Dusty Stellar Birth and Death in the Metal-poor Galaxy NGC 6822

Near-IR photometric data has been adopted from Sibbons et al. (2012), which used the Wide Field CAMera (WFCAM) on the 3.8 m United Kingdom Infrared Telescope (UKIRT) located on Maunakea to image NGC 6822 in J, H, and K. This work was completed during two runs, in 2005 April and 2006 November, as part of a larger survey project of Northern Hemisphere Local Group galaxies' AGB star content. The infrared detectors of WFCAM possess a scale of 04 per pixel over an observational area of 0.75 deg². A four-tile mosaic of 3 deg² centered on the optical coordinates of NGC 6822 (α = 19^h44^m56^s, δ = −14°48'06'') was obtained, with combined exposure times per pixel equaling 300 seconds in J and 540 s in H and K. In total, this catalog provided near-IR data for a total of 210,979 sources.

Data reduction followed standard steps and was completed using the WFCAM pipeline at the Institute of Astronomy in Cambridge. Astrometric and photometric calibrations were performed based on the 2MASS point-source catalog (Irwin et al. 2004; Hodgkin et al. 2009). Photometric measures are based on aperture photometry, with zero points calibrated against 2MASS but not transformed into the 2MASS system, such that published magnitudes and colors are of the WFCAM instrumental system (transformation equations can be found in Hodgkin et al. 2009). Because reddening values have been found to vary across NGC 6822, with E(B − V) = 0.24 mag in the outer regions and E(B − V) = 0.45–0.54 mag in the center (Massey et al. 1995; Hernández-Martínez et al. 2009), no corrections were made for internal reddening. The extinction map of Schlegel et al. (1998) was utilized for implementing foreground component corrections.

We have adopted Spitzer (Werner et al. 2004) mid-IR photometric data from Khan et al. (2015), who utilized archival photometric data (Khan et al. 2010, 2013) originally employed to target stars of certain subclasses with specific photometric properties. The more recent study was designed to inventory all point sources, in effort to study seven nearby galaxies with recent star formation using the Infrared Array Camera (IRAC; Fazio et al. 2004) at 3.6, 4.5, 5.8, and 8.0 μm and the Multiband Imaging Photometer (MIPS; Rieke et al. 2004) at 24 μm. This catalog contains in total 30,745 point sources from NGC 6822, created by dual-band selection of sources with >3σ detections within a matching radius (<0.5 pixels or 0.5–1.0 pixel; 90% of 4.5 μm sources match to a 3.6 μm source within 0.5 pixels) at both 3.6 and 4.5 μm, complemented with measurements of the 5.8, 8.0, and 24 μm bands through a combination of PSF and aperture photometry. It was developed as a resource archive for studying mid-IR-luminous sources and transients, as well as for planning observations with JWST, which will provide much deeper observational power than Spitzer. With a resolution nearly an order of magnitude better than that of Spitzer (Gardner et al. 2006), data from JWST will also benefit from reduced confusion.

This catalog is ideal for identifying dusty sources in NGC 6822. With similar Spitzer IRAC band coverage, complemented with the 24 μm MIPS band, the Khan et al. (2015) source list enables a similar wavelength range of study to SAGE. It provides the J2000.0 coordinates (R.A. and decl.) for each object, along with their Vega-calibrated apparent magnitudes (m_λ), the associated 1σ uncertainties (σ_λ), and for the 3.6–8.0 μm bands, the differences between the PSF and aperture photometry magnitudes (δ_λ). Khan et al. (2015) note that, for the 24 μm band photometry, the lower spatial resolution suggests limited utility, such that the relevant aperture can be contaminated by cold interstellar dust emission and objects other than the intended target.

In an effort to identify dusty sources in NGC 6822, we have combined the near- and mid-IR catalogs adopted from Sibbons et al. (2012) and Khan et al. (2015), respectively, into a joined master catalog. We use the CDS crossmatch utility⁷ to find the nearest Sibbons et al. (2012) neighbor to each of the 30,745 sources in the Khan et al. (2015) table within a 1'' radius. We obtain near-IR counterparts for 14,534 sources. This is a very small fraction of the full Sibbons et al. (2012) source list, due to the large difference in survey areas between the two studies. The near-IR counterparts include 1735 sources classified by Sibbons et al. (2012) as AGB stars (1099 M type, 636 C type). The 1'' radius allows us to recover all but nine (six M type, three C type) of the Sibbons et al. (2012) AGB candidates within the Spitzer footprint. Table 1 describes the columns in the joined catalog, and the full catalog is available in electronic form with this paper. JHK photometry adopted from Sibbons et al. (2012) is presented in the WFCAM instrumental system; equations to convert to the 2MASS system are given in Hodgkin et al. (2009).

Our first step in source classification is to compute the location of the TRGB. This is an important step in the identification of dusty and evolved star candidates. This prominent feature in the magnitude distributions of old- and intermediate-aged stellar populations corresponds to a conspicuous discontinuity between RGB and AGB populations (Salaris & Girardi 2005; Sibbons et al. 2012; Madore et al. 2018). From the Padova stellar evolution models (Marigo et al. 2008, 2013), the study of Bruzual et al. (2013) created simple stellar population models of the Magellanic Clouds, finding that >90% of TP-AGB stars are brighter than the TRGB (Boyer et al. 2015a). Establishing the location of the TRGB is therefore useful as a first-order boundary used to segregate nondusty and nonevolved stars from our overall joined sample.

We determined the location of the TRGB using a KDE technique. The overall distribution of the magnitudes of all sources with photometric errors less than 0.1 mag is modeled using a Gaussian kernel via the Gaussian_kde module available in the Python SciPy library. In order to find the position of the TRGB in magnitude space, we compute the smoothed first derivative of the KDE by applying a Savitsky–Golay filter. We then identify the location of the minimum in the derivative with the location of the TRGB. In order to estimate the uncertainty associated with this estimate, we draw a large number of samples of the entire distribution using the photometric uncertainties. The results of this Monte Carlo + KDE (MCKDE) method are shown for CMD1 and CMD2 in Figure 3 as blue curves; the discontinuity between RGB and AGB populations is discernible as the point of steepest slope. The value of the derivatives to the fits is represented as red curves for both CMD1 and CMD2 in Figure 3. Where the absolute value of this curve is maximized corresponds to where in magnitude space lies the location of the separation between RGB and AGB stars, represented by the vertical green line.

Zoom In Zoom Out Reset image size

For statistical robustness, the location of the TRGB was determined via Savitsky–Golay filter analysis by iterating the process 1000 times for each CMD in order to find the mean value. These superimposed results are illustrated as Figure 3 for both CMD1 and CMD2. (Because of the complexity associated with the different structure of CMD3, a TRGB was not similarly determined for it; more details in Section 3.2.) This Monte Carlo method of the KDE method (hereafter "MCKDE") provides the final value of the TRGB as well as a robust estimate of its uncertainty. For CMD1, the MCKDE method determined the TRGB at K = 17.36 ± 0.04 mag (see Figure 3, left), while for CMD2, the TRGB was found to be [3.6] = 17.16 ± 0.06 mag (see Figure 3, right).

The distribution of sources in each of the three CMDs presents structure representative of the underlying stellar astrophysics (Woods et al. 2011; Ruffle et al. 2015; Jones et al. 2017b). For a given individual source, the combination of magnitude and color reflects both the physical properties and current evolutionary phase of the star. Such parameters are generally not obviously and easily demarcated along lines of constant magnitude or color. The spur of points extending up and to the right from the main clustering of sources in both CMD1 and CMD2, for example, cannot be well isolated and categorized through strictly vertical and horizontal demarcating boundary lines (e.g., Nikolaev & Weinberg 2000; Weinberg & Nikolaev 2001; Cioni & Habing 2005; Blum et al. 2006; Cioni et al. 2006; Boyer et al. 2011, 2015c). These clustered points in color–magnitude space clearly represent a distinct stellar type.

The presence of opacity differences, as well as underlying absorption features from various molecular bands and dust, have a profound effect on the photometric properties of the various stellar types. Consequently, these populations become grouped together on the diagnostic CMDs, making it possible to distinguish them from one another. For example, C-rich stars appear red due in part to characteristic C₂ bands between 1.15 and 1.3 μm (Groenewegen et al. 2009), as well as HCN and C₂H₂ absorption features at 3.5 and 3.8 μm within the Spitzer IRAC 3.6 μm band (Marengo et al. 2007). In the IRAC 8.0 μm band, silicate features such as SiO and SO₂ are prominent in O-rich evolved stars, as they are quite sensitive to the amounts of dust being produced. For the near-IR, Wright et al. (2009) present a catalog of AGB star absorption features from molecular bands useful for modeling integrated emission for evolved stars. In addition, Blum et al. (2006) showed that, utilizing star formation history models of the LMC using the 2MASS CMD from Cioni et al. (2006) and new stellar models from Marigo et al. (2003), C- and O-rich AGB stars were well defined and separated in their CMDs (similar to the ones used in this study). A strategy to group such sources under one classification which more closely resembles these structures of points is therefore to define boundaries which follow the natural shape and direction seen in the diagrams.

In order to define such boundaries between stellar types in color space, we implemented KDE analyses of the distribution of sources in each of the three CMDs. With the location of the TRGB established for CMD1 and CMD2, small successive spans of magnitude incorporating at least 1000 points (for statistical robustness) were defined to progressively lower magnitude. The vertically stacked magnitude spans used in the analyses of CMD1 and CMD2 are sufficiently capable of separating the different stellar types (see Figure 4, left and center, respectively). Establishment of color-cut boundaries for CMD3, however, required supplemental restraint, due to its more complicated structure. Notably, the separation between some stellar types, particularly that which segregates background galaxies from dusty sources, run horizontally across the CMD. Because this direction runs congruent with the magnitude-only spans from which the boundary lines are derived, it is difficult for the KDE procedure to determine an optimal border location. In addition to vertically stacked magnitude spans, we therefore additionally implemented horizontally stacked spans in color space, again possessing a minimum 1000 sources. With a comparison of the demarcation lines defined by both the vertically stacked color spans and horizontally stacked magnitude spans, we define functional fits to the segregation boundaries of this longer wavelength CMD (see Figure 4, right).

Zoom In Zoom Out Reset image size

In each of these spans of magnitude (and/or color for CMD3), a KDE fit was established (see Figure 5, left) which defined the shape of the distribution of sources (see Figure 5, right). In the left plot of this figure, within the boundaries of the CMD established by the red lines, the blue points represent only the data within the given range. We note that some points in the magnitude ranges are not included in the KDE analysis (i.e., left as gray in the plots) as they introduced unnecessary noise to the fit that had no effect on the results. In the right plot of this figure, a histogram of the data (blue columns) is presented as comparison with the KDE fit (black line). Where the fit presents a local minimum is coincident with gaps in the histogram. Locations of the local minima in each span are representative of breaks in the numbers of various stellar populations and are therefore representative of the physically distinct separations which exist between different stellar types. When these local minima from the contiguous successive magnitude spans are joined, rough demarcation lines are formed (see Figure 6). Linear fits made to these points establish the functional forms of the color-cut boundaries which were used to classify the sources in NGC 6822.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

These categorizations, illustrated in Figure 7, include MS stars (Region 1; violet dots), RSGs (Region 2; gray dots), O-rich AGB star candidates (Region 3; blue dots), C-rich AGB star candidates (Region 4; red dots), dust-enshrouded AGB star and YSO candidates (Region 5; green dots), and background galaxies (Region 6; gold dots). Stellar type categorizations used here should be considered as approximations drawn from a statistical sample of sources. Furthermore, intrinsic variations of individual objects will cause some overlap. A small degree of contamination between regions is to be expected in the diagnostic CMDs. Final classification criteria implement region designations from multiple CMDs, which helps to reduce erroneous misidentifications (see Section 3.3 for details). In addition, we are not able to take into account the myriad variety of all possible stellar types. For example, short-lived blue core helium-burning sequence stars (BHeBs; also referred to as "blue loop") present a potential source of contamination at shorter wavelengths (Regions 1 and 2; Dalcanton et al. 2009). Inspection of Hubble Space Telescope (HST) data from Holtzman et al. (2006) indicates that BHeB contamination is no more than a few percent. Finally, we note that any foreground sources are most likely Galactic MS stars, and therefore are also expected to appear within shorter-wavelength regions. While some foreground stars will contaminate Region 2, we expect the majority to fall within Region 1. Henceforth, objects identified within this region in any CMD will be referred to collectively as "foreground/MS stars."

Zoom In Zoom Out Reset image size

For CMD1 and CMD2, the color-cut boundaries are confined within the TRGB boundaries established in Section 3.1, corresponding to K = 17.36 mag and [3.6] = 17.16 mag, respectively. In addition, a second boundary for both CMD1 and CMD2 exists at K and [3.6] values of 13.0 mag, whereby the number density of sources becomes too low for KDE bins to effectively constrain the source distributions. The functional coefficients for the boundary lines separating the different stellar types in CMD1 are presented in Table 2 and are illustrated in Figure 7 (left). CMD2, which incorporates a combination of near- and mid-IR photometry, presents a distribution of sources similar in appearance to that of CMD1, but exhibits some distinct differences. Boundary-line functional coefficients which segregate the different stellar types are presented in Table 3 and are illustrated in Figure 7 (center). Finally, utilizing only mid-IR data from Spitzer IRAC bands, CMD3 presents a structure visibly different from those seen with CMD1 and CMD2. Rather than establishing a boundary line at the TRGB, color cuts were defined through analysis of KDE local minima analyses in both vertically and horizontally stacked spans of magnitude and color space, respectively. This produced the complicated set of functional fits necessary to demarcate the various stellar types (see Figure 7, right). Functional coefficients of the stellar type boundary lines are presented in Table 4.

Table 2.  Stellar Type Category Boundary Function Coefficients for CMD1, Following the Form K = m × (J − K) + b

Regions Separated	Slope	y-intercept	Left Extent	Right Extent
by Boundary Line	m	b	(J − K)=	(J − K)=
1 and 2	−9.586	21.892	0.472	0.665
	−128.318	100.790	0.665	0.684
2 and 3	−15.880	31.275	0.876	1.151
3 and 4	−2.568	19.735	0.923	1.265
	3.094	12.572	1.265	1.267
	−4.380	20.994	1.127	1.264
	−1.420	17.253	1.264	2.094
4 and 5	−3.301	21.511	1.257	1.403
	−0.934	18.189	1.403	2.094
	$\infty$	⋯	2.094	2.094

Note. A slope of $\infty$ is representative of a vertical line.

Download table as:  ASCIITypeset image

Table 3.  Stellar Type Category Boundary Function Coefficients for CMD2, Following the Form [3.6] = m × (J − [3.6]) + b

Regions Separated	Slope	y-intercept	Left Extent	Right Extent
by Boundary Line	m	b	($J-$[3.6])=	(J − [3.6])=
1 and 2	−12.033	26.524	0.779	0.847
	2.388	14.315	0.847	0.668
	−5.969	19.898	0.668	0.766
	−25.236	34.656	0.766	0.858
2 and 3	−5.025	22.208	1.005	1.291
	19.022	−8.849	1.291	1.149
3 and 4	−2.287	20.335	1.390	1.581
	11.269	−1.095	1.581	1.460
	−1.158	17.046	1.460	2.637
4 and 5	−24.317	61.302	1.816	1.855
	−0.922	17.901	1.855	2.637
	$\infty$	⋯	2.637	2.637

Download table as:  ASCIITypeset image

Table 4.  Stellar Type Category Boundary Function Coefficients for CMD3, Following the Form [8.0] = m × ([3.6]–[8.0]) + b

Regions Separated	Slope	y-intercept	Left Extent	Right Extent
by Boundary Line	m	b	([3.6]–[8.0])=	([3.6]–[8.0])=
1 and 2/3	0	15.529	$\infty$	−0.475
	−2.029	14.565	−0.475	2.485
	$\infty$	⋯	9.522	9.522
2 and 3	0.749	14.332	0.084	1.273
2 and 6	−0.735	16.226	$\infty$	1.273
3 and 5/6	−2.251	18.151	1.273	2.485
	$\infty$	⋯	12.556	12.556
5 and 6	0.229	14.277	1.562	$\infty$

Download table as:  ASCIITypeset image

The stellar classification categories established from our analyses are presented in Table 5. These columns represent an addendum to the master joined catalog (Table 1) and are included as part of the downloadable data set. The full overall data set is available in electronic, machine-readable format. Included are an object ID number, the R.A. and decl. coordinates from Sibbons et al. (2012), and the coordinate disparity between the UKIRT and Spitzer photometry measured in arcseconds, where applicable. The Khan et al. (2015) data (Spitzer [3.6], [4.5], [5.8], [8.0], and [24]) present Vega-calibrated apparent magnitudes (m_λ) and the associated 1σ uncertainties (σ_λ) For the 5.8, 8.0, and 24 μm bands, a 1σ uncertainty of "−9999" corresponds to the photometric flux being a 3σ upper limit. Sibbons et al. (2012) data (UKIRT J, H, and K) were calibrated astrometrically and photometrically to the 2MASS point-source catalog (Irwin et al. 2004; Hodgkin et al. 2009). Photometric measurements are based on aperture photometry, using zero points calibrated against (but not transformed into) the 2MASS system.

Finally, for each source in the master joined catalog, we include an identifier value signifying the color-cut region within which the source is located for each of the three diagnostic CMDs (see Section 3.2). As with the CMD classification regions of Figure 7, "1" signifies that the source is located in the color-cut region of foreground/MS stars. "2" represents RSG candidates, while sources marked "3" belong to O-rich AGB star candidates. C-rich AGB star candidates are notated with "4," while "5" are dust-enshrouded sources which include dusty AGB stars and YSOs. Finally, sources marked "6" are background galaxy candidates, and any object denoted as "−9999" were not classified in the given CMD (did not inhabit an available color-cut boundary region). The final column, "TYPE," gives the source classification as defined by the procedure outlined in Section 3.3. "RELIABLE" classifications are denoted by one of the following descriptors: "RSG," "O-rich," "C-rich," or "YSO." Furthermore, "CANDIDATE" sources are indicated as "RSG?," "O-rich?," "C-rich?," or "Dusty?." In addition, we note "Star," "Galaxy," and "Unclassifiable" sources based on the categorizations provided by our CMD analyses, though we note that our selection criteria are not specified for these kinds of objects. Finally, sources for which a classification was not established are presented as "Classless!."

From the established color-cut boundaries, we have designated a type classification for every source in each of the three CMDs (see Figure 7). Guided by the KDE analyses described in the previous section, the demarcation lines indicate whether a given source is:a foreground/MS star (Region 1; violet points); an RSG candidate (Region 2 (CMD1 and CMD2 only); gray points); an O-rich AGB star candidate (Region 3 (CMD1 and CMD2 only); blue points); a C-rich AGB star candidate (Region 4; red points); a dust-enshrouded star, typical of extreme AGB star candidates and YSOs (Region 5; green points); or a background galaxy (Region 6 (CMD3 only); gold points). For CMD3 only as well, Region 2 (cyan points) corresponds to all M-type stars, which include both RSGs and O-rich AGB star candidates. If a source fails to be categorized within any of the regions described above, it is left as classless (represented as faint gray points in CMD1 and CMD2 of Figure 7).

The stellar type classification demographics derived from each of the three CMDs are presented in Table 6. Remarkably, the stellar type demographics from the analyses of each CMD produce numbers which are generally self-consistent. This is true despite the KDE analyses for each of the three CMDs being completed entirely independently of one another. The only obvious exception is the elevated number of dust-enshrouded sources identified in CMD3 as compared to CMD1 and CMD2. This disparity is attributable to the longer wavelengths of the photometric bands employed, resulting in an increased sensitivity to the dusty objects which primarily emit in this range.

Utilizing the stellar type categorizations as determined by each CMD individually, an overall classification for every source in the master catalog was determined. These compared the categorization results from each CMD to those from each of the other two CMDs in an effort to reach a consensus. High-confidence classifications by robust combinations of individual CMD results were classified as "RELIABLE," while lower-confidence classifications were marked as "CANDIDATE." From a total overall catalog of N = 30,745 sources, n = 3179 (10.3%) have been classified as a "RELIABLE" dusty or evolved star. Our selection criteria stipulate relative consistency in stellar type categorization across the three CMDs. Because some differences in categorization can result as a consequence of the different sensitivities afforded to the various wavelength regimes (e.g., CMD1 can better identify foreground stars, while CMD3 is better for classifying very dusty sources), such small inconsistencies will not disqualify a source from a "RELIABLE" label.

A source is considered a "RELIABLE" M-type star candidate when it is classified as an RSG or an O-rich AGB star in CMD1 and CMD2, and so long as it is not categorized as a dust-enshrouded source in CMD3. Of the M-type stars, RSGs are those which inhabit Region 2 in both CMD1 and CMD2, while O-rich AGB stars are those which inhabit Region 3 in both CMD1 and CMD2, or else is categorized as Region 3 in one of these two CMDs and as Region 2 in the other. Categorization within Region 3 in at least one of CMD1 or CMD2 favors a source to be an O-rich AGB star rather than an RSG. Previous study of the LMC employing CMDs equivalent to our CMD1 (Nikolaev & Weinberg 2000) and CMD2 (Blum et al. 2006) identified features blueward of the O-rich AGB star sequence, categorizing the relevant sources as RSGs. These classifications were then confirmed through spectroscopic cross-correlation. Boyer et al. (2011) extended this technique to the SMC to assess a distinction between RSGs and O-rich AGB stars, which was later revisited by Yang et al. (2018, 2019). With theoretical support from the modeling of Dell'Agli et al. (2015), we have adopted this prior proven technique for understanding the distinction between RSGs and O-rich AGB stars in CMDs applied to other galaxies.

In total, n = 2342 sources have been classified as "RELIABLE" M-type star candidates, with n = 1292 of these being designated as RSGs and n = 1050 as O-rich AGB stars. "RELIABLE" C-type star candidates are those which are categorized as C-rich in at least two of the three diagnostic CMDs, and are additionally never categorized as dust-enshrouded in any of the three CMDs. There are n = 560 sources classified as "RELIABLE" C-type star candidates. A source is considered a "RELIABLE" YSO candidate only when it is categorized as a dust-enshrouded object in each of the three CMDs. In addition, because YSOs are definitively restricted to regions of recent star formation, which are found in clusters, we have included a secondary requirement that a YSO is categorized as "RELIABLE" only if it is found within one of seven star formation regions where the number density of YSO candidate sources is highest (e.g., Jones et al. 2019). This yields a final total of "RELIABLE" YSO candidates of n = 277. A spatial distribution plot of all "RELIABLE" sources is presented as Figure 8, while classification criteria are presented in Table 7.

Other types of sources, including background galaxies and foreground/MS stars, were similarly classified based on their categorizations from the three diagnostic CMDs. Because our methods are not optimized for such sources, we refrain from detailed explanations of these objects' classifications. Rather, we note that such sources were identified outside of the RSG, AGB star, and YSO candidate color cuts. Finally, sources that are determined to be reliably unclassifiable are those which do not receive categorization into any of the possible stellar types in any of the diagnostic CMDs.

A further n = 260 (0.8%) objects have been classified as "CANDIDATE" sources, indicating that the confidence level in their combined stellar type categorizations is lower than for the "RELIABLE" targets. This is typically due to more pronounced inconsistencies in classification defined across the three CMDs than was allowable for the "RELIABLE" targets, while retaining photometric characteristics consistent with a particular stellar type. While very few such "CANDIDATE" sources populate our catalog as compared to "RELIABLE" sources, we felt their inclusion was warranted as a preliminary means of classification, with expectation for more sophisticated techniques applied later. The combination of diagnostic CMD categorizations for "CANDIDATE" sources follow similar structure to those of "RELIABLE" sources. These selection criteria are summarized in Table 8. In total, our catalog of "CANDIDATE" objects includes n = 65 RSGs, n = 59 O-rich AGB stars, n = 27 C-rich AGB stars, and n = 109 extremely dusty objects. We note that a source that would qualify as a "RELIABLE" YSO but lies outside of the star-forming regions is thus classified as a "CANDIDATE" C-type star. These dust-enshrouded sources, most predominantly selected from the longer wavelength color cuts of CMD3, include both YSOs and perhaps a rare class of the dustiest extreme AGB stars ("x-AGB" stars; see Blum et al. 2006; Boyer et al. 2011, 2015a). These x-AGB stars have been found to produce an extremely large proportion of dust for their small numbers in metal-poor, star-forming environments.

Finally, CMD categorization region combinations for background galaxies (n = 13,765, 44.8%) and foreground/MS stars (n = 1931, 6.3%) are summarized in Table 9, totaling 15,687 sources (51.1%). The remaining n = 1660 (5.4%) of sources are left as "CLASSLESS." These objects' photometric properties provided no obvious identity based on categorical consistency within the three CMDs and were therefore left unclassified. Comprising a relatively small minority of the overall sample, we expect future categorization with more sophisticated identification techniques to shed light on these sources. The demographic results of the "RELIABLE" and "CANDIDATE" catalogs are summarized in Table 10, along with all "OTHER" sources. Source classification demographics are illustrated as pie charts for both the entire joined catalog and only "RELIABLE" dusty and evolved stars in Figure 9.

Contamination of the various source catalogs developed through this study has been somewhat mitigated via our color-cut specifications; however, some number of inadvertent contaminants must be expected. Near- and mid-IR colors of red supergiant stars (RSGs) are similar to those on the AGB, and at the distance of NGC 6822, unresolved background galaxies, including star-forming galaxies and active galactic nuclei, have colors that significantly overlap those of dusty sources and YSOs (Jones et al. 2019). Due to the lower luminosity of the populations, magnitude-based cutoffs used effectively at selecting AGB stars and YSOs in the Magellanic Clouds are less efficient in systems at greater distances, such as NGC 6822.

Estimating conservative upper limits for source contamination, we assume a random and homogeneous distribution of background galaxies and foreground stars across our field of interest. We have adopted an area of 10.80 arcmin² on the outskirts of the galaxy, away from known star formation regions, to estimate the approximate level of contamination in our source cuts, similar to the method implemented by Jones et al. (2019). With 476 point sources in this off-target box identified from the master catalog, this region has a total point-source density of 44.07 sources per arcmin². We calculate the maximum contamination percentage for each of the RSG, O-rich AGB star, C-rich AGB star, and dust-enshrouded source candidates (including YSOs) for each of the three diagnostic CMDs (where appropriate), as well as our "RELIABLE" source lists. Results are summarized in Table 11. From this analysis, we estimate a maximum contamination percentage for our "RELIABLE" source lists to be 52.93% for RSGs, 29.52% for O-rich AGB stars, and 20.76% for C-rich AGB stars. Because our selection criteria for "RELIABLE" YSOs require inclusion within star-forming regions (see Figure 8), this off-target contamination study necessarily yields zero erroneous sources. CMDs of the master catalog sources contained within the off-target comparison region are presented as Figure 10. Overplotted are objects from the "RELIABLE" catalogs, including RSGs and both O- and C-rich AGB stars. Note that because no YSOs appear in this off-target comparison region, none appear in these CMDs.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

In addition to a statistical investigation of contamination sources, we queried the SIMBAD and VizieR Astronomical Databases using the CDS X-Match Service⁸ for associations of our RSG, AGB star, and YSO candidates with other known astronomical objects. Within a 2'' radius, 646 RSGs, 840 O-rich AGB stars, 530 C-rich AGB stars, and 148 YSOs were matched to sources with literature classifications. These mostly comprise AGB star candidates identified using optical or near-IR colors (Letarte et al. 2002; Kang et al. 2006; Sibbons et al. 2012; Whitelock et al. 2013). In addition, there are matched classifications indicative of youth, e.g., Hα-emitting objects (Massey et al. 2007), CO-bright clumps (Schruba et al. 2017), and compact H II regions (Hernández-Martínez & Peña 2009), and presence within a star-forming region (Melena et al. 2009). Finally, a handful sources have been matched to spectroscopically confirmed red giant stars (Kirby et al. 2017).

Of the RSGs, 3 have been matched to red giants, 4 have been matched to Hα emission-line stars, and 23 to star-forming regions. Twenty-three sources had conflicting classifications (all but three of which are classified as both star-forming regions and AGB stars). For O-rich AGB stars, 2 have been matched to red giants, 4 have been matched to Hα emission-line stars, 15 to star-forming regions, and 1 to a CO-bright clump. Conflicting classifications occurred for 21 sources. C-rich AGB stars match to three red giants, five Hα emission-line stars, nine star-forming regions, and two CO-bright clumps. Seventeen sources presented conflicting classifications. Finally, 1 YSO was matched to a red giant (likely the result of a chance superposition of one optically bright and one IR-bright source in the crowded field), 7 matched to H II regions, 2 to planetary nebulae (PNe), 20 to Hα emission-line stars, 10 to star-forming regions, 10 to Hα emission regions from UV data, and 80 to CO-bright clumps. We note that mid-IR Spitzer colors for YSOs and PNe are indistinguishable and require spectral disentanglement and are thus potentially significant sources of contamination (Jones et al. 2017b). There were 31 instances of conflicting classifications.

Finally, we employed the TRILEGAL service⁹ to better understand foreground sources of contamination. This tridimensional model of the Galaxy (Girardi et al. 2012) is used to populate the field of view to NGC 6822 with artificial foreground stars, which are then run through the same color-cut classification criteria as are applied to our master catalog sources. Because completeness of the data is not well defined at blue wavelengths, this technique was implemented to grasp the likelihood of foreground contamination by O- and C-rich AGB stars and YSOs. CMDs of all sources obtained from the TRILEGAL Galaxy model are presented as Figure 11. This data was subsequently processed through the same color-cut classification criteria as applied to the "RELIABLE" dusty and evolved star catalogs, producing artificial contaminants which have been overplotted. In total, we found 41 simulated O-rich AGB star contaminants, 0 such C-rich AGB stars, and 9 YSOs.

Zoom In Zoom Out Reset image size

With a census of "RELIABLE" O- and C-rich AGB star candidates determined through our color-cut classifications, we can investigate the stellar metallicity characteristics of this galaxy. The ratio of the number of C-rich (C-type) to O-rich (M-type) AGB stars can be used as a gauge of the star formation environment. For a lower initial ambient metallicity, fewer dredge-up events are necessary to form a C-rich atmosphere, and thus a higher measured value of the overall C/M ratio is expected compared to a solar-metallicity environment (Groenewegen 2006 and references therein; Sibbons et al. 2012; Boyer et al. 2017; Boyer 2019). In addition, a lower metallicity environment will push the AGB evolutionary track toward higher temperatures. This has the further result of reducing the numbers of M-type stars while simultaneously increasing the number of K spectral type RGB stars (Iben & Renzini 1983; Marigo et al. 1999; Sibbons et al. 2012). By the relation of Battinelli & Demers (2005) and refined by Cioni (2009),

$$\begin{eqnarray}&&[\mathrm{Fe}/{\rm{H}}]=-1.39\pm 0.06-0.47\pm 0.10\times \mathrm{log}({\rm{C}}/{\rm{M}}).\end{eqnarray} \tag{ 1 }$$

The C/M ratio for NGC 6822, computed over the entire field of view for our study, is 0.533 ± 0.028. We therefore find a metallicity for NGC 6822 of [Fe/H] = −1.262 ± 0.098. Corrected for maximum source contamination (see Table 11), the C/M ratio becomes 0.599 ± 0.036, which gives a metallicity [Fe/H] = −1.286 ± 0.095. The study of Sibbons et al. (2012) report a C/M ratio of 0.23 ± 0.01 over their full-observed area, which produces a metallicity estimate of [Fe/H] = −1.14 ± 0.08 utilizing the same Cioni (2009) relation. Within 4 kpc of the center, however, they report a C/M ratio of 0.48 ± 0.02, equivalent to [Fe/H] = −1.24 ± 0.07. For a distance to NGC 6822 of 490 kpc, a radius 4 kpc from the galaxy center is equivalent to 2806, which is larger than the field of view of our study as dictated by the extent of the Spitzer data. This second metallicity from Sibbons et al. (2012) is therefore a much more appropriate comparison to our own reported value, which agrees within the respective errors.

In addition, we have investigated the AGB stars in NGC 6822 for possible structure in the metallicity distribution of this galaxy. First, by segregating the AGB stars along the central N−S bar, we investigate the variability of metallicity across the galactic axis of rotation. We adopt the coordinates for the center of NGC 6822 from Sibbons et al. (2012) as R.A.=19^h44^m56^s, decl. = −14^d48^m06^s, and the angle of the axis rotation of 330° adopted from McConnachie (2012). On the left side of NGC 6822, we find a C/M ratio of 0.507 ± 0.038, which corresponds by Equation (1) to [Fe/H] = −1.251 ± 0.105. On the right side of the galaxy, we find a C/M ratio of 0.561 ± 0.041, giving a metallicity of [Fe/H] = −1.272 ± 0.100. When corrected for maximum source contamination, the left-side C/M ratio becomes 0.570 ± 0.049, corresponding to a metallicity of [Fe/H] = −1.275 ± 0.102. The contamination-corrected values for the right side are a C/M ratio of 0.631 ± 0.053, which gives a metallicity [Fe/H] = −1.296 ± 0.097. The measured stellar abundances of these two halves of NGC 6822 fall neatly on either side of the overall metallicity of [Fe/H] = −1.262 ± 0.098 ([Fe/H] = −1.286 ± 0.095 when corrected for contamination) discussed earlier; however, the magnitude of the difference is minuscule in comparison with the respective errors. We therefore conclude that the rotation of NGC 6822 has no impact on the measured stellar metallicity of the system.

Second, we have isolated the C- and M-type AGB star populations in six concentric rings of width equal to the quarter-light radius (1325; ∼0.189 kpc), initiating from the center of the galaxy. Within these rings, we compute the C/M ratio and the associated value of [Fe/H] out to a distance of 795 (∼1.133 kpc), equivalent to the one-and-one-half-light radius. We find a minimum value of [Fe/H] = −1.334 ± 0.111 in the most central annulus, with metallicities increasing with expanding radius to a maximum of [Fe/H] = −1.225 ± 0.129 in the outermost ring ([Fe/H] = −1.358 ± 0.112 to −1.249 ± 0.129 when contamination corrected). These results are illustrated in Figure 13 and summarized in Table 12. The rate of increase we find for NGC 6822 is quite minimal, however, and all values of metallicity agree within their respective errors, therefore giving only a very slight suggestion of a radial metallicity gradient. We therefore do not suggest the reality of a radial metallicity gradient in NGC 6822.

Zoom In Zoom Out Reset image size

Overall, these values of the stellar metallicity depend on the implicit assumption that all C- and O-rich AGB star candidates from our "RELIABLE" catalog are actually AGB stars. Computations utilizing maximum contamination percentage corrected values are the only avenue in this analysis. While our classification methodology requires candidacy from multiple CMDs to be marked as "RELIABLE" for the sake of redundancy, the presence of contaminant RGBs in our M-type star catalog cannot be ignored. Our C/M ratios and [Fe/H] values must therefore be considered as lower limits. The true nature of these sources requires more sophisticated analysis than color-cut segregations can provide. Future studies employing spectral energy distribution (SED) fitting promise to refine this value even further with higher-confidence identifications of the relevant stellar populations.

The bright, well-known star-forming complexes of NGC 6822 (e.g., Hubble X, Hubble V, and Hubble I and III situated along the north side, and Hubble IV situated on the south side of the galaxy), distinguished by optically conspicuous Hα emission (see Figure 1), have been studied extensively across several wavelength regimes. Hubble (1925) first presented optical plates of NGC 6822, identifying conspicuous star-forming regions. Hα fluxes of the two brightest H II regions, Hubble V and Hubble X, were obtained by Kennicutt (1979), which were later observed with HST by O'Dell et al. (1999). Hodge et al. (1988) compiled an atlas of H II regions via optical survey, while Hodge et al. (1989) provided an Hα luminosity function and H II region size distribution. Gallagher et al. (1991) and Israel et al. (1996) studied the star-forming regions in the far-IR. CO emission was studied to trace molecular gas by Israel et al. (2003) and later, on a larger scale, by Schruba et al. (2017). Cannon et al. (2006) presented Spitzer imaging of emission regions in NGC 6822 to study the nature of IR, Hα, H I, and radio continuum emission. Gouliermis et al. (2010) employed optical photometry and stellar density maps to show the hierarchical stellar structure in the galaxy, linking some stellar concentrations with IR-bright complexes identified with Spitzer imaging from Cannon et al. (2006). Finally, several studies have used the massive H II regions of NGC 6822 as specimens of extragalactic star formation, employing Spitzer spectroscopy to study photodissociation, fine-structure lines, and abundance ratios (e.g., Lee et al. 2005; Hunter & Kaufman 2007; Rubin et al. 2016).

The spatial distribution of "RELIABLE" YSO candidates is illustrated in the lower right of Figure 12. Sources are tightly clustered in compact regions consistent with the well-known star-forming regions of NGC 6822. In addition to these, we have identified comparable numbers of dusty sources grouped in other tight clusters. Located coincident with the central N−S bar, these collections of red-excess objects appear to be massive, embedded regions of star formation that have been otherwise hidden from previous optical surveys. Only via concentrated effort with near- and mid-IR photometry are these obscured clusters of YSO candidates revealed. The most central star-forming region, dubbed Spitzer I by Jones et al. (2019), is located at R.A. = 19^h44^m5897, decl. = −14^d47^m3739 and is home to 80 of the "RELIABLE" YSOs identified in this study. Cannon et al. (2006) categorized this region as the infrared source "NGC 6822 11," cross-listed as Hubble VI and VII, while Gratier et al. (2010) studied the molecular cloud complexes of this region in CO. Immediately adjacent to Spitzer I is the smaller Spitzer II, located at R.A. = 19^h45^m0980, decl. = −14^d48^m3477, and within which we have found 18 YSOs. Finally, next to Hubble IV in the south of NGC 6822 is Spitzer III, located at R.A. = 19^h44^m3908, decl. = −14^d51^m4412, and which contains 14 YSOs.

The high concentration of YSO candidates within these star-forming regions, along with the strong likelihood of source confusion due to crowding and resolution effects, suggests a larger number of individual YSOs than our selection criteria are capable of singling out. The observational signature of Spitzer I may in fact indicate the presence of a proto-super-star cluster (SSC; M_* ≳ 10⁵ M_⊙). SSCs are representative of a star formation mode where stellar surface densities exceed that of H II regions and OB associations by orders of magnitude (Nayak et al. 2019) and may be early-universe analogs to modern-day globular clusters. The most well-known SSC is the cluster R136, which resides within the massive LMC star-forming region 30 Doradus. A parallel star-forming complex known as N79, identified to possess Hα-emitting sources first by Henize (1956), occupies space symmetrically across the bar of the LMC, discernible in the distribution of H I gas (see Figure 1 of Ochsendorf et al. 2017). Within the southernmost of three giant molecular clouds in N79 is the object H72.97−69.39, originally thought to be the most luminous YSO in the entire LMC (Seale et al. 2014), but recently identified as a proto-SSC (Ochsendorf et al. 2017; Nayak et al. 2019).

In contrast to the optically luminous and conspicuous 30 Doradus, the N79 region is weak in optical star formation tracers such as Hα emission. Ochsendorf et al. (2017) have found N79 to be extremely active, however, hosting a densely packed population of YSOs. Similarly, while roughly equivalent in physical size to Hubble V, the brightest H II region in NGC 6822 (radius ≈120 pc), our identification of 80 YSOs associated with Spitzer I is ∼50% more than were found in Hubble V (n = 55), demonstrating a high density of young stars. Spitzer I is rich in molecular gas (inferred from 8 μm emission; Sandstrom et al. 2012) and is probably the youngest star-forming region in NGC 6822 (Jones et al. 2019). Furthermore, analyses of star formation tracers reveal a higher level of IR flux as compared to UV or Hα, providing a strong indication that the SFR is expected to increase. Similarly to the LMC object H72.97−69.39 in N79, Spitzer I has yet to reach its peak level of star formation activity and may potentially become an SSC. While the question of exactly how SSCs form remains a mystery, future detailed studies of H72.97−69.39 and Spitzer I may shed light on this extreme star formation phenomenon.

The recent work of Jones et al. (2019) employed the joined master catalog developed in this study to specifically characterize the YSO population of NGC 6822. Utilizing an independent set of color-cut criteria, Jones et al. (2019) have identified 90 YSOs within the region of space affiliated with Spitzer I. Their classification system further employed SED modeling, finding the Spitzer I YSOs exhibiting properties characteristic of the earliest stages of formation. It is rich in molecular gas and has substantial emission in 8 μm as well as some emission in 24 μm, but is faint in Hα and UV. Combined with a large H I mass, the large population of high-confidence YSO detections strongly indicate that this star-forming region is young and very embedded, exhibiting the highest SFR known in NGC 6822 (Jones et al. 2019). Future observations with JWST promise to characterize this potential proto-SSC with exceptional detail.