PAndAS' CUBS: DISCOVERY OF TWO NEW DWARF GALAXIES IN THE SURROUNDINGS OF THE ANDROMEDA AND TRIANGULUM GALAXIES^*

Two new dwarf galaxies were directly identified on matched-filter surface density maps of RGB candidate stars selected to have colors consistent with the average (generally) metal-poor locus of M31 dwarf spheroidals. Satellites in the very sparsely populated outer halo regions correspond to stellar overdensities that are striking enough for them to be easily spotted, even down to magnitudes of M_V ∼ −6.5. This was the case for And XII (Martin et al. 2006) and And XX (McConnachie et al. 2008) and also here with Andromeda XXII. An automated search is certainly warranted to better quantify the detection limits of the survey in uncovering new dwarf galaxies. However, it is beyond the scope of this initial discovery paper and will be presented elsewhere.

The locations of the two new satellites are shown on the map of the PAndAS survey (Figure 1). As with many previous detections, these two systems are noticeably close to the edge of the survey, suggesting that the survey limit of 150 kpc in projected distance from M31 does not represent the true extent of Andromeda's satellite system (see also Figure 2 of McConnachie et al. 2009). In the spirit of previous conventions, we dub these two satellites Andromeda XXI and Andromeda XXII (And XXI and And XXII; see the discussion below, in the Appendix).

Zoom In Zoom Out Reset image size

Both systems appear as overdensities of stars on the sky, as is visible in the left panels of Figures 2 and 3. These stars are also aligned along a RGB that would be at, or close to, the distance of M31 or M33. The color-magnitude diagrams (CMDs) within 2 half-light radii of the dwarfs (determined in Section 3.3) are shown in the middle panels of these figures and, when compared to the CMD of reference fields chosen in an annulus covering the same area at a distance of ∼20' from the dwarfs' centers (right panels), indeed reveal an alignment of stars that follow the typical shape of a RGB. Isolating these stars enhances the contrast of the overdensity of stars on the sky (large symbols in the left panels).

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

And XXI is typical of the relatively bright dwarf galaxies that we have found before (such as And XV or And XVI). The overdensity of stars in the CMD region 24.5 < i < 25.0 and g − i ∼ 1.0 corresponds to its horizontal branch but the degrading star/galaxy separation at these magnitudes, as well as the rapidly dropping completeness, makes it difficult to extract any accurate information from this region of the CMD. With only ∼20 stars visible along its RGB, And XXII is one of the faintest M31 satellites found so far.

All the parameters of the two dwarf galaxies determined in this section of the paper are summarized in Table 1.

Table 1. Derived Properties of the Satellites

Parameter	And XXI	And XXII
α (J2000)	23h54m47.7s ± 1.0s	01h27m40.0s ± 0.6s
δ (J2000)	42°28'15'' ± 15''	28°05'25'' ± 6''
(l, b) (°)	(111.9 − 19.2)	(132.6 − 34.1)
E(B − V)a	0.094	0.076
(m − M)0	24.67 ± 0.13	[24.50] ≲25.07b
D (kpc)	859 ± 51	[794] ≲1033b
rM31 (kpc)	152 ± 31	[224] ≲315
rM33 (kpc)	⋅⋅⋅	[42] ≲227
〈[Fe/H]〉	−1.8	[ − 1.8] ≳ −2.5
rh (arcmin)	3.5 ± 0.3	0.94 ± 0.10
rh (pc)	875 ± 127	[217] ≲282
P.A. (North to East) (°)	110 ± 15	114 ± 15
= 1 − b/a	0.20 ± 0.07	0.56 ± 0.11
MV	−9.9 ± 0.6	[ − 6.5 ± 0.8] ≳ −7.0 ± 0.8
LV (L☉)	7.8 ± 1.9 × 105	[0.3 ± 0.1 × 105] ≲ 0.6 × 105
μV,0 (mag arsec−2)	27.0 ± 0.4	26.7 ± 0.6

Notes. ^aSchlegel et al. (1998). ^bGiven the highly unreliable distance estimate of And XXII from its TRGB, parameters between square brackets are determined assuming the average distance of M31/M33. The limit on all parameters, determined from the distance limit measured from the brightest star along And XXII's RGB, are also given in the same column.

Download table as:  ASCIITypeset image

It is difficult to accurately estimate distances to dwarf galaxies without photometric observations that reach down to their horizontal branches. Indeed, the tip of the RGB (TRGB), which is often used as a distance indicator since its magnitude in the I band only slightly varies with the properties of the dwarfs' stellar populations (e.g., Bellazzini et al. 2001), is not sampled well enough in faint systems that suffer from "CMD shot noise": these systems do not contain enough stars for their CMD to be reliably sampled by their few stars, especially along their RGB (Martin et al. 2008). In the case of And XXI, this is not so much of an issue as it contains a reasonable number of RGB stars and the TRGB is therefore well defined. From the luminosity function (LF) of stars within two half-light radii of the dwarf's center (dashed ellipse on the map of Figure 2) and within the two dashed lines of the dwarf's de-reddened CMD presented in Figure 4, we estimate I_0,TRGB = 20.63 ± 0.04 by applying the algorithm of McConnachie et al. (2004).¹² The complete LF is shown on the top-right panels of the same figure. Along with the absolute TRGB magnitude of M_I = −4.04 ± 0.12 (Bellazzini et al. 2001), we determine a distance modulus of (m − M)₀ = 24.67 ± 0.13, or a distance of D = 859 ± 51 kpc.¹³

Zoom In Zoom Out Reset image size

In And XXII, on the other hand, the brightest possible member RGB star that can be seen in the CMD of stars within two half-light radii is much less likely to be close to the TRGB and thus to yield a reliable distance. It can only provide an upper limit to the distance modulus of this dwarf. This star, with I₀ = 21.03 (see the top-right panel of Figure 4) converts to a maximum distance modulus of (m − M)_0,max = 25.07 or D_max = 1033 kpc. This limit to And XXII's distance clearly makes this dwarf galaxy a likely member of the M31/M33 satellite system. The g-band LF of And XXII also shows a tentative detection of its horizontal branch, at a distance of ∼940 kpc, very near the limit of the survey. Although this measurement is very uncertain, the absence of a clear horizontal branch at brighter magnitude would imply that And XXII is at least as distant as And XXI. For the moment, we will assume a distance to And XXII that is the same as that to M31/M33 ((m − M)₀ ≃ 24.50 or 794 kpc) until deeper photometric data provide a better constraint, keeping in mind that the dwarf is likely to be more distant than this.

Figure 4 also presents the metallicity distribution (MD) of the stars that were used to construct the LF of the two satellites. The metallicity estimates of individual stars were obtained by comparison with the Dotter et al. (2008) set of 13 Gyr isochrones¹⁴ with [α/Fe] = +0.0,¹⁵ placed at the distance moduli determined above. Each star is given a metallicity determined by the ratio of the distances to its two closest isochrones. These have a metallicity ranging from [Fe/H] = −2.5 to [Fe/H] = +0.5, with steps of 0.1 dex. There is no significant contamination of these MD from the field population (dotted histograms), except for the very few, most metal-rich stars to appear in And XXI's MD. When this contamination is removed, both galaxies are determined to be quite metal poor, with an average metallicity of 〈[Fe/H]〉 = −1.8. The corresponding isochrones, overlaid on the CMDs of the same figure, show good agreement with the location of the observed RGB stars that revealed the satellites. Of course, if the distance modulus of And XXII is larger than the (m − M)₀ = 24.50 value adopted here, that would translate into a lower metallicity for this system, with a minimum 〈[Fe/H]〉 = −2.5 for the maximum distance obtained above. It should also be noted that, given the photometric uncertainties and the small color differences between metal-poor isochrones, the determination of photometric metallicities does not give a good handle on the metallicity spread in these systems and the width of the MDs should not be interpreted at face value.

To determine the structural properties of the two dwarf galaxies, we apply the maximum likelihood technique developed by Martin et al. (2008) that determines the best exponential profile (and background level) to fit the spatial distribution of a galaxy's stars. Compared to the original paper, the algorithm has been modified to account for incompleteness in the surroundings of the satellites, produced by gaps between the MegaCam CCDs, the survey edges, and the presence of saturated bright stars. The first effect is patently important in the case of And XXI, which has a CCD gap running through its center. The best exponential profile is determined iteratively, after filling holes from the parameters of the best previous model. This step is performed by first binning the survey region around the dwarfs with small 0.1 × 0.1 arcmin² pixels. Each pixel that falls over the edge of the survey, in a CCD gap, or behind a bright star is filled with a number of stars randomly generated from a Poisson distribution of mean the model density at this pixel's central position, computed from the best model from the previous iteration. These stars are then randomly distributed within the pixel. Convergence on the structural parameters model is usually achieved within five iterations but we proceed with eight iterations to secure convergence. Given the stochasticity introduced by filling the holes with artificial stars, we determine the contribution of the gap to the uncertainties on the model parameters via a Monte Carlo approach, performing 100 fits for each galaxy.

So as to limit the impact of background/foreground contamination, we apply the maximum likelihood fit to these stars that fall within the selection box overlaid on the CMDs of Figures 2 and 3. These have a faint limit fixed at i = 24.0 so that fainter regions where star/galaxy separation becomes an issue are avoided. The centroid (α₀, δ₀), half-light radius (r_h), ellipticity (defined as = 1 − b/a with a and b being, respectively, the major and minor axes scale lengths of the system) and position angle (P.A.) for which convergence was achieved are listed in Table 1.

In Figure 5, the fitted exponential profiles are compared to the data, which are binned following the best centroid, ellipticity, and position angle determined by the fit. The binned data points have been corrected for spatial incompleteness produced by CCD gaps, survey edges, and bright stars. For both dwarfs, the agreement between profile and data is quite good. There are regions—in particular in the inner parts of the And XXI profile and the outer parts of the And XXII one—where deviations from the model are significant but these correspond to regions were a large completeness correction was applied.

Zoom In Zoom Out Reset image size

To measure the total magnitude of each dwarf galaxy, we first determine the flux contribution of all stars brighter than i₀ = 23.5 (where we know the survey to be complete¹⁶; Ibata et al. 2007) and within its half-light radius, following the best structural parameters determined above. We then correct this number for background/foreground contamination by counting the flux within the annulus between 16' and 20', scaled to the area covered within the half-light radius. Both these fluxes are also corrected for incompleteness coming from bright stars, CCD gaps, and survey edges. For And XXI and And XXII, we measure partial magnitudes of M_V = −7.7 and −4.3, respectively. These numbers then need to be corrected to include the flux contribution of all stars fainter than i₀ = 23.5. In order to do this, we use the PandAS observations of another Andromeda satellite, And III, that has also been observed in the survey under the same conditions as And XXI and And XXII. In the case of this satellite, we measure a difference of 2.2 mag between the flux of stars brighter than the i₀ = 23.5 limit and the total magnitude of the dwarf M_V = −10.2 ± 0.3 measured by McConnachie & Irwin (2006). This leads to total magnitudes of M_V = −9.9 and −6.5 for the two new dwarfs when we account for this difference. We estimate the uncertainties on these values to be ±0.6 for And XXI and ±0.8 for And XXII, stemming from the uncertainties on the structural parameters (±0.2), the distance estimates (±0.1 and ±0.3), the CMD shot noise of these sparsely populated dwarf galaxies (±0.2 and ±0.4), and the total magnitude of And III used for the calibration (±0.3) all added in quadrature, to which we add a systematic uncertainty of ∼0.2 mag to account for possible differences between the And XXI and And XXII luminosity functions and that of And III.

From the total magnitude, we determine the surface brightness of the dwarfs, following Equation (6) of Martin et al. (2008) with the number of stars N_* being replaced by the total luminosity of the dwarf, L_V, listed in Table 1. This equation accounts for the structure of the dwarf and in particular its ellipticity, and yields central surface brightnesses of μ_V,0 = 27.0 ± 0.4 and 26.7 ± 0.4 mag arcsec⁻², respectively for And XXI and And XXII. Their low surface brightnesses make them invisible on photographic plate and, contrary to the brighter And XVIII (McConnachie et al. 2008), they do not appear on the Digitized Sky Survey (DSS).

The measured magnitude and surface brightness of both new dwarf galaxies are within the range of values measured for dwarf galaxies previously found and, as such, And XXI and And XXII are very typical Andromeda satellites. The structures of these two dwarf galaxies are also fairly typical of that of M31's satellites. Although the ellipticity of And XXII is quite high, it is not the only elliptical dwarf spheroidal galaxy known to orbit Andromeda. And III, for instance, has a similarly large ellipticity ( = 0.52 ± 0.02; McConnachie & Irwin 2006).

With a half-light radius of almost 1 kpc, And XXI supports the size difference that has previously been noted to distinguish Andromeda from Milky Way satellites: an M31 dwarf galaxy is typically 2–3 times larger than its Galactic counterpart (McConnachie & Irwin 2006). This distinction still holds for fainter dwarf galaxies like the ones recently discovered. And XXI is the fourth largest dwarf spheroidal galaxy known in the Local Group after And XIX (r_h = 1.7 ± 0.1 kpc; McConnachie et al. 2008), And II (r_h = 1.12 ± 0.02 kpc; based on the exponential profile fit of McConnachie & Irwin (2006) for consistency) and Sagittarius, which is clearly in the final throes of its tidal disruption by the Milky Way. It is also interesting to notice that two of these galaxies (And XIX and And XXI) are recent discoveries, hinting that some other large satellites could still elude us, especially given that we are biased against finding large faint galaxies, as they are more diffuse and therefore have a lower surface brightness. Galaxies as large and diffuse as And XIX would also be very hard to find around the Milky Way but, on the contrary, it is unlikely that Milky Way satellites that share the properties of And II and And XXI (r_h ∼ 1 kpc, M_V ≲ −9.0) could have been missed in the quarter of the Galactic sky covered by the SDSS (Koposov et al. 2008; Walsh et al. 2009). The largest Galactic satellite known to date was indeed discovered in the SDSS, but its half-light radius is only a mere 564 ± 36 pc. Such a distinction is puzzling and can be used, in conjunction with spectroscopic data, to provide strong constraints on the type of dark matter haloes that these satellites should inhabit (Peñarrubia et al. 2008).

For And XXII, the most intriguing question is whether it is a satellite of M33 or M31. M33 is a small spiral galaxy with a mass approximately one tenth the mass of M31. It resides at a distance of 204 kpc from the center of M31, presumably well inside the M31 dark halo whereas measurements of its line-of-sight velocity and proper motion suggest that it is on a highly elongated orbit around M31. The implication is that M33 is a satellite of M31 and likely passed much closer to M31 in the past than its present distance. This association is supported by the existence of a strong warp in the M33 HI disk (Rogstad et al. 1976) as well as the presence of a stellar stream extending from its stellar disk (McConnachie et al. 2009).

If And XXII is shown to be a satellite of M33, then it would expand the known size of the M33 system by 50%, from its recently discovered outermost globular cluster (Huxor et al. 2009). The existence of a satellite at this radius would also place a constraint on the strength of an M33–M31 interaction since such an interaction naturally strips material from the outer parts of M33. Finally, we point out that And XXII may be one of the few examples of a satellite's satellite, a phenomenon that naturally arises in the hierarchical clustering scenario of structure formation.

And XXII is, in projection, much closer to M33 than to M31. However, since the distance to And XXII has not been determined, the true And XXII/M31 and And XXII/M33 distances are not known. This is illustrated in the top panel of Figure 6 that shows these distances as a function of the heliocentric distance to And XXII, D_{And XXII}.

Zoom In Zoom Out Reset image size

Cosmological simulations suggest that the density of satellites traces the dark matter density (Sales et al. 2007). If true, the ratio of the M33 halo density to the M31 halo density at a given location provides a rough measure of the probability for a dwarf at this location to be a satellite of M33 relative to the probability that it is a satellite of M31. This ratio is shown in the bottom panel of Figure 6, as a function of the heliocentric distance to And XXII. Here, we have assumed that the halos of M31 and M33 have Navarro–Frenk–White density profiles (Navarro et al. 1997) defined by a virial mass, M_vir, and a virial radius, R_vir. By definition, the mean density within the virial radius is a factor Δ times the background density where we take Δ ≃ 95 (Macciò et al. 2008), as appropriate for WMAP5 values of the cosmological parameters (Komatsu et al. 2009). We assume M_vir = 2 × 10¹² M_☉ for M31 and a mass that is 10 times lower for M33, consistent with current observational constraints (McConnachie et al. 2006). Concentration parameters are chosen as c = 10 and c = 9, respectively, following the relations of Neto et al. (2007) and Macciò et al. (2008), which leads to virial radii of R_vir = 327 kpc for M31 and R_vir = 152 kpc for M33. Figure 6 suggests that if the distance to And XXII is found to be between 720 kpc and 930 kpc, then it is more likely to be a satellite of M33 than of M31.¹⁷

An alternative argument can be made based on the notion of a Hill sphere which roughly defines the region where M33 rather than M31 dominates the dynamics of small systems. In the limit of circular orbits (admittedly, a bad assumption for this problem) the radius of the Hill sphere for M33 is given by r_Hill = D_M33-M31(M_M33/3M_M31)^1/3. The range of distances where And XXII is likely to be an M33 satellite is then narrower (757 < D_And XXII < 856 kpc).

While our arguments are admittedly simplistic, they illustrate the importance an accurate distance measurement for And XXII will have in settling this issue. A measurement of the line-of-sight velocity for And XXII will also provide valuable information as will detailed modelling of the M31–M33 interaction.