FORTY-SEVEN MILKY WAY-SIZED, EXTREMELY DIFFUSE GALAXIES IN THE COMA CLUSTER

While there have been tremendous advances in deep, high-resolution imaging surveys over the past decades (e.g., Scoville et al. 2007; Heymans et al. 2012), the low surface brightness sky remains relatively unexplored. The Dragonfly Telephoto Array (Abraham & van Dokkum 2014) was developed with the specific aim of detecting low surface brightness emission. It is comprised of eight Canon 400 mm f/2.8 II telephoto lenses which all image the same part of the sky, forming what is effectively a 40 cm f/1.0 refractor. Four of the lenses are equipped with a Sloan Digital Sky Survey (SDSS) g filter and four with an SDSS r filter. The lenses are attached to cameras that provide an instantaneous field of view of 26 × 19, sampled with 28 pixels.

The main science program of Dragonfly is a deep imaging survey of a sample of nearby galaxies (see van Dokkum et al. 2014; Merritt et al. 2014). In the late spring of 2014 we interrupted this survey to observe the Coma cluster. The main goal of the Coma observation is to accurately measure the luminosity and color of the intra-cluster light (ICL). We are also looking for streams and tidal features, inspired by the beautiful deep imaging of the Virgo cluster of Mihos et al. (2005).

A visual inspection of the reduced images revealed a large number of faint, spatially resolved objects. The nature of these objects was not immediately obvious, as they are not listed in existing catalogs of faint galaxies in the Coma cluster (e.g., Ulmer et al. 1996; Adami et al. 2006). Furthermore, they seemed to be too large to be part of the cluster: typical dwarf galaxies have effective radii of a few hundred parsecs, which corresponds to much less than a Dragonfly pixel at the distance of Coma (D_A = 98 Mpc; D_L = 103 Mpc).⁴

Expecting that the objects would turn out to be isolated dwarf galaxies in the foreground of the cluster, we decided to perform a (mostly) objective selection with the aid of SDSS and archival Canada–France–Hawaii Telescope (CFHT) data, as described in the next section. Surprisingly, as we show below, the objects turn out to be associated with the Coma cluster after all, and represent a class of very large, very diffuse galaxies. Only a handful of similar objects were known from previous studies (Impey et al. 1988; Bothun et al. 1991; Dalcanton et al. 1997).

2.1. Candidates in the Dragonfly Data

2.2. Rejection Using SDSS and CFHT

2.3. A Population of Large, Diffuse Galaxies

We are left with 47 objects, listed in Table 1, that are clearly detected in the Dragonfly imaging, are spatially extended, are not detected in the SDSS, and do not resolve into multiple objects in the higher resolution CFHT data. Four typical examples spanning a range of apparent brightness are shown in Figure 1. The galaxies are clearly detected but barely resolved in the Dragonfly data, and very faint, fuzzy blobs in the CFHT data.

Zoom In Zoom Out Reset image size

Table 1. Positions and Properties

Id	R.A.	Decl.	μ(g, 0)	reff	Mg	b/a
	(J2000)	(J2000)	(mag arcsec−2)	(kpc)	(mag)
DF1	12h59m141	29°07'16''	$25.1_{-0.5}^{+0.5}$	$3.1_{-0.6}^{+0.9}$	$-14.6_{-0.2}^{+0.3}$	0.82 ± 0.03
DF2	12h59m095	29°00'25''	$24.4_{-0.6}^{+0.6}$	$2.1_{-0.4}^{+0.6}$	$-14.3_{-0.2}^{+0.2}$	0.71 ± 0.03
DF3	13h02m165	28°57'17''	$24.5_{-0.5}^{+0.5}$	$2.9_{-0.7}^{+0.8}$	$-14.2_{-0.2}^{+0.3}$	0.40 ± 0.04
DF4	13h02m334	28°34'51''	$25.7_{-0.6}^{+0.6}$	$3.9_{-0.5}^{+1.0}$	$-14.3_{-0.2}^{+0.2}$	0.71 ± 0.03
DF5	12h55m105	28°33'32''	$24.9_{-0.6}^{+0.6}$	$1.8_{-0.4}^{+0.4}$	$-13.5_{-0.2}^{+0.2}$	0.71 ± 0.03
DF6	12h56m297	28°26'40''	$25.5_{-0.5}^{+0.5}$	$4.4_{-1.1}^{+1.6}$	$-14.3_{-0.3}^{+0.4}$	0.47 ± 0.03
DF7	12h57m017	28°23'25''	$24.4_{-0.5}^{+0.5}$	$4.3_{-0.8}^{+1.4}$	$-16.0_{-0.2}^{+0.2}$	0.76 ± 0.03
DF8	13h01m304	28°22'28''	$25.4_{-0.5}^{+0.5}$	$4.4_{-0.9}^{+1.5}$	$-14.9_{-0.3}^{+0.3}$	0.73 ± 0.05
DF9	12h56m228	28°19'53''	$25.6_{-0.7}^{+0.7}$	$2.8_{-0.4}^{+0.5}$	$-14.0_{-0.1}^{+0.1}$	0.92 ± 0.03
DF10	12h59m163	28°17'51''	$24.4_{-0.6}^{+0.6}$	$2.4_{-0.4}^{+0.6}$	$-14.7_{-0.2}^{+0.2}$	0.83 ± 0.03
DF11	13h02m255	28°13'58''	$24.2_{-0.6}^{+0.6}$	$2.1_{-0.3}^{+0.4}$	$-14.8_{-0.1}^{+0.2}$	0.98 ± 0.03
DF12	13h00m091	28°08'27''	$25.2_{-0.6}^{+0.6}$	$2.6_{-0.9}^{+0.6}$	$-14.1_{-0.2}^{+0.5}$	0.88 ± 0.03
DF13	13h01m562	28°07'23''	$25.3_{-0.6}^{+0.6}$	$2.2_{-0.5}^{+0.6}$	$-13.7_{-0.2}^{+0.3}$	0.83 ± 0.03
DF14	12h58m078	27°54'46''	$25.3_{-0.7}^{+0.7}$	$3.8_{-0.1}^{+0.8}$	$-14.4_{-0.1}^{+0.1}$	0.51 ± 0.07
DF15	12h58m163	27°53'29''	$25.5_{-0.1}^{+0.1}$	$4.0_{-0.1}^{+5.5}$	$-14.9_{-0.4}^{+0.1}$	0.99 ± 0.29
DF16	12h56m524	27°52'29''	$24.8_{-0.8}^{+0.8}$	$1.5_{-0.2}^{+0.1}$	$-13.2_{-0.1}^{+0.2}$	0.82 ± 0.10
DF17	13h01m583	27°50'11''	$25.1_{-0.5}^{+0.5}$	$4.4_{-0.9}^{+1.5}$	$-15.2_{-0.2}^{+0.3}$	0.71 ± 0.03
DF18	12h59m093	27°49'48''	$25.5_{-0.6}^{+0.6}$	$2.8_{-0.5}^{+0.6}$	$-13.4_{-0.1}^{+0.2}$	0.47 ± 0.03
DF19	13h04m051	27°48'05''	$25.9_{-0.5}^{+0.5}$	$4.4_{-0.9}^{+1.6}$	$-14.5_{-0.3}^{+0.3}$	0.78 ± 0.03
DF20	13h00m189	27°48'06''	$25.5_{-0.8}^{+0.8}$	$2.3_{-0.1}^{+0.3}$	$-13.0_{-0.1}^{+0.1}$	0.53 ± 0.11
DF21	13h02m041	27°47'55''	$23.5_{-0.7}^{+0.7}$	$1.5_{-0.2}^{+0.3}$	$-14.6_{-0.1}^{+0.2}$	0.82 ± 0.03
DF22	13h02m578	27°47'25''	$25.1_{-0.6}^{+0.6}$	$2.1_{-0.3}^{+0.4}$	$-13.8_{-0.1}^{+0.2}$	0.84 ± 0.03
DF23	12h59m238	27°47'27''	$24.8_{-0.6}^{+0.6}$	$2.3_{-0.3}^{+0.5}$	$-14.3_{-0.2}^{+0.2}$	0.89 ± 0.03
DF24	12h56m289	27°46'19''	$25.2_{-0.7}^{+0.7}$	$1.8_{-0.4}^{+0.4}$	$-12.5_{-0.2}^{+0.2}$	0.38 ± 0.03
DF25	12h59m487	27°46'39''	$25.2_{-0.5}^{+0.5}$	$4.4_{-0.7}^{+1.4}$	$-14.5_{-0.2}^{+0.2}$	0.43 ± 0.03
DF26	13h00m206	27°47'13''	$24.1_{-0.6}^{+0.6}$	$3.3_{-0.4}^{+0.8}$	$-15.4_{-0.2}^{+0.2}$	0.63 ± 0.03
DF27	12h58m573	27°44'39''	≳ 26.5	⋅⋅⋅	⋅⋅⋅	⋅⋅⋅
DF28	12h59m304	27°44'50''	$24.4_{-0.6}^{+0.6}$	$2.7_{-0.4}^{+0.6}$	$-14.9_{-0.2}^{+0.2}$	0.79 ± 0.03
DF29	12h58m050	27°43'59''	$25.3_{-0.2}^{+0.2}$	$3.1_{-0.1}^{+1.6}$	$-14.6_{-0.1}^{+0.1}$	0.99 ± 0.13
DF30	12h53m151	27°41'15''	$24.4_{-0.5}^{+0.5}$	$3.2_{-0.6}^{+0.9}$	$-15.2_{-0.2}^{+0.2}$	0.70 ± 0.03
DF31	12h55m062	27°37'27''	$25.0_{-0.5}^{+0.5}$	$2.5_{-0.6}^{+0.7}$	$-14.1_{-0.2}^{+0.3}$	0.75 ± 0.03
DF32	12h56m284	27°37'06''	$24.8_{-0.6}^{+0.6}$	$2.8_{-0.3}^{+0.6}$	$-14.2_{-0.1}^{+0.1}$	0.52 ± 0.03
DF33	12h55m301	27°34'50''	$25.1_{-0.7}^{+0.7}$	$1.9_{-0.1}^{+0.2}$	$-13.4_{-0.1}^{+0.1}$	0.69 ± 0.03
DF34	12h56m129	27°32'52''	$26.0_{-0.6}^{+0.6}$	$3.4_{-0.4}^{+0.5}$	$-13.6_{-0.1}^{+0.1}$	0.66 ± 0.03
DF35	13h00m357	27°29'51''	$25.6_{-0.4}^{+0.4}$	$2.7_{-0.3}^{+1.0}$	$-13.9_{-0.2}^{+0.2}$	0.89 ± 0.09
DF36	12h55m554	27°27'36''	$25.0_{-0.6}^{+0.6}$	$2.6_{-0.4}^{+1.0}$	$-14.3_{-0.4}^{+0.3}$	0.80 ± 0.14
DF37	12h59m236	27°21'22''	$24.5_{-0.7}^{+0.7}$	$1.5_{-0.2}^{+0.3}$	$-13.7_{-0.2}^{+0.2}$	0.83 ± 0.03
DF38	13h02m001	27°19'51''	$24.2_{-0.6}^{+0.6}$	$1.8_{-0.3}^{+0.4}$	$-14.3_{-0.1}^{+0.2}$	0.84 ± 0.03
DF39	12h58m104	27°19'11''	$25.5_{-0.5}^{+0.5}$	$4.0_{-0.7}^{+1.3}$	$-14.7_{-0.2}^{+0.2}$	0.77 ± 0.05
DF40	12h58m011	27°11'26''	$24.6_{-0.6}^{+0.6}$	$2.9_{-0.5}^{+0.7}$	$-14.6_{-0.2}^{+0.2}$	0.56 ± 0.03
DF41	12h57m190	27°05'56''	$24.9_{-0.5}^{+0.5}$	$3.4_{-0.5}^{+0.9}$	$-14.7_{-0.1}^{+0.1}$	0.64 ± 0.03
DF42	13h01m191	27°03'15''	$25.0_{-0.6}^{+0.6}$	$2.9_{-0.4}^{+0.6}$	$-14.1_{-0.1}^{+0.1}$	0.52 ± 0.03
DF43	12h54m514	26°59'46''	$24.2_{-0.8}^{+0.8}$	$1.5_{-0.2}^{+0.2}$	$-13.8_{-0.2}^{+0.2}$	0.82 ± 0.10
DF44	13h00m580	26°58'35''	$24.5_{-0.5}^{+0.5}$	$4.6_{-0.8}^{+1.5}$	$-15.7_{-0.2}^{+0.2}$	0.65 ± 0.03
DF45	12h53m537	26°56'48''	$24.4_{-0.5}^{+0.5}$	$1.9_{-0.4}^{+0.6}$	$-14.2_{-0.2}^{+0.2}$	0.80 ± 0.03
DF46	13h00m473	26°46'59''	$25.4_{-0.6}^{+0.6}$	$3.4_{-0.6}^{+1.0}$	$-14.4_{-0.2}^{+0.2}$	0.74 ± 0.04
DF47	12h55m481	26°33'53''	$25.5_{-0.5}^{+0.5}$	$4.2_{-0.7}^{+1.4}$	$-14.6_{-0.2}^{+0.1}$	0.66 ± 0.04

Download table as:  ASCIITypeset image

We had expected that the objects would be randomly distributed in the 3° × 3° field that has both Dragonfly and CFHT coverage, as their apparent sizes seemed too large for a distance of 100 Mpc. However, as shown in Figure 1 they are strongly clustered toward the center of the image. A Monte Carlo implementation of the Clark–Evans test gives a probability of 0.04% that the distribution is spatially random. Moreover, the apparent east–west elongation of the distribution is similar to that of confirmed Coma cluster members (e.g., Doi et al. 1995). We conclude that most or all of the LSBs are, in fact, at the distance of the Coma cluster and are resolved in the Dragonfly data because they are intrinsically very large. As we show in Section 4 this conclusion is supported by Hubble Space Telescope (HST) Advanced Camera for Surveys (ACS) imaging of one of the galaxies.

3.1. Structure

We used GALFIT (Peng et al. 2002) to measure structural parameters of the galaxies from the CFHT images. The fits were performed on the summed g+i images, with neighboring objects masked. To increase the stability of the fits, the Sérsic index and sky background were not allowed to vary. All galaxies were fit three times, with the Sérsic index held fixed at n = 0.5, n = 1, and n = 1.5. The average χ² is lowest for n = 1 (exponential), but for individual galaxies the three fits are generally equally good. We therefore use the n = 1 results for all objects and determine the uncertainties in the structural parameters of individual galaxies from the full range of fits. Three examples of fits are shown in Figure 2. Forty-six galaxies were successfully fit; the S/N of one object (DF27) is too low for a stable fit.

Zoom In Zoom Out Reset image size

The distribution of the galaxies in the surface brightness—size plane is shown in Figure 3, under the assumption that they are all at the distance of the Coma cluster. The central surface brightnesses, calculated from the circularized effective radii and the total fit magnitudes, range from μ(g, 0) = 24–26 mag arcsec⁻². The effective radii, measured along the major axis, range from 1.5 kpc to 4.5 kpc. At fixed surface brightness the newly found galaxies are much larger than typical dwarf elliptical galaxies in the Virgo cluster (open circles; Gavazzi et al. 2005). The median central surface brightness 〈μ(g, 0)〉 = 25.0 mag arcsec⁻² (≈25.4 mag arcsec⁻² in the B band) and the median effective radius 〈r_eff〉 = 2.8 kpc. An interesting point of comparison is the disk of the Milky Way. Bovy & Rix (2013) derive a mass-weighted exponential scale length of 2.15 ± 0.14 kpc, corresponding to r_eff = 3.6 kpc. Twelve of the newly found objects are larger than this, although for individual objects the difference is typically not significant. We note that the gap between SDSS and the Dragonfly data in Figure 3 is due to the selection limits of the surveys. The newly found galaxies are simply the low surface brightness, large size extension of the general galaxy population, and samples such as that of Thompson & Gregory (1993) would fill in the gap.

Zoom In Zoom Out Reset image size

The axis ratio distribution is shown in the right panel of Figure 3. The galaxies are remarkably round, with a median axis ratio of 0.74. We note that there is no obvious selection effect against inclined disks, as the galaxies are barely resolved in the Dragonfly data. Randomly oriented thin disks would have a uniform b/a distribution, and this can be ruled out.

3.2. Stellar Content

We searched the HST Archive for serendipitous observations of the newly found galaxies. Three of the 47 galaxies have been observed by HST. Two of the observations are short (200–300 s) WFPC2 exposures, which show only hints of the objects. The third comprises 8 orbit, multi-band ACS imaging of DF17, whose properties are close to the median of the sample. The ACS data include g₄₇₅, V₆₀₆, and I₈₁₄ parallels to a Cepheid program with the WFC3/UVIS camera (GO-12476, PI: Cook; Macri et al. 2013). The data were obtained from the archive and reduced using standard techniques (van Dokkum 2001).

A color image, created from the V₆₀₆ and I₈₁₄ images, is shown in Figure 4. DF17 is large and spheroidal and does not have obvious spiral arms, star forming regions, or tidal features. We fit the ACS data with a Sérsic profile, leaving all parameters free. The best fitting parameters are r_eff = 70, n = 0.6, μ₄₇₅ = 25.8, and b/a = 0.71. The effective radius, surface brightness, and axis ratio are in excellent agreement with the n = 0.5 fit to the CFHT image.

Zoom In Zoom Out Reset image size

The fact that the galaxy is not resolved into stars implies a lower limit to its distance. We created model images of DF17, following the methodology described in van Dokkum & Conroy (2014). Stars were drawn randomly from a Poisson distribution, weighted by their expected frequency in a 10 Gyr old stellar population with a metallicity [Fe/H] = −1.6. This stellar population reproduces the observed V₆₀₆ − I₈₁₄ color (V₆₀₆ − I₈₁₄ = 0.40). The models are constrained to reproduce the observed two-dimensional light distribution of DF17 and its observed total magnitude of I₈₁₄ = 19.3, with the distance as the only free parameter. The model images were convolved with the ACS point-spread function and placed in the ACS image, after subtracting the best-fitting GALFIT model of the galaxy.

The results are shown in the bottom panels of Figure 4. Out to well beyond the Virgo cluster (16 Mpc) the ACS camera easily resolves individual stars in LSBs, as also shown by Caldwell (2006). Only at distances ≳ 50 Mpc do the models take on the same smooth appearance as the data, and we conclude that the ACS observations support the interpretation that the galaxies are associated with the Coma cluster. The effective radius of DF17 is then 3.4 kpc, almost identical to that of the disk of the Milky Way.

We have identified a significant population of low surface brightness, red, nearly round objects in a wide field centered on the Coma cluster. Based on their spatial distribution and the analysis of one example observed with ACS, we infer that most or all of the objects are associated with Coma. Their inferred sizes are similar to those of L_* galaxies and the disk of the Milky Way, even though their stellar masses are a factor of ∼10³ lower.

The galaxies do not resemble "classical" LSBs such as those described by, e.g., van der Hulst et al. (1993), Bothun et al. (1997), and van den Hoek et al. (2000). Typical LSBs have blue, gas-rich disks, and are thought to be normal spiral galaxies with a low stellar content and low star formation rate for their rotation velocity (see, e.g., Schombert et al. 2013, and references therein). They are also significantly brighter than the objects found in this Letter: the lowest surface brightness object in the compilation of Bothun et al. (1997) has μ(0, B) ≈ 24.0 mag arcsec⁻², corresponding to μ(0, g) ≈ 23.6 mag arcsec⁻². Many have bulges; for example, Malin I has a central surface brightness ≲ 16 mag arcsec⁻² if its bulge is taken into account (Lelli et al. 2010).

Visually and structurally, the newly found galaxies are more similar to dwarf spheroidal galaxies such as those found in the Local Group, around M101, and in the Virgo and Coma clusters than to classical LSBs: they have similar Sérsic indices, axis ratios, and surface brightness (e.g., Thompson & Gregory 1993; Geha et al. 2003; Gavazzi et al. 2005; McConnachie 2012; Merritt et al. 2014; Toloba et al. 2014). However, the term "dwarf" is not appropriate for these large objects. Dwarf spheroidals have typical sizes of a few hundred parsecs (e.g., McConnachie 2012; Lieder et al. 2012), and in the Local Group and other nearby groups only a few have an effective radius exceeding 1 kpc (e.g., Kim et al. 2011; McConnachie 2012; Chiboucas et al. 2013; Merritt et al. 2014). The largest known low luminosity Local Group galaxy is And XIX, with a size of 1.6 kpc (McConnachie et al. 2008). The Coma objects are much larger, with sizes typical of ∼L_* spiral and elliptical galaxies (e.g., Shen et al. 2003).

The closest analogs to the Coma objects are several very large low surface brightness objects in the Virgo and Fornax clusters, first identified by Impey et al. (1988). There are four galaxies in the Impey et al. sample with central surface brightness ≳ 25 mag arcsec⁻² and r_eff > 2.5 kpc; the largest of these, V1L5 and V4L7, have r_eff = 3.7 kpc. As the Impey et al. survey area is four times smaller than ours the number of such galaxies in Virgo and Coma could be similar. Although the distances to these particular objects are not confirmed, Caldwell (2006) used HST/ACS imaging to show that at least one galaxy with a central surface brightness of μ(g, 0) ≈ 27.2 and an effective radius of 1.5 kpc is part of the Virgo cluster. We propose the term "ultra-diffuse galaxies," or UDGs, for galaxies with r_e ≳ 1.5 kpc and μ(g, 0) ≳ 24 mag arcsec⁻². We stress that this term does not imply that these objects are distinct from the general galaxy population; these are simply the largest and most diffuse objects in a continuous distribution.

As shown in Figure 5 no UDGs are found in the central regions of the cluster, consistent with earlier results for slightly brighter diffuse spheroidals in Coma (Thompson & Gregory 1993). This could mean that they are only able to survive at large radii (see, e.g., Bothun et al. 1991; Gregg & West 1998; Martel et al. 2012). We can estimate what the mass of the galaxies needs to be to survive a passage within ∼300 kpc of the core of the cluster, which is where the closest-in UDGs are located. The criterion for survival is that the total mass m_tot within the tidal radius r_tide = 2r_e = 6 kpc is at least m_tot > 3M(r_tide/R)³, with M the mass of the cluster within radius R. Using the mass profile of A2667 (Newman et al. 2013) as a proxy for that of Coma, we find m_tot ≳ 3 × 10⁹ M_☉, or a dark matter fraction within the tidal radius of ≳ 98%. We note that there may be UDGs closer to the cluster core, as crowding and the ICL limit our ability to detect them (see Ulmer et al. 1996; Adami et al. 2006, 2009).

Zoom In Zoom Out Reset image size

It is not clear how UDGs were formed. It seems unlikely that they are the product of galaxy harassment (Moore et al. 1996) or tidal stirring (Mayer et al. 2001) of infalling galaxies: these processes tend to shrink galaxies, as the stars at larger radii are less bound than the stars at small radii (see, e.g., Mayer et al. 2001). A likely end-product of cluster-induced tidal effects are the ultra-compact dwarfs (Drinkwater et al. 2003), which have similar total luminosities and stellar masses as UDGs but stellar densities that are a factor of ∼10⁷ higher.⁵ We note, however, that the morphological evolution of infalling galaxies is difficult to predict, as it probably depends sensitively on the shape of the inner dark matter profile (e.g., Peñarrubia et al. 2010). An intriguing formation scenario is that UDGs are "failed" ∼L_* galaxies, which lost their gas after forming their first generation(s) of stars at high redshift (by ram pressure stripping or other effects). If this is the case they may have very high dark matter fractions, which could also help explain their survival in the cluster. Future studies of these objects, as well as counterparts in other clusters and in the field (see Dalcanton et al. 1997), may shed more light on these issues.

We thank the anonymous referee for an excellent and constructive report. We also thank the staff at New Mexico Skies for their support and Nelson Caldwell for comments on the manuscript. Support from NSF grant AST-1312376 is gratefully acknowledged.