0.05	L
		+05d18m5529
Sex B	109067	09h59m5570	20.917 ± 0.020	20.443 ± 0.018	19.937 ± 0.016	16.48 ± 0.03	15.68 ± 0.06	Lx
		+05d19m5304
Sex B	116156	09h59m5343	21.706 ± 0.059	21.232 ± 0.043	20.644 ± 0.034	16.93 ± 0.05	16.18 ± 0.05	L
		+05d18m5176
Sex B	123385	09h59m5109	21.060 ± 0.076	20.822 ± 0.073	20.463 ± 0.061	17.02 ± 0.06	16.31 ± 0.05	L
		+05d19m2661
Sex A	90034	10h10m5954	17.860 ± 0.004	17.835 ± 0.004	17.404 ± 0.003	16.21 ± 0.03	15.94 ± 0.05	H
		−04d40m5869
Sex A	91324	10h10m5908	22.921 ± 0.098	22.707 ± 0.103	21.936 ± 0.068	17.98 ± 0.05	17.37 ± 0.05	L
		−04d43m5930
Sex A	94328	10h10m5803	22.413 ± 0.065	21.965 ± 0.057	21.112 ± 0.038	16.98 ± 0.05	16.22 ± 0.05	Lx
		−04d43m0422
Sag	44334	19h29m5794	18.034 ± 0.004	17.723 ± 0.004	17.256 ± 0.003	14.83 ± 0.03	14.17 ± 0.03	H
		−17d40m1743
Sag	45478	19h29m5729	21.245 ± 0.050	20.966 ± 0.053	20.559 ± 0.041	16.55 ± 0.06	15.99 ± 0.08	L
		−17d40m1128

Note. Near-IR HST and Spitzer magnitudes for candidate M-type, dusty AGB stars. Magnitudes are corrected for extinction (see the text). This includes stars in the M-type region of Figure 4 and with >4σ excess in [3.6]–[4.5] color. The Spitzer magnitudes are from the DUSTiNGS survey (Paper I & II). The "Note" column marks M-type AGB stars identified with high-confidence (H; HST colors indicate water absorption) and low-confidence (L; star is near the border of C and M stars, is near the contamination cutoff in Figure 8, or is affected by crowding). An "x" denotes a DUSTiNGS x-AGB variable (those with >3σ variability between two Spitzer epochs, i.e., a likely AGB star). Non-x-AGB stars may also be variable, but were observed by DUSTiNGS with an unfavorable cadence.

^aThe IRAC magnitudes listed for 112431 are from DUSTiNGS epoch 2 because the star is not as red in epoch 1 (<4σ excess; Paper I).

Download table as:  ASCIITypeset image

AGB stars remain O-rich M-type stars both at the low and high ends of the AGB mass range due either to insufficient dredge-up or to HBB, though the exact mass limits depend on the metallicity. Most of the M-type stars we identified in Section 2.4.2 are low-mass AGB stars and are expected to produce only modest amounts of dust (e.g., McDonald et al. 2009, 2011; Boyer et al. 2015a). The significant IR excesses of the 26 M-type dusty stars identified here suggests that they are instead HBB AGB stars and are therefore more massive than their C-rich counterparts. Moreover, the dusty M-type stars are mostly confined to within each galaxy's half-light radius (McConnachie 2012). This suggests they are more massive than the dusty C stars, which are located throughout the spatial coverage of the HST data including in fields that are entirely outside the half-light radii (i.e., Fields A in NGC 147 and Sag DIG). Age gradients are expected in dwarf star-forming galaxies (e.g., Aparicio & Tikhonov 2000; Hidalgo et al. 2013, see also Paper III) and this is expected from models of the effects of feedback and stellar migration (e.g., Stinson et al. 2009; El-Badry et al. 2016). Overall, we identify about 1% of the AGB stars as massive M type, similar to the fraction seen in the SMC (Boyer et al. 2011). If confirmed, their high masses and low metallicities make them the closest known analogs to high-redshift dusty AGB stars.

The brightest among the massive AGB candidates may instead be RSG stars, which are expected to destroy all or most of their dust when they explode as supernovae (Lakićević et al. 2015; Temim et al. 2015). It can be difficult to distinguish M-type AGB stars from RSGs—luminosity is often used as a diagnostic, with the classical AGB limit near ${M}_{\mathrm{bol}}\,=-7.1$ mag. However, HBB AGB stars have been known to surpass this luminosity (e.g., García-Hernández et al. 2009). Water absorption is another possible diagnostic; Lancon et al. (2007) notes that water absorption in the near-IR is present in Galactic RSGs with cool effective temperatures (${T}_{\mathrm{eff}}\lt 3100$ K for $\mathrm{log}g=-1$), but is weak compared to the strength of water features in Mira variables. Messineo et al. (2014) use the Rayner et al. (2009) IRTF spectral library to compute a water absorption index and find that the index is large for all Mira-variable AGB stars, and small for semiregular variable AGB stars and RSGs. It follows that the H₂O absorption sequence in the HST CCD (Figures 4 and 5) can be used as a diagnostic to separate probable AGB and RSG stars. M-type stars on the water absorption sequence are probable AGB stars, while those at or below the "knee" in Figure 4 may be either less-evolved AGB stars or RSGs. Variability is an additional diagnostic; stars that were identified as variable in DUSTiNGS are probable AGB stars, given their large infrared amplitudes.

Based on their H₂O and variability signatures, IC 10 has several candidate dusty M-type AGB stars likely undergoing HBB: 10 identified with high confidence and 5 with low confidence (Table 7). There are two additional high confidence dusty M-type candidates in NGC 147 (#103322) and in Sextans A (#90034). Low-confidence dusty M-type stars either show weak H₂O absorption, fall near the contamination cutoff in Figure 8, or fall on the dividing line between C- and M-type stars in Figure 5, and thus a C-rich chemistry cannot be ruled out. Other low-confidence dusty M-type stars are affected by crowding, which may influence the near-IR colors by up to 0.5 mag (as measured by Dolphot; Section 2); these include #91324 and #94328 in Sextans A, #45478 in Sag DIG, and #96092 in IC 10.

The candidates in Sextans A and Sag DIG (especially #44334 and #90034) are of particular interest. The gas-phase ISM metallicities (and thus the metallicities of even the most massive stars) are low for both galaxies ($12+\mathrm{log}({\rm{O}}/{\rm{H}})\lesssim 7.5$), suggesting that stars with primordial abundances can still form a significant dust mass. Whether this occurs via single-star evolution or through a more exotic avenue remains unclear. On the other hand, the high near-IR luminosities of both stars suggests that they may in fact be RSGs, though #90034 in Sextans A does show evidence for water absorption and Star #44334 is a known long-period variable (P = 950 days; Whitelock et al. 2018). Both of these characteristics point to an AGB nature. Three additional confirmed O-rich stars in Sag DIG that were identified by Momany et al. (2014) are included in our near-IR HST coverage. These three stars have very red F606W–F814W colors, suggesting substantial circumstellar dust, but we do not detect dust around these stars. Their nature is unclear.

We have too small a sample to draw definitive conclusions, but there is nonetheless no hint of increased dust production at higher metallicities. The dusty M-type stars have remarkably similar properties in the Spitzer data, including colors, magnitudes, and pulsation amplitudes. Spectroscopy can definitively classify these sources, especially at wavelengths covering ice features common in YSOs (3–15 μm) or in the optical, where the Li I 6707 Å and 8126 Å and Rb I 7800 Å lines indicate HBB in AGB stars.

3.2.1. Super AGB Stars

Based on the results in the SMC (see Figure 6; Srinivasan et al. 2016), O-rich AGB stars redder than [3.6]–[4.5] ≈ 0.1 mag have dust-production rates that are an order of magnitude or more larger than C-rich stars with the same color. In our fields, the dusty C-rich stars outnumber the dusty O-rich sources by factors of about 2–10, but the total dust mass may be more evenly split between silicates and carbon grains if the M-type dust-production rates behave similarly to those in the SMC. The rarity of dust-producing M-type stars makes the balance between O-rich and C-rich dust highly stochastic. Observations with the James Webb Space Telescope will provide more reliable estimates of the dust-production rates for entire stellar populations in dwarf galaxies well beyond the Local Group (Boyer 2016).