CARBON STARS IN THE SATELLITES AND HALO OF M31

Our spectroscopic and photometric data were obtained over ∼10 yr as part of the SPLASH survey. In this paper, we focus on the subset of SPLASH data that excludes the bright disk of M31. This includes fields targeting the dwarf spheroidals (dSphs), dwarf ellipticals (dEs), the smooth virialized halo, halo substructure, and M32. The extent of these observations is shown in Figure 1. These data include 14,143 stellar spectra taken with the DEIMOS multi-object spectrograph on the Keck II 10 m telescope. These spectra are spread across 151 individual DEIMOS masks, targeting ∼60 separate fields (red rectangles in Figure 1).

Zoom In Zoom Out Reset image size

The SPLASH spectroscopic selection functions vary significantly from field to field, as the observations were conducted with specific science goals in mind that varied from field to field rather than following the strict guidelines of an overarching homogenous survey. As a result, we will not go into details regarding those selection functions in this section. Instead, we will discuss the selection functions to the extent that they affect the identification of carbon stars in Section 3.2.

2.1.1. dSphs

SPLASH has observed 16 of the dSphs in the M31 system: And I, And II, And III, And V, And VII, And IX, And X, And XI, And XII, And XIII, And XIV, And XV, And XVI, And XVIII, And XXI, and And XXII. The properties of these dSphs relevant to this analysis are shown in Table 1.

Table 1.  Properties of M31 Satellites Observed by SPLASH

Satellite	R.A.a	Decl.a	[Fe/H]a	MVa	${(m-M)}_{0}$ a	vsys (km s−1)b	σ (km s−1)b	[α/Fe]c	ITRGBd
And I	00:45:39.8	+38:02:28	−1.45 ± 0.04	−11.7 ± 0.1	24.36 ± 0.07	−376 ±  2.2	10.2 ± 1.9	0.278 ± 0.164	20.36
And II	01:16:29.8	+33:25:09	−1.64 ± 0.04	−12.4 ± 0.2	24.07 ± 0.06	−192.4 ± 0.5e	7.8 ± 1.1e	0.033 ± 0.09	20.10
And III	00:35:33.8	+36:29:52	−1.78 ± 0.04	−10.0 ± 0.3	24.37 ± 0.07	−344.3 ± 1.7	0.3 ± 1.4	0.205 ± 0.16	20.40
And V	01:10:17.1	+47:37:41	−1.6 ± 0.3	−9.1 ± 0.2	24.44 ± 0.08	−397.3 ± 1.5	10.5 ± 1.1	0.119 ± 0.09	20.54
And VII	23:26:31.7	+50:40:33	−1.40 ± 0.30	−12.6 ± 0.3	24.41 ± 0.10	−307.2 ± 1.3	13.0 ± 1.0	0.296 ± 0.09	20.59
And IX	00:52:53.0	+43:11:45	−2.2 ± 0.2	−8.1 ± 1.1	24.42 ± 0.07	−209.4 ± 2.5	10.9 ± 2.0	...	20.59
And X	01:06:33.7	+44:48:16	−1.93 ± 0.11	−7.6 ± 1.0	24.23 ± 0.21	−164.1 ± 1.7	6.3 ± 1.4	0.505 ± 0.24	20.33
And XI	00:46:20.0	+33:48:05	−2.00 ± 0.20	−6.9 ± 1.3	${24.40}_{-0.50}^{+0.20}$	−427.5 ± 3.5f	${7.6}_{-2.8}^{+4.0}$ f	...	20.51
And XII	00:47:27.0	+34:22:29	−2.10 ± 0.20	−6.4 ± 1.2	24.70 ± 0.30	−557.1 ± 1.7f	0.0 + 4.0f	...	20.88
And XIII	00:51:51.0	+33:00:16	−1.90 ± 0.20	−6.7 ± 1.3	${24.80}_{-0.40}^{+0.10}$	−185.4 ± 2.4	5.8 ± 2.0	...	20.89
And XIV	00:51:35.0	+29:41:49	−2.26 ± 0.05	−8.4 ± 0.6	24.33 ± 0.33	−480.6 ± 1.2	5.3 ± 1.0	...	20.51
And XV	01:14:18.7	+38:07:03	−1.80 ± 0.20	−9.4 ± 0.4	24.00 ± 0.20	−323.0 ± 1.4	4.0 ± 1.4	...	20.02
And XVI	00:59:29.8	+32:22:36	−2.10 ± 0.20	−9.2 ± 0.4	23.60 ± 0.20	−367.3 ± 2.8	3.8 ± 2.9	...	19.72
And XVIII	00:02:14.5	+45:05:20	−1.8 ± 0.1	−9.7	25.66 ± 0.13	−332.1 ± 2.7	9.7 ± 2.3	...	21.76
And XXI	23:54:47.7	+42:28:15	−1.80 ± 0.20	−9.9 ± 0.6	24.67 ± 0.13	−361.4 ± 5.8	7.2 ± 5.5	...	20.75
And XXII	01:27:40.0	+28:05:25	−1.8	−6.5 ± 9.9	24.50	−126.8 ± 3.1	${3.54}_{-2.49}^{+4.16}$	...	20.56
NGC 147	00:33:21.1	+48:30:32	−1.1 ± 0.1	−14.6 ± 0.1	24.15 ± 0.09	−193.1 ± 0.8g	16 ± 1g	0.356 ± 0.10	20.35
NGC 185	00:38:58.0	+48:20:15	−1.3 ± 0.1	−14.8 ± 0.1	23.95 ± 0.09	−208.8 ± 1.1g	24 ± 1g	0.120 ± 0.09	20.12
NGC 205	00:40:22.1	+41:41:07	−0.8 ± 0.2	−16.5 ± 0.1	24.58 ± 0.07	−246 ± 1h	35 ± 5h	...	20.73
M32	00:42:41.8	+40:41:55	−0.25	−16.4 ± 0.2	24.53 ± 0.21	$-{196.9}_{-4.9}^{+5}$ i	${29.9}_{-4.6}^{+5.2}$ i	0.425 ± 0.32	21.15

Notes.

^aMcConnachie (2012). ^bTollerud et al. (2012), unless otherwise noted. ^cVargas et al. (2014). ^dCalculated in this work. ^eHo et al. (2012). ^fCollins et al. (2013). ^gGeha et al. (2010). ^hGeha et al. (2006). ⁱHowley et al. (2013).

Download table as:  ASCIITypeset image

The majority of these spectra were targeted using Washington photometry (M, T₂, and DDO51) taken with the Mosaic Camera on the Kitt Peak National Observatory (KPNO) 4 m Mayall telescope. The DDO51 filter is centered on Mg absorption features (Mgb) that are highly dependent on surface gravity, thus allowing for the discrimination of M31 giant stars from MW foreground dwarf stars (Majewski et al. 2000; Gilbert et al. 2006). We selected spectroscopic targets using the M − DDO51 versus M − T₂ color–color diagram. The spectra in And X were targeted using Sloan Digital Sky Survey (SDSS) imaging (Adelman-McCarthy et al. 2006). SDSS imaging was also used to supplement the Washington photometry imaging for And II. The spectra in And XV and And XVI were targeted using archival Canada–France–Hawaii Telescope (CFHT) imaging. Finally, the spectra in And XVIII and And XXII were targeted using B- and V-band imaging from the Large Binocular Telescope.

All spectra were observed with the 1200 line mm⁻¹ grating with a central wavelength of 7800 Å. This configuration has a dispersion of 0.33 Å pixel⁻¹ and a wavelength range from Hα to the Ca ii triplet at 8500 Å. The typical integration time was 3600 s per mask.

For further descriptions of the original observations, we refer the reader to Majewski et al. (2007, And XIV), Kalirai et al. (2009, And X), Kalirai et al. (2010, And I and III), Ho et al. (2012, And II), and Tollerud et al. (2012, the remainder). There are 3778 stellar spectra from dSph fields. All have associated Washington M, T₂, and DDO51 photometry (either used to target or taken after spectroscopic observations).

2.1.2. dEs

2.1.3. M32

2.1.4. M31 Halo

The fields outlined above have been observed in a variety of filters: M, T₂, V, I, R, CN, and TiO. To homogenize our data set, we perform a series of photometric transformations to ensure that, whenever possible, stars have the equivalent of V-, I-, and R-band photometry. The Washington photometry filters M and T₂ are nearly identical to Johnson–Cousins V and I. To convert from one to the other, we apply the transformation established by Majewski et al. (2000). To transform V to R, we use the transformation between R − I and V − I established for red and "very red" stars by Battinelli & Demers (2005). This relationship is color dependent, with the break point at $V-I=1.7$. After transforming V − I to R − I, we add the I-band magnitude to extract R on its own. To transform R to V, we apply the same equations in the opposite direction. We do not further transform V and I into M and T₂, simply because the Washington photometry filters are less commonly used in the literature and we do not require them for comparison. Typical uncertainties of the transformed photometry are 0.09 mag in V and 0.02 mag in R. In contrast, the typical uncertainties of the raw photometry are 0.04 mag in V and 0.05 mag in R.

We also calculate synthetic CN and TiO magnitudes. CN and TiO are narrowband filters centered at 8120.5 and 7778.4 Å, respectively. The CN filter is centered on the CN band in carbon stars and a continuum region of M stars, while the TiO filter is centered on the TiO band in M stars and a continuum region of C stars. CN–TiO color can thus distinguish between C- and O-rich TP-AGB stars. Indeed, many of the photometric surveys of AGB stars in the Local Group have used a broadband color and CN–TiO color to identify carbon stars (the four-band photometry system, FBPS; Wing 1971). Broadband photometry has too low spectral resolution to contain the detailed spectral information provided by narrowband filters, so we instead compute synthetic CN and TiO magnitudes by weighting the spectra with CFHT/CH12k CN and TiO throughput curves (see a full discussion of this method in Hamren et al. 2015).

We would like to confirm whether the carbon stars identified in this paper are members of an M31 satellite, the M31 halo/extended disk, or the halo of the MW. To do this, we take advantage of the membership criteria established by the SPLASH team.

Membership in the halo and substructure fields is determined via the likelihood estimates from Gilbert et al. (2006), which uses radial velocity, the equivalent width of the Na i line at 8190.5 Å (${\mathrm{EW}}_{\mathrm{Na}{\rm{I}}}$), position in M-DDO51 versus $M-{T}_{2}$ color–color space, position in I versus V − I color–magnitude space, and estimated [Fe/H] to determine the probably that a given star belongs to the M31 RGB or the MW foreground. Likelihood classes go from −3 (secure MW classification) through −1 (marginal MW classification) and 1 (marginal M31 classification) to 3 (secure M31 classification). In this work we will take those stars with likelihood greater than or equal to zero (i.e., those stars more likely to be associated with the M31 than the MW) to be M31 members. Likelihoods have also been calculated for many stars in the dSphs.

Membership in the M31 dSphs has been calculated by Ho et al. (2012, And II) and Tollerud et al. (2012, the remainder). In And II, the authors use a kinematical restriction, requiring member stars to be within 3σ of the systemic velocity of the dSph (−228 km s⁻¹ < v < −157 km s⁻¹). They then apply a photometric and spectroscopic cut, V − I < 2.5 and ${\mathrm{EW}}_{\mathrm{Na}{\rm{I}}}\lt 4$, to further eliminate foreground MW contamination. In the remaining dSphs, the authors calculate membership probability using the distance of the star from the center of the dSph, the distance from the fiducial isochrones in T₂ versus $M-{T}_{2}$ space, the equivalent width of Na i, and the half-light radius of the dSph.

In NGC 147 and NGC 185, membership criteria have been established by Geha et al. (2010), using a modification of the method set forth in Gilbert et al. (2006). The authors' metric uses line-of-sight velocity, ${\mathrm{EW}}_{\mathrm{Na}{\rm{I}}}$, position with respect to isochrones in I versus V − I space, and Ca ii triplet-based spectroscopic metallicity.

Membership to NGC 205 is more difficult to determine: its proximity to M31 leads to contamination from the M31 disk, as well as the M31 and MW halos. As a rule of thumb, stars with velocities more negative than the systemic velocity of the dE (v_sys = −246 ± 35 km s⁻¹) are likely M31 halo stars. Stars with velocities much less negative than the systemic velocity of the dE are likely to be foreground contamination (Geha et al. 2006). While it is not used in Geha et al. (2006), EW${}_{\mathrm{Na}{\rm{I}}}$ can also be used to distinguish member stars from foreground dwarfs (e.g., Gilbert et al. 2006). Here we will apply the condition that EW${}_{\mathrm{Na}{\rm{I}}}\lt 3$.

It is even more difficult to conclusively determine membership in M32, which is superimposed on the M31 disk. Howley et al. (2013) define M32 candidate members to be stars with −275 ≤ v ≤ −125 km s⁻¹. Stars with velocities more negative than this range likely belong to either M31's disk or inner spheroid. Stars with velocities less negative than this range are likely MW foreground.

The optical spectra of carbon stars are distinguished by prominent CN features at ∼7000–8200 Å and C₂ bands at ∼6100–6600 Å. Previous authors have used the CN band to identify carbon stars, either by using narrowband filters centered on CN and its oxygen-rich counterpart TiO (e.g., Nowotny et al. 2003; Battinelli & Demers 2004a, 2004b; Wing 2007) or by cross-correlating spectra against templates with and without CN (Hamren et al. 2015).

In this work, we identify carbon stars using the spectroscopic classification statistic demonstrated by Hamren et al. (2015) in the disk of M31. This method identifies carbon stars by cross-correlating the spectrum in question with a suite of MW templates. Spectra that are best fit by a carbon star template are flagged as likely carbon stars and visually examined for final confirmation. We refer the reader to Hamren et al. (2015) for further details about the template observations and testing of the classification metric.

To make use of the high signal-to-noise ratio (S/N) carbon templates, which were observed with DEIMOS's 600-line grating, we first rebin our spectra to match the template spectra's 0.65 Å pixel⁻¹ dispersion. We then apply the classification statistic to the full SPLASH satellite/halo sample of 14,143 stars and identify 41 carbon stars. The full list of carbon stars, including their SPLASH ID number, position, magnitude, and velocity, is presented in Table 2. For the carbon stars found in the satellites, their position within the galaxy (whose center is denoted by a white star) is shown in Figure 2. These maps are omitted for stars in the halo fields, as there is no discernible structure or center to use as a useful reference point.

Zoom In Zoom Out Reset image size

Using the membership criteria outlined in Section 2.3, we find that all carbon stars in the dSph fields are unambiguous members of their respective satellite. The same is true for the carbon stars in NGC 185. Only 2 of the 12 carbon stars identified in NGC 147 fields satisfy the full set of membership criteria outlined in Section 2.3. However, if we take into account the rotational velocity of the dE and radial-dependent velocity dispersion (Geha et al. 2006), then all the stars are within 3σ of the systemic velocity. We will thus mark all carbon stars in NGC 147 as members.

Four of the five carbon stars identified in NGC 205 have kinematics consistent with the dE. Star 83287 has a velocity of −423.8 km s⁻¹, making it more likely to be a member of the M31 halo. In M32, only star 240022 has a velocity consistent with the cE. Stars 160934 and 183365 have velocities that may indicate either foreground carbon dwarfs or stars in the tail of the M31 spheroid's kinematic distribution. Given that these masks are located on a relatively high surface brightness part of M31, picking up a foreground star (let alone a rare dC star) is unlikely. These two stars are thus more likely to belong to the M31 spheroid. Stars 263107 and 233791 have velocities consistent with either the M31 spheroid or disk.

We also identify four carbon stars in fields with known substructure, associated with M31's northeast or western shelves. These shelves are debris from the giant southern stream and are known to contain AGB stars (e.g., Tanaka et al. 2010).

The presence, or lack thereof, of carbon stars in a particular field is highly dependent on the spectroscopic selection function. The identification of carbon stars was never one of the science goals of the SPLASH survey, and so the regions in color–magnitude space in which they are typically found were not always prioritized.

To estimate the number of carbon stars we might expect to observe in each field, we begin with the assumption that our carbon stars are all TP-AGB stars brighter than the tip of the red giant branch (TRGB). To estimate the total number of TP-AGB stars in each field (N_AGB), we count the number of spectral targets brighter than the TRGB that have V − I > 1.5 (equivalently, $R-I\gt 0.73$). These limits are based on the region in color–magnitude space where carbon-rich TP-AGB stars are typically found, with a slightly bluer color limit designed to encompass the majority of our sample.

In the satellites, we can calculate the I-band magnitude of the TRGB using the calibration from Bellazzini et al. (2004) adjusted by the distance moduli listed in Table 1. This calibration requires [M/H] measurements, so we convert the [Fe/H] measurements in Table 1 using Equation (1) from Ferraro et al. (1999). We use the [α/Fe] measurements from Vargas et al. (2014) where available and assume [α/Fe] = 0.28 (the same value used by Ferraro et al. 1999) for the remainder. Finally, we correct the calculated I_TRGB for foreground extinction using the dust maps from Schlafly & Finkbeiner (2011). Our calculated I_TRGB are listed in Table 1. Because the i' photometry of M32 was specifically calibrated such that i'_TRGB and I_TRGB are equivalent, our calculated I_TRGB for M32 is still comparable to the i' photometry. For the halo fields, we assume that I_TRGB is located at M_bol = −4 and adopt the distance modulus of M31 (24.47; Stanek & Garnavich 1998), giving us I_TRGB = 20.47.

In the fields with potential TP-AGB stars, we can estimate how many carbon stars we would expect to observe using a theoretical C/M ratio—the number ratio of carbon- to oxygen-rich TP-AGB stars. We calculate the C/M ratio in each field using the well-calibrated relationship between log(C/M) and [Fe/H] from Cioni (2009). In satellite fields, we adopt the [Fe/H] values listed in Table 1. In halo fields, we compute the projected line-of-sight distance from M31 and then derive the [Fe/H] value using the metallicity gradient from Gilbert et al. (2014).

The probability of detecting a C star in a TP-AGB population (assuming that the region in color and magnitude space described above is populated entirely by C- and M-type AGB stars) is

$$\begin{eqnarray}&&{P}_{c}=\displaystyle \frac{1}{{({\rm{C}}/{\rm{M}})}^{-1}+1}.\end{eqnarray} \tag{ 1 }$$

Multiplying this probability by the number of likely TP-AGB stars gives us the number of carbon stars expected in each field.

Table 3 outlines this analysis. For each field, it lists the theoretical C/M ratio, the number of TP-AGB stars (N_AGB), the number of predicted carbon stars (N_CP), and the number of observed carbon stars (N_CO). We also include the projected radial distance for all fields, although it was only used in the analysis of halo fields.

Table 3.  Selection Function Data

Field	${N}_{{\rm{masks}}}$	${D}_{{\rm{M31,proj}}}$ a	Theoretical C/M	${N}_{{\rm{AGB}}}$ b	${N}_{{\rm{C,pred}}}$ c	${N}_{{\rm{C,obs}}}$ d
		(kpc)
SE8	1	3.8	0.001	0	0.00	0
SE9	1	6.0	0.001	0	0.00	0
f109	1	8.9	0.002	1	0.00	0
f1	2	12.2	0.002	1	0.00	0
NW1V	1	13.1	0.002	1	0.00	0
f116	1	13.2	0.002	1	0.00	0
f115	1	14.5	0.002	3	0.01	0
f207	1	16.3	0.002	0	0.00	0
NW1dV	1	18.0	0.003	2	0.01	0
25 kpc	1	20.0	0.003	0	0.00	0
NE3	1	20.6	0.003	0	0.00	0
f123	1	21.0	0.003	0	0.00	0
f2	2	21.3	0.003	2	0.01	0
NW2V	2	21.5	0.003	1	0.00	0
NE1	1	22.8	0.003	0	0.00	1
NW2dV	1	24.2	0.004	2	0.01	0
M32	6	5.7	0.004	0	0.00	5
NE6	1	24.8	0.004	0	0.00	0
f130	2	25.3	0.004	3	0.01	0
NW3V	1	26.8	0.004	1	0.00	1
a0	3	29.8	0.005	4	0.02	0
NE4	1	32.9	0.006	0	0.00	1
g1	1	34.5	0.006	0	0.00	0
a3	3	35.4	0.006	4	0.03	0
mask4	1	36.8	0.007	0	0.00	0
f135	1	37.9	0.007	0	0.00	0
NE2	1	39.1	0.008	0	0.00	1
A240	3	55.2	0.017	0	0.00	0
A338	3	56.1	0.018	1	0.02	0
m4	5	56.6	0.018	4	0.07	0
a13	4	58.0	0.019	5	0.10	0
NW9V	2	65.4	0.028	1	0.03	0
n205	3	32.6	0.056	34	1.79	7
a19	4	79.5	0.056	1	0.05	0
A220	3	85.2	0.075	2	0.14	0
m6	5	85.3	0.075	0	0.00	0
A310	3	87.1	0.082	1	0.08	0
A040	3	90.3	0.096	0	0.00	0
b15	5	91.9	0.104	1	0.09	0
NW13dV	1	92.3	0.106	0	0.00	0
NW14V	1	95.5	0.125	0	0.00	0
NW15V	1	100.3	0.158	0	0.00	0
N147	4	101.4	0.242	7	1.36	12
m8	2	116.8	0.356	1	0.26	0
N185	5	97.4	0.643	4	1.57	4
d7	2	220.1	1.050	3	1.54	0
d1	2	44.8	1.342	8	4.58	0
d5	4	109.8	2.798	2	1.47	1
m11	4	158.6	2.827	2	1.48	0
A170	2	159.9	3.014	0	0.00	0
d2	12	140.5	3.403	11	8.50	5
A080	2	164.8	3.838	1	0.79	0
A305	2	169.5	4.842	0	0.00	0
d3	3	68.4	6.758	5	4.36	0
d18	1	113.0	7.453	1	0.88	0
d22	3	245.1	7.453	0	0.00	0
d15	2	93.7	7.453	0	0.00	0
d13	5	126.8	12.165	0	0.00	0
d10	2	76.7	14.091	0	0.00	2
d11	1	102.2	19.855	0	0.00	0
d16	2	138.7	32.407	0	0.00	0
d12	1	114.0	32.407	1	0.97	0
d9	2	36.8	52.894	5	4.91	1
d14	1	159.8	70.968	1	0.99	0

Notes.

^aThe projected distance (in kpc) of each field from the center of M31. ^b The number of likely AGB stars observed in each field (see text for details). ^c The predicted number of carbon stars in each field given the total number of observed AGB stars and the theoretical C/M ratio. ^d The observed number of carbon stars in each field.

Download table as:  ASCIITypeset images: 1 2

There are 25 fields in which there are zero predicted C stars and zero observed C stars. Of these 25, five are dSphs (And XI, And XIII, And XV, And XVI, and And XXII). These five dSphs are some of the least massive of our sample, where the majority of bright stars would have been put on a mask. There are four fields in which we do identify some carbon stars despite there being none predicted: And X, NE2, NE4, and NE1. These carbon stars are either fainter than the TRGB or bluer than $V-I=1.8$, and they will be discussed in detail later in this paper.

In most of the remaining fields, we observe roughly the number of carbon stars that we expect (to within a factor of two). However, in And II, NGC 147, NGC 185, and NGC 205 we observe far more carbon stars than are predicted, despite the fact that our calculated C/M ratios match those in the literature. As in And X and the northeast shelf fields, many of these stars are fainter than the TRGB and/or bluer than $V-I=1.8$. In And I and And III, we predict several carbon stars (∼4.5 in both fields) but do not observe any.

We identify 41 carbon stars in the satellites and halo of M31. Many are located in fields in which significant numbers of intermediate-age AGB stars have already been identified, including the dEs NGC 147, NGC 185, and NGC 205 and the dSph And II (e.g., Nowotny et al. 2003; Kerschbaum et al. 2004; Battinelli & Demers 2004a, 2004b). We also identify several carbon stars in the vicinity of M32, which is known to have a substantial intermediate-age population (Davidge 2014; Jones et al. 2015). However, the majority of the stars in our sample have kinematics suggesting membership to M31 rather than M32 itself. We identify a small number of carbon stars in the dSphs And V, And IX, and And X. However, this combined sample of four stars, three of which are fainter than the TRGB, indicates no significant intermediate-age population in these galaxies. Finally, we see four carbon stars in regions known to contain substructure from the giant southern stream.

Analysis of the carbon star photometry also reveals that a significant fraction of our sample is fainter than the TRGB. These may be extrinsic in origin, and we will explore their nature fully in the next section.

Our sample of carbon stars appears to be largely dust-free and hydrostatic. Their optical colors are fairly well fit by the dust-free hydrostatic models from Aringer et al. (2016), though the models do not extend blue enough to fully match the colors of the faint carbon stars. The introduction of MIR colors complicates this picture, as the dust-free models do not match the observations well in optical versus MIR color–color space. However, none of the carbon stars in our sample appear in the DUSTiNGS variable star/extreme-AGB star catalog, which further supports the idea that they are not heavily affected by dust and dynamics (though again, two epochs are insufficient to rule out variability completely). Finally, very few of our carbon stars exhibit Hα emission, which is associated with the presence of shock waves induced by pulsation in the stellar atmosphere. Of the carbon stars in the solar neighborhood, 70% of the Miras, 66% of the SRa type, and 20% of SRb- and Lb-type stars show Hα in emission (Mauron et al. 2014). The carbon stars in the satellites and halo of M31 are comparatively quite quiet, which would be consistent with a significant population being extrinsic.

Application of PCA to the carbon star spectra illustrates the effects of temperature and metallicity. We find that the depth of the broad CN bands (governed by the second eigenspectrum) correlates most strongly with metallicity, while the depth and shape of the break in the spectrum at ∼7900 Å trace T_eff (governed by the fourth eigenspectrum). Comparing the coefficients of the eigenspectra governing these features illustrates that the carbon stars in M32 and the M31 disk/halo are typically more metal-rich than the carbon stars in the substructure and dSphs.

In many of our satellites we see a significant number of carbon stars fainter than the TRGB. There are three possible explanations. The first is that these are extrinsic carbon stars, whose carbon was obtained by mass transfer from a carbon-rich AGB star onto a binary companion rather than through dredge-up of He-burning products during the TP-AGB phase (de Kool & Green 1995; Frantsman 1997; Izzard & Tout 2004). The second possible explanation is that they are genuine TP-AGB stars experiencing the post-flash luminosity dip (Boothroyd & Sackmann 1988) or in a minimum of their dynamic phase (Nowotny et al. 2011). Finally, they may be bright AGB stars extincted by circumstellar dust. Both Sections 4 and 5 treat the bright (super-TRGB) and faint (sub-TRGB) carbon stars as separate populations, to help determine which of the above explanations is most likely.

From the photometry, we can rule out dust as a likely explanation for the faint carbon stars. If the faint carbon stars are heavily extincted, then they should appear redder than the brighter population in V − I. However, as noted in the previous section, our sample appears to be well fit by the dust-free hydrostatic models. In addition, both the one- and two-dimensional color distributions (Figures 4 and 5, respectively) indicate that the faint carbon stars are typically bluer in V − I than the bright carbon stars.

If the faint carbon stars are extrinsic, we would expect them to be present down to the SPLASH detection limit, while carbon stars on the TP-AGB would be clustered around the TRGB. The observed distribution of I-band magnitudes suggests the former: Figure 3 illustrates that the faint carbon stars are present down to 1.5 mag below the TRGB (though this varies from field to field given inhomogeneous detection limits).

We would also expect extrinsic carbon stars to be warmer than their intrinsic counterparts. Both Figures 4 and 5 illustrate that many of the faint carbon stars are indeed bluer in $(V-I)$ than the bright carbon stars, despite comparable CN–TiO color. They are also typically bluer than the dust-free hydrostatic AGB models from Aringer et al. (2016). The effects of dust and dynamics are not likely to make C stars so optically blue ($V-I\lt 1.5$), so this suggests that these stars do not in fact belong to the AGB.

Spectroscopically, many of the known differences between intrinsic and extrinsic carbon stars fall beyond our wavelength coverage (e.g., G band of CH at ∼4300 Å, ${}^{12}{\rm{C}}{/}^{13}{\rm{C}}$ ratio). The major difference between the bright and faint carbon stars in our sample is the eigencoefficients of the second eigenspectrum (EC₂) (see Figure 10). The faint carbon stars typically have negative EC₂ values, while the bright carbon stars have positive EC₂ values. Observationally, this translates to the faint carbon stars having weaker CN bands and appearing more metal-poor than the bright carbon stars. This is fully consistent with observations of extrinsic carbon stars in the Local Group, which are often characterized by their lack of metals (e.g., the CEMP stars). Within the subset of faint carbon stars, we find that EC₂ anticorrelates with absolute I-band magnitude. Observationally, this means that the brightest/most massive of the faint carbon stars have the strongest CN. Finally, the eigencoefficients of the fifth eigenspectrum (EC₅) also differ between the bright and faint samples, perhaps due to more noise in the fainter stars.

However, all of the figures discussed in this section also illustrate that the colors and magnitudes of the faint carbon stars cover a significant range. Some are fully consistent with the bright carbon stars. It is probable that the faint population consists of a mix of intrinsic and extrinsic carbon stars. Models based on the Magellanic Clouds suggest that <10% of TP-AGB stars are fainter than the TRGB (e.g., Marigo & Girardi 2007; Melbourne & Boyer 2013) at any given time. This places an upper limit on the number of intrinsic interlopers in the faint population at 1.8 stars.

At first glance, we appear to observe more faint carbon stars in our less luminous galaxies. To parameterize this, we calculate ${N}_{{FC}}/{N}_{C}$, the fraction of faint (sub-TRGB) carbon stars in the full sample of carbon stars, for each galaxy. To place these findings within a broader context, we also calculate ${N}_{{FC}}/{N}_{C}$ in other Local Group satellites. To compare similar groups of stars, we limit ourselves to optical carbon star surveys, leaving aside surveys in the NIR and MIR and serendipitous carbon (or CH) star discoveries. We assemble C-star counts in IC 1613 (Albert et al. 2000), Leo I, Sagittarius dIrr (Demers & Battinelli 2002), DDO 210/Aquarius, Pegasus (Battinelli & Demers 2000), NGC 6822 (Letarte et al. 2002), Phoenix (Martínez-Delgado et al. 1999), Draco, Ursa Minor (Shetrone et al. 2001), the LMC (Kontizas et al. 2001), And III, And VI, And VII, and Cetus (Harbeck et al. 2004). These data are provided in Table 5.

Table 5.  Fraction of Faint Carbon Stars in the Local Group

Field	NC	${N}_{{FC}}/{N}_{C}$	Ilimit − ITRGB	Reference
IC 1613	195	${0.087}_{0.01}^{0.02}$	1.6	Albert et al. (2000)
Leo I	27	${0.148}_{0.04}^{0.06}$	1	Demers & Battinelli (2002)
Aquarius	3	${0.000}_{0.00}^{0.16}$	0.85	Battinelli & Demers (2000)
Pegasus	40	${0.150}_{0.04}^{0.05}$	1.1	Battinelli & Demers (2000)
Sagittarius dIrr	33	${0.242}_{0.05}^{0.06}$	1.6	Demers & Battinelli (2002)
LMC	7760	${0.017}_{0.00}^{0.00}$	...	Kontizas et al. (2001)
NGC 6822	907	${0.163}_{0.01}^{0.01}$	1.9	Letarte et al. (2002)
Phoenix	2	${0.500}_{0.23}^{0.23}$	...	Martínez-Delgado et al. (1999)
Draco	6	${1.000}_{0.09}^{0.00}$	...	Shetrone et al. (2001)
Ursa Minor	7	${1.000}_{0.07}^{0.00}$	...	Shetrone et al. (2001)
And III	1	${1.000}_{0.36}^{0.00}$	1.6	Harbeck et al. (2004)
And VI	2	${0.500}_{0.23}^{0.23}$	1.6	Harbeck et al. (2004)
And VII	5	${0.400}_{0.15}^{0.17}$	1.4	Harbeck et al. (2004)
Cetus	3	${0.667}_{0.22}^{0.16}$	1.5	Harbeck et al. (2004)
And X	2	${1.000}_{0.22}^{0.00}$	1.92	This work
M32	1	${0.000}_{0.00}^{0.36}$	0.25	This work
And IX	1	${1.000}_{0.36}^{0.00}$	2.2	This work
NGC 205	6	${0.167}_{0.08}^{0.14}$	0.3	This work
And V	1	${0.000}_{0.00}^{0.36}$	1.71	This work
NGC 147	158	${0.301}_{0.03}^{0.03}$	2.6	This work + Nowotny et al. (2003)
NGC 185	157	${0.325}_{0.03}^{0.03}$	3.0	This work + Nowotny et al. (2003)
And II	10	${0.300}_{0.10}^{0.12}$	2.5	This work + Kerschbaum et al. (2004)

Download table as:  ASCIITypeset image

The majority of these fields (IC 1613, Pegasus, Aquarius, NGC 6822, Leo I, Sag dIrr) rely on optical FBPS to identify carbon stars. All survey the full galaxy area and extend to 1–2 mag fainter than the TRGB. The approximate extent below the TRGB is listed in Table 5. This is roughly equivalent to the SPLASH coverage and detection limit, so we will not worry tremendously about differences in depth for these fields. However, all require C stars to have $R-I\gt 0.9$ (though Letarte et al. [2002] also look at "bluer" C stars with $0.8\lt R-I\lt 1.1$). This makes them biased against the faint carbon stars, which we find to be bluer. As a result, the ${N}_{{FC}}/{N}_{C}$ that we calculate in these fields are likely lower limits. In each field we compute ${I}_{{\rm{TRGB}}}$ following the same prescription outlined in Section 3.2 and count the number of C stars fainter than that limit.

We consider the carbon star samples in Phoenix, Draco, Ursa Minor, the LMC, And III, And VI, And VII, and Cetus to be representative of the full C-star population. The two carbon stars in Phoenix were confirmed spectroscopically by Da Costa (1994), though our data come from Martínez-Delgado et al. (1999). Subsequent work by Menzies et al. (2006) did not identify any other carbon stars visible in the optical, so we consider this sample of two to be complete. The C-star census presented in Shetrone et al. (2001) for Draco and Ursa Minor relies on the identification of carbon stars via photographic plates (Aaronson et al. 1982; Azzopardi et al. 1986). These stars have been thoroughly vetted (e.g., Dominguez et al. 2004; Abia et al. 2008), and no new carbon stars have been identified. The Kontizas et al. survey of the LMC was conducted with objective-prism plates and visual identification of carbon stars using the Swan C₂ bands at 4737 and 5165 Å. Because this is a spectroscopic rather than photometric identification, there is not a strong bias against faint C stars. This survey is likely incomplete in the most crowded central regions, but we do not expect the extrinsic and intrinsic carbon stars to have different spatial distributions, and this incompleteness is not likely to skew our counts. Thus, the ${N}_{{FC}}/{N}_{C}$ fraction we calculate in the LMC is likely representative. These four fields rely on spectroscopic identification of carbon stars, so their photometric limits are excluded from Table 5. Finally, Harbeck et al. (2004) use FBPS to identify carbon stars in And III, And VI, And VII, and Cetus, but do not apply a limit in V − I. They thus sample the bright and faint carbon stars uniformly.

Because Shetrone et al. (2001) provide V-band magnitudes of the carbon stars in Draco and Ursa Minor, we compute ${V}_{{\rm{TRGB}}}$ using relationships from Bellazzini et al. (2004) and Mager et al. (2008). For Phoenix and the LMC, for which I-band magnitudes have been provided, we once again calculate the TRGB magnitude using the method described in Section 3.2. Harbeck et al. (2004) have provided designations of C versus dC (which translate to super- and sub-TRGB), which we use to calculate ${N}_{{FC}}/{N}_{C}$ in their four fields.

In addition to the new fields discussed above, we also compile C-star populations in And II, NGC 147, and NGC 185 from the literature. We combine our C-star samples observed by SPLASH with these samples from the literature. The observations of And II (Kerschbaum et al. 2004), NGC 147, and NGC 185 (Nowotny et al. 2003) also use FBPS, but with V and i₀ as the broadband colors rather than R and I. Both authors' observations extend ∼0.5 mag below the TRGB, and they require C stars to have ${(V-i)}_{0}\gt 1.16$ and (TiO–CN)₀ < −0.3. This limit in ${(V-i)}_{0}$ is equivalent to $(R-I)\gt 0.53$, which encompasses the full color range over which we observe carbon stars. As a result, we consider this a representative sample. As Bellazzini et al. (2004) do not provide a TRGB calibration for i₀, we use the values calculated by Kerschbaum et al. (2004), ${i}_{{\rm{TRGB}}}=20.5$, and Nowotny et al. (2003), ${i}_{{\rm{TRGB}}}=19.96$ for NGC 185 and ${i}_{{\rm{TRGB}}}=20.36$ for NGC 147, when computing ${N}_{{FC}}/{N}_{C}$.

Finally, two of the SPLASH fields (M32 and NGC 205) were surveyed so shallowly that they too should be considered lower limits on ${N}_{{FC}}/{N}_{C}$. We calculate this limit in SPLASH fields using the full SPLASH sample (independent of membership or strength of carbon features).

Figure 11 plots ${N}_{{FC}}/{N}_{C}$ in the galaxies listed in Table 5 as a function of the absolute magnitude and metallicity of the host satellite (obtained from McConnachie 2012). Also plotted are the $1\sigma$ binomial proportion confidence intervals, save in the cases discussed above where the calculated fraction represents a lower limit. With the exception of the LMC, NGC 147, and NGC 185, the small sample sizes translate to large uncertainties. A trend is still evident; the fraction of faint carbon stars decreases as the galaxy gets brighter and its metallicity increases. The serendipitous detections of CH and dC stars in MW globular clusters adhere to this trend, increasing the number of faint, low-metallicity systems in which ${N}_{{FC}}/{N}_{C}=1$.

Zoom In Zoom Out Reset image size

There are two possible explanations for this trend. The first presumes that the faint carbon stars are extrinsic and owe their composition to a binary companion. If the fraction of stars in binary systems increases in smaller satellites (as it has been shown to increase in smaller globular clusters; Milone et al. 2012), then so will the fraction of faint carbon stars. However, this explanation would also require a sizable population of AGB stars in earlier generations to provide the binary companions. The second explanation presumes that the entire carbon star sample is made up of intrinsic TP-AGB stars. In this case, the trend could be explained by the effect of metallicity on carbon star formation. At low metallicity, dredge-up efficiency is higher, the range of masses over which carbon stars can form is larger, and TDU begins earlier (Karakas et al. 2002; Marigo et al. 2013; Karakas 2014). It is possible that at low metallicities these effects combine to produce carbon stars that are fainter than the TRGB. Indeed, both explanations may be occurring simultaneously. The more massive companion in an earlier binary pair is likely to go through the AGB. In metal-poor galaxies, that AGB star is more likely to be carbon-rich and the low-mass companion is more likely to become carbon enhanced.