EIGHT ULTRA-FAINT GALAXY CANDIDATES DISCOVERED IN YEAR TWO OF THE DARK ENERGY SURVEY

Observations: The Y2Q1 data set consists of 26,590 DECam exposures taken during the first two years of DES observing that pass DES survey quality cuts. Slightly less than half of the DES survey area was observed during the first season (Y1; 2013 August 15–2014 February 9), with typically two to four overlapping exposures, referred to as "tilings," in each of the grizY filters. The second season (Y2; 2014 August 7–2015 February 15) covered much of the remaining survey area to a similar depth. Exposures taken in the griz bands are 90 s, while Y-band exposures are 45 s.

Image Reduction: The DES exposures were processed with the DES data management (DESDM) image detrending pipeline (Sevilla et al. 2011; Desai et al. 2012; Mohr et al. 2012; R.A. Gruendl et al. 2015, in preparation). This pipeline corrects for cross-talk between CCD amplifier electronics, bias level variations, and pixel-to-pixel sensitivity variations (flat-fielding). Additional corrections are made for nonlinearity, fringing, pupil, and illumination. Both the Y1 and Y2 exposures were reduced with the same image detrending pipeline used to process Y1A1.

Single-epoch Catalog Generation: Astronomical source detection and photometry were performed on a per exposure basis using the PSFex and SExtractor routines (Bertin & Arnouts 1996; Bertin 2011) in an iterative process (Mohr et al. 2012; R.A. Gruendl et al. 2015, in preparation). As part of this step, astrometric calibration was performed with SCAMP (Bertin 2006) by matching objects to the UCAC-4 catalog (Zacharias et al. 2013). The SExtractor source detection threshold was set at a signal-to-noise ratio (S/N) of S/N > 10 for the Y1 exposures, while for Y2 this threshold was lowered to S/N > 5 (the impact of this change on the stellar completeness is discussed in Section 2.2). During the catalog generation process, we flagged problematic images (e.g., CCDs suffering from reflected light, imaging artifacts, point-spread function (PSF) mis-estimation, etc.) and excluded the affected objects from subsequent analyses. The resulting photometric catalogs were ingested into a high-performance relational database system. Throughout the rest of this paper, photometric fluxes and magnitudes refer to SExtractor output for the PSF model fit.

Photometric Calibration: Photometric calibration was performed using the stellar locus regression technique (SLR: Ivezić et al. 2004; MacDonald et al. 2004; High et al. 2009; Gilbank et al. 2011; Coupon et al. 2012; Desai et al. 2012; Kelly et al. 2014). Our reference stellar locus was empirically derived from the globally calibrated DES Y1A1 stellar objects in the region of the Y1A1 footprint with the smallest E(B − V) value from the Schlegel et al. (SFD; 1998) interstellar extinction map. We performed a 1'' match on all Y1 and Y2 objects with S/N > 10 observed in r-band and at least one other band. We then applied a high-purity stellar selection based on the weighted average of the spread_model quantity for the matched objects ($| \mathrm{wavg}\_\mathrm{spread}\_\mathrm{model}\_r| \lt 0.003$; see below). The average zero point measured in Y1A1, ZP_grizY = {30.0, 30.3, 30.2, 29.9, 28.0}, was assigned to each star as an initial estimate. Starting from this coarse calibration, we began an iterative procedure to fix the color uniformity across the survey footprint. We segmented the sky into equal-area pixels using the HEALPix scheme (Górski et al. 2005). For each ∼0.2 deg² (resolution nside = 128) HEALPix pixel, we chose the DES exposure in each band with the largest coverage and ran a modified version of the Big MACS SLR code (Kelly et al. 2014)⁴¹ to calibrate each star from the reference exposure with respect to the empirical stellar locus. These stars became our initial calibration standards. We then adjusted the zero points of other CCDs so that the magnitudes of the matched detections agreed with the calibration set from the reference exposure. CCDs with fewer than 10 matched stars or with a large dispersion in the magnitude offsets of matched stars (σ_ZP > 0.1 $\mathrm{mag}$) were flagged. For each calibration star, we computed the weighted-average magnitude in each band using these new CCD zero points; this weighted-average magnitude was used as the calibration standard for the next iteration of the SLR. In the first iteration, we assigned SLR zero points to the calibration stars based on the HEALPix pixel within which they reside. In subsequent iterations, we assigned SLR zero points to the calibration stars based on a bi-linear interpolation of their positions onto the HEALPix grid of SLR zero points. After the second iteration, the color zero points were stable at the 1–2 $\mathrm{mmag}$ level. The absolute calibration was set against the 2MASS J-band magnitude of matched stars (making use of the stellar locus in color-space), which were de-reddened using the SFD map with a reddening law of A_J = 0.709 × E(B − V)_SFD from Schlafly & Finkbeiner (2011). The resulting calibrated DES magnitudes are thus already corrected for Galactic reddening by the SLR calibration.

Unique Catalog Generation: There is significant imaging overlap within the DES footprint. For our final high-level object catalog, we identified unique objects by performing a 05 spatial match across all five bands. For each unique object, we calculated the magnitude in each filter in two ways: (1) taking the photometry from the single-epoch detection in the exposure with the largest effective depth,⁴² and (2) calculating the weighted average (wavg) magnitude from multiple matched detections. During the catalog generation stage, we kept all objects detected in any two of the five filters (this selection is later restricted to the g- and r-band for our dwarf galaxy search). The process of combining catalog-level photometry increases the photometric precision, but does not increase the detection depth.

The resulting Y2Q1 catalog covers ∼5100 deg² in any single band, ∼5000 deg² in both the g- and r-bands, and ∼4700 deg² in all five bands. The coverage of the Y2Q1 data set is shown in Figure 1. The Y2Q1 catalog possesses a median relative astrometric precision of ∼40 $\mathrm{mas}$ per coordinate and a median absolute astrometric uncertainty of ∼140 $\mathrm{mas}$ per coordinate when compared against UCAC-4. The median absolute photometric calibration agrees within ∼3% with the de-reddened global calibration solution of Y1A1 in the overlap region in the griz bands. When variations in the reddening law are allowed, this agreement improves to ≲1%. The color uniformity of the catalog is ∼1% over the survey footprint (tested independently with the red sequence of galaxies). The Y2Q1 catalog has a median point-source depth at an S/N = 10 of g = 23.4, r = 23.2, i = 22.4, z = 22.1, and Y = 20.7.

We selected stars observed in both the g- and r-bands from the Y2Q1 unique object catalog following the common DES prescription based on the wavg_spread_model and spreaderr_model morphological quantities (e.g., Chang et al. 2015; Koposov et al. 2015a). Specifically, our stellar sample consists of well-measured objects with $| \mathrm{wavg}\_\mathrm{spread}\_\mathrm{model}\_r|$ $\lt \;0.003+\mathrm{spreaderr}\_\mathrm{model}\_r$, flags_{g, r} < 4, and magerr_psf_{g, r} < 1. By incorporating the uncertainty on spread_model, it is possible to maintain high stellar completeness at faint magnitudes while achieving high stellar purity at bright magnitudes. We sacrificed some performance in our stellar selection by using the r-band spread_model, which has a slightly worse PSF than the i-band; however, relaxing the requirement on i-band coverage increased the accessible survey area.

The stellar completeness is the product of the object detection efficiency and the stellar classification efficiency. As described in Section 2.1, different source detection thresholds were applied to the individual Y1 and Y2 exposures, and accordingly, the detection efficiency for faint objects in Y2Q1 varies across the survey footprint. We used a region of the DES Science Verification data set (SVA1), imaged to full survey depth (i.e., ∼10 overlapping exposures in each of grizY), to estimate the object detection completeness using the individual Y1 exposures. For the Y2 exposures, we instead compared to publicly available catalogs from the Canada–France–Hawaii Telescope Lensing Survey (CFHTLenS; Erben et al. 2013). In both cases, the imaging for the reference catalog is ∼2 $\mathrm{mag}$ deeper than the Y2Q1 catalog. These comparisons show that the Y2Q1 object catalog is >90% complete at about 0.5 $\mathrm{mag}$ brighter than the S/N = 10 detection limit.

We estimated the stellar classification efficiency by comparing to CFHTLenS and by using statistical estimates from the Y2Q1 data alone. For the CFHTLenS matching, we identified probable stars in the CFHTLenS catalog using a combination of the class_star output from SExtractor and the fitclass output from the lensfit code (Heymans et al. 2012). Specifically, our CFHTLenS stellar selection was (fitclass = 1) OR (class_star > 0.98), and our galaxy selection was (fitclass = 0) OR (class_star < 0.2). Comparing against the DES stellar classification based on spread_model, we found that the Y2Q1 stellar classification efficiency exceeds 90% for r < 23 $\mathrm{mag}$, and falls to ∼80% by r = 24 (Figure 2). We independently estimated the stellar classification efficiency by creating a test sample of Y2Q1 objects with high stellar purity using the color-based selection r − i > 1.7. We then applied the spread_model morphological stellar selection to evaluate the stellar completeness for the test sample. The stellar selection efficiency determined by the color-based selection test is in good agreement with that from the CFHTLenS comparison.

Zoom In Zoom Out Reset image size

For a typical field at high Galactic latitude, the stellar purity is ∼50% at r = 22.5 $\mathrm{mag}$, and falls to ∼10% at the faint magnitude limit (Figure 2). In other words, the majority of faint objects in the "stellar sample" are misclassified galaxies rather than foreground halo stars.⁴³ This transition can be seen in the color–magnitude distribution diagnostic plots for individual candidates (Section 6.1).

Our most straightforward and model-independent search technique used a simple isochrone filter to facilitate visual inspection of the stellar density field. The specific algorithm was a variant of the general approach described in Section 3.1 of Bechtol et al. (2015) and follows from the methods of Koposov et al. (2008), Walsh et al. (2009), and Kim & Jerjen (2015a).

We first increased the contrast of putative satellite galaxies by selecting stars that were consistent with the isochrone of an old ($\tau$ = 12 $\mathrm{Gyr}$) and metal-poor ($Z$ = 0.0001) stellar population (Dotter et al. 2008). Specifically, we selected stars within 0.1 $\mathrm{mag}$ of the isochrone locus in color–magnitude space, enlarged by the photometric measurement uncertainty of each star added in quadrature. After isochrone filtering, we smoothed the stellar density field with a 2' Gaussian kernel. For each density peak in the smoothed map, we scanned over a range of radii (1'–18') and computed the Poisson significance of finding the observed number of stars within each circle centered on the density peak given the local field density. This process was repeated for stellar populations at a range of distance moduli (16 $\mathrm{mag}$ ≤ $m\;-\;M$ ≤ 24 $\mathrm{mag}$) and diagnostic plots were automatically generated for each unique density peak. Our nominal magnitude threshold for this search was g < 24 $\mathrm{mag}$; however, in regions on the periphery of the survey with large variations in coverage we used a brighter threshold of g < 23 $\mathrm{mag}$.

Our second search strategy utilized a matched-filter algorithm (Balbinot et al. 2011, 2013; Luque et al. 2015). We began by binning objects in our stellar sample with 17 $\mathrm{mag}$ < g < 24 $\mathrm{mag}$ into 1' × 1' spatial pixels and g versus g−r color–magnitude bins of 0.01 $\mathrm{mag}$ × 0.05 $\mathrm{mag}$. We then created a grid of possible simple stellar populations modeled with the isochrones of Bressan et al. (2012) and populated according to a Chabrier (2001) initial mass function to use as signal templates. The model grid spanned a range of distances (10 kpc < D < 200 kpc), ages (9 < log₁₀ $\tau$ < 10.2), and metallicities (Z = {0.0002, 0.001, 0.007}). The color–magnitude distribution of the field population was empirically determined over larger 10° × 10° regions. For each isochrone in the grid, we then fit the normalization of the simple stellar population in each spatial pixel to create a corresponding density map for the full Y2Q1 area.

The set of pixelized density maps do not assume a spatial model for the stellar system, and can be used to search for a variety of stellar substructures (e.g., streams and diffuse clouds). To search specifically for dwarf galaxies, we convolved the density maps with Gaussian kernels having widths ranging from 0' (no convolution) to 9' and used SExtractor to identify compact stellar structures in those maps. This convolution and search was performed on each map for our grid of isochrones. The resulting seeds were ranked according to their statistical significance and by the number of maps in which they appeared, i.e., the number of isochrones for which an excess was observed.

Our third search technique was a maximum-likelihood-based algorithm that simultaneously used the full spatial and color–magnitude distribution of stars, as well as photometric uncertainties and information about the survey coverage. The likelihood was constructed from the product of Poisson probabilities for detecting each individual star given its measured properties and the parameters that describe the field population and putative satellite (Section 3.2 of Bechtol et al. 2015). When searching for new stellar systems, we used a spherically symmetric Plummer model as the spatial kernel, and a composite isochrone model consisting of four isochrones of different ages, $\tau$ = {12 $\mathrm{Gyr}$, 13.5 $\mathrm{Gyr}$}, and metallicities, $Z$ = {0.0001, 0.0002}, to bracket a range of possible stellar populations (Bressan et al. 2012). We scanned the Y2Q1 data testing for the presence of a satellite galaxy at each location on a multi-dimensional grid of sky position (07 resolution; HEALPix nside = 4096), distance modulus (16 mag < $m\;-\;M$ < 23 $\mathrm{mag}$), and kernel half-light radius (r_h = {003, 01, 03}). For the likelihood search, we restricted our magnitude range to 16 $\mathrm{mag}$ < g < 23.5 $\mathrm{mag}$ as a compromise between survey depth and stellar completeness. The statistical significance of a putative galaxy at each grid point was expressed as a test statistic ($\mathrm{TS}$) based on the likelihood ratio between a hypothesis that includes a satellite galaxy versus a field-only hypothesis (Equation (4) of Bechtol et al. 2015). As a part of the model fitting, the satellite membership probability is computed for every star in the region of interest. These photometric membership probabilities can be used to visualize which stars contribute to the candidate detection significance (Section 5) and have previously been used to select targets for spectroscopic follow-up (Simon et al. 2015; Walker et al. 2015).

• 1.  
  DES J2204–4626 (Grus II, Figure 6): As the most significant new candidate, DES J2204–4626 has ∼161 probable member stars with g < 23 $\mathrm{mag}$ in the DES imaging. The large physical size of this system (93 $\mathrm{pc}$) allows it to be tentatively classified as an ultra-faint dwarf galaxy. A clear red giant branch (RGB) and MSTO are seen in the color–magnitude diagram of DES J2204–4626. Several likely blue horizontal branch (BHB) members are seen at g ∼ 19 $\mathrm{mag}$. There is a small gap in survey coverage ∼05 from the centroid of DES J2204–4626, but this is accounted for in the maximum-likelihood analysis and should not significantly affect the characterization of this rich satellite.DES J2204–4626 is from the LMC (46 kpc) and from the Galactic center (49 kpc). It is slightly closer to the SMC (33 kpc) and intriguingly close to the the group of stellar systems in Tucana (∼18 kpc). While it is unlikely that DES J2204–4626 is currently interacting with any of the other known satellites, we cannot preclude the possibility of a past association.
• 2.  
  DES J2356–5935 (Tucana III, Figure 7): DES J2356– 5935 is another highly significant stellar system that presents a clearly defined main sequence and several RGB stars. The physical size and luminosity of DES J2356–5935 (r_1/2 = 44 $\mathrm{pc}$, M_V = −2.4 $\mathrm{mag}$) are comparable to that of the recently discovered dwarf galaxy Reticulum II. At a distance of 25 kpc, DES J2356–5935 would be among the nearest ultra-faint satellite galaxies known, along with Segue 1 (23 kpc), Reticulum II (32 kpc), and Ursa Major II (32 kpc). The relative abundance of bright stars in DES J2356–5935 should allow for an accurate spectroscopic determination of both its velocity dispersion and metallicity. DES J2356–5935 is reasonably close to the LMC and SMC (32 kpc and 38 kpc, respectively), and measurements of its systemic velocity will help to elucidate a physical association.As mentioned in Section 5, there is an additional linear feature in the stellar density field around DES J2356–5935 that strongly influences the fitted ellipticity of this galaxy candidate. In Figure 8, we show the 6 ° × 6° region centered on DES J2356–5935, which contains this linear feature. Selecting stars with the same isochrone filter used to increase the contrast of DES J2356–5935 relative to the field population, we found a faint stellar overdensity extending ∼2° on both sides of DES J2356–5935. This feature has an FWHM of ∼03 (i.e., projected dimensions of 1.7 kpc × 130 $\mathrm{pc}$ at a distance of 25 kpc) and is oriented ∼85° east of north. The length of the feature and the smooth density field observed with an inverted isochrone filter support the conclusion that this is a genuine stellar structure associated with DES J2356–5935.One interpretation is that DES J2356–5935 is in the process of being tidally stripped by the gravitational potential of the Milky Way. The relatively short projected length of the putative tails may imply that DES J2356–5935 is far from its orbital pericenter (where tidal effects are strongest) and/or that stripping began recently. Despite the possible presence of tidal tails, the main body of DES J2356–5935 appears to be relatively round. While the proximity of DES J2356–5935 might make it an important object for indirect dark matter searches, evidence of tidal stripping would suggest that a large fraction of its outer dark matter halo has been removed. However, the mass within the stellar core (and therefore the dark matter content in the central region) can still likely be determined accurately (Oh et al. 1995; Muñoz et al. 2008; Peñarrubia et al. 2008).
• 3.  
  DES J0531–2801 (Columba I, Figure 9): DES J0531– 2801 is the second most distant of the Y2 DES candidates (182 kpc) and is detected as a compact cluster of RGB stars. Both the BHB and red horizontal branch (RHB) appear to be well-populated. The distance, physical size (103 $\mathrm{pc}$), and luminosity (M_V = −4.5 $\mathrm{mag}$) of DES J0531–2801 place it in the locus of Local Group dwarf galaxies, comparable to Leo IV and CVn II (Belokurov et al. 2007). DES J0531–2801 is isolated from the other new DES systems and is likely not associated with the Magellanic system.
• 4.  
  DES J0002–6051 (Tucana IV, Figure 10): DES J0002– 6051 has the largest angular size of the candidates (r_h = 91), which at a distance of 48 kpc corresponds to a physical size of r_1/2 ∼ 127 $\mathrm{pc}$. This large half-light radius is inconsistent with the sizes of known globular clusters (Harris 1996, 2010 edition), thus making it very likely that DES J0002–6051 is a dwarf galaxy. The measured ellipticity of DES J0002–6051, = 0.4, is also consistent with a galactic classification. DES J0002–6051 is found to be 27 kpc from the LMC and 18 kpc from the SMC. DES J0002–6051 is one of the proposed members of the Tucana group, with a centroid separation of 7 kpc. Measurements of the radial velocity and proper motion of DES J0002–6051 will provide strong clues as to whether it was accreted as part of a system of satellites. Similar to DES J2356–5935 and DES J2204–4626, the MSTO of DES J0002–6051 is well-populated and clearly visible. Several possible member stars can also be seen along the HB.
• 5.  
  DES J0345–6026 (Reticulum III, Figure 11): DES J0345– 6026 appears to be similar to DES J2204–4626 in its structural properties, but is more distant (heliocentric distance of 92 kpc). Like DES J2204–4626, DES J0345– 6026 can be tentatively classified as a dwarf galaxy based on its physical size and low surface brightness. There is some indication of asphericity for this object; however, the ellipticity is not significantly constrained by the DES data. DES J0345–6026 has a sparsely populated RGB, a few possible RHB members, and two possible BHB members. The MSTO for DES J0345–6026 is slightly fainter than our fitting threshold of g < 23 $\mathrm{mag}$, thus its age is poorly constrained.
• 6.  
  DES J2337–6316 (Tucana V, Figure 12): DES J2337– 6316 is the second new system in the Tucana group and is the closest to the group's centroid (3 kpc). At a heliocentric distance of 55 kpc, DES J2337–6316 is also located 29 kpc from the LMC and 14 kpc from the SMC. The physical size (17 $\mathrm{pc}$) and luminosity (M_V ∼ −1.6 $\mathrm{mag}$) place DES J2337–6316 in a region of the size–luminosity plane close to Segue 1, Willman 1, and Kim 2. Segue 1 and Willman 1 have sizes similar to the most extended globular clusters, but are approximately an order of magnitude less luminous. On the other hand, deep imaging of Kim 2 has led Kim et al. (2015b) to conclude that it is very likely a star cluster. DES J2337–6316 is one of the few new objects for which the DES data places a constraint on ellipticity. The high ellipticity, = 0.7, supports a galactic classification for DES J2337–6316. Most of the high probability member stars for DES J2337–6316 are on the main sequence and fainter than g ∼ 22.5 $\mathrm{mag}$. As with other similarly faint objects, a few stars situated close to the lower RGB are also likely members. A system similar to DES J2337–6316 would have been difficult to detect in SDSS with a threshold at r < 22.
• 7.  
  DES J2038–4609 (Indus II, Figure 13): DES J2038–4609 is the first of the two lower-confidence candidates. The detection of this object comes predominantly from a tight clump of BHB stars at g ∼ 22 $\mathrm{mag}$ and g−r ∼ 0 $\mathrm{mag}$. Three of the potential HB members are clustered within a spatial region of radius ∼10''. Several additional HB and RGB stars are assigned non-zero membership probabilities by the likelihood fit, enlarging the best-fit size of this system. DES J2038–4609 is located at a boundary between the Y1 and Y2 imaging, and the effect of the deeper Y2 source detection threshold is visible when extending to magnitudes of g > 23 $\mathrm{mag}$. The best-fit physical size of DES J2038–4609 (181 $\mathrm{pc}$) would place it among the population of Local Group galaxies. Deeper imaging is needed to better characterize this candidate since the MSTO is fainter than the current DES detection threshold.
• 8.  
  DES J0117–1725 (Cetus II, Figure 14): DES J0117–1725 is the second lower-confidence candidate, being the least luminous (M_V = 0.0 ± 0.68 $\mathrm{mag}$) and most compact (r_1/2 = 17 $\mathrm{pc}$) of the new stellar systems. DES J0117–1725 is nearly two orders of magnitude fainter than globular clusters with comparable half-light radii. At a heliocentric distance of 30 kpc, DES J0117–1725 is the second nearest of the new systems. DES J0117–1725 is detected predominantly by its main sequence stars and has one potential HB member. The lack of strong features in the color–magnitude distribution of stars associated with DES J0117–1725 leads to a large degeneracy between the distance, age, and metallicity of this system. If determined to be a dwarf galaxy, DES J0117–1725 would be the least luminous known galaxy; however, the current classification of this object is ambiguous.There are several gaps in the Y2Q1 coverage ∼01 from DES J0117–1725; however, DES J0117–1725 is sufficiently compact that few member stars are expected in the region of missing coverage. These gaps are incorporated into the likelihood analysis and should have a minimal impact on the fit. The region around DES J0117–1725 was imaged in only a single r-band exposure; however, there are two to three overlapping exposures in each of the g- and i-bands. We have verified that the stellar overdensity is also apparent in the g−i filter combination. Nonetheless, the properties of DES J0117–1725 should be interpreted with caution until additional imaging is acquired.

Many studies of the Milky Way satellite galaxy population require knowledge of the detection efficiency for satellites as a function of luminosity, heliocentric distance, physical size, and sky position in order to account for observational selection effects. To quantify our search sensitivity with the Y2Q1 data set, we follow the approach that Bechtol et al. (2015) applied to the DES Y1A1 coadd data. Briefly, we generated many realizations of satellite galaxies with different structural properties that are spread uniformly throughout the survey footprint, excluding regions of high stellar density near to the LMC. We then applied the map-based search algorithm to estimate the likelihood of a high-confidence detection. In the present implementation, the detection significance was evaluated as the Poisson probability of finding the "observed" number of stars within a circular region matched to the half-light radius of the simulated satellite given the effective local field density after the isochrone selection (see Section 3.1). Realizations that yield at least 10 detectable stars and produce a >5.50σ overdensity within the half-light radius are considered "detected."

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

An essential input to the completeness calculation is the magnitude threshold for individual stars. For the searches described in Section 3, we used deeper magnitude thresholds (e.g., g < 24 $\mathrm{mag}$) in order to maximize the discovery potential. A more conservative threshold, g < 23 $\mathrm{mag}$, is selected to estimate the completeness in detecting new stellar systems based on the following considerations. First, the Y2Q1 stellar sample is expected to be nearly complete for both Y1 and Y2 exposures at magnitudes g ≲ 23 $\mathrm{mag}$. Figure 3 shows that the measured stellar density in this magnitude range varies smoothly over the Y2Q1 footprint. At fainter magnitudes, discontinuities in the field density can occur in regions where exposures from Y1 and Y2 overlap (typically, more faint objects are detected in the Y2 exposures). Second, the field population with g ≳ 23 $\mathrm{mag}$ in high Galactic latitude regions is likely dominated by misclassified galaxies rather than halo stars, and therefore the assumption of isotropy on arcminute angular scales is compromised due to galaxy clustering. The adopted magnitude threshold matches that used in Bechtol et al. (2015), and in this case, the sensitivity estimates obtained in the prior work can be approximately scaled to the enlarged effective solid angle of the Y2Q1 data set, ∼5000 deg².

In Table 4, we list the expected detection efficiencies for the DES ultra-faint galaxy candidates using magnitude thresholds of g < 23 $\mathrm{mag}$ and g < 22 $\mathrm{mag}$, which roughly correspond to the stellar completeness limits of Y2Q1 and SDSS, respectively. For the brighter threshold, several of the DES candidates have substantially reduced detection probabilities.

Table 4.  Expected Detection Efficiencies for Ultra-faint Galaxy Candidates

Name	MV	Distance	rh	Efficiency	Efficiency
		(kpc)	(arcmin)	(g < 23 $\mathrm{mag}$)	(g < 22 $\mathrm{mag}$)
DES J2204–4626 (Gru II)	−3.9	53	6.0	1.00	0.39
DES J2356–5935 (Tuc III)	−2.4	25	6.0	1.00	0.94
DES J0531–2801 (Col I)	−4.5	182	1.9	0.95	0.86
DES J0002–6051 (Tuc IV)	−3.5	48	9.1	0.98	0.05
DES J0345–6026 (Ret III)	−3.3	92	2.4	0.33	0.05
DES J2337–6316 (Tuc V)	−1.6	55	1.0	0.94	0.01
DES J2038–4609 (Ind II)	−4.3	214	2.9	0.26	0.00
DES J0117–1725 (Cet II)	0.0	30	1.9	0.38	0.01
Ret II	−3.6	32	6.0	1.00	1.00
DES J0344.3–4331 (Eri II)	−7.4	330	1.8	1.00	1.00
DES J2251.2–5836 (Tuc II)	−3.9	58	7.2	1.00	0.07
Hor I	−3.5	87	2.4	0.66	0.14
DES J0443.8–5017 (Pic I)	−3.7	126	1.2	0.85	0.47
DES J0222.7–5217 (Eri III)	−2.4	95	0.4	0.41	0.04
DES J2339.9–5424 (Phe II)	−3.7	95	1.2	0.99	0.77
Gru I	−3.4	120	2.0	0.21	0.04
Hor II	−2.6	78	2.1	0.26	0.01

Note. Expected detection efficiencies are provided for two magnitude thresholds roughly corresponding to the stellar completeness depths of DES Y2Q1 (g < 23 $\mathrm{mag}$) and SDSS (g < 22 $\mathrm{mag}$). The detection efficiencies are estimated from multiple realizations of each candidate within the DES footprint using the best-fit luminosity M_V, heliocentric distance, and azimuthally averaged half-light radius r_h, which were then analyzed with the map-based detection algorithm described in Section 3.1. The average detection probability ratio between the two threshold choices is ∼50%, implying that roughly half of the DES candidates would have been detected if they were located in the SDSS footprint.

Download table as:  ASCIITypeset image

We note that all of the faint stellar systems that were discovered in the Y1A1 data set are also significantly detected using the Y2Q1 data set, suggesting that our present search has sensitivity comparable to previous studies.

The detection of 17 candidate ultra-faint galaxies in the first two years of DES data is consistent with the range of predictions based on the standard cosmological model (Tollerud et al. 2008; Hargis et al. 2014). These predictions model the spatial distribution of luminous satellites from the locations of dark matter subhalos in cosmological N-body simulations of Local Group analogs. Significant anisotropy in the distribution of satellites is expected (Tollerud et al. 2008; Hargis et al. 2014) and likely depends on the specific accretion history of the Milky Way (Deason et al. 2015).

The proximity of the DES footprint to the Magellanic Clouds is noteworthy when considering an anisotropic distribution of satellites. The possibility that some Milky Way satellites are or were associated with the Magellanic system has been discussed for some time (Lynden-Bell 1976; D'Onghia & Lake 2008; Nichols et al. 2011; Sales et al. 2011), and the recent DES results have renewed interest in this topic (Deason et al. 2015; Koposov et al. 2015a; Wheeler et al. 2015; Yozin & Bekki 2015). The existence of structures on mass scales ranging from galaxy clusters to ultra-faint satellite galaxies is a generic prediction of hierarchical galaxy formation in the cold dark matter paradigm. However, specific predictions for the number of observable satellites of the LMC and SMC depend on the still unknown efficiency of galaxy formation in low-mass subhalos. Identifying satellites of satellites within the Local Group would support the standard cosmological model, and may provide a new opportunity to test the impact of environment on the formation of the least luminous galaxies (Wetzel et al. 2015).

Based on a study of Milky Way–LMC analogs in N-body simulations, Deason et al. (2015) propose a method to evaluate the probability that a given satellite was at one time associated with the LMC based upon the present three-dimensional separation between the two (Table 3). This prescription predicts that two to four of the Y2 DES candidates were associated with the LMC before infall, in addition to the two to four potentially associated satellites found in Y1 DES data. Deason et al. further suggest that a grouping of satellites near the LMC in both physical separation and velocity would imply that the Magellanic system was recently accreted onto the Milky Way since members of this group would disperse within a few Gyr.

As a population, the new DES satellite galaxy candidates do appear to be unevenly distributed within the survey footprint; 15 are located at δ₂₀₀₀ < −40°, close to the Magellanic Clouds (Figure 3). A Kolmogorov–Smirnov test, shown in Figure 15, significantly rejects the hypothesis of a uniform distribution (p = 4.3 × 10⁻⁴). One possible explanation for non-uniformity is that some of the new systems are satellites of the LMC or SMC. Below we introduce a simple model to test this hypothesis.

Zoom In Zoom Out Reset image size

We began by modeling the probability distribution for detecting a satellite galaxy as a function of sky position. In this analysis, we excluded classical Milky Way satellites and considered only satellites that have similar characteristics to those discovered by DES (i.e., are detectable at the DES Y2Q1 depth). The underlying luminosity function and the completeness as a function of luminosity, heliocentric distance, and physical size were all simplified into a two-dimensional probability distribution.

We modeled this two-dimensional probability distribution with three components: (i) an isotropic distribution of satellites associated with the main halo of the Milky Way, (ii) a population of satellites spatially associated with the LMC halo, and (iii) a population of satellites spatially associated with the SMC halo. The first component was uniform on the sky, and the latter two components were spatially concentrated around the Magellanic Clouds. We assumed spherically symmetric three-dimensional distributions of satellites centered around each of the Magellanic Clouds having power-law radial profiles with slope α = −1 and truncation radius r_t = 35 kpc. These distributions were then projected onto the sky to create two-dimensional predicted density maps.

Several underlying assumptions went into this simple model. First, we assumed that the Magellanic components were concentrated enough that variations in the completeness with respect to heliocentric distance would not substantially alter the projected density maps. Second, we assumed that the truncation radii of the LMC and SMC are the same, but later test the sensitivity of our results to this choice. Third, we assumed that smaller satellites of the Milky Way do not have their own associated satellites. We note that a power-law radial profile with slope α = −1 can be viewed as a very extended Navarro–Frenk–White profile (Navarro et al. 1997). Since the DES footprint does not cover the central regions of the Magellanic Clouds, this shallow profile yields a conservative estimate on the fraction of satellites associated with the Magellanic system.

We performed an unbinned maximum-likelihood fit of our three-component model to the distribution of observed satellites. The parameters of the likelihood function are the total number of satellites and the relative normalizations of the three components. The likelihood is defined as the product of the probabilities to detect each observed satellite given its sky position. When presenting our results, we marginalized over the ratio of the LMC component to the SMC component. This ratio is not strongly constrained by our analysis; however, there is a slight preference for a larger SMC component.

The left panel of Figure 16 shows the constraints using the sky positions of the seventeen DES satellite galaxy candidates. We find that the DES data alone reject a uniform distribution of satellites at >3σ confidence when compared to our best-fit model. This again implies that there is either a clear overdensity of satellites around the Magellanic Clouds or that the initial assumption of isotropy around the main halo is incorrect. In Figure 15, we show the distribution of declinations for our best-fit three-component model and find good agreement with the observations (p = 0.7).

Zoom In Zoom Out Reset image size

We next tested the sensitivity of our result to the assumed distribution of satellites around the LMC and SMC. The central panel of Figure 16 shows the 1σ contours for three different values of r_t = {25, 35, 45 kpc}, and two values of the slope, α = {−1, −2}. Despite slight changes in the likelihood contours, in each case >10% of Milky Way satellites are likely to be spatially associated with the Magellanic Clouds.

We further examined the satellite population by simultaneously considering the 16 ultra-faint satellites observed in the SDSS DR10 footprint.⁴⁹ To fully combine DES and SDSS observations we would need to model the detection efficiency of SDSS relative to DES as a function of satellite distance, size, and luminosity. However, since the SDSS DR10 footprint only overlaps with the isotropic component in our model, we can again fold the complications of detection efficiency into a constant detection ratio. Our fiducial calculation assumes that 50% of the satellites discovered in DES would be detectable by SDSS, which agrees with the relative detection efficiency of the 17 DES satellites at the depth of SDSS (Table 4). The right panel of Figure 16 shows that the DES+SDSS constraints are in agreement with those from DES data alone, though the fraction of LMC and SMC satellites is more tightly constrained by including information from a large region widely separated from the Magellanic Clouds. The combined results imply that there are ∼100 satellites over the full sky that are detectable at DES Y2Q1 depth, and that 20%–30% of these might be associated with the Magellanic Clouds. Importantly, this prediction does not attempt to model the full population of Milky Way satellites beyond the DES Y2Q1 sensitivity and ignores the diminished detection efficiency close to the Galactic plane.

The Pan-STARRS team has recently identified three candidate Milky Way satellite galaxies in their 3π survey (Laevens et al. 2015a, 2015b). If this search is complete to the depth of SDSS over its full area, it is notable that so few candidates have been found. This observation alone could imply anisotropy without requiring the influence of the Magellanic Clouds. However, much of the Pan-STARRS area is located at low Galactic latitudes where the elevated foreground stellar density and interstellar extinction may present additional challenges. Under the assumption that Pan-STARRS covers the full sky with ${\delta }_{2000}$ > −30°, ∼2000 deg² overlap with DES Y2Q1. This area of the DES footprint includes two new candidates, one of which has a large enough surface brightness to likely have been detected at SDSS depth (DES J0531–2801, see Table 4). These two candidates are located in a region of the sky that would be observed at a relatively high airmass by Pan-STARRS and may suffer from decreased detection efficiency.

From this analysis, we conclude that the distribution of satellites around the Milky Way is unlikely to be isotropic, and that a plausible component of this anisotropy is a population of satellites associated with the Magellanic Clouds. However, several alternative explanations for anisotropy in the Milky Way satellite distribution exist. For example, Milky Way satellites could be preferentially located along a three-dimensional planar structure, as has been suggested by many authors, starting with Lynden-Bell (1976). This proposed planar structure encompasses the Magellanic Clouds and many of the classical and SDSS satellites. Pawlowski et al. (2015) suggest that the satellites discovered in Y1 DES data are also well aligned with this polar structure. We note that the Y2 discoveries presented here include several objects near the SMC and may reduce the fraction of objects in close proximity to the proposed plane. An additional possibility is that the satellites are associated with the orbit of the Magellanic System and are not isotropically distributed around the Magellanic Clouds themselves (Yozin & Bekki 2015). The DES footprint covers only a fraction of the region surrounding the Magellanic Clouds and additional sky coverage may yield more satellites with similar proximity to the Magellanic system and/or help to distinguish between these various scenarios. Measurements of the relative motions of the satellites and further theoretical work will also help clarify the physical relationships between these stellar systems.