ULTRA-COMPACT DWARFS IN THE COMA CLUSTER

Spectroscopic redshifts provide the most reliable method for establishing membership in galaxy clusters, but are prohibitively time consuming considering the abundance of faint objects in cluster fields. Given this limitation, studies of faint cluster galaxies often use indirect means such as selecting probable members by color, surface brightness, and morphology. Such indirect methods, however, require assumptions about the properties of cluster members. Consequently, these methods are inherently biased. Entire populations of galaxies may be left out of cluster member samples.

It had previously been assumed that compact high surface brightness objects were either background ellipticals or foreground stars. This was due in part to a well-defined surface-brightness–magnitude relation in which dwarf elliptical galaxies trend toward lower surface brightnesses at fainter magnitudes (Caldwell & Bothun 1987; Ferguson & Sandage 1988; Impey et al. 1988; Binggeli & Cameron 1991; Mieske et al. 2004a). Only a decade ago, spectroscopic surveys of the Fornax cluster, including an all-object 2dF spectroscopic survey, revealed a new class of object termed ultra-compact dwarfs (UCDs; Hilker et al. 1999; Drinkwater et al. 2000; Phillipps et al. 2001). These faint objects are offset from the magnitude–surface-brightness relation of dE galaxies having high surface brightnesses and very small sizes.

This discovery spurred dedicated searches for UCDs in other nearby groups and clusters. A large population of about 60 UCDs has since been established to exist in Fornax (Goudfrooij et al. 2001; Drinkwater et al. 2005; Firth et al. 2008; Gregg et al. 2009). A similarly large population has recently been identified in the Hydra cluster (Misgeld et al. 2011), while somewhat smaller populations have been found in the core of the Virgo (∼25 UCDs; Haşegan et al. 2005; Jones et al. 2006; Firth et al. 2008) and Centaurus clusters (Mieske et al. 2007a, 2009). A single UCD was confirmed in the Dorado group (Evstigneeva et al. 2007a). Candidate UCDs have also been identified in the Coma, Hydra, A1689, and AS0740 clusters and NGC 1023 group (Adami et al. 2009; Madrid et al. 2010; Wehner & Harris 2007; Mieske et al. 2004b; Blakeslee & Barber DeGraaff 2008; Mieske et al. 2007b). Only a handful of UCDs have been detected in low-density environments, either as companions to isolated field galaxies or in poor groups (Hau et al. 2009; Da Rocha et al. 2011; Norris & Kannappan 2011). None have been found in the poor Local or nearby M81 Group (Chiboucas et al. 2009) environments.

UCDs are intrinsically faint and compact objects that are unresolved in ground-based surveys and have magnitudes between −13.5 < M_V < −10.5. Effective radii for these objects span the range 7 pc < r_e < 100 pc (Mieske et al. 2007a; Jones et al. 2006; Mieske et al. 2008b), which is larger than typical globular clusters with sizes ∼2–5 pc (Larsen et al. 2001; Jordán et al. 2005) but smaller than dwarf ellipticals which have sizes of a few 100 parsecs (Drinkwater et al. 2003). At the distance of the Coma cluster, the UCD half-light radii range corresponds to 0.01–0.2 arcsec. Thus, with the Advanced Camera for Surveys (ACS) ∼0.1 arcsec resolution, the larger ones are just resolved. The brighter UCDs also have exceptionally red optical colors, lying redward of the red sequence by up to ∼0.2 mag in Virgo and Fornax, redder than typical globular clusters by 0.1 mag, and redder than the nuclei of dE,N galaxies (Mieske et al. 2006). The red color may be driven by high metallicities (Mieske et al. 2006; Chilingarian et al. 2008). With these distinct properties, it is clear that UCDs constitute a separate class of object from the abundant dE and dSph galaxies found in clusters and groups.

Measurements of mass-to-light ratios (M/L) have been made for a few of these objects and found to range from 2 to 9 (Mieske et al. 2008b; Haşegan et al. 2005; Evstigneeva et al. 2007b; Hilker et al. 2007). The larger values could be evidence for the presence of dark matter but alternative explanations include (1) inflated mass estimates produced by tidal heating which increases the velocity dispersion (Fellhauer & Kroupa 2006) or (2) bottom (Mieske et al. 2008a) or top heavy initial mass functions (IMFs; Dabringhausen et al. 2009; Murray 2009).

A number of potential formation mechanisms have been proposed to explain the nature and origin of these enigmatic objects (Drinkwater et al. 2000; Fellhauer & Kroupa 2002; Bekki et al. 2003; Haşegan et al. 2005; Chilingarian et al. 2008; Evstigneeva et al. 2008). UCDs may simply be globular star clusters comprising the extreme bright tail of the globular cluster luminosity function or super star cluster end products from the mergers of massive star clusters themselves formed during galaxy mergers. Alternatively, they may be the visible nuclei of dwarf ellipticals with exceptionally low surface brightness envelopes, primordial objects formed of the small-scale peaks in the initial power law spectrum, or remnant nuclei of tidally stripped nucleated dwarf elliptical or late type galaxies which have been "threshed" during tidal encounters with giant galaxies and with the cluster potential. While this latter explanation is sometimes favored, other possibilities have not been ruled out. In particular, evidence exists showing that UCDs may be more metal-rich and, in some cases, older than the nuclei of dE,N from which they purportedly originated (Chilingarian et al. 2008; Mieske et al. 2006). Because dE,N can continue to form stars in their cores through gas accretion while threshed nuclei, having lost their outer halo, suffer from strangulation and a cessation of star formation, it is difficult to reconcile the higher metallicities and, at the same time, older ages with a threshing mechanism. These studies instead support a super star cluster scenario originating from the merger of young massive clusters at early times. In addition, UCD sizes tend to be larger than both globular clusters and the nuclei of dwarf galaxies (De Propris et al. 2005). However, finding age and metallicity measurements for UCDs and dwarf nuclei in Virgo that are compatible, Paudel et al. (2010) argue that the stripping scenario is a viable option. Yet other researchers suggest that with a continuum of luminosities and with colors similar to metal-rich globular clusters, UCDs may simply constitute a bright extension of the globular cluster sequence (Wehner & Harris 2007).

If UCDs originate as remnant dE nuclei via a threshing mechanism they would be expected to populate cores of massive and dynamically evolved clusters. Compared to lower density environments, they would be expected in greater numbers and with a broader distribution within the massive Coma cluster. Alternatively, if UCDs are giant globular clusters or super star clusters they should be associated with individual galaxies and have properties similar to those of the host galaxy's globular cluster and stellar populations. Indeed, evidence has shown UCD populations in Fornax and Virgo are strongly clustered, having smaller velocity dispersions than cluster dwarfs or even the central giant galaxy globular cluster systems (Gregg et al. 2009; Mieske et al. 2004a; Jones et al. 2006; Firth et al. 2008). Firth et al. (2008) have furthermore found that the majority of UCDs lie within the fields of the dominant giant galaxies in Virgo and Fornax while very few have been discovered to lie in intracluster regions. Therefore, we might expect to find a population of UCDs within the Coma cluster, associated with either the cluster potential or individual giant galaxies depending on formation mechanism.

This work is part of the larger Hubble Space Telescope (HST)/ACS Coma Cluster Treasury Survey (Carter et al. 2008), a two-passband imaging survey designed to cover 740 arcmin² in the Coma cluster to a depth of I_C ∼ 26.6 mag for point sources. These data are being used to perform comprehensive structural, photometric, and morphological studies of the Coma cluster members. Only 28% of the originally proposed areal coverage was completed due to the failure of ACS, but it includes much of the core region. The goal of the Keck/Low-Resolution Imaging Spectrometer (LRIS) study discussed in this work was to measure redshifts and establish membership for galaxies at the faint end of the cluster luminosity function. We specifically targeted two samples of faint galaxies. The first, which is discussed in Chiboucas et al. (2010), targeted galaxies in the magnitude range 19 < R < 22 (− 16 < M_R < −13) having low surface brightness and membership previously estimated through indirect means. The second sample, which we discuss here, targeted candidate UCDs with R < 24 (M_R < −11). Candidates were chosen in the core region of the Coma cluster where HST/ACS data had already been obtained.

In Section 2 we describe the observations and data reduction procedure, in Section 3 we detail the properties and distribution of confirmed cluster member UCDs, and in Section 4 we discuss these results and potential origins. A summary is presented in Section 5. Throughout this paper, we assume a distance to the Coma cluster of 100 Mpc and a distance modulus μ = 35 (see Carter et al. 2008, Table 1).

2.1. Target Selection

2.2. Observations and Data Reduction

Observations and data reduction are described in Chiboucas et al. (2010), but we summarize here. Because we are observing faint targets, down to R < 24, along with low surface brightness galaxies, the spectroscopic design was intended to maximize light throughput at the expense of resolution. As we intend only to measure redshifts in order to distinguish between stars, Coma cluster members at z = 0.023, and background objects, and the Coma cluster has a velocity dispersion of ∼1000 km s⁻¹, a resolution of 200 km s⁻¹ is adequate. The spectral range was chosen to include the 4000 Å Balmer break and Ca H and K lines at the blue end (∼4060 Å at the redshift of Coma) as these are some of the strongest spectral features in quiescent, low-redshift, faint galaxies.

We use the Keck/LRIS in multi-object spectroscopy mode which has very high sensitivity in the blue. A dichroic was used to split light at 5600 Å between red and blue chips. On the blue side, we used the 400 line mm⁻¹ grism blazed at 3400 Å providing a dispersion of 1.07 Å pix⁻¹ and wavelength coverage of 4384 Å. With 1.2 arcsec slitlets, we achieved a resolution of 7.8 Å FWHM. On the red side we chose the 400 line mm⁻¹ grating blazed at 8500 Å with wavelength coverage 3950 Å and 1.92 Å pix⁻¹ dispersion. The red-side data are used primarily to identify high-redshift objects with emission line spectra.

Four masks were observed over two nights, 2008 April 2–3. Another two masks were observed 2009 March 28–30 in poor conditions. Each mask covers a field of view of 5' × 8' with an average 35 slitlets. Total integrations for each mask during the first run were 8 × 1500 s with the exception of the blue side of one mask for which we obtained 6 × 1500 s. During the second run we were plagued with thick clouds and very poor seeing. Total integrations were ∼8 × 1500 s but effective exposure times are much less. Signal-to-noise Å⁻¹ (measured around 5000 Å) for the UCD candidate spectra range from ∼20 to less than 1. Secure redshifts were measured from spectra with signal-to-noise ratio (S/N) >4.0, while less secure measurements came from 2.5 < S/N <5 spectra. Table 1 provides a summary of the observations. In Figure 1, we overlay the locations of the completed ACS fields and the six LRIS masks on an image of the central region of the Coma cluster. The original and expanded set of candidate UCDs are shown as green and brown/cyan points, respectively.

Zoom In Zoom Out Reset image size

Table 1. Keck/LRIS Observations

Mask	Date	α (J2000.0)	δ (J2000.0)	P.A. (deg)	Seeing (arcsec)	Nobj
1	2008 Apr 2	13 00 40.69	28 01 54.86	1.5	0.7	35
2	2008 Apr 3	13 00 17.82	28 01 26.81	1.5	1.0	32
3a	2008 Apr 3	13 00 24.62	27 56 26.11	80.0	1.0	38
4	2008 Apr 2	12 59 40.10	27 59 06.28	121.0	0.8	39
5b	2009 Mar 28–29	13 00 43.95	27 59 47.08	85.0	0.8–1.4	43
6	2009 Mar 30	12 59 53.57	27 57 50.93	−94.0	1–1.3	42

Notes. ^aExposure times for both red and blue chips were 8 × 1500 sec with the exception of Mask 3 for which we obtained only 6 × 1500 sec on the blue side. ^bObservations were taken in thick and variable cloud cover.

Download table as:  ASCIITypeset image

Data were reduced using the standard procedures in IRAF. Images were overscan corrected and corrected for different gains. Halogen flats for each mask were combined and a normalized flat image was generated with APFLATTEN. Following division by this flat, the eight individual exposures for each mask were median combined using sigma clipping. Slit spectra were rectified by tracing the slit gaps for each mask and fitting these with fourth-order Legendre polynomials. GEOMAP was run to compute the two-dimensional surface for the full set of slit gaps in a mask and, using that transformation, GEOTRAN was then executed to generate images with straightened slits. Arc spectra were rectified for each slitlet in the same manner.

The usual IRAF tasks IDENTIFY, FITCOORD, and TRANSFORM were used to wavelength calibrate the object spectra from arc lamp spectra. Since arc spectra were taken only once per night, sky lines in each object spectra were used to correct for offsets from the lamp wavelength calibrations. Unfortunately, the prominent 5577 Å sky line fell on the edge of our spectra, or depending on the location in the mask, off the observed blue-side spectra altogether. Due to uncertainties in the applied shift, we therefore expect systematic errors of up to 100 km s⁻¹ in the radial velocity measurements from our 2008 run targets. Observations taken in 2009 were bracketed with arc lamp exposures and are not expected to suffer from these systematic offsets.

To extract one-dimensional spectra, we used APALL to identify the spectrum center, width, and sky regions. RVSAO/XCSAO was used, along with absorption and emission line template spectra, to measure redshifts. A flux standard was observed during a later run with the same configuration, used for relative flux calibration.

Tables 2 and 3 list spectroscopic redshift measurements for the UCD candidate samples. We assume that objects with radial velocities between 4000 km s⁻¹ < v_r < 10, 000 km s⁻¹, within 3σ of the cluster mean, are cluster members. In total, we find 27 compact sources with redshift measurements consistent with Coma cluster measurement. Another 14 also have measured redshifts that would place them in the Coma cluster, but these come from very low S/N (<5) spectra and we consider these measurements to be highly uncertain. Out of the first set of 47 targeted candidates, 19 proved to be members, 6 were highly uncertain members, 4 turned out to be background galaxies, and 6 proved to be foreground stars. The remaining 12 targets had spectra with too low S/N to even attempt redshift measurements. Spectra for these 19 confirmed UCDs are shown in Figures 2–4. Thumbnail images of these confirmed UCDs are provided in Figure 6. The second round of observations based on our expanded candidate sample turned up seven background galaxies and six stars along with eight members. Spectra for these, taken in poor conditions, are shown in Figure 5 and thumbnails are presented in Figure 7. In Figure 1, we display the spatial distribution of all confirmed and questionable UCDs along with the full set of candidates. Spectroscopically determined stars and background galaxies are also highlighted.

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

In Figure 8, we show color–magnitude diagrams for all objects in our field. Boxes delineate the original candidate color selection criteria. Candidates, confirmed members, background galaxies, and stars are denoted by different symbols. dEs confirmed as members from LRIS spectra taken concurrently with the UCD observations are also shown. The dE red sequence is obvious in B − V. UCDs meanwhile exhibit a very large spread in colors ranging from the red sequence on the blue side to 0.4 mag redder than the red sequence in B − V. One discrepant point with very red colors is likely affected by photometric errors. Two cE galaxies observed during our initial LRIS campaign, along with several others taken from Price et al. (2009), also lie redward of the red sequence by about 0.2 mag.

Zoom In Zoom Out Reset image size

We find that, in good conditions, S/N starts to become too low to measure redshifts for these high surface brightness objects at R ∼ 23.3, and we fail to measure secure redshifts altogether at R = 23.5. Therefore, brighter than R = 23.3 where we can measure redshifts for nearly 100% of our sources, we find that 66% (18/27) of the original targeted candidates are bona fide Coma cluster members. This is an exceptionally high success rate given that the UCD candidates were selected primarily based only on fairly loose color criteria. Those targeted, however, were weighted toward objects with profiles slightly broader than pure PSFs and we believe that this boosted our success rate.

In Figure 9, we compare the FWHM and SExtractor classification of the UCDs as measured in the ACS F814W images to that of confirmed stars and galaxies. The ACS measurements were performed with SExtractor as described in Hammer et al. (2010). It can be seen that the majority of confirmed UCDs have an FWHM at a given magnitude that lies between the stellar minimum and the larger background galaxies. One may expect that many of the candidates with FWHM in the range of the confirmed background galaxies will also prove to be background objects, while the remainder of these points with sizes larger than that of the stellar minimum may turn out to be further UCD cluster members. These partially resolved objects are classified by SExtractor in our images as falling between stars (1) and galaxies (0). Candidates in this range would also be considered likely members, along with objects classified as stars but which have slightly broader profiles. However, with these low number statistics, we cannot rule out the possibility that there are large numbers of unresolved UCDs.

Zoom In Zoom Out Reset image size

3.1. Properties

3.1.1. Photometric Properties

In Figure 10, we show optical color–magnitude diagrams for confirmed members of the Coma cluster (UCDs and dEs). In all cases, a red sequence of normal dE galaxies, exhibiting some scatter, is apparent. The confirmed UCDs, which lie at the faint end of this sequence, have a much larger spread in colors of up to 0.4 mag in both g − I and B − V. Based on the g − I ACS photometry, UCD colors range from just redward of the red sequence to nearly 0.4 mag redder. Also shown in these plots are seven confirmed Coma cluster cEs from Price et al. (2009). These objects at brighter magnitudes also lie redward of the red sequence by about 0.2 mag. Although there is a ∼1.5 mag gap between the cE and red UCD populations, these objects do share similarly red colors and could perhaps form a single sequence of extremely red cluster objects. The colors may indicate that these objects have similar stellar populations and perhaps evolutionary histories.

Zoom In Zoom Out Reset image size

We plot UCDs and other object types in magnitude–surface-brightness space in Figure 11. Central surface brightness versus R-band total magnitude comes from ground-based photometry (Adami et al. 2006) for all objects in our survey region while 〈μ〉_e versus F814W are measured from our ACS data. LRIS confirmed normal dwarf galaxies are found to lie along the well-known magnitude–surface-brightness relation for dE galaxies (Caldwell & Bothun 1987; Ferguson & Sandage 1988; Binggeli & Cameron 1991; Boselli et al. 2008). At much higher surface brightness, a second almost parallel sequence of UCDs is found, bounded at high surface brightness by the seeing disk in the ground-based data where the UCDs are merged with the stellar sequence. The ACS data, on the other hand, do a much better job at separating stars and UCDs, although some overlap is still present. In the region between the dE and UCD sequences are confirmed and probable background galaxies (see Chiboucas et al. 2010). Brightward of the UCD sequence, the set of 7 cE galaxies also lie at very high surface brightness, well separated from the normal galaxies. Two of these objects for which we obtained LRIS spectra are filled in. Although the UCDs and cEs are separated by a gap of 1.5 mag, their surface brightness characteristics also suggest that these two object types may follow a continuous sequence and form a single class of object.

Zoom In Zoom Out Reset image size

3.1.2. Structural Properties

Because the UCDs are marginally resolved in the ACS data, we are able to measure the sizes of many of these objects. Other studies have shown that it is possible to measure globular cluster effective radii if the stellar PSF is accurately known and the cluster size (r_h) is larger than about 10% of the PSF FWHM (see Harris et al. 2010, and references therein). The stellar PSF for the ACS/WFC is about 0.1 arcsec, so we would expect to be able to measure sizes down to about 5 pc. Recently, Madrid et al. (2010) identified candidate UCDs near NGC 4874 based on size measurements of stellar-like objects in these ACS data. From a comparison of measurements in two different bands, they determined that they could accurately measure sizes down to r_h = 9.2 pc.

To measure the structural parameters for these UCDs, including physical sizes, we perform surface brightness profile fitting on the objects in our F814W ACS images. Two methods are used to fit these barely resolved objects. We first model each object with a Sersic function using the two-dimensional fitting algorithm Galfit (Peng et al. 2002). Galfit fits the light profile of galaxies with axisymmetric analytical functions and is robust for well resolved, extended objects. Often used for profile fitting of globular clusters, Ishape (Larsen 1999) is a routine in the baolab data reduction package designed specifically for fitting marginally resolved sources. Both convolve a PSF with model profiles and both algorithms determine best fits through χ² minimization.

To run Galfit, the user must provide a PSF for convolution with a model, an uncertainty map which is critical for determining when the best-fit solution is reached, the choice of profile to fit, and initial guesses for each parameter to be fit. The procedure we use is similar to that explained in detail in Hoyos et al. (2011). Briefly, to create the uncertainty images, we make use of the Multidrizzle output inverse variance images and in addition take into account the Poisson noise from the sources themselves. To generate PSFs appropriate for Galfit input, TinyTim (Krist 1995) PSFs are generated for a grid of locations on the ACS chips for each filter. These are then added at the corresponding shifted locations to blank individual distorted flat-fielded (FLT) images. The images are then run through Multidrizzle with the same configuration used to produce the final combined science images. PSF images 3 arcsec in size are extracted from this drizzle combined image. For the UCD fits, we perform the fitting in image sections 300 × 300 pixels in size. All objects other than the central UCD are masked. We fit for all parameters of the Sersic function (total magnitude, effective radius R_e, index n, axis ratio b/a, and position angle) along with the sky value. We found that measured values often depended on initial guesses. Furthermore, the Sersic index n and R_e are strongly correlated. We therefore ran Galfit with initial guesses for the Sersic index ranging from 1.4 to 6.4. We take the fit with the lowest χ² value as the best fit. Since Galfit measurement uncertainties are in most cases underestimates, we take for measurement uncertainties the standard deviation of the measurement values from fits with five sets of initial guesses, including two cases where we have held the Sersic index n fixed at 2.5 and where we included an upper limit constraint of 7.6 for the index.

Ishape requires 10 times subsampled PSFs so we generate an empirical PSF based on 26 real stars in one of our ACS images using the routines in the IRAF daophot package. We fit each UCD with a set of five profile shapes: a Moffat profile with power index 2.5, a Sersic function where the index n is fitted for, and King profiles with three distinct values for the concentration parameter (defined as r_t/r_c where r_t is the tidal radius and r_c is the core radius): 15, 30, and 100. We chose to fit these as elliptical functions. Output includes the FWHM along both major and minor axes, position angle, and χ² for the fit. To convert FWHM to effective radii, we take the circularized FWHM from the geometric mean of the semimajor and semiminor axes and convert to R_e using the appropriate concentration-dependent conversion factors described in Larsen (2001), and provided in the user guide. We take as final measurements the parameters from the fit producing the best residuals and lowest χ². These proved consistently to be from King models with concentration parameters 30 and 100. During testing, it was discovered that PSF size and fitting radius affected the size measurements by amounts greater than the small quoted errors. Since the quoted errors are therefore expected to be underestimates, which also assume that a particular model is a good match to the true profile shape, we therefore use the standard deviation of measurements from fits with the five different models using two different fitting radii each (10 and 25 pixels) as a more realistic measure of the uncertainty in size. Final sizes are taken to be those based on 25 pixel fitting radii which typically produced the fits with the smallest residuals.

Size measurements are presented in Table 4. To convert effective sizes from arcsec to parsec we assume a distance of 100 Mpc. We find sizes ranging from 5 to 125 pc with a median effective radius (for both Galfit and Ishape measurements) of 23 pc. Overall, the consistency between Galfit and Ishape measurements of R_e is quite good. We show the comparison in Figure 12. For the largest object, 151072, Galfit finds a size about twice as large as Ishape. This may be an indication that the object has an extended halo which affects the Sersic function fits more than the King profile fits. Although we only fit single component model profiles for these objects we note that a number of the objects, including some of the larger UCDs, may be better fit with two components. For objects best fit with very large n, this may be caused by a lower surface brightness envelope surrounding a high surface brightness core forcing a fit with larger wings. In particular, object 150000, although highly obscured in the diffuse light of a nearby bright galaxy, upon closer inspection appears to have a large low surface brightness envelope. Objects 121666 and 195526 also may require a second component fit. In Figure 13, we show the residuals from both Galfit and Ishape fits for four UCDs which may be better fit with two components. Object 150000 in the top panel has 19% of the flux remaining in the residual when fit with a single Sersic function. UCD 151072 in the bottom panel has residuals at the level of 5% when fit with a King profile. Figure 14 displays residuals for four cases where the UCDs are well fit with single component models by both Galfit and Ishape.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Table 4. Effective Radii

ID	GALFIT	(Sersic)	F814W	b/aa	ISHAPE	Index	Re (pc)	b/aa
	Re (pc)	n			Profile
191006	7.1 ± 2.3	5.2	22.51	0.82	King	30	7.8 ± 2.6	0.68
192636	33.3 ± 3.8	3.1	22.28	0.86	King	30	29.2 ± 6.5	0.93
120985	40.5 ± 3.2	1.7	22.17	0.90	King	30	42.8 ± 12.2	0.88
121666	42.0 ± 13.3	3.9	21.54	0.99	King	100	38.3 ± 3.6	0.88
195526	37.6 ± 14.9	5.7	21.66	0.91	King	100	32.5 ± 5.3	0.99
195614	22.0 ± 2.5	6.9	22.23	0.84	King	100	22.9 ± 5.2	0.89
196790	20.9 ± 2.4	6.6	21.74	0.86	King	100	22.6 ± 6.1	0.87
182204	14.7 ± 4.0	5.4	23.17	0.65	King	100	8.6 ± 4.6	0.91
242857	23.7 ± 3.0	4.7	22.30	0.79	King	30	24.8 ± 5.2	0.79
160141	7.0 ± 3.1	5.1	22.66	0.76	King	100	6.0 ± 2.9	0.89
161244	4.8 ± 2.9	3.7	22.60	1.00
163341	23.3 ± 2.7	7.6	22.24	0.98	King	100	19.5 ± 6.7	0.87
163400	66.9 ± 16.4	6.3	23.11	0.74	King	100	54.4 ± 7.1	0.80
92415	24.3 ± 3.8	7.6	22.67	0.91	King	100	22.0 ± 6.4	0.87
163575	17.1 ± 4.1	5.3	22.53	0.74	King	100	12.8 ± 6.6	0.90
150880	12.8 ± 5.9	4.4	22.74	0.95	King	100	11.0 ± 2.4	0.77
151072	125.5 ± 18.8	6.6	21.29	0.98	King	100	68.1 ± 6.4	0.89
150000	36.0 ± 8.3	7.9	22.38	0.92	King	100	24.3 ± 2.8	0.72
81669	18.2 ± 6.0	2.6	22.95	0.95	King	30	17.7 ± 5.4	0.97
1041346	27.7 ± 7.5	4.4	21.89	0.94	King	100	28.3 ± 4.5	0.85
1039188	25.6 ± 4.2	5.2	21.97	0.89	King	100	26.4 ± 6.0	0.88
1041508	10.2 ± 3.9	6.1	22.91	0.74	King	100	14.4 ± 4.6	0.86
1043225	20.9 ± 7.0	4.2	22.92	0.64	King	100	13.0 ± 4.2	0.74
1042830	19.7 ± 4.8	4.1	21.87	0.97	King	100	15.2 ± 6.8	0.94
2000005	7.1 ± 3.3	4.1	23.07	0.85	King	100	6.7 ± 2.5	0.98
1044251	34.7 ± 8.4	4.5	22.15	0.84	King	30	32.6 ± 6.7	0.85
1044847b

Notes. ^aTypical uncertainties in b/a were 0.07 and 0.05 for Galfit and Ishape, respectively. However, for such small objects, we don't consider the axial ratio measurements to be very reliable. ^bThis object is outside of our ACS coverage. Although this object has been observed in archival WFPC2 images, the larger pixel scale makes size measurements for such small objects impossible.

Download table as:  ASCIITypeset image

Three of the objects less well fit by single component profiles turn out to be the three brightest UCDs in the sample. If the poor fits are due to an extra envelope of material surrounding the UCD core, we could be seeing remnant tidal debris from the stripping of nucleated, more massive galaxies. Extended halos have also been found around UCDs in Virgo and Fornax from HST imaging (Evstigneeva et al. 2008) and around massive GCs in the Milky Way and other nearby galaxies (McLaughlin & van der Marel 2005; McLaughlin et al. 2008). They may therefore be a common property of star clusters and are not necessarily evidence for the threshing mechanism. In fact, the lack of such signatures around most of the fainter UCDs in our sample may simply be due to the lower S/N and the small sizes of these objects. If the brightness and extent of the envelope is correlated with UCD size and brightness, extended halos around fainter objects could remain undetectable in our imaging. Object 150000, on the other hand, has an intermediate core size and brightness but more extended low surface brightness envelope, distinguishing it from the rest of the objects in the UCD sample.

The Sersic index n measurements listed in Table 4 range from 1.7 to 7.6 with the majority having n > 4. Since typically only giant elliptical galaxies are best fit with such large n, and since R_e and n are known to be correlated, we compare those results to fits with forced n = 2, more typical of globular clusters in the Milky Way (McLaughlin et al. 2008). For most UCDs the resultant R_e was little changed. Excluding four cases, the average difference in R_e between the best fit listed in the table and a fit with forced n = 2 is only 0.1 pc. This suggests that for such small objects, the shape parameter has little effect on the fit, and cannot be well constrained. For four objects, the measured size with n = 2 was significantly smaller (163400, 151072, 150000, and 1043225 with sizes 41.1 pc, 67.6 pc, 22.7 pc, and 11.9 pc, respectively), more in line with the Ishape size measurement. In at least two of these cases, the larger R_e (and Sersic index) from the unconstrained fit is likely due to the presence of an outer envelope.

3.1.3. Ages and Metallicities

From our spectra, we measure absorption line index strengths in the Lick/IDS system and compare to single stellar population (SSP) models to derive luminosity weighted metallicity and age estimates for our confirmed UCDs. We have chosen to use the models of Schiavon (2007) and the publicly available EZ-Ages code (Graves & Schiavon 2008) for deriving the ages and abundances. An advantage of EZ-Ages is that a separate code within this package measures line indices from our spectra. These models also include the effects of non-solar abundance patterns.

Line index strengths from our LRIS blue-side spectra are measured after first degrading the spectra to the Lick resolution. These line strengths are fed into the EZ-Ages code which first determines an initial estimate for the age and metallicity from the Hβ and Fe indices using grid inversion, and then calculates the [Mg/Fe] ratio to obtain an alpha element abundance measurement. Age and metallicity are re-derived from Hβ versus Mgb, where the Mgb index is highly sensitive to alpha element abundance. If significantly different from the first age and metallicity measurements, the [Mg/Fe] ratio is increased incrementally and iterations proceed until a user supplied tolerance is reached. The process is then repeated for other Lick indices. The models span a wide range of age and metallicities, but will fail to produce derived quantities in cases where measured indices fall outside of the model range.

As initial input for the fitting process, we use a Salpeter IMF and solar scaled isochrones. Velocity dispersions are also required but due to the low resolution of our spectra, we have not attempted to measure these for our UCDs. However, velocity dispersions have been measured for a number of UCDs in the Virgo and Fornax clusters (Haşegan et al. 2005; Mieske et al. 2008b; Evstigneeva et al. 2007b; Hilker et al. 2007) and found to range between 9 < σ < 42 km s⁻¹. We therefore simply assume a small σ_v of 20 km s⁻¹. We confirm that including velocity dispersions up to 65 km s⁻¹ (the mean value found by Price et al. (2009) for seven cE galaxies and presumably a very high upper limit for our much smaller UCDs) affects derived values insignificantly within the uncertainties provided by EZ-Ages. As we did not obtain any Lick standards we cannot correct for any systematic offsets. Although the 〈Fe〉 index (an average of Fe 5270 Å and Fe 5335 Å measurements) is often used to estimate metallicity, the Fe 5335 Å line is near the edge of our blue-side spectra or missing altogether. We therefore make use only of the Fe 5270 Å line index. The strong Mgb line is used for the Mg measurements.

We initially ran EZ-Ages on our UCDs individually, but as most have S/N <15, we found results unreliable, having exceptionally large uncertainties for the derived ages and metallicities. We therefore stack multiple UCD spectra. For this to be useful, the objects must have similar stellar populations and spectral properties. Otherwise, this would only produce a spectrum with a random mix of line strengths. Therefore, we stack spectra of similar type objects based on physical properties such as color and location in cluster, and on properties of the individual line index measurements. Different sets of UCD spectra are combined. We have stacked the spectra from the five brightest UCDs, six red (V − I > 1.05) UCDs, nine blue (V − I < 1.05) UCDs, five red UCDs around NGC 4874, eight with weak Hβ(< 2.5), and eight with strong Hβ(> 2.5). Several UCDs do not have V − I measurements and are not included in the color selected samples, and we do not include any UCDs from our second observing run as the S/N was particularly low. We plot Hβ and Fe 5270 Å line index measurements from the composite spectra in Figure 15. We also include line index measurements for three individual UCDs with high S/N spectra which suggest population ages or abundances that are different from the majority of the UCDs in our sample. The uncertainties for these are large, however, with error bars which span much of the grid. For clarity, we therefore do not include error bars for these individual objects. Overlaid in the figure are grids for α/Fe = 0 and 0.3 with lines of constant age and [Fe/H].

Zoom In Zoom Out Reset image size

Spectra for non-UCDs come from Chiboucas et al. (2010). Included in this figure are measurements for elliptical, dE, and dE,N galaxies. To produce these, we have stacked spectra for 2 small elliptical galaxies, 10 low surface brightness dE galaxies, 4 dE,N with prominent nuclei, and 24 dE,N with small/faint nuclear star clusters. For the four objects with prominent nuclei, we re-extracted the two-dimensional spectra with smaller apertures meant to include light primarily from the nucleus although some contamination from the spheroidal component of these dE,N may still be present. We have also measured Fe 5270 Å and Hβ for one of two cEs for which we have obtained spectra (object 91543 in Chiboucas et al. 2010). This object was also observed with Hectospec on the MMT and line index measurements are presented in Price et al. (2009). Both measurements are included in this plot.

Table 5 presents line index measurements and derived ages and metallicities ([Fe/H]), along with [Mg/Fe], an indicator of alpha element abundance. For comparison, we include in this table [Fe/H] estimates from the V − I colors using an empirical relation based on Galactic globular clusters from Barmby et al. (2000). For the composite spectra cases, we use an average V − I color.

Table 5. Age and Metallicity

Type	N	Hβ	Fe 5270	Mgb	Age	[Fe/H]	[Mg/Fe]	V − Ia	[Fe/H](V − I)b
		(Å)	(Å)	(Å)	(Gyr)	(dex)	(dex)	(mag)	(dex)
UCDs- red	6	1.53 ± 0.23	2.40 ± 0.32	2.42 ± 0.29	14.5max5.4	−0.62max0.22	−0.100.110.12	1.18	−0.41
UCDs- blue	9	2.41 ± 0.31	0.99 ± 0.47	1.33 ± 0.36				0.95	−1.40
UCDs- bright	5	2.01 ± 0.34	1.84 ± 0.45	2.73 ± 0.40	8.7max4.6	−0.800.420.41	0.440.510.51
UCDs- red,N4874	5	1.73 ± 0.25	2.22 ± 0.33	2.60 ± 0.31	12.6max4.6	−0.640.260.26	0.140.230.23	1.18	−0.40
UCDs- strong Hβ	8	2.99 ± 0.28	1.09 ± 0.41	1.98 ± 0.32	4.01.00.7	−1.24max0.57	0.760.300.27	0.97	−1.31
UCDs- weak Hβ	8	1.20 ± 0.28	1.92 ± 0.39	2.02 ± 0.39				1.08	−0.84
UCDs- metal-rich	2	1.63 ± 0.43	3.08 ± 0.54	2.52 ± 0.52	13.2max7.7	−0.130.310.26	−0.320.210.20	1.17	−0.47
UCD/dE,N	2	2.48 ± 0.22	0.76 ± 0.31	1.03 ± 0.27	5.21.51.1	−1.020.240.12	−0.070.180.18	1.07	−0.87
dE,N(brt)	4	2.26 ± 0.19	1.85 ± 0.27	1.28 ± 0.24	6.01.91.6	−0.730.230.23	−0.130.150.16	1.08	−0.82
dE,N(fnt)	24	2.42 ± 0.14	1.84 ± 0.20	1.04 ± 0.17	4.61.21.0	−0.660.170.16	−0.240.100.10	1.06	−0.91
dEs	10	2.19 ± 0.29	1.29 ± 0.39	0.82 ± 0.33	7.43.62.0	−1.16max0.29	−0.180.210.19
Individual:
150000		3.01 ± 0.62	1.57 ± 0.85	1.82 ± 0.75	2.12.20.7	−0.400.60max	0.040.300.20
196790		1.95 ± 0.66	2.71 ± 0.91	4.18 ± 0.83	7.3max4.5	−0.080.340.25	0.240.270.27	1.25	−0.12
121666		1.84 ± 0.44	3.48 ± 0.57	4.27 ± 0.51	7.6maxmax	0.170.16max	0.020.120.12	1.17	−0.47
cE (91543)		2.08 ± 0.12	3.04 ± 0.15	4.22 ± 0.13	4.81.41.3	0.080.070.09	0.200.090.08	1.33	0.22

Notes. Where age and metallicity values are absent, this is due to measured line indices falling outside of model ranges. ^aAverage values are used for combined spectra. ^bEstimated from the relation between (V − I) color and [Fe/H] provided by Barmby et al. (2000).

Download table as:  ASCIITypeset image

Given the large uncertainties, it is clear that our age and metallicity measurements should not be taken at face value. Furthermore, we find that the addition of single UCDs in composite spectra in some cases produce a large change in measured values. However, we can draw some conclusions from this exercise. The most obvious is that the UCDs are not a homogeneous population. Rather, they exhibit a spread in age and metallicity significant at the ⩾2.5σ level. This was already expected based on the large color spread we find for these UCDs. Line index measurements confirm this range is due to real differences in stellar populations. Redder UCDs are found to be both older and more metal-rich than the bluest UCDs. Blue UCDs have on average intermediate ages and low metallicity. The majority have sub-solar metallicity, although a couple individual cases show evidence for more metal-rich stellar populations. Red UCDs around NGC 4874 are found to be no different from the overall set of red UCDs.

With the exception of a subset of red UCDs, most UCDs appear to have high [Mg/Fe] abundances, suggesting possible super-solar alpha element abundances. This would indicate that the stellar populations in UCDs formed in quick bursts of star formation. Because of the large uncertainties for this derived quantity, we caution the reader against drawing any firm conclusions. However, we note with interest that the derived values for most subsets of UCDs are similar to those of globular clusters which typically have super-solar alpha element abundances. See, e.g., Puzia et al. (2005) who find a mean [α/Fe] = 0.47 ± 0.06 with a dispersion of 0.26 dex for globular clusters in early type galaxies. The set of strong Hβ UCDs are found to be very young, metal-poor, and highly alpha element enriched. The relation between star formation time scale and alpha element abundance in Thomas et al. (2005) would suggest a very rapid burst of star formation with a timescale of only ∼0.25 Myr. The extreme values for this set of objects is likely strongly influenced by the inclusion of object ID 150000. This object appears to be surrounded by a very faint extended envelope and may not be the same class of object as these other UCDs.

Comparing dE to dE,N derived ages and metallicities, we do not find any differences within the uncertainties. Red UCDs are generally older than the dE and dE,N galaxies, significant at greater than 2σ. There is a slight hint that the UCDs are more metal-rich as well, but this is insignificant within the errors. Blue UCDs are similar to the dE and dE,N within the errors, having slightly lower metallicities and similar ages. The measured indices we find for the one cE are quite different from the UCDs, having super-solar metallicity and a young to intermediate age. Other cEs from Price et al. (2009) have metallicities as low as those found for the red UCDs.

3.2. Distribution

Looking at the spatial distribution displayed in Figure 1, we find that all confirmed UCDs lie toward the central core region of the cluster along a band in declination at δ ∼ 27.97. We do not find any members lying far from this, although it can be seen that very few UCD candidates were targeted away from this band. Three crosses (denoting background galaxies or foreground stars) lie north of 28.06 and another three south of 27.94. With these small number statistics it is hard to say too much about the true spatial distribution of the full UCD population. However, if we look at the original set of good candidates (from which we achieved a 66% confirmation success rate) in conjunction with the confirmed members, we find a pronounced linear structure running east–west (E–W) from NGC 4874, past the other central giant elliptical NGC 4889, and continuing toward IC 4051. For a comparison, we take a sample of 10,000 points distributed randomly over the fields which were available during the initial selection of candidates, assuming a uniform distribution. A two-dimensional Kolmogorov–Smirnov (K–S) test finds that the good candidates have a distribution different from a uniformly distributed sample at a 99.95% confidence while a one-dimensional K–S test finds that the candidate population differs in declination at 99.997% confidence. In contrast, the candidates have a different distribution in R.A. than the random sample at only 67.18% confidence. As it is hard to imagine how the color selection criteria could have biased the candidate list spatially, we suspect this is a real structure in the Coma cluster core. It is possible, however, that this apparent linear structure is produced primarily by UCDs associated with giant galaxies in the cluster core.

We also see what appears to be a concentration of confirmed UCDs around the cD galaxy NGC 4874. A majority seven of nine objects around NGC 4874 have red colors (V − I > 1.05), while UCDs east of this grouping have a greater percentage of blue V − I colors (at least 10 of 16 objects). We also find that all UCDs near NGC 4874 have v_r > 6800 km s⁻¹. Objects eastward of this display a larger range in measured radial velocities.

Turning to the radial velocity distribution, we note that line-of-sight velocity dispersions for UCD populations in other clusters have been found to be smaller than for other cluster galaxy types (Mieske et al. 2007a; Gregg et al. 2009; Firth et al. 2008), indicative of strong clustering. In the Coma cluster, Edwards et al. (2002) have measured velocity dispersions for the giant and dwarf galaxy populations of 979 ± 30 and 1096 ± 45 km s⁻¹, respectively. The Coma cluster has a mean radial velocity of 6925 km s⁻¹. For the faint, low surface brightness sample of 51 dwarf cluster members observed with LRIS, we find 〈v_r〉 = 6970  ±  178 with σ_v = 1269  ±  126 km s⁻¹ (Chiboucas et al. 2010), slightly higher than previous measurements. For the UCD population, we find a mean radial velocity of 〈v_r〉 = 6887 ± 207 and a velocity dispersion σ_v = 1072 ± 146 km s⁻¹, comparable to what had been found for giant and brighter dwarf galaxies, and smaller (although by <2σ) from what we find for dE and dE,N galaxies in the same region. Histograms showing the radial velocity distribution of the UCDs along with that of the normal dwarfs observed in the same masks are presented in Figure 16. Peculiar velocities of prominent galaxies are indicated. The left histogram shows that most UCDs have velocities similar to, and centered on, the cluster mean, with just a few outliers.

Zoom In Zoom Out Reset image size

The isolation in color, space, and velocity noted in the spatial distribution is also evident in the radial velocity histograms (Figure 16) where there appears to be a distinction between the UCD groups split either by color or R.A. with hints of associations with specific major galaxies. For example, 12 UCDs with R.A. <195 have 〈v_r〉 = 7296 ± 160 km s⁻¹ with σ_v = 558 ± 119 as compared to 15 objects at R.A. >195 with 〈v_r〉 = 6559 ± 325 km s⁻¹ and σ_v = 1257 ± 238, a difference of about 3σ in the velocity dispersion. Table 6 lists mean radial velocities and dispersions for different UCD subsamples.

Table 6. Velocity Distribution

Sample	Number	〈vr〉	err	σv	err
		(km s−1)		(km s−1)
Full LSB	51	6970	178	1269	126
Full UCD	27	6887	207	1072	146
(V − I) > 1.05	11	6992	379	1260	269
(V − I) > 1.10	8	6907	325	922	231
(V − I) < 1.05	14	6707	231	861	163
α < 195	12	7296	160	558	114
α > 195	15	6559	325	1257	230
N4874 proximity	9	7257	87	254	60
N4889 proximity	5	6526	148	332	105

Download table as:  ASCIITypeset image

Putting these distribution trends together in Figure 17, we show UCD projected distance from three prominent Coma giants versus radial velocity and see evidence that many of the UCDs are associated with giant ellipticals. In particular nine UCDs in velocity and spatial proximity to NGC 4874 have 〈v_r〉 = 7257 ± 87 with σ_v = 254 ± 60 km s⁻¹. This is very similar to the radial velocity, v_r = 7220 km s⁻¹, of NGC 4874 and over 3σ from the cluster mean, indicating these objects are more likely to be associated with the cD galaxy NGC 4874 than the general cluster potential. A study of Coma cluster globular clusters by Peng et al. (2011) finds that the GC population of NGC 4874 extends out to ∼130 kpc before the intracluster GC population starts to dominate. This corresponds to about 4.5 arcmin.

Zoom In Zoom Out Reset image size

We calculate whether the nine UCDs near NGC 4874 are likely to be bound to the cD galaxy. The mass of NGC 4874 has been estimated at 1.4 × 10¹³ M_☉ from X-ray observations (Vikhlinin et al. 1994). We take 1.0 × 10¹³ M_☉ as a lower limit. The escape velocity for NGC 4874 is then given by

$$\begin{equation} v_{{\rm esc}} = \sqrt{(}2{\it GM} / r_{{\rm sep}}). \end{equation} \tag{ 1 }$$

For the UCD at the largest separation from NGC 4874 (121666, 3.7 arcmin), we find v_esc = 863 km s⁻¹ at a projected distance of 108 kpc. The difference in radial velocities for the two objects is 369 km s⁻¹, lower than the escape velocity and therefore consistent with being a bound satellite. As in Firth et al. (2007), we confirm that this object also lies within the tidal radius of NGC 4874 as imposed by the other central giant galaxy NGC 4889. Vikhlinin et al. (1994) find similar masses for NGC 4889 and NGC 4874. From

$$\begin{equation} r_t = (m/3M)^{1/3} D \end{equation} \tag{ 2 }$$

(Binney & Tremaine 1987), where r_t is the tidal radius, m and M are the masses of NGC 4874 and NGC 4889, respectively, and D is the separation, we find a tidal radius of ∼5 arcmin. All nine UCDs are within this projected distance, and with a velocity dispersion of only 254 km s⁻¹, we expect these UCDs to be bound to NGC 4874.

Although clustering in both velocity and spatially around NGC 4889 is not as clear, the five UCDs in closest proximity, within 4.5 arcmin of NGC 4889 at v_r = 6495 km s⁻¹, have 〈v_r〉 = 6526 ± 148 km s⁻¹ with σ_v = 332 ± 105. This is also nearly 3σ from the cluster mean. The lack of confirmed UCDs within 2 arcmin of this giant elliptical is strictly due to the fact that the ACS field centered on this galaxy was never observed.

Three very low velocity UCDs which are spatially coincident with IC 4051 have 〈v_r〉 = 4706  ±  66 km s⁻¹ with σ_v = 114  ±  47, similar to the very low peculiar velocity of 4779 km s⁻¹ of this giant. This galaxy, which is not a central dominant giant, is remarkable for having one of the highest globular cluster specific frequencies (S_N = 12.7) in the Coma cluster (Marín-Franch & Aparicio 2002). Two of the UCDs have a fairly large projected separation from this galaxy. It is possible that while these UCDs may not be bound to this giant, they may be part of a cluster sub-structure which includes IC 4051.

In addition, two objects with particularly large radial velocities are found within 4 arcmin of IC 3998 having a similarly large radial velocity. We also identify several UCDs which are coincident with the galaxies IC 4042 and IC 4041 toward the eastern side of the cluster. As the systemic velocities for these two galaxies are similar to that of the cluster mean, it is possible that these UCDs simply belong to the general cluster potential. Searching for all possible associations of UCDs with giant galaxies, we find nearly all UCDs have potential host giant galaxies (Figure 17).

In the left panel of Figure 18, we show the cumulative distribution of confirmed UCDs as a function of distance from the nearest of one of the three brightest galaxies in the core region (NGC 4874, NGC 4889, and IC 4051). For comparison we determine the expectation for a uniform distribution. This is done by randomly distributing 25,000 points over the LRIS footprint and normalizing the resultant minimum distance cumulative distribution. We find that the UCDs are more concentrated around the giant galaxies than a spatially uniform distribution would predict. The right panel is similar but includes velocity information. In this case, the cumulative velocity difference from the nearest of one of the three giants is shown. Velocities are attached to the random sample of 25,000 points by assuming a Gaussian distribution for the velocities having 〈v_r〉 = 6925 km s⁻¹ and σ = 1000 km s⁻¹. Again we find that the UCDs are more strongly clustered around the giant galaxies. A clear 2/3 of the confirmed UCDs are associated with one of 3 central giants.

Zoom In Zoom Out Reset image size

With the exception of the two central cD/giant ellipticals, other UCDs may not be bound to their respective neighboring galaxies. However, if this is the case, the velocity and location groupings we find suggest that either these are previously bound clusters in the process of dissolution to the cluster potential, or that they belong to substructures present in the Coma cluster.

From an initial observing run, we identified 19 members selected primarily based on color along with some information on profile size. For targets with R < 23.3, our completeness limit, we had an unexpectedly high 66% success rate for these initial observations. A follow-up run with an expected lower confirmation rate due to intentionally loose selection criteria confirmed a further eight UCDs from poor weather observations. These results suggest that the Coma cluster harbors a large population of UCDs, at least in the core region, although this high success rate must be due at least in part to the fact that we can marginally resolve these objects with the superb resolving capabilities of HST/ACS.

The spatial distribution of UCDs are compared to that of the Coma cluster globular cluster and dE,N populations in Figure 19. Candidate globular clusters (Peng et al. 2011) are mapped as points in the top plot. There is a clear concentration of globular clusters visible around NGC 4874. An overdensity is also apparent surrounding the missing ACS field that would have included NGC 4889. On the eastern side of the cluster, there is a slight overdensity at δ = 28.02 deg near IC4051 which lies just off our ACS footprint. Intracluster globular clusters not associated with individual galaxies are spread throughout the core region. After smoothing the globular cluster distribution, Peng et al. (2011) find an excess which forms a band through the core between NGC 4874 in the west to IC 4051 and NGC 4908 in the east. Similarly, the UCDs confirmed to date exhibit a concentration around NGC 4874 along with an E–W flattened distribution through the core forming a linear structure in the same region as this apparent band of globular clusters. Considering candidate UCDs lends further support for the presence of a band-like structure running through the core. However, with the small number statistics for UCDs, it is difficult to tell if this is a real structure, or if the UCDs are simply following the distribution of more massive galaxies in the core region.

Zoom In Zoom Out Reset image size

Meanwhile the dE,Ns exhibit a slight excess near NGC 4874 and NGC 4889, but otherwise appear to be spread uniformly through the ACS footprint, with no indication of either an excess or deficit in the eastern side of the core where the remainder of the UCDs have been confirmed. In Figure 20, we show the cumulative distribution of confirmed and candidate UCDs, dE,N, and GCs as a function of projected distance from NGC 4889. For comparison, we take a sample of 10,000 points distributed randomly over the observed core ACS footprint. The UCDs are more centrally concentrated than any other population. The dE,N are also more concentrated than expected for a uniform distribution, while the globular clusters have a shallower distribution until ∼8 arcmin in distance from NGC 4889 most likely due to the large concentration around NGC 4874. However, taking the cumulative distributions as a function of distance in declination only, we find that both GCs and UCDs are much more concentrated toward the central declination, while dE,N follow the uniform distribution. This difference between the populations is greater when considering declination only, most likely because of the strong clustering of GCs and UCDs around the central giants, and perhaps due to the excess of compact systems noted along a band in declination.

Zoom In Zoom Out Reset image size

In future work, we plan to do more detailed photometric, structural, and distribution comparisons between Coma UCDs and dE,N galaxies. For now, we use the luminosity relation found by Côté et al. (2006) between Virgo dE,N and their respective nuclei to compare the magnitudes of UCDs with those expected for the Coma dE,N nuclei. We find that within the Coma footprint, there are only ∼20–30 dE,N within the magnitude range 15 < F814W < 20 which could host nuclei with magnitudes in the range 21 < F814W < 23.5. We have confirmed 27 UCDs and identified over 100 other candidates. We expect the ratio of UCDs/N (21 < F814W < 23.5) to be at least 2–3. If we assume that all UCDs are produced from threshing of dE,N, the efficiency of stripping within the Coma core region must be very high. However, given the requirement for highly eccentric orbits, threshing models do not predict such high efficiency rates (Bekki et al. 2003). In this scenario, one might also expect to find the number of dE,N to be depleted toward the core of the cluster. Instead, the dE,N are found to have a fairly uniform distribution with perhaps a slight excess in the central core.

The confirmed UCDs display a wide range in color extending to exceptionally red objects. One explanation for the extreme red colors is that a dE,N which initially lies along the red sequence undergoes threshing and is stripped of at least 4 mag of material by the tidal field of the cluster without affecting the color of the remnant nucleus. We find evidence for an alternative explanation when we compare UCD and globular cluster colors. Globular clusters exhibit a well-known bimodal color distribution likely due to a metallicity bimodality (West et al. 2004). This generates a large range in globular cluster colors. In Figure 21, we display the g − I color distribution of the confirmed UCDs with that of the Coma cluster globular cluster candidates from Peng et al. (2011). Small circles represent globular clusters from visit 19, the field which includes NGC 4874. A histogram of globular clusters fainter than I > 24.7 contains only one obvious peak around g  −  I = 0.95. If we look only at visit 19 globular clusters brighter than I < 24.7, the second peak becomes apparent. The histograms in this figure are scaled to fit within the boundaries of this plot. The total number of globular clusters represented by this lower histogram is ∼500. We fit a double Gaussian to the lower histogram and find peaks at 0.93 and 1.17 with σ = 0.09 for both. For the full sample of GCs, Peng et al. (2011) find peaks at ∼0.9 and 1.15. For inner GCs around NGC 4874 they find a slightly redder blue peak at 0.94. The UCDs are found to span the same wide range in color as the GCs. With so few confirmed UCDs, it is difficult to determine whether they also exhibit a bimodality in color. To calculate average colors, we assume the Gaussian distributions for the globular clusters in order to assign a weight to each UCD for grouping with the red and blue distributions. We find weighted averages of blue and red objects of g − I = 0.95 and 1.15, very similar to those of the globular cluster population. Since we are assuming the same distribution, we cannot argue that they share this bimodality. However, this does indicate that the UCD distribution is not inconsistent with that of the globular clusters.

Zoom In Zoom Out Reset image size

One secure conclusion that we can draw from this plot is that there is no discontinuity between the two populations; the UCDs simply extend to brighter magnitudes while exhibiting the same spread in color as GCs. In fact, due to the lack of any discontinuity, the Peng et al. (2011) globular cluster catalog includes many of our confirmed and candidate UCDs. It is worth noting that the red GC peak is more prominent when considering bright GCs into the range of UCDs in the field around NGC 4874, where we also find a large fraction of red UCDs. The possible extension of globular clusters to brighter magnitudes has also been observed for the cD galaxy NGC 3311 in Hydra where the red globular cluster sequence is seen to extend upward in luminosity into the range of UCDs (Wehner et al. 2008).

Both NGC 4874 and IC4051 harbor large globular cluster populations, with S_N ∼ 9.0 and 12.0, respectively (Marín-Franch & Aparicio 2002; Harris et al. 2009). We find a large population of UCDs that is almost certainly bound to NGC 4874 and another group of three UCDs likely to be associated with IC 4051. Unfortunately, this latter galaxy lies east of our ACS footprint. Confirmation of a large UCD population around this non-central, non-cD galaxy would lend strong support for a star cluster origin.

We test whether the UCDs can be accommodated by the bright tail of the NGC 4874 globular cluster distribution. Gaussian fits to Peng et al. (2011) GC candidates in visit 19 find I_peak = 26.2 with σ = 1.2. However, incompleteness sets in before the peak. We therefore take the globular cluster luminosity function (GCLF) values found recently for M87 in Virgo, I_{AB, peak} = 26.9 with σ = 1.37 (Peng et al. 2009, 2011). This produces fewer counts than UCD candidates at the very bright end by >3σ for F814W < 22. Fainter than F814W > 22 (M_F14W > −13) the UCD population is fully consistent with being drawn from a Gaussian distribution of GCs. In fact, many of the UCD candidates are included in the GC candidate list. We also take the M87 form of the GCLF and randomly distribute in magnitude according to this Gaussian distribution the ∼3000 globular cluster candidates found in the same ACS field as NGC 4874. From 1000 simulations, we find that the brightest expected globular cluster is M_I = −13.0 ± 0.4. The brightest confirmed UCD in this field has M_F814W = −13.2, consistent with the expectations.

We compare the sizes of the Coma cluster UCDs with other compact objects in Figure 22. Globular clusters in the range −10 < M_V < −8 have nearly constant half-light radii ∼2.5 ± 1.5 pc independent of luminosity. Faintward of this, the observed sizes increase, with effective radii as large as 20 pc being found for clusters as faint as −5. Brightward of M_V = −10, the few very bright globular clusters start to increase in size with increasing luminosity. These objects merge seamlessly into the range of the so-called UCDs which display a trend of increasing size and decreasing surface brightness with increasing luminosity. The Coma UCDs follow the same luminosity–size relation as found for Virgo and Fornax UCDs. Compact ellipticals and normal dEs are found toward the bright extension of this sequence, although cEs would fall off toward higher surface brightnesses while dEs have lower surface brightness. Nuclei of early type galaxies in the Virgo cluster tend to have smaller sizes at a given magnitude than the UCDs (Côté et al. 2006). This was also noted for nuclear star clusters in low mass dwarf galaxies by Georgiev et al. (2009), who suggest that nuclei may expand in size upon being freed from the strong gravitational potential within a galaxy, e.g., in the case of galaxy threshing.

Zoom In Zoom Out Reset image size

A transition between objects which have sizes independent of magnitude and those which follow a magnitude—size relation has been previously noted at M_V ∼ −11. This transition occurs at the same magnitude as a break in the metallicity distribution (Mieske et al. 2006). It furthermore corresponds to a mass of ∼2.5 × 10⁶ M_☉ where a break in the log σ–log M scaling relation has been noted between globular clusters and larger systems. Higher mass-to-light ratios, in the range 6 < M/L_V < 9, have been found for objects above this transition magnitude (Haşegan et al. 2005). As these cannot be explained with canonical IMFs and baryonic matter, they suggest the fundamental difference between globular clusters and UCDs is the presence of dark matter. This supports the threshing model in which remnant cores of threshed dE,N retain some of their dark matter halo. The nuclei themselves may have formed from the coalescence of globular clusters through orbital decay within dwarf galaxies (Bekki et al. 2004; Lotz et al. 2002). This would produce complex stellar populations and a range of colors/metallicities, along with larger sizes and masses consistent with what is found for massive globular clusters and perhaps UCDs (Georgiev et al. 2009; Hilker 2006).

However, another possibility for the larger sizes and velocity dispersions of the UCDs is a difference in physics at the time of formation. Murray (2009) shows that star clusters above M ⩾ 10⁶ M_☉ would be optically thick to IR radiation at the time of formation, and that the balance between radiation pressure and gravity sets up a size—mass relationship. If the break in the IMF is set by the Jeans mass, they argue that optically thick clusters will have top heavy IMFs. Top heavy IMFs could be the origin of the high M/L ratios, since massive stars will age quicker and leave more stellar remnants (Dabringhausen et al. 2009).

Taken together, the properties of the Coma cluster UCDs appear to point toward a star cluster origin for a majority of these objects. The UCDs exhibit strong spatial and velocity correlations with the major galaxies in the core. In particular a large fraction of the UCDs reside within the halos of the massive galaxies, at least in the case of the two cD/giant ellipticals in the cluster core. Other UCDs may be associated with some of the other giant galaxies in the core region. The UCDs are furthermore correlated in color and metallicity with the host galaxy. We find a large population of red UCDs around NGC 4874 and a bluer population around NGC 4889. Unfortunately due to the missing ACS field, we do not know whether a radial gradient in color/metallicity could be present around NGC 4889. The range of UCD colors is identical to that of globular clusters and there is no evidence for discontinuity between the two populations in luminosity. The spatial distribution shows remarkable similarities to globular clusters with a large number of confirmed and candidate UCDs found around the same galaxies hosting large globular cluster populations. Both UCDs and globular clusters also show some evidence for a structure of compact objects running E–W through the core region.

From the age–metallicity diagram (Figure 15), we find that many of the UCDs are not extremely old, and thus not primordial in origin. Red UCDs are more metal-rich than blue ones and must have undergone self-enrichment or been formed from pre-enriched gas. Red, metal-rich globular clusters are more often found around massive galaxies (Peng et al. 2006; Harris et al. 2009). The luminosity of globular clusters in a system also correlates with the luminosity of the host galaxy, with brighter galaxies hosting populations with brighter globular clusters (Hilker 2009). If UCDs simply extend the globular cluster sequence to brighter magnitudes, it then follows that UCDs should be found around the brightest and most massive galaxies such as the central Coma cD/giant elliptical galaxies and should include a significant number of red objects, as we find.

The commonality in UCD color/metallicity properties with location could be a consequence of the formation of UCDs (and GCs) in a small number of discrete star formation events, each event characterized by the metallicity of gas available for star formation. These star formation events could be induced by, e.g., cataclysmic wet mergers at early times. Since each galaxy has a unique history it follows that the UCD and GC populations should vary from galaxy to galaxy. The affiliation by color with host suggests that the UCDs were born in a single event, or only a couple events.

Some of the blue UCDs are associated with NGC 4889. Others may follow the globular cluster intracluster distribution noted by Peng et al. (2011). Some of these blue intracluster globulars may be metal-poor clusters stripped from infalling dwarf galaxies. In this scenario, it is possible that the largest compact stellar systems may be the remnant nuclei of some of these disrupted dwarf galaxies. However, Peng et al. (2011) find ∼20% of intracluster globular clusters are red and suggest that at least part of this intracluster population may be stripped GCs from the halos of massive galaxies. In this scenario, the UCDs could also simply be giant globular clusters stripped from these more massive galaxies.

The findings so far cannot rule out threshing as an origin for UCDs. The threshing model predicts that UCDs will be found in greater abundance and have a distribution concentrated toward the center of the cluster potential and around cluster super giant galaxies (Bekki et al. 2003; Bekki 2007; Thomas et al. 2008). This is generally what we find, although detailed modeling would be needed to understand the E–W linear structure of compact objects if real. The differentiation of color by host is more difficult to explain within the framework of a threshing scenario. One possible explanation is if the color variations we find are due to radial gradients in metallicity in which more metal-rich dE,N are found closer to the central giants (Mieske et al. 2006), thereby producing an excess of metal-rich UCDs with smaller orbital radii around the central giants. Another possibility for explaining the large fraction of red UCDs around NGC 4874 is if more massive (and by implication, more metal-rich) galaxies require smaller pericenter radii and interaction with more massive galaxies for efficient stripping. Since the sizes of dE,N nuclei are known to scale with the brightness of the host dwarf galaxy, we would then expect that larger UCDs would be redder and that these would be found preferentially close to the central giants. From Figure 23 we see only hints of such trends. After two outlying UCDs are excluded, we find that larger objects tend to be redder, but with a correlation coefficient of only 0.36. We also find that larger objects tend to be closer to NGC 4874, but this trend is also very weak, with a correlation coefficient of −0.26. No correlation is found when comparing UCD size with distance from NGC 4889, or with the distance from the closest of either NGC 4889 or NGC 4874, although the missing ACS panel centered on NGC 4889 may be the cause of this.

Zoom In Zoom Out Reset image size

Some similarities between the UCDs and cE galaxies, in particular in the color and surface brightness characteristics, are seen. There may be some correlation in spatial distribution as well, although larger numbers of cEs would be needed to establish this. Since cE galaxies are expected to form from tidal stripping of more massive galaxies, a relation between the two populations would suggest a common threshing origin. However, we have thus far been unable to fill a ∼1.5 mag luminosity gap that currently separates these two object types.

We have identified a few confirmed UCDs which may have very faint, low surface brightness envelopes. One of these, object 151072, has obvious surrounding structure, the largest measured size of our confirmed UCDs, and a fairly blue color. Another object, 150000, has a hint of a large very low surface brightness envelope and a young derived age. Since it is expected that destructive processes such as threshing should play a role in rich clusters like Coma it might be expected that at least some fraction of these objects are remnants from galactic stripping. Norris & Kannappan (2011) make a statistical argument for a bright upper limit for star cluster formation. They suggest that compact objects with M_V > −13 may be a mixture of objects formed through star cluster formation processes and those formed via a stripping mechanism. Compact objects brighter than this limit cannot be formed as star clusters or through the merger of star clusters simply because there are no globular cluster systems which are abundant enough to be populated that far into the bright tail of the GCLF. Object 151072 is our brightest UCD with M_V = −12.9, close to this suggested limit. Two other very bright UCDs also display hints of surrounding structure. Object 150000, while not particularly bright, does not resemble Coma cluster GCs in stellar population or structural properties. Thus, although our findings suggest that the majority of Coma cluster UCDs have a star cluster origin, several of the confirmed UCDs which have slightly different properties may in fact have a galactic origin. Indeed, a number of other recent studies have also found evidence for multiple formation channels for UCDs (Hilker 2006; Da Rocha et al. 2011; Taylor et al. 2010; Chilingarian et al. 2011; Norris & Kannappan 2011), all finding both cases where a star cluster origin is preferred and cases where stripping of a more massive galaxy is the more likely origin.

The results presented here are based on UCDs located within the core of the Coma cluster. In future work it will be important to search outer regions as well, to completely define the spatial and velocity distribution and range of UCD properties. Future work will discuss the full candidate population and compare in greater detail to the properties of Coma cluster dE,N.

Using LRIS on Keck, we have spectroscopically confirmed 27 UCDs within the core region of the Coma cluster. With an initial UCD confirmation success rate greater than 60% for M_F814W < −11.7, we believe that the rich, dense, evolved environment of the Coma cluster core hosts a large population of UCDs. However, the high success rate is also due in part to the fact that UCDs are marginally resolved in ACS imaging. We find properties of the UCD population consistent with what has been found for UCDs in other clusters, having the same range in luminosities, sizes, colors, and metallicities and having similar distributions.

The confirmed Coma cluster UCD population has magnitudes M_I > −13.5, colors 0.8 < (g − I) < 1.3, sizes primarily in the range 7–40 pc, and metallicities −1.3 ≲ [Fe/H] ≲ −0.6. They are distributed through the central core between NGC 4874 in the west and IC 4051 in the east. We find strong spatial and velocity correlations with the major cluster galaxies. A subset of UCDs is almost certainly bound to cD galaxy NGC 4874. Other UCDs may be associated with the other central giant, NGC 4889, and with some of the other giants in the core region. Notably, the UCDs also exhibit color/metallicity correlations with location in the cluster. NGC 4874 hosts a large population of red, metal-rich UCDs while NGC 4889 appears to host primarily blue UCDs, although a radial gradient cannot be ruled out. There is also a subset of blue UCDs which may lie in the intracluster region. The affiliations with host by color suggests formation in discrete star formation events with metallicity determined by that of the gas pool available during such an event, e.g., during the same early time cataclysmic wet mergers that are believed responsible for producing globular cluster populations.

We suggest that these objects could be related to globular clusters. Not only do they share similar colors and lie along the extension of the bright tail of the GCLF, but they also have a similar distribution in the cluster. NGC 4874 has a very high GC specific frequency and hosts a significant UCD population as determined by the large number of both confirmed and candidate UCDs surrounding this galaxy. IC 4051 is another interesting galaxy lying east of our ACS survey region having an unusually high GC specific frequency and a non-central location in the cluster. We have identified several UCDs associated with this galaxy and find a slight excess nearby in our full candidate list. Confirmation of a large UCD population around IC 4051 would provide further evidence for a GC–UCD relationship. Intracluster GCs have been found distributed throughout the core region with a possible excess running in an E–W band. We find a similar structure with our confirmed UCDs. Compared to dE,N, possible progenitors in a threshing scenario, the UCDs are more concentrated around massive galaxies while dE,N are dispersed through the core.

Although most UCDs appear to be oversized globular clusters, a few of these objects exhibit more evidence for a threshing origin. With hints of diffuse surrounding envelopes, bluer colors, younger ages, and/or exceptionally large sizes, these objects may in fact be remnant nuclei from stripping of dE,N or more massive galaxies. Since destructive processes are rampant in dense clusters, it would almost be surprising if such stripping never occurred.

A larger sample is required to fully characterize the properties of these objects and probe the full spatial and velocity distribution of this population. Establishing the origin of these objects would provide important clues for understanding galaxy and cluster evolution. Objects which are remnant nuclei from a previously larger population of dE,N would be useful probes for understanding cluster destructive processes and could help mitigate the missing satellite problem. Objects which prove to be star clusters would be valuable probes of galaxy merger and cluster merger and dynamical histories.

We thank the anonymous referee for helpful suggestions which have improved this paper. Based on observations made with the NASA/ESA Hubble Space Telescope, obtained at the Space Telescope Science Institute, which is operated by the Association of Universities for Research in Astronomy, Inc., under NASA contract NAS 5-26555. These observations are associated with program GO10861. Support for program GO10861 was provided by NASA through a grant from the Space Telescope Science Institute, which is operated by the Association of Universities for Research in Astronomy, Inc., under NASA contract NAS 5-26555. Some of the data presented herein were obtained at the W. M. Keck Observatory, which is operated as a scientific partnership among the California Institute of Technology, the University of California and the National Aeronautics and Space Administration. The Observatory was made possible by the generous financial support of the W. M. Keck Foundation. The authors recognize and acknowledge the very significant cultural role and reverence that the summit of Mauna Kea has always had within the indigenous Hawaiian community. We are most fortunate to have the opportunity to conduct observations from this mountain.