Structural Properties of Ultra-Compact Dwarf Galaxies in the Fornax and Virgo Clusters
E.A. Evstigneeva, M.J. Drinkwater, C.Y. Peng, M. Hilker, R. De Propris, J.B. Jones, S. Phillipps, M.D. Gregg, A.M. Karick
aa r X i v : . [ a s t r o - ph ] A p r Draft version October 24, 2018
Preprint typeset using L A TEX style emulateapj v. 03/07/07
STRUCTURAL PROPERTIES OF ULTRA-COMPACT DWARF GALAXIESIN THE FORNAX AND VIRGO CLUSTERS
E. A. Evstigneeva, M. J. Drinkwater
Department of Physics, University of Queensland, QLD 4072, Australia; [email protected], [email protected]
C. Y. Peng
NRC Herzberg Institute of Astrophysics, 5071 West Saanich Road, Victoria, BC, Canada V9E 2E7; [email protected]
M. Hilker
European Southern Observatory, Karl-Schwarzschild-Str. 2, 85748 Garching bei M¨unchen, Germany; [email protected]
R. De Propris
Cerro Tololo Inter-American Observatory, Casilla 603, La Serena, Chile; [email protected]
J. B. Jones
Astronomy Unit, School of Mathematical Sciences, Queen Mary University of London, Mile End Road, London E1 4NS;[email protected]
S. Phillipps
Astrophysics Group, Department of Physics, University of Bristol, Tyndall Avenue, Bristol BS8 1TL; [email protected] andM. D. Gregg , A. M. Karick Department of Physics, University of California, Davis, CA 95616, USA; gregg,[email protected]
Draft version October 24, 2018
ABSTRACTWe present a detailed analysis of high-resolution two-band
Hubble Space Telescope
Advanced Camerafor Surveys imaging of 21 ultra-compact dwarf (UCD) galaxies in the Virgo and Fornax Clusters. Theaim of this work is to test two formation hypotheses for UCDs—whether they are bright globularclusters (GCs) or stripped (“threshed”) early-type dwarf galaxies—by direct comparison of UCDstructural parameters and colors with GCs and galaxy nuclei. We find that the UCD surface brightnessprofiles can be described by a range of models and that the luminous UCDs in particular can notbe described by standard King models with tidal cutoffs as they have extended outer halos. This isnot expected from traditional King models of GCs, but is consistent with recent results for massiveGCs. The total luminosities, colors and sizes of the UCDs (their position in the color-magnitude andluminosity-size diagrams) are consistent with them being either luminous GCs or threshed nuclei ofboth early-type and late-type galaxies (not just early-type dwarfs). For the most luminous UCDs weestimate color gradients over a limited range of radius. These are systematically positive in the senseof getting redder outwards: mean ∆( F W − F W ) = 0 .
14 mag per 100 pc with rms = 0 .
06 magper 100 pc. The positive gradients found in the bright UCDs are consistent with them being eitherbright GCs or threshed early-type dwarf galaxies (except VUCD3). In contrast to the above resultswe find a very significant ( > .
9% significance) difference in the sizes of UCDs and early-type galaxynuclei: the effective radii of UCDs are 2 . +0 . − . times larger than those of early-type galaxy nuclei atthe same luminosity. This result suggests an important test can be made of the threshing hypothesisby simulating the process and predicting what size increase is expected. Subject headings: galaxies: clusters: individual: Fornax Cluster – galaxies: clusters: individual: VirgoCluster – galaxies: star clusters – galaxies: dwarf – galaxies: fundamental param-eters – galaxies: structure – galaxies: formation INTRODUCTION
Ultra-compact dwarf (UCD) galaxies are a class of stel-lar system originally discovered in the Fornax Cluster(Hilker et al. 1999, Drinkwater et al. 2000). They ap- Institute for Geophysics and Planetary Physics, Lawrence Liv-ermore National Laboratory, L-413, Livermore, CA 94550, USA pear star-like in ground-based photographic survey im-ages, but have recession velocities consistent with clustermembership. UCDs have spectra typical of old stellarpopulations but they are generally far more luminousthan ordinary Milky Way (MW) globular clusters (GCs),and much more compact than similarly luminous dwarfspheroidal galaxies. Evstigneeva et al.UCDs have now been detected in the Virgo (Ha¸seganet al. 2005, Jones et al. 2006) and Centaurus Clusters(Mieske et al. 2007) and, possibly, in Hydra I (Wehner &Harris 2007) and Abell 1689 (Mieske et al. 2004). On theother hand, they are much less common in groups (e.g.Evstigneeva et al. 2007a) and the general field (Liskeet al. 2006), suggesting that UCDs are somehow relatedto the cluster environment. We have now surveyed thecentral region of the Fornax Cluster for less luminousobjects: we find that the central region of the clustercontains a large population of UCDs (60 objects to amagnitude limit of b J = 21 . M V ∼ − . HubbleSpace Telescope (HST) “snapshot” program, allowing usto obtain images of a statistically useful subset of ourentire sample. Our primary aim was to test the first andthe third hypotheses above by direct comparison of UCDstructural parameters with GCs and galaxy nuclei alsomeasured by HST.The structure of the paper is as follows. In Section 2we describe the data sample and in Section 3 we describethe HST imaging and image modeling. In Section 4 wepresent analysis of the structure and colors of the mostextensive and complete sample of UCDs in the Fornaxand Virgo Clusters observed to date. The sample in-cludes six Virgo UCDs initially presented by Evstigneeva et al. (2007), as we have made several improvements tothe image analysis. In Section 5 we summarize our re-sults and findings. Throughout this paper we adopt dis-tance moduli of 30.92 for the Virgo Cluster and 31.39 forthe Fornax Cluster (Freedman et al. 2001), correspond-ing to the distances of 15.28 Mpc to Virgo and 18.97 Mpcto Fornax. UCD GALAXY SAMPLE
In this section we briefly describe the properties we useto define UCD galaxies and how we obtained the samplefor our HST observations. More details are given in therespective discovery papers listed below.
Definition of the UCD galaxy type
As noted above, the UCD galaxies originally discoveredin the Fornax Cluster were unresolved in ground-basedphotographic imaging, but had redshifts consistent withcluster membership. The first UCDs found were muchmore luminous than any known GCs, with apparent mag-nitudes in the range 17 . < b J < . − . < M V < − .
9. At these luminositiesthe UCDs were clearly distinct from any known galacticor stellar system (Drinkwater et al. 2003). Note that,although the “unresolved” criterion depends on the im-age quality, there is a very large gap in parameter spacebetween the UCDs and normal dwarf galaxies, so it onlyserves to remove clearly normal galaxies from the sam-ples. We have since extended our searches to fainter lim-its ( b J < . M V < − . UCD targets
As noted above, our HST snapshot proposal was moti-vated by the availability of a large sample of UCDs with arange of luminosities from the Fornax and Virgo Clusters.Although the UCDs were originally discovered (Drinkwa-ter et al. 2000) through the “all-object” approach of theFornax Cluster Spectroscopic Survey (Drinkwater et al.2000a), our subsequent UCD searches were more selec-tive. In particular we imposed color selection to avoidthe reddest stellar objects as none of the original UCDswere this red.In the Virgo Cluster we carried out a targeted searchspecifically aimed at detecting luminous
UCDs with sim-ilar properties to the originally discovered Fornax UCDs.This was very successful with 9 UCDs detected in just afew hours of observing time (Jones et al. 2006). TheVirgo objects were selected in the magnitude range16 . < b J < . − . < M V < − .
1) andthe color range b J − r F < . V − I < . ∼
270 kpc) radius regionof the Virgo Cluster to be about 14.tructural Properties of UCDs 3In the Fornax Cluster we extended the search to muchfainter limits using very similar approach. In our fi-nal observations with the 2dF system we selected unre-solved objects in the magnitude range 16 . < b J < . − . < M V < − .
3) and the color range b J − r F < . V − I < .
6) and we limited thearea searched to a 0.9-degree radius from the center of theFornax Cluster. This approach was again very successfulwith a total of 60 UCDs detected in these limits (Gregget al. 2008). Allowing for the incompleteness of our spec-troscopic observations, we estimate the true populationof UCDs in the central 0.9-degree ( ∼
300 kpc) regionof the Fornax Cluster to be about 105. Note that thisis to a fainter limit ( b J < .
5) than our Virgo sample( b J < . HST OBSERVATIONS AND IMAGE MODELING
We obtained images of 21 of the requested Fornax andVirgo UCDs in the course of HST snapshot program10137. The data were taken with the Advanced Cam-era for Surveys (ACS), High Resolution Channel (HRC),through the F606W and F814W filters. Exposure timeswere 870 sec in F606W and 1050 sec in F814W. The HRCscale is 0 . ′′ pixel − . For the image analysis we used MultiDrizzle (*.mdz) files retrieved from the HSTarchive.To measure the total magnitudes, we plotted curves ofgrowth (integrated magnitude vs. circular aperture ra-dius) to find an aperture radius large enough to encloseall the light from an object. The instrumental F606Wand F814W magnitudes were transformed into Landolt V and I band following Sirianni et al. (2005). The result-ing V magnitudes and V − I colors are listed in Table 1.The images of Fornax and Virgo UCDs were modeledusing the two-dimensional fitting algorithm GALFIT (Peng et al. 2002) and assuming empirical King, S´ersicand Nuker models for the luminosity profile.The empirical King profile is characterized by the coreradius, R c , and the tidal radius, R t , and has the followingform (Elson 1999): I ( R ) = I (cid:20) R/R c ) ) α − R t /R c ) ) α (cid:21) α , (1)where I is the central surface brightness. We tried boththe standard model with α = 2 and generalized modelwith variable α . See http://stsdas.stsci.edu/multidrizzle.
The S´ersic power law has the following form (S´ersic1968): I ( R ) = I eff exp " − k (cid:18) RR eff (cid:19) n − ! , (2)where R eff is the half-light (effective) radius, I eff is sur-face brightness at the effective radius, n is the concen-tration parameter ( n = 4 for de Vaucouleurs profile and n = 1 for exponential profile) and k is a constant whichdepends on n .The Nuker law is as follows (Lauer et al. 1995): I ( R ) = I b β − γα (cid:18) RR b (cid:19) − γ (cid:20) (cid:18) RR b (cid:19) α (cid:21) γ − βα . (3)It is a double power law, where β is the outer power lawslope, γ is the inner slope, and α controls the sharpness ofthe transition (at the “break” radius R b ) from the innerto the outer region. I b = I ( R b ).The UCDs are barely resolved—even with theHST/ACS resolution—so to obtain their intrinsic lumi-nosity profiles, we must correct for the telescope point-spread function (PSF). We derived artificial PSFs forthe images in each filter using the TinyTim software and MultiDrizzle as described in Evstigneeva et al.(2007). The size of the PSFs was chosen to be 3 . ′′ × . ′′ (140 ×
140 pixel), slightly larger than the minimum sizerecommended by
TinyTim . For the brightest UCDs,which are also the most extended, we used PSFs of alarger size (the same as for the color gradient analysisin Section 4.3.1) to correct for the extended PSF halo inthe F814W filter.
GALFIT models an analytic profile convolved withthe PSF and determines the best-fitting profile parame-ters by minimizing residuals between the model and orig-inal two-dimensional image. We limited all the models toa maximum fitting radius defined by the point where theUCD light profile reaches the background noise level inthe V image (the models are not constrained beyond thatpoint). The image of Fornax UCD3 is complicated by thepresence of a background object in projection (possibly aspiral galaxy, see Evstigneeva et al. 2007). We thereforerestricted the model fitting to the half of the image leastaffected by the background source.The sky (background) was estimated and subtractedfrom the UCD images before running GALFIT . The skywas initially subtracted by
MultiDrizzle . We thenapplied additional background corrections, determinedfrom empty parts of the images. So we held the sky valuefixed to zero when fitting the images. It is important tohold the sky fixed when fitting standard models to anobject, because the model function used may not be op-timal and the model mismatch can push the sky arounda little. On the other hand, if the object is fitted as wellas possible (by multi-component models) and if the skyregion is large enough to fit, then the sky can be allowedto vary as a free parameter. This is what we did for thecolor gradient analysis in Section 4.3.1. Thus, in this sec-tion, for the structural modeling, we held the sky valuefixed (did not fit the sky with
GALFIT ). We, however,did the tests of changing the sky by hand (subtracting Evstigneeva et al.the sky values found by GALFIT in Section 4.3.1) andredoing the fit. It did not affect the structure parame-ters in Tables 1 & 2 very much: the changes were withinthe uncertainties given in the tables. However, the skycorrections can be critical for the outer color profile, forthe tiny color gradients we find in UCDs.The quality of the
GALFIT model fits is shown inFigure 1. For this figure we used the
ELLIPSE taskin
IRAF to produce one-dimensional surface brightnessprofiles for the objects and (PSF-convolved)
GALFIT models.To choose the best model for each object (see the lastcolumn of Table 2), we used χ ν values of the fits (Peng etal. 2002). In the case of faint UCDs ( m V, ∼ . − . m V, ∼ . − . α = 2 or S´ersic and the outer envelope byS´ersic). This does not necessarily mean that bright andfaint UCDs are intrinsically different. If we had deeperobservations for the fainter UCDs, we would possibly beable to detect outer halos in them were any present. Thedetection of extended halos around the luminous UCDs isnot consistent with their having the standard King pro-files traditionally associated with GCs, but recent workof McLaughlin & van der Marel (2005) and McLaughlinet al. (2008) has shown that extended halos are a gen-eral characteristic of massive GCs in the Milky Way andsome of its satellites and NGC5128.In Table 1 we quote effective radius R eff , model magni-tude M mod V, and ellipticity ǫ . The ellipticity value is thebest model value (see last column of Table 2). The R eff and M mod V, values were obtained from generalized King(or standard King with α = 2, if it fits better) modelsfor one-component UCDs and King+S´ersic models fortwo-component UCDs, via numerical integration (to in-finity) of the V and I luminosity profiles. These modelsgive the most stable estimates for R eff as discussed inEvstigneeva et al. (2007). The M mod V, values are onlyslightly different from the observational M V, values, ob-tained by integrating the actual image pixel values. Infurther analyses we use observational magnitudes ( M V, ).The choice of magnitude, however, does not change ourfinal results and conclusions (e.g. equations (4)–(7) staythe same, as well as the size difference between UCDs andnuclei, found below). Table 1 also contains the parame-ters for the four bright Fornax UCDs from Evstigneevaet al. (2007) (HST/STIS data), as we have made someimprovements to the image analysis and we use thesedata in the analyses below.In Table 2 we list more model parameters such as S´ersicindex n ; King central surface brightness µ V, , core radius R c , concentration c and α parameter; Nuker inner slope γ , outer slope β and break radius R b . We do not listall the parameters for all the models, but only the mostsignificant ones needed for qualitative analysis, mainly tocompare UCDs with each other. We recommend cautionin using the actual values of these parameters in detailedanalyses. One of the main reasons for this is that theobjects are barely resolved. For example, for about half of the UCDs, the King model fits appear good, but thecore radii of these models R c < R c valuesare uncertain. As a result, we can not trust the valuesof central surface brightness and concentration obtainedfrom these King models. The central surface brightnessof UCDs can also be derived from S´ersic models: we cancalculate µ V, from GALFIT ’s m V,tot , R eff and n valuesanalytically by using formulae for the S´ersic function.However, for models with high S´ersic indices ( n > ANALYSIS AND DISCUSSION
Structural parameters
The results of the modeling show that the UCDs havea range of S´ersic indices n and King concentrations c ,as well as a range of central slopes (seen from Nuker in-ner slopes γ ): from flat “King” cores to central cusps.This suggests that convolving some cuspy models withthe PSF allows such models to fit the seeing-blurred cen-ters of some UCD profiles. We, however, can not makevery strong conclusions regarding central cusps or coresin the UCDs, taking into account the limits of our data(described in the previous section).We tried to look for correlations of the model parame-ters in Table 2 with the UCD luminosity, size and color,but did not find any.We compared the distribution of UCD ellipticities(in Table 1) with those for MW GCs (Harris 1996),NGC5128 GCs (Holland et al. 1999, Harris et al. 2002)and M31 GCs (Barmby et al. 2007). The two-sampleK-S test shows that the UCD ellipticities are consistentwith extragalactic GC distributions (NGC5128 and M31GCs), but significantly different from the MW GC dis-tribution. The Wilcoxon test gives the same result. Itis interesting to note that Harris et al. (2002) found theMW GC ellipticity distribution to be significantly differ-ent from the NGC5128 and M31 GC distributions. Har-ris et al. (2002), however, do not place too much weighton their result because of uncertainties about possible se-lection effects and different methods for ellipticity mea-surements (see also Barmby et al. 2007).No correlation of ellipticity with luminosity, size orcolor was found for UCDs.In Figure 2 we present the luminosity-size diagram forUCDs. For comparison, we also show “dwarf globulartransition objects” from Ha¸segan et al. (2005), early-typegalaxy nuclei from Cˆot´e et al. (2006) and GCs (see cap-tion of Figure 2). The UCDs and GCs form a continuousdistribution across the plane, but UCDs seem to be dif-ferent to typical GCs in the sense that the UCDs havesizes correlated with luminosities whereas the GCs donot. This difference has been reported previously (e.g.by Ha¸segan et al. 2005 for “dwarf globular transition ob-jects” and five bright Fornax UCDs).We obtain the following luminosity-size relation for theUCDs in our sample (fitting a linear least squares re-gression to the data for both Fornax and Virgo UCDs intructural Properties of UCDs 5Table 1):log R eff = − . ± . − . ± . M V (4)or R eff ∝ L . ± . V . (5)If we exclude the two brightest objects, which may bedifferent from all other UCDs (they are much brighterthan all other UCDs and have the largest envelopes), weobtain:log R eff = − . ± . − . ± . M V (6)or R eff ∝ L . ± . V . (7)The fact that the UCDs show a luminosity-size rela-tion while the GCs do not is not a result of any selectioneffects. To illustrate this point, we show in Figure 2 theselection boundaries for our UCD sample. The sampleselection is only restricted at the faint magnitude end (byour survey flux limit corresponding to M V = − . R eff > bright GC observations by Barmby et al.(2007). They report an increasing lower bound on R eff inthe mass vs. R eff plane for the most massive GCs (masses ≥ . × M ⊙ ). Barmby et al. interpreted it as an exten-sion of a similar relation for early-type galaxies followingfrom the existence of a “zone of exclusion” (ZOE) in thefundamental plane ( κ -space), discussed by Burstein etal. (1997). According to Burstein et al. (1997), no stel-lar system violates the rule κ + κ <
8, which meansthat the maximum global luminosity density of stellarsystems varies as mass − / . Assuming constant mass-to-light ratios, this is equivalent to R min ∝ L . , the samerelation (within uncertainties) as we find for UCDs. Soif we consider UCDs as a part of GC family, perhaps itwould be more correct to talk about the increasing lowerboundary on R eff rather than about luminosity-size re-lation for them, which could be explained by the ZOE.However, the existence of a ZOE for galaxies is not quiteunderstood yet.The nuclei of early-type galaxies also show aluminosity-size correlation as noted by Cˆot´e et al. (2006).Fitting a linear least squares regression to all the nucleibrighter than M V = − . M V ∼ −
16 mag (to approximately match theUCD luminosity range), and excluding both unresolvedand offset nuclei, we found the following relation:log R eff = − . ± . − . ± . M V (8)or R eff ∝ L . ± . V , (9)where R eff is the mean of the two pass bands, g ′ and z ′ .This relation is very similar (especially the slope) to theUCD luminosity-size relations above. This is consistentwith the threshing hypothesis for UCD formation fromdisrupted early-type galaxies. We do not show the nucleiof late-type galaxies (or “nuclear star clusters”) in Fig-ure 2. They are nearly identical to the early-type galaxynuclei in the sense of luminosities and sizes (e.g. Cˆot´e et al. 2006 and references therein), so they would notadd new information to the figure. The similarity, how-ever, means that UCDs—if formed by disruption—couldbe the remnant nuclei of both early-type and late-typegalaxies, not just dE,Ns as it was initially suggested.From inspection of Figure 2 and the luminosity-sizerelations above we note that the UCDs are significantlylarger than the corresponding early-type galaxy nuclei atthe same luminosity. This is consistent with an earliercomparison we made of the brighter Fornax UCDs withsome fainter nuclei (de Propris et al. 2005), but we arenow able to confirm the size difference is not due to thedifference in luminosity of those samples.Taking the luminosity-size relation fitted for the galaxynuclei as a baseline, we calculate that the UCDs are anaverage factor of ∆ log R eff = 0 . ± .
03 larger (we donot include the two brightest UCDs with the largest en-velopes here). This difference is significant at a confi-dence level > .
9% (using the T-test) and correspondsto the effective radii of the UCDs being on average10 . ± . = 2 . +0 . − . times larger than the nuclei of thesame luminosity. In Figure 3 we show the histograms of R eff /L . V for nuclei and UCDs separately (in the same lu-minosity range as was used to derive relations (7) & (9)),which better emphasizes the difference between UCDsand early-type galaxy nuclei. Although this observationis superficially inconsistent with a process whereby thesegalaxy nuclei are stripped to form UCDs, the change insize may be a result of the stripping process. The nu-clei may be more concentrated (denser) than the UCDsdue to a truncation effect caused by their location withinthe prominent stellar envelope (or dark matter halo) ofa galaxy which does not affect the isolated UCDs. Con-versely, the stripping process itself may well result indynamical heating of the nuclei, so they expand as theirhost galaxies are disrupted. The simulations of the strip-ping process by Bekki et al. (2001, 2003) do indicatesome expansion of the remnant core, but no quantita-tive results are provided. Given our new observationalresult, it will be important to test this in detail againstsimulations of the threshing process to see if it is con-sistent with the size difference we have found betweenUCDs and the early-type galaxy nuclei. As an additionalconstraint on this process we note that the cores of thelargest UCDs (which have small stellar envelopes) havemagnitudes and sizes similar to the other UCDs (whichdo not have envelopes): − . ≥ M V ≥ − .
39 mag,6 . ≤ R eff ≤ . Color-magnitude diagrams
The total colors of the UCDs from our ACS survey(Table 1) can be compared to those of other hot stel-lar systems in the same environment. We therefore con-structed color-magnitude diagrams (CMDs) for GCs, nu-clei of early-type galaxies and UCDs in the Virgo andFornax Clusters. These are shown in Figures 4 & 5. Allmagnitudes and colors were transformed to the Johnson-Cousins V &( V − I ) system to facilitate the comparisonof our data to many other works in this filter set. Evstigneeva et al.The GC data were taken from the ACS Virgo ClusterSurvey (M87 and M49 GCs, Cˆot´e et al. 2004, Peng etal. 2006) and the ACS Fornax Cluster Survey (NGC1399and NGC1404 GCs, Jord´an et al. 2007). The magnitudesand colors for the nuclei of Virgo early-type galaxies and“dwarf globular transition objects” were also taken fromthe ACS Virgo Cluster Survey (Cˆot´e et al. 2006, Ha¸seganet al. 2005). The transformations from the ACS survey g ′ and z ′ AB magnitudes to the V and I magnitudeswere performed by using theoretical single stellar popu-lation (SSP) models of Bruzual & Charlot (2003). Wederived transformation equations for three different agebins to account for possible intermediate age populationsin nuclei and “dwarf globular transition objects”. For aChabrier initial mass function (IMF) and a metallicityrange of -2.2 to 0.6 dex, the transformation equations forthe three age bins are as follows:11 −
13 Gyr : V = g ′ AB − . − . g ′ − z ′ ) AB (10)( V − I ) = 0 .
445 + 0 . g ′ − z ′ ) AB ; (11)6 − V = g ′ AB − . − . g ′ − z ′ ) AB (12)( V − I ) = 0 .
449 + 0 . g ′ − z ′ ) AB ; (13)2 − V = g ′ AB + 0 . − . g ′ − z ′ ) AB (14)( V − I ) = 0 .
451 + 0 . g ′ − z ′ ) AB . (15)For the objects in Figures 4 & 5, we chose the transfor-mation equations obtained for the very old ages. Theage effect on the transformations, however, is not strong,as shown by the color-magnitude relations for the Virgonuclei in Figure 4, and the choice of age does not affectour main conclusions below.The tick marks on the bottom of the CMDs in Fig-ures 4 & 5 indicate the peak colors for the blue and redGC populations in Virgo and Fornax derived by Larsenet al. (2001). They are in a good agreement with the av-erage transformed colors for the blue and red GCs fromthe ACS surveys. This shows the reliability of our trans-formations.The nucleus V & ( V − I ) values for Fornax dwarf ellip-ticals were taken from Lotz et al. (2004). For the FornaxUCDs, we present our ACS sample plus additional ob-jects from Gregg et al. (2008) with colors from Karick etal. (2008): the original Sloan gri magnitudes were trans-formed to V I via the transformation equations of Lupton(2005) .Looking at the CMDs, we observe that the UCDsand “dwarf globular transition objects” are spread overthe same color range as GCs. The brightest UCDs( M V < − . M V ∼ −
10 mag is a pure selection and incom-pleteness effect. On the one hand, the GCs from the ACSsurveys only cover a small area around galaxies and are selected by their apparent sizes and magnitudes. UCDswould have been rejected by these surveys as extendedbackground galaxies. On the other hand, the redshift-selected UCDs cover a large area in the clusters butsuffer from the completeness limit of spectroscopic sur-veys. Our Fornax Cluster survey has a magnitude limitof M V ∼ − . .An additional feature is visible in the Fornax CMD.There is a group of UCDs (with M V ∼ − . V − I ≤ . ∼ − β line indices (Mieske et al.2006), a hint to the contribution of young stellar popu-lations.In the Virgo CMD, we plot the color-magnitude re-lation for dwarf galaxy nuclei (host galaxy magnitude B T > . V − I ∼ .
25 mag, at the samelocation where the brightest and also the reddest UCDsare found. We can not distinguish UCDs from nucleiof early-type galaxies by simply using magnitudes andcolors.In the Fornax CMD, we also show nuclear star clus-ters (NCs) of late-type galaxies, although the galaxiesare not Fornax members. The data are from the work ofRossa et al. (2006), based on previous works by Walcheret al. (2005, 2006) and B¨oker et al. (2002, 2004). Thedata were corrected for the foreground extinction but theintrinsic extinction of the host galaxy is more problem-atic. Only few NCs in the Rossa et al. and B¨oeker et al.lists have estimates of the internal reddening (and werecorrected for it). As we mentioned in the previous sec-tion, NCs of late-type galaxies and nuclei of early-typegalaxies are nearly identical in the sense of luminositiesand sizes. The only difference is that the majority ofNCs have young ages, they are younger than early-typegalaxy nuclei. Looking at Figure 5, one can see thatthere are some blue (young) NCs that are close to thelocation of blue GCs and UCDs or at least should passthis location when aging. There are quite a few NCswith very blue colors ( V − I < . M V < −
10 mag), which can not be seen inFigure 5 due to the scale of the figure. There are alsomany red NCs that probably are contaminated by inter-nal reddening (so they might actually be bluer). The blueNCs look very bright, on average brighter than early-typegalaxy nuclei and UCDs. The bright magnitudes can becaused by young ages. Younger stellar populations arenot only bluer but also much brighter. They will fadewith time. Red bright NCs might be intrinsically veryluminous/massive. They come mostly from MW-typespirals. As above (in Section 4.1), the main conclusionwe can draw out of the comparison of UCDs with NCsis that some of the UCDs, both blue and red, could be We note that some UCDs in Figure 5 have lower luminositiesthan the nominal survey limit. This is because the original surveyselection was based on photographic photometry with relativelyhigh uncertainties. tructural Properties of UCDs 7threshed nuclei of late-type galaxies.
Color profiles and color gradients
The analysis of radial color gradients for the UCDs isvery challenging, mainly because they are tiny and verycompact objects, hardly resolved even with the HST.So the corrections for telescope PSF effects become ex-tremely important. In this section we therefore reanalysethe images to obtain the best possible estimates of the ra-dial color profiles and devise a series of tests to quantifyhow the PSF and other issues affect the color profiles.Note that we present results for the brightest UCDs only(Virgo UCDs and Fornax UCD3 & UCD6) as the colorgradients for the fainter UCDs could not be estimatedwith any reliability.
Measurement of color gradients
We obtain the radial color profiles by fitting UCD im-ages with PSF-convolved models using
GALFIT . Su-perficially it may appear easier to derive color profilesdirectly from PSF-deconvolved images. However, wechoose not to do this, as a deconvolution process (e.g. theLucy-Richardson algorithm) amplifies noise in the imageand, in our case, produces very large artificial fluctua-tions in the color profiles. Instead we derive color profilesfrom
GALFIT models. We do not use the models ob-tained in Section 3 as these were designed to obtain thebest parameters for standard models where possible. Thedifferences to the approach in Section 3 are as follows:1. Multiple component models (e.g.King+S´ersic+S´ersic) are used to better fitthe UCDs. No constraints that the model pa-rameters must have physically meaningful valuesare applied. Our aim is to match the UCDsurface brightness (SB) profiles as well as possible.Multiple S´ersic models are the best for almost allthe UCDs.2. The PSFs are presented out to a radius of 10 ′′ inthe modeling. We also do several experiments withthe PSFs to make sure that the color gradients wefind for UCDs are robust.3. We fit the background directly rather than holdingit fixed.We start by using GALFIT to fit more general multi-component models to each UCD and filter combination.From the best-fitting parameters we generate the two-dimensional model (using
GALFIT again), which is inprinciple free of PSF effects. Then we calculate SB pro-files for the models in each filter using the
IRAF task
ELLIPSE . The color profile is then determined as thedifference of the SB profiles in the two filters. The re-sulting color profiles are presented in the top panels ofFigures 7–16.In the rest of this subsection we discuss the most im-portant issues which affect the color profiles, such as ac-curacy of the
GALFIT model fits to the data, PSF ef-fects and sky subtraction.
1. Accuracy of the
GALFIT model fits to the data
The object-minus-model residuals are shown in thebottom panels of Figures 7–16. The residuals get largerwith radius. This is normal and expected. If we plot the error bars on the UCD surface brightness profiles pro-duced by
ELLIPSE (errors on measuring the mean fluxalong each isophote), we find that they are are compa-rable to the size of the residuals. The most importantpoint here is to fit the objects so that the residuals donot show any systematic trend (so that the residuals fluc-tuate evenly around zero). In some cases, the residualfluctuations are larger than the
ELLIPSE errors, butthis could be because the
ELLIPSE error bars are onlystatistical and do not include any other possible errors.
2. PSF effects
The PSF depends on the object color. In the samebroad-band filter, the PSF (PSF FWHM) of a red starmay be noticeably larger than that of a more blue star.We therefore generated three PSFs using
TinyTim (and
MultiDrizzle , as described in Section 3): one with theaverage UCD color, one with the reddest and one withthe bluest possible with
TinyTim color.The PSFs were made of a very large size (20 ′′ × ′′ or 800 ×
800 pixel, larger than the UCDs) to attemptto address the PSF halo problem. The ACS/HRC chiphas a defect that creates a halo surrounding the PSFat wavelengths > . µ m and the relative proportion offlux within this halo increases with wavelength. The halois large (many arc seconds) and can contain 10-20% ofthe total flux. TinyTim models this effect, but unfortu-nately not very well .Stellar images (“real” PSFs) are better representationsof the true PSF, but they also have some disadvantages.All the stellar images we found in the HST archive suit-able for the UCD image modeling (i.e. images of a singlestar which was centrally located and non-saturated) aremuch bluer than the UCDs. Another disadvantage ofthese PSFs is that they have a very low S/N at largeradii and, therefore, can not correctly capture the outerPSF halo. Brighter stars, which have higher S/N at largeradii, are saturated in the center and, hence, can not beused for modeling, but we can use them for the compar-ison with the
TinyTim
PSFs, to check how reliable the
TinyTim
PSFs are at large radii. We have managed tofind in the archive a couple of high S/N stars with redcolors, similar to UCD colors.In Figure 6 we plot the color profiles for
TinyTim
PSFs, a low S/N star (with blue color) and a high S/Nstar (with red color), and normalize all of them at ∼ TinyTim
PSFs look very similar (except atvery large radii,
R >
100 pixels) and give very similarresults for the UCD color profiles in Figures 7 & 8. Sofor the UCD image modeling, we can safely use just one
TinyTim
PSF with the average UCD color. Another im-portant conclusion we can draw from Figure 6 is that thecolor profiles of the high S/N star (with red color similarto UCD colors) and
TinyTim
PSFs are nearly the samein the region ∼ −
54 pixels. The match between thecolor profiles of the low S/N star (with blue color) and
TinyTim
PSFs is a bit worse in the same radius range(compared to the high S/N PSF).In Figures 7–16 we present the UCD color profiles, ob-tained with the
TinyTim
PSF (with the average UCDcolor and of a very large size) and the low S/N star, butdue to the above reasons we prefer to trust the
TinyTim
PSF more. Evstigneeva et al.Having done all of this, we conclude that the colorgradient is most reliable in the region ∼ −
54 pixels( ∼ . −
100 pc for Virgo and ∼ −
124 pc for For-nax). The uncertainties in the PSF structure becomesignificant outside this radius range. In addition to thePSF effects, we can not trust the color gradient beyond R ∼ −
80 pc for most of the UCDs due to the no-ticeable deviation between the
GALFIT model and thedata. The color gradient may be incorrectly amplifiedbecause of it. To highlight the region where we considerthe color gradient is reliable (the “trusted region” ), wedraw two vertical lines in all the panels: at the outerradius where object-minus-model residuals start to devi-ate from zero significantly, and at the inner radius wherethere may be uncertainties in the structure of the PSFcore. For VUCD3, for example, the “trusted region” is ∼ . −
75 pc (Figure 7).
3. Sky subtraction
The sky (background) was initially subtracted fromthe UCD images by
MultiDrizzle . We then appliedadditional background corrections, determined fromempty parts of the images. Finally, we fitted thebackground when fitting UCDs with
GALFIT . Forthe latter, we did the fitting over very large area,masking out the HRC “occulting finger” and all thebackground/foreground objects in the image, exceptobjects appearing very close to the UCDs. We fitted andsubtracted such objects from the image, which is moreaccurate than masking. Note that besides the mean skylevel,
GALFIT finds the gradient in the sky. The skylevel is determined very well, as seen from the absenceof a trend in the residuals in Figures 7–16 (except forVUCD2).The results of the color profile fitting are shown inFigures 7–16. The main observational conclusion is that,with two exceptions, all the UCDs appear to have smallpositive color gradients in the sense of getting redderoutwards: mean ∆( F W − F W ) = 0 .
14 mag per100 pc with rms = 0 .
06 mag per 100 pc. The two ex-ceptions are VUCD2 and UCD3. It is hard to make verystrong conclusions regarding VUCD2 because of the largefluctuations in the color profile and a slight systematictrend (upwards) in the residuals. We would also considerthe UCD3 color gradient as unreliable, since we did notmodel and subtract the background spiral (see Section 3and Evstigneeva et al. 2007). It is very hard to separatethis faint object from the UCD reliably. We did all themodeling for the half of the UCD, which is less affectedby the spiral, but we can not prove that it is completelyunaffected. There still can be some flux from this spiral,projected onto the UCD3 center and making it slightlybluer.
Interpretation
We stress that this discussion only applies to thebrightest UCDs for which it was possible to estimate thecolor gradients.First we consider the hypothesis that UCDs are veryluminous GCs.Djorgovski et al. (1991) report color gradients in MWGCs. They find that the clusters with post core collapsemorphology (with central cusps) have positive color gra-dients (in the same sense as the gradient in the UCDs), while clusters with King model morphology (with flatcores) do not show any color gradient. Djorgovski et al.interpret this as evidence that the dynamical evolutionof clusters can modify their stellar populations. Core col-lapse tends to affect more massive stars, with lower massstars (red sub dwarfs) tending to move to larger averageradii. The Djorgovski et al. study may not, however, berelevant to UCDs. Their measurements only extend to1–8 pc from the cluster center and, more importantly,relaxation and mass segregation (less massive red subdwarfs sitting outside) would not work for systems aslarge as UCDs. Their half-mass relaxation times are ∼ R > ω Cen,NGC2808, NGC1851) are found to contain multiple stel-lar populations (e.g. Milone et al. 2008). ω Cen is theonly GC for which we have information about the ra-dial distribution of stellar populations (Hilker & Richtler2000, Sollima et al. 2007). Hilker & Richtler (2000) stud-ied the two major populations (of RGB stars) in ω Cen (ithas at least three sub-populations) and found that theyounger, more metal-rich population is more centrallyconcentrated than the older, more metal-poor popula-tion. This gives a negative color gradient—opposite towhat we find for UCDs. A possible explanation for ω Cenis that a more centrally concentrated population formedat a later time from enriched material. Assuming thesame scenario for UCDs, we can test if it agrees withthe UCD color gradients. For Virgo UCD ages (8 − − .
35 to − .
17 dex) and col-ors ( V − I = 0 . − .
13 mag) (Evstigneeva et al. 2007)and using Maraston (2005) SSP models, we consider thefollowing model populations: a more metal rich ([Z/H]=-0.33 dex), younger (8 Gyr) population with V − I = 1 . V − I = 0 .
98 mag. The colorsare obtained for a Salpeter IMF (the Kroupa IMF givesvery similar results). The sum of these two populationswill give total ages, metallicities and colors similar tothose observed in the UCDs. The color gradient, how-ever, is negative—opposite to what is observed in theUCDs (although the same as in ω Cen). This suggeststructural Properties of UCDs 9that the star-formation histories in UCDs and ω Cen aredifferent, but this still does not prove that UCDs aredifferent to GCs. Recent work by Milone et al. (2008),for example, says that three different GCs with multi-ple stellar populations ( ω Cen, NGC2808, NGC1851) allhave different star-formation histories. Milone et al. alsonote that “the star-formation history of a GC can varystrongly from cluster to cluster” and they point out thatthe multiple stellar populations have only been detectedin the most massive MW GCs. They suggest that “clus-ter mass might have a relevant role in the star-formationhistory of GCs.”From these results we conclude that UCDs may con-tain multiple stellar populations without contradictingthe hypothesis that they are bright GCs. However a lotmore has to be done in terms of both observational workand theory/simulations (as was concluded by D’Antona& Caloi 2007) to understand the origin of multiple stel-lar populations in UCDs and GCs. Obviously, somethinghappens in very massive/luminous GCs so that they havemultiple stellar populations. It might be that their grav-itational potential is strong enough to keep the gas thatwould otherwise be completely expelled by stellar windsand supernovae explosions (see e.g. Baumgardt et al.2008) and that the second (and other) generations ofstars formed from this gas.We now consider the alternative hypothesis, thatUCDs are the nuclei of threshed galaxies.It could be argued that the color gradients in UCDs areconsistent with the threshing hypothesis for the forma-tion of UCDs from dE,Ns. Lotz et al. (2004) showed forFornax cluster dE,Ns, using aperture photometry, thattheir stellar envelopes are 0.1-0.2 mag redder in V − I than their nuclei. In this case, residual stars from theenvelope after the disruption process would naturally ex-plain the color gradient in UCDs. We can test this withmore recent data from the ACS Virgo Cluster Surveyfor early-type galaxies. We compared the colors of thenuclei (Cˆot´e et al. 2006) and their underlying galaxies(Ferrarese et al. 2006): the nuclei are bluer than theirunderlying galaxies for dEs ( M B > − . M B < − . simple stripping of dEs (withno other processes involved, like effects of gas removalon the chemical evolution of nuclei), the UCD color gra-dients are in agreement with the threshing hypothesis(with both dE color gradients and dE core/envelope col-ors). There is a problem with the brightest and reddestUCDs, however. If we have a look at the CMDs (Fig-ures 4 & 5), VUCD3, VUCD7 and UCD3 are all located in the region where only giant elliptical galaxy nucleiare found. If we interpret the fact that VUCD7 andUCD3 have envelopes as incomplete stripping and con-sider only the cores of these UCDs in the CMDs, thecores will then be in the same region as dE nuclei. (Thecores of all the UCDs with envelopes have magnitudesand colors as follows: − . ≥ M V ≥ − .
39 mag,0 . ≤ V − I ≤ .
18 mag.) Thus, VUCD7 and UCD3 maystill be threshed dE galaxies. VUCD3 does not have anenvelope but it is very bright and very red, so it does looklike a giant elliptical galaxy nucleus. However, VUCD3has a positive color gradient which we might not expectin a giant elliptical galaxy. It may be possible to arguethat VUCD3 is a threshed late-type galaxy (if we stillprefer the threshing hypothesis), since late-type galaxieshave mostly positive color gradients (e.g. Taylor et al.2005).It may also be possible that UCDs are genuine,but anomalous, dwarf galaxies (requiring no threshing).Their positive color gradients are consistent with dEgalaxy color gradients—the pure age gradients, explainedby galactic winds (as above). If we assume that UCDshave younger stellar populations in the center (7 Gyr)and older populations in the outer regions (15 Gyr),then for a metallicity of [Z/H]=-0.33 dex (same for thetwo populations), Maraston (2005) SSP models witha Salpeter IMF will give us the color variation from V − I = 1 .
06 mag in the center to V − I = 1 .
16 magoutside; or for a metallicity of [Z/H]=-1.35 dex: from V − I = 0 .
90 mag in the center to V − I = 0 .
99 mag out-side. This is consistent with what we see in the UCDs.Our conclusion is that the positive color gradientsfound in bright UCDs (2 Fornax UCDs and 8 VirgoUCDs) are consistent with them being either brightGCs or threshed dE galaxies (except for VUCD3).However, the spectroscopic ages, metallicities, and α -abundances for Virgo UCDs, obtained in our previouswork (Evstigneeva et al. 2007), are not consistent withthe formation of UCDs by the simple removal of the enve-lope from the nuclei of dE galaxies. Since spectroscopicages and metallicities are more powerful tools than colors(colors are degenerate), we have to reject the threshinghypothesis for the Virgo UCD origin. Hence, the VirgoUCDs are more consistent with being bright GCs. Asfor the two Fornax UCDs, no firm conclusions on theirorigins can be drawn without having spectroscopic age,metallicity, and α -abundance estimates for them. At themoment, we can only say that their color gradients areconsistent with all three hypotheses: threshed dE galax-ies, bright GCs, and genuine dwarf galaxies. SUMMARY
In this paper we have presented analysis of the struc-ture and colors of the most extensive and complete sam-ple of UCDs in the Fornax and Virgo Clusters observedto date: 13 Fornax UCDs with magnitudes in the range − . ≥ M V ≥ − .
33 mag and 8 Virgo UCDs withmagnitudes − . ≥ M V ≥ − .
42 mag. The sampleincludes 6 Virgo UCDs initially presented by Evstigneevaet al. (2007), as we have made several improvements tothe image analysis. The main results of our analysis areas follows:1. We have modelled the images of Fornax and Virgo0 Evstigneeva et al.UCDs using the two-dimensional fitting algorithm
GALFIT , assuming empirical King, S´ersic andNuker models (or King+S´ersic and S´ersic+S´ersic)for the luminosity profile. We find that for the faintUCDs ( m V, ∼ . − . m V, ∼ . − . n and King concentrations c , as well as a range of central slopes (seen fromNuker inner slopes γ ): from flat “King” cores tocentral cusps. This suggests that convolving somecuspy models with the PSF allows such models tofit the seeing-blurred centers of some UCD profiles.We, however, can not make very strong conclusionsregarding central cusps or cores in the UCDs, tak-ing into account the resolution and other limits ofour data.We do not find any correlations between the modelparameters mentioned above and the UCD lumi-nosity, size or color. Furthermore, no correlationof ellipticity with luminosity, size or color has beenfound for the UCDs.The two-sample K-S and Wilcoxon tests showthat the UCD ellipticities are consistent with ex-tragalactic GC distributions (NGC5128 and M31GCs), but significantly different from the MW GCdistribution.2. We have shown that UCDs and GCs form a contin-uous distribution across the luminosity-size plane,but UCDs seem to be different to typical GCs in thesense that the UCDs have sizes correlated with lu-minosities whereas the GCs do not. We obtain thefollowing luminosity-size relation for the UCDs inour sample: R eff ∝ L . ± . V or R eff ∝ L . ± . V (if we exclude two brightest objects). However, theluminosity-size relation we observe for the UCDsis similar to observations for most luminous GCs:Barmby et al. (2007) report an increasing lowerbound on R eff in the mass vs. R eff plane for themost massive GCs.The nuclei of early-type galaxies (from Cˆot´e et al.2006) with luminosities similar to the UCDs showa luminosity-size correlation with the same slopeas the UCD relation, but we find that the effec-tive radii of the UCDs are systematically ∼ . > . F W − F W ) = 0 .
14 mag per 100 pcwith rms = 0 .
06 mag per 100 pc. The two excep-tions are VUCD2 and UCD3: the color gradientestimates for them are unreliable.The positive color gradients found in the brightUCDs are consistent with them being either brightGCs or threshed dE galaxies (except VUCD3).However, the spectroscopic ages, metallicities, and α -abundances for Virgo UCDs, obtained in our pre-vious work (Evstigneeva et al. 2007), are not con-sistent with the formation of UCDs by the sim-ple removal of the envelope from the nuclei of dEgalaxies. Since spectroscopic ages and metallicitiesare more powerful tools than colors, we have to re-ject the threshing hypothesis for the Virgo UCDorigin. Hence, the Virgo UCDs are more consis-tent with being bright GCs. As for the two FornaxUCDs, no firm conclusions on their origins can bedrawn without having spectroscopic age, metallic-ity, and α -abundance estimates for them. At themoment, we can only say that their color gradi-ents are consistent with them being threshed dEgalaxies, bright GCs, and genuine dwarf galaxies.The aim of our investigation was to test two forma-tion hypotheses for UCDs—whether they are bright GCsor threshed early-type dwarf galaxies—by direct com-parison of UCD structural parameters and colors withGCs and galaxy nuclei. In most of the measurementstructural Properties of UCDs 11we have made (profiles; color-magnitude relations; colorgradients), the UCDs display properties consistent witha threshing origin and with what might be expected forluminous GCs. We therefore conclude that these struc-tural parameters and colors are not able to distinguishsufficiently between the different formation hypotheses.The one exception to this conclusion is the differencewe find in the size-luminosity relation between UCDs andthe nuclei of early-type galaxies. This significant differ-ence (the UCDs are 2.2 times as large as nuclei at thesame luminosity) suggests an important numerical testof the threshing hypothesis: it should be relatively easyto predict the increase in the size of nuclei resulting fromthe threshing hypothesis and compare that to our newobservational results. This work has been supported by a Discovery Projectgrant from the Australian Research Council. Part of thework reported here was done at the Institute of Geo-physics and Planetary Physics, under the auspices of theU.S. Department of Energy by Lawrence Livermore Na-tional Laboratory in part under Contract W-7405-Eng-48 and in part under Contract DE-AC52-07NA27344.The authors wish to thank Harry Ferguson (STScI) andAnton Koekemoer (STScI) for their assistance with theHST data analysis, Holger Baumgardt (University ofBonn) for helpful discussions and the referee for theirvery careful reading of our paper and many helpful sug-gestions. Facilities:
HST (ACS)
REFERENCESBarmby, P., McLaughlin, D.E., Harris, W.E., Harris, G.L.H., &Forbes, D.A. 2007, AJ, 133, 2764Baumgardt, H., Kroupa, P., & Parmentier, G. 2008, MNRAS, inpress, arXiv:0712.1591Bekki, K., Couch, W.J., & Drinkwater, M.J. 2001, ApJ, 552, L105Bekki, K., Couch, W.J., Drinkwater, M.J., Shioya, Y. 2003,MNRAS, 344, 399B¨oker, T., Laine, S., van der Marel, R.P., Sarzi, M., Rix, H.-W.,Ho, L.C., & Shields, J.C. 2002, AJ, 123, 1389B¨oker, T., Sarzi, M., McLaughlin, D.E., van der Marel, R.P., Rix,H.-W., Ho, L.C., & Shields, J.C. 2004, AJ, 127, 105Burstein, D., Bender, R., Faber, S., & Nolthenius, R. 1997, AJ,114, 1365Bruzual, G., & Charlot, S. 2003, MNRAS, 344, 1000Cˆot´e, P., et al. 2004, ApJS, 153, 223Cˆot´e, P., et al. 2006, ApJS, 165, 57D’Antona, F. & Caloi, V. 2007, in Vesperini E., Gierzs M. &Sills A., eds., Proceedings of IAU Symposium 246 “DynamicalEvolution of Dense Stellar Systems”, in press, arXiv:0709.4601De Propris, R., Phillipps, S., Drinkwater, M.J., Gregg, M.D., Jones,J.B., Evstigneeva, E., & Bekki, K. 2005, ApJ, 623, L105De Propris, R., Colless, M., Driver, S.P., Pracy, M.B., & Couch,W. 2005a, MNRAS, 357, 590Djorgovski, S., Piotto, G., Phinney, E.S., & Chernoff, D.F. 1991,ApJ, 372, L41Drinkwater, M.J., Jones, J.B., Gregg, M.D., & Phillipps, S. 2000,PASA, 17, 227Drinkwater, M.J., et al. 2000a, A&A, 355, 900Drinkwater, M.J., Gregg, M.D., Hilker, M., Bekki, K., Couch, W.J.,Ferguson, H.C., Jones, J.B., & Phillipps, S. 2003, Nature, 423,519Drinkwater, M.J., Gregg, M.D., Couch, W.J., Ferguson, H.C.,Hilker, M., Jones, J.B., Karick, A., & Phillipps, S. 2004, Publ.Astron. Soc. Australia, 21, 375Elson, R.A.W. 1999, in Globular Clusters, CambridgeContemporary Astrophysics, ed. C. Mart´ınez Roger, I. P´erezFournon, & F. S´anchez (Cambridge: Cambridge Univ. Press)Evstigneeva, E.A., Gregg, M.D., Drinkwater, M.J., & Hilker, M.2007, AJ, 133, 1722Evstigneeva, E.A., Drinkwater, M.J., Jurek, R., Firth, P., Jones,J.B., Gregg, M.D., & Phillipps, S. 2007a, MNRAS, 378, 1036Fellhauer, M., & Kroupa, P. 2002, MNRAS, 330, 642Ferrarese, L. et al. 2006, ApJS, 164, 334Freedman, W.L., et al. 2001, ApJ, 553, 47Graham, A.W., Erwin, P., Trujillo, I., & Ramos, A.A. 2003, AJ,125, 2951Gregg, M.D., et al. 2008, AJ, submittedHarris, W.E. 1996, AJ, 112, 1487Harris, W.E., Harris, G.L.H., Holland, S.T., & McLaughlin, D.E.2002, AJ, 124, 1435Hilker, M., Infante, L., Vieira, G., Kissler-Patig, M., & Richtler, T.1999, A&AS, 134, 75 Hilker, M., & Richtler, T. 2000, A&A, 362, 895Hilker, M., Baumgardt, H., Infante, L., Drinkwater, M.,Evstigneeva, E., & Gregg, M. 2007, A&A, 463, 119Holland, S., Cˆot´e, P., & Hesser, J.E. 1999, A&A, 348, 418Ha¸segan, M., et al. 2005, ApJ, 627, 203Jones, J. B., et al. 2006, AJ, 131, 312Jord´an, A., et al. 2007, ApJS, 169, 213Karick, A., Gregg, M.D., & Drinkwater, M.J. 2008, AJ, submittedLarsen, S.S., Brodie, J.P., Huchra, J.P., Forbes, D.A., & Grillmair,C.J. 2001, AJ, 121, 2974Lauer, T.R., Ajhar, E.A., Byun, Y.-I., Dressler, A., Faber, S.M.,Grillmair, C., Kormendy, J., Richstone, D., & Tremaine, S. 1995,AJ, 110, 2622Liske, J., Driver, S.P., Allen, P.D., Cross, N.J.G., & De Propris, R.2006, MNRAS, 369, 1574Lotz, J.M., Miller, B.W., & Ferguson, H.C. 2004, ApJ, 613, 262Maraston. C. 2005, MNRAS, 362, 799McLaughlin, D.E., & van der Marel, R.P. 2005, ApJS, 161, 304McLaughlin, D.E., Barmby, P., Harris, W.E., Forbes, D.A., &Harris, G.L.H. 2008, MNRAS, 384, 563Mieske, S., Hilker, M., & Infante, L. 2002, A&A, 383, 823Mieske, S. et al. 2004, AJ, 128, 1529Mieske, S., Hilker, M., Infante, L., & Jord´an, A. 2006, AJ, 131,2442Mieske, S., Hilker, M., Jord´an, A., Infante, L., & Kissler-Patig, M.2007, A&A, 472, 111Milone, A.P. et al. 2008, ApJ, 673, 241Peng, C.Y., Ho, L.C., Impey, C.D., & Rix, H.-W. 2002, AJ, 124,266Peng, E.W., et al. 2006, ApJ, 639, 95Rossa, J., van der Marel, R.P., B¨oker, T., Gerssen, J., Ho, L.C.,Rix, H.-W,. Shields, J.C., & Walcher, C.-J. 2006, AJ, 132, 1074Schlegel, D.J., Finkbeiner, D.P., & Davis, M. 1998, ApJ, 500, 525S´ersic, J. L. 1968, Atlas de Galaxias Australes (C´ordoba: Obs.Astron., Univ. Nac. C´ordoba)Sirianni, M., et al. 2005, PASP, 117, 1049Sohn, Y.-J., Byun, Y.-I., & Chun, M.-S. 1996, Ap&SS, 243, 379Sohn, Y.-J., Byun, Y.-I., Yim, H.-S., Rhee, M.-H., & Chun, M.-S.1998, JASS, 15, 1Sollima, A., Ferraro, F.R., Bellazzini, M., Origlia, L., Straniero, O.,& Pancino, E. 2007, ApJ, 654, 915Taylor, V.A., Jansen, R.A., Windhorst, R.A., Odewahn, S.C., &Hibbard, J.E. 2005, ApJ, 630, 784Vader, J.P., Vigroux, L., Lachi`eze-Rey, M., & Souviron, J. 1988,A&A, 203, 217Walcher, C.-J., et al. 2005, ApJ, 618, 237Walcher, C.-J., B¨oker, T., Charlot, S., Ho, L.C., Rix, H.-W., Rossa,J., Shields, J.C., & van der Marel, R.P. 2006, ApJ, 649, 692Wehner, E., & Harris, W. 2007, ApJ, 668, L35
Fig. 1.—
Surface brightness profiles, measured in the F606W images, and model fits to 4 UCDs. The plots for the other 17 UCDs areavailable in the electronic edition of the Journal (Figures 1.1.–1.5). The magnitudes are AB magnitudes. The open circles represent theUCD profile, the dashed line represents the best-fitting model, convolved with the PSF. The PSF for each object is shown with the dottedline. ∆ µ V plots show the residuals for each fit: the difference (in magnitudes) between the UCD profile and the profile of the best-fittingmodel, convolved with the PSF. tructural Properties of UCDs 13 Fig. 2.—
Luminosity – Size diagram. Fornax and Virgo UCDs: this work (Table 1). MW GCs: McLaughlin & van der Marel (2005),photometry is based on Wilson models. M31 GCs: Barmby et al. (2007), photometry is based on King models. NGC5128 GCs: Holland etal. (1999), Harris et al. (2002). Virgo DGTOs: Ha¸segan et al. (2005), certain DGTO candidates, R eff is the mean of the two pass bands, g ′ and z ′ . Virgo early-type galaxy nuclei: Cˆot´e et al. (2006), all resolved nuclei, R eff is the mean of the two pass bands, g ′ and z ′ . Thevertical dotted line shows the magnitude limit of our 2dF/HST UCD observations. The horizontal lines show the HST/ACS resolutionlimit: the lower line is for the Virgo distance and upper line is for Fornax (below these lines R eff < Fig. 3.—
Size comparison of UCDs and early-type galaxy nuclei. The sizes are all scaled relative to the size-luminosity relation fitted forthe nuclei in Equation 9. tructural Properties of UCDs 15
Fig. 4.—
Color-magnitude diagram for the Virgo Cluster objects. Filled circles are Virgo UCDs from the present work. Open circles are“certain and probable” Virgo “dwarf globular transition objects” from Ha¸segan (2005). Grey dots are M87 and M49 GCs from the ACSVirgo Cluster Survey. Triangles are nuclei of early-type galaxies from the ACS Virgo Survey: open triangles – nuclei of galaxies brighterthan B T = 13 . B T = 13 . B T = 13 . Fig. 5.—
Color-magnitude diagram for the Fornax Cluster objects. Filled circles are Fornax UCDs from the present work. Open circlesare Fornax UCDs from Gregg et al. (2008). Grey dots are NGC1399 and NGC1404 GCs from the ACS Fornax Cluster Survey. Filledtriangles are dwarf elliptical nuclei from Lotz et al. (2004). Ticks are for the blue and red GC peaks of NGC1399 and NGC1404 (Larsenet al. 2001). Asterisks are nuclear star clusters (NCs) of late-type galaxies (Rossa et al. 2006). tructural Properties of UCDs 17
Fig. 6.—
PSF color profiles in comparison with the VUCD3 color profile (contains all PSF effects). All the PSF profiles are normalizedat ∼ ∼
11 pc at the Virgo Cluster distance).
Fig. 7.—
VUCD3 color profile. tructural Properties of UCDs 19
Fig. 8.—
VUCD5 color profile.
Fig. 9.—
VUCD1 color profile. tructural Properties of UCDs 21
Fig. 10.—
VUCD2 color profile.
Fig. 11.—
VUCD4 color profile. tructural Properties of UCDs 23
Fig. 12.—
VUCD6 color profile.
Fig. 13.—
VUCD7 color profile. tructural Properties of UCDs 25
Fig. 14.—
VUCD8 color profile.
Fig. 15.—
UCD3 color profile. tructural Properties of UCDs 27
Fig. 16.—
UCD6 color profile.
TABLE 1UCD photometry.
Object R.A.(J2000) Dec.(J2000) m V, M V, ( V − I ) R eff M mod V, ǫ (h : m : s) ( ◦ : ′ : ′′ ) (mag) (mag) (mag) (pc) (mag)UCD3 3:38:54.10 -35:33:33.6 18.06 -13.33 1.25 86.5 ± a UCD6 3:38:05.09 -35:24:09.6 18.81 -12.58 1.07 10.3 ± ± ± ± ± ± ± ± ± ± ± ± ± ± a VUCD3 12:30:57.40 +12:25:44.8 18.33 -12.59 1.27 20.0 ± ± ± ± ± a VUCD8 12:32:14.61 +12:03:05.4 18.97 -11.95 1.06 23.5 ± c c c b c a Note . — The V band apparent magnitude, m V, , is determined as described in Section 3 and is corrected forforeground dust extinction (Schlegel et al. 1998). The absolute magnitude, M V, , is computed assuming distancemoduli of 31.39 and 30.92 mag for the Fornax and Virgo Clusters, respectively (Freedman et al. 2001). The( V − I ) color is reddening-corrected. The half-light radius value, R eff , is the mean of the two pass bands, V and I . The R eff and M mod V, values were obtained from generalized King (or standard King) models for one-componentUCDs and King+S´ersic models for two-component UCDs (see Section 3). The ellipticity value, ǫ , is the bestmodel value (see last column of Table 2), mean of the two pass bands. a The first number is for the core, the second number is for the halo. b The difference to the analysis in Evstigneeva et al. (2007) is that we now derive R eff and M mod V, from ageneralized King model. It gives the more stable estimate for R eff , than the two-component King+S´ersic modelobtained in Evstigneeva et al. (2007). c The colors are from Karick et al. (2008). tructural Properties of UCDs 29
TABLE 2UCD structural parameters.
Object S´ersic Generalized King a Nuker b Best model h n µ V, R c d c e α R b d β γ UCD3 f ± ± ± ± ± g ± ± ± ± ± ± ± g ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± g ± ± ± f ± g ± ± ± ± ± ± ± g ± ± ± ± ± ± ± ± ± ± ± ± ± ± g ± ± ± ± ± ± ± g ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± f ± ± ± ± ± ± ± ± ± ± ± f ± ± ± ± ± ± ± ± ± ± ± ± ± ± f ± ± ± ± ± ± ± f ± ± ± ± ± ± ± ± ± ± ± Note . — All the parameters (except µ V, ) are the means of the two pass bands, V and I . a If the standard King model ( α = 2 .
0) fits the object better than the generalized King model ( α is a free parameter), we provide parameters for thestandard King model. b The Nuker model parameters are provided only if this model is reasonably good for the object and there is a good agreement between two pass bands. c µ V, is measured in mag arcsec − . d R c and R b are measured in pc. e Concentration c = log ( R t /R c ). f For two-component UCDs, S´ersic and King model parameters are given for the central (core) component. g R c < h N – Nuker, K – King ( α = 2), GK – Generalized King ( αα