New Constraints on Early-Type Galaxy Assembly from Spectroscopic Metallicities of Globular Clusters in M87
Alexa Villaume, Aaron J. Romanowsky, Jean Brodie, Jay Strader
DDraft version May 31, 2019
Preprint typeset using L A TEX style emulateapj v. 12/16/11
NEW CONSTRAINTS ON EARLY-TYPE GALAXY ASSEMBLY FROM SPECTROSCOPIC METALLICITIESOF GLOBULAR CLUSTERS IN M87
Alexa Villaume
Department of Astronomy & Astrophysics, University of California Santa Cruz, 1156 High Street, Santa Cruz, CA 95064, USA
Aaron J. Romanowsky
Department of Physics & Astronomy, San Jose State University, One Washington Square, San Jose, CA 95192, USA andUniversity of California Observatories, 1156 High Street, Santa Cruz, CA, 95064, CA, USA
Jean Brodie
Department of Astronomy & Astrophysics, University of California Santa Cruz, 1156 High Street, Santa Cruz, CA 95064, USA
Jay Strader
Center for Data Intensive and Time Domain Astronomy, Department of Physics and Astronomy, Michigan State University, EastLansing, MI 48824, USA
Draft version May 31, 2019
ABSTRACTThe observed characteristics of globular cluster (GC) systems, such as metallicity distributions, arecommonly used to place constraints on galaxy formation models. However, obtaining reliable metallic-ity values is particularly difficult because of our limited means to obtain high quality spectroscopy ofextragalactic GCs. Often, “color–metallicity relations” are invoked to convert easier-to-obtain photo-metric measurements into metallicities, but there is no consensus on what form these relations shouldtake. In this paper we make use of multiple photometric datasets and iron metallicity values derivedfrom applying full-spectrum stellar population synthesis models to deep Keck/LRIS spectra of 177GCs centrally located around M87 to obtain a new color–metallicity relation. Our new relation differssubstantially from previous relations in the blue, and we present evidence that the M87 relation differsfrom that of the Milky Way GCs, suggesting environmental dependence of GC properties. We useour color–metallicity relation to derive a new GC metallicity-host galaxy luminosity relation for redand blue GCs and find a shallower relation for the blue GCs than what previous work has found andthat the metal-poor GCs are more enriched than what was previously found. This could indicate thatthe progenitor satellite galaxies that now make up the stellar halos of early-type galaxies are moremassive and formed later than previously thought, or that the properties of metal-poor GCs are lessdependent on their present-day host, indicating a common origin.
Keywords: INTRODUCTION
Although ΛCDM cosmology gives us the broad frame-work that galaxies form hierarchically, the details of howgiant early-type galaxies (ETGs) form is still a matterof debate. Areas of ongoing uncertainty include the as-sembly of ETGs such as the epoch of the last mergerand what kind of progenitor galaxies now constitute thestellar halos of ETGs. In particular, while cosmologi-cal simulations point to massive progenitor satellites asbuilding the stellar halos of present day giant ETGs (see,for example, Figure 13 of Pillepich et al. 2018), observa-tional constraints suggest dwarf galaxies as the progeni-tors (Figure 2 of Forbes et al. 2015).Globular clusters (GCs) are nearly ubiquitous aroundgalaxies and have been determined to be old ( ∼
10 Gyr)in a variety of systems (see references in Brodie andStrader 2006). Those properties as well as their lumi-nosity ( − < M V < −
10) make them potentially usefultracers of galaxy formation and assembly. However, thepromise of GCs in this capacity has yet to be fully real-ized, in part, because of our limited means to understand the present-day physical properties of GC systems.van den Bergh (1975) first used the likely connectionbetween a galaxy’s star-formation episodes and its GCpopulation to suggest a link between galaxy luminosityand the metallicities of its GCs. This relation was con-firmed by Brodie and Huchra (1991), and subsequentlythe paradigm of bimodality has overtaken the extragalac-tic GC field. Bimodality was first established throughoptical color distributions from
Hubble Space Telescope photometry (Gebhardt and Kissler-Patig 1999; Kunduand Whitmore 2001a; Larsen et al. 2001). Since thenGC systems around ETGs are treated as composed fromtwo subpopulations and separately track the subpop-ulation characteristics with host galaxy characteristicsto place constraints on galaxy formation scenarios (e.g.,Cˆot´e et al. 2002; Strader et al. 2005; Rhode et al. 2005; Liand Gnedin 2014). Recently though, Harris et al. (2017)presented observational evidence that the most massiveETGs, brightest cluster galaxies, can have broad uni-modal distributions in addition to bimodal distributions.GCs are thought to contain coeval stars with old agesand mostly homogenous metallicities and so broadband a r X i v : . [ a s t r o - ph . GA ] M a y colors of GCs are generally considered to reflect their un-derlying mean metallicity. The simplicity of this logic be-lies the fact that there is no consensus on how broadbandcolors should be transformed into metallicities (parame-terized as the “color–metallicity relation”). The core ofalmost all astronomical problems is translating observedcharacteristics into physically meaningful properties andunderstanding GC systems is no exception. We havevery limited means to obtain spectroscopy – our best ob-servational tool for deriving physical stellar populationcharacteristics – of individual GCs around the largest el-liptical galaxies. This is a result of a two-fold problem:at the distances of elliptical galaxies, GCs are faint, andthe largest elliptical galaxies can host systems of tens ofthousands of GCs. This means that in extragalactic workwe often only have access to coarse observational charac-teristics of individual GCs, such as broadband photome-try.The problems associated with obtaining the metallic-ity distribution are illustrated through the difference be-tween the Harris et al. (2006) and Peng et al. (2006)color–metallicity relations. Peng et al. (2006) used HST /ACS photometry of GCs around Virgo Clustergalaxies from Jord´an et al. (2004) and metallicities gath-ered from the few spectroscopic studies of extragalac-tic GCs available at the time (Cohen et al. 1998, 2003).Peng et al. (2006) found a color–metallicity relation witha significant break when transitioning to the blue GCs,but, crucially their relation was based almost entirely onMilky Way GCs at the metal-poor end. Harris et al.(2006) derived a linear relation between B − I colors andmetallicities for Milky Way GCs to interpret the broad-band colors they obtained for Virgo Cluster GC systems.Peng et al. (2006) and Harris et al. (2006) reported es-sentially the same color distributions for the Virgo GCsystems but different metallicity distributions.Despite their differences, both Peng et al. (2006) andHarris et al. (2006) maintained evidence for metallicitybimodality but that paradigm was challenged by Yoonet al. (2006). Yoon et al. (2006) introduced the idea ofgenerating synthetic color–metallicity relations to trans-form the overall color distributions of GC systems tometallicity distributions. They found that highly non-linear color–metallicity relations, like those that resultfrom inclusion of helium-rich hot horizontal branch stars,can transform unimodal metallicity distributions into bi-modal color distributions.Contrary to Yoon et al. (2006) and their follow-upwork (Lee et al. 2019), studies that directly model thespectroscopic observations of GCs consistently find bi-modal metallicity distributions (Alves-Brito et al. 2011;Usher et al. 2012; Brodie et al. 2012). Despite the near-consensus regarding bimodality, the differences in variouscolor–metallicity relations (see also Usher et al. 2012)highlight that there may be physical properties beyondmetallicity that affect the broadband colors of GCs.Full-spectrum stellar population synthesis (SPS) mod-eling provides a way to move past these problems. Mod-ern full-spectrum models allow for variations in abun-dance patterns (Conroy et al. 2014) over a variety of ages(Choi et al. 2014) and metallicities (Conroy et al. 2018).In addition to fully accounting for possible variationsin many stellar population parameters, we have shownthat full-spectrum fitting allows us to extract informa- tion from data in a lower signal-to-noise (S/N) regimethan traditional index fitting (Conroy et al. 2018).It is exactly this last property of full-spectrum SPSmodels that enables us to make use of the Straderet al. (2011) database of spectroscopy of individual GCsaround M87. In this paper we present the most com-prehensive and accurate compendium of metallicities forindividual GCs around M87 (which we describe in Sec-tion 2). We use these metallicities to derive a new color–metallicity relation in Section 3. We discuss the implica-tions of the new color–metallicity relation in Section 4. STELLAR POPULATION SYNTHESIS MODELING
We make use of the Keck/LRIS spectroscopic sub-sample of the dataset described in Strader et al. (2011)( ∼ − alf ) described in Conroy et al. (2018). The mostrelevant update of the Conroy et al. (2018) models withregards to this work is the expansion of stellar parame-ter coverage of the models with the Spectral PolynomialInterpolator (SPI, Villaume et al. 2017a) . With SPI weused the optical MILES stellar library (S´anchez-Bl´azquezet al. 2006), the Extended IRTF stellar library (E-IRTF,Villaume et al. 2017a), and a large sample of M Dwarfspectra (Mann et al. 2015) to create a data-driven modelwhich we can use to generate stellar spectra as a functionof effective temperature, surface gravity, and metallicity.The empirical parameter space is set by the E-IRTFand Mann et al. (2015) samples which together span − . (cid:46) [Fe / H] (cid:46) +0 . . (cid:46) log T eff (cid:46) .
5. Topreserve the quality of interpolation at the edges of em-pirical parameter space we augment the training set witha theoretical stellar library (C3K). The alf models allowfor variable abundance patterns by differentially includ-ing theoretical element response functions. In Conroyet al. (2018) we fitted the Schiavon et al. (2005) spec-troscopic sample of Milky Way GCs and compared the alf -inferred [Fe/H] values with a compilation of [Fe/H]values from the literature (see Roediger et al. 2014, fordetails). Over a range of − . (cid:46) [Fe / H] (cid:46) +0 . / H] lit ∝ . / H] alf ).The LRIS sample is in the low signal-to-noise (S/N)regime with ∼ −
30 encompassing the range of themedian S/N over each spectrum (bottom-right panel in https://github.com/AlexaVillaume/SPI_Utils D e c ( J ) g i ) i NGVSLRIS 5 10 15 20 25Median SNR (LRIS)0.02.55.07.510.012.515.017.5
Figure 1. (Upper) Image of M87 from the Burrell Schmidt Deep Virgo Survey (Mihos et al. 2017) with spatial distributions of the NGVSphotometry (yellow), ACSVCS photometry (green), and LRIS spectroscopy (blue), (Left) Color–magnitude diagram for the M87 GCs fromthe NGVS catalog from Oldham and Auger (2016) (grey), the culled sample of the LRIS data set (red) from Strader et al. (2011). (Right)Histogram of median signal-to-noise ratio values for the individual spectra of the LRIS sample. P o s t e r i o r ( F u ll - Sp e c t r u m F i t ) P o s t e r i o r ( I n d e x F i t ) Figure 2.
Comparing [Fe/H] posteriors for metal-rich GCs (left column) and metal-poor GCs (right column) where the spectra were fittedusing the full spectrum (top row) and Lick indices (row). In all cases the full-spectrum fits provide better constraints on [Fe/H] than indexfits. For the metal-rich GCs, the index fits have broader tails than the full-spectrum fits of the same GCs. For the metal-poor GCs, theindex fits do not result in well-behaved posterior distributions on [Fe/H]. The GCs in this figure correspond to the GCs in Figures 3 and 4. A r b i t r a r y F l u x S/N ~ 20S/N ~ 104000 4200 4400 4600 4800 5000 52001050510 R e s i d u a l ( % ) Å )1050510 R e s i d u a l ( % ) Figure 3.
Top: Comparison of metal-rich ([Fe/H] > +0 .
1) LRIS spectra (black) and best-fit models for a high-S/N (H51142, ( g − z ) NGVS =+1 .
38, brown) observation and a low-S/N (H51943, ( g − z ) NGVS = +1 .
33, green). Middle: Comparison of residuals between best-fit modeland data for H51075 and uncertainty of flux from the input spectrum (grey). Bottom: Same as middle panel but for H51943.
Figure 1). In this modest S/N regime it is difficult toobtain accurate stellar population parameters (S´anchez-Bl´azquez et al. 2011). To obtain an accurate color–metallicity relation we need the metallicities of individual
GCs and therefore stacking spectra is not a good optionfor this particular problem.We fit objects using both full-spectrum (left) and tra-ditional line-index methods (right). For our line-indexfits we use the canonical set of Lick indices (Faber et al.1985; Burstein et al. 1986; Worthey et al. 1994): H δ F ,CN , Ca4227, G4300, H γ F , Fe4383, Fe4531, C β ,Fe5015, Mg b , Fe5270, Fe345, and Fe5406. For the full-spectrum fits we fit in simple-mode over the wavelengthregions: 3900 − − − δ , H β , and Mg b , which are well-characterized bythe best-fit model.After we fit every spectrum we visually inspected theresiduals between the observed spectrum and the best-fit model. From this inspection we identified cases wherethe best-fit model is clearly a poor fit to the data. We re-moved these clusters from our subsequent analysis, bring-ing our final sample to 177 GCs. Of the 23 GCs weculled from our final metallicity sample, 20 have NGVSphotometry, and 15 of those are considered to be blue( g − z < . A r b i t r a r y F l u x S/N ~ 25S/N ~ 104000 4200 4400 4600 4800 5000 52001050510 R e s i d u a l ( % ) Å )1050510 R e s i d u a l ( % ) Figure 4.
Same as for Figure 3 but for the metal-poor GCs ( < − .
5) H38032 (( g − z ) NGVS = +0 .
70, brown) and H42981 (( g − z ) NGVS =+0 .
69, green). metal-poor GCs. However, with our remaining blue GCswe are still adequately covering the metal-poor param-eter space. The posteriors for the [Fe/H] values for thefinal sample of GCs are available at https://github.com/AlexaVillaume/m87-gc-feh-posteriors . RESULTS
Comparison to Previous Work
Cohen et al. (1998) previously did stellar populationanalysis on a spectroscopic sample of M87 GCs (Cohenand Ryzhov 1997) using indices to determine metallic-ity values. To aid our analysis we matched the Cohenet al. (1998) sample to the Oldham and Auger (2016)NGVS-based photometry catalog. We matched the Co-hen et al. (1998) sample to the data presented in Haneset al. (2001), which provided right ascension and decli-nation values for all the GCs in the Strom et al. (1981)catalog that Cohen and Ryzhov (1997) selected their tar-gets from.Then we used the position values to match with theOldham and Auger (2016) catalog with a max sepa-ration of 1 (cid:48)(cid:48) . We dereddened the Oldham and Auger(2016) photometry using the Fitzpatrick (1999) extinc-tion law and extinction values taken from the Schlegelet al. (1998) dust map using the NASA/IPAC InfraredScience Archive ( A g = 0 . A i = 0 . A z = 0 . R g = 3 . R i = 2 . R z = 1 . g − z ) colors of the two sample. Inthe right panel we compare the cumulative metallicitydistributions of both samples. We see that ∼
40% of theobjects in our sample are fainter than the faintest GCincluded in the Cohen et al. (1998) sample. The range ofcolors spanned by each sample are similar but the Cohenet al. (1998) sample has a different overall distributionthan our sample. More importantly, we see that fromthe way the curves change from color to metallicity thatthe Cohen et al. (1998) color–metallicity relation will bedifferent than ours. Furthermore, the Cohen et al. (1998)metallicities are, on the whole, lower than our metallici-ties. We discuss the nature of this last difference in moredetail in Section 4.1.
Updated color–metallicity Relationships C u m u l a t i v e D i s t r i b u t i o n F un c t i o n ( N o r m a li z e d ) Complete Cohen+ 1998 sampleCohen+ 1998 matched to NGVS
Figure 5.
Comparing the complete sample of M87 GC metallic-ities from Cohen et al. (1998) (140 GCs, blue) with the samplewhen matched to the Oldham and Auger (2016) photometry (101GCs, orange) to demonstrate that we are not biasing the Cohenet al. (1998) metallicity distribution by matching to photometry.
We use two photometric datasets of the M87 GC sys-tem: the Oldham and Auger (2016) catalog of groundbased photometry using the NGVS survey data (Fer-rarese et al. 2012) and photometry from the ACS VirgoCluster Survey (ACSVCS) from Jord´an et al. (2009). Weuse the g - and z -band filters from each survey but it isimportant to note that the filters are not identical be-tween the two instruments (see Figure 7) and so thecolor–metallicity relationships for the two instrumentswill be slightly different.Our sample of 177 spectroscopically-derived [Fe/H] val-ues overlaps with 172 objects from the NGVS catalog butonly 37 of the GCs with spectroscopically-derived metal-licities overlap with the ACSVCS catalog. To mitigateany problems that might arise from such a sparse samplewe leverage the fact that the underlying alf models ex-tend over a wider wavelength range than the LRIS dataand are flux calibrated (see Villaume et al. 2017a; Conroyet al. 2018, for discussion).We used the flux-calibrated models that correspond tothe inferred stellar parameters for each individual GCto compute synthetic photometry for both the ACSVCSand NGVS bandpasses. In Figure 7 we show the rela-tion between the synthetic photometry using the differ-ent filter systems. We also show our best-fit line to thedata (excluding the outliers marked with the open cir-cles) so that the colors of GCs can be transformed fromone system to the other. GCs identified as outliers bythe regression model are marked with open circles. Theoutliers from this relation are just the result of numer-ical problems for these particular clusters in generatingmodels over the available wavelength range. As can beseen in Figure 7, the overwhelming majority of the GCsfollow a tight relation between the ACSVCS filter systemand the NGVS system.In Figure 8 we show the color–metallicity relations us-ing the NGVS (left) and ACSVCS (right) photometryfor both the observed (top) and the synthetic (bottom) g − z colors. We fit all four color–metallicity relations us- Slope σ slope Intercept σ intercept σ residuals ACSVCS (obs) 1 .
79 0 . − .
77 0 .
31 2 . .
96 0 . − .
88 0 .
10 2 . .
12 0 . − .
92 0 .
13 2 . .
20 0 . − .
90 0 .
11 2 . Table 1
Median values of posterior distributions of best-fit line parameterswith standard deviations for each fit. We also show the standarddeviation of the residual [Fe/H] distributions, σ residuals . ing linear regression in a Bayesian framework with outlierpruning and uncertainty weighting (see Hogg et al. 2010,for details) and show the best-fit lines for each relationand 100 samples drawn from the posteriors in each panel(orange lines).We demonstrate that there is good agreement betweenthe relations using observed and synthetic NGVS pho-tometry. This is important because this assures us ofthe quality of the synthetic color–metallicity relation forthe ACS photometry. The relation using the observedACSVCS photometry has large uncertainties because ofthe sparsity of the sample.Any outliers detected by the fitting algorithm are high-lighted by red open circles in each panel. The regressionfits do not include those points. Linearity is a good rep-resentation of the data in all four cases. We fit the datawith a quadratic relation which was not statistically pre-ferred over the linear relation in any case. In Table 1we list the median and standard deviation of slope andintercept values of each relation.In Figure 9 we show the normalized histograms of theresiduals between the observed [Fe/H] values and the val-ues predicted by the best-fit color–metallicity relationsdivided by the observed [Fe/H] uncertainties. In eachpanel we show a standard normal distribution and in-dicate in the legend the measured mean and varianceof the residual distribution. The residuals have a largervariance than what is expected from a standard normaldistribution. This is likely because the color–metallicityrelations have genuine spread since GC systems are anamalgamation of different stellar populations.In the right panels of Figure 8 we also show the Penget al. (2006) relation. Our relation is consistent withPeng et al. (2006) for the red ( g − z > .
0) GCs but differssignificantly for the blue GCs. We already noted in theprevious section that the Cohen et al. (1998) metallicitiesused by Peng et al. (2006) are more metal-poor as a wholethan the metallicities that we have derived for the M87GCs. Peng et al. (2006) also supplemented their samplewith Milky Way GCs.To understand how the presence of Milky Way GCsmight have affected the color–metallicity relation we lookat how the Milky Way GCs compare to the M87 GCsin Figure 10. We generated synthetic photometry forthe Milky Way GCs to obtain ACS g − z colors for theclusters. We show the color–metallicity relation usingboth the [Fe/H] values we derived from our fits to theSchiavon et al. (2005) spectroscopy (brown circles) and[Fe/H] values compiled from various literature sources(Roediger et al. 2014, open green circles). We also showthe M87 GCs (black points). We show the best-fit linesfor the Milky Way GC color–metallicity relation (coloredlines) and the Peng et al. (2006) relation (dashed blackline). In the left panel we show the blue GCs and in the M i % ( < M i , N G V G S ) Cohen+ 1998Villaume+ 2019 g z ) % ( < g z ) % ( < [ F e / H ]) Milky Way GCs
Figure 6.
Left: Comparing the cumulative magnitude functions for the matched Cohen et al. (1998) sample and the sample from thiswork. Middle: Same as left but for ( g − z ). Right: Same as left and middle but for [Fe/H]. Also in the right panel we show the distributionof [Fe/H] values for Milky Way GCs from our full-spectrum fits to the Schiavon et al. (2005) data, which shows that the Milky Way GCsare typically more metal-poor than the M87 GCs. A C S V C S s y n t h e t i c p h o t o m e t r y Figure 7.
Relation between synthetic NGVS and ACSVCS pho-tometry for the spectroscopic sample. Since the two surveys areon slightly different filter systems we present a way to transformcolors between each: ( g − z ) ACSVCS = 1 . g − z ) NGVS − . right panel we show the red GCs.We see in Figure 10 that the blue Milky Way GCs havea different color–metallicity relation than the M87 GCs.The color–metallicity relations for the Milky Way GCsare closer to the Peng et al. (2006) relation, which makessense because it is the Milky Way GCs that drive the blueend of Peng et al. (2006) relation. Moreover, Peng et al.(2006) used the Harris (1996) compilation of Milky WayGC [Fe/H] values and we show that the color–metallicityrelation using [Fe/H] values from literature is even closerto the Peng et al. (2006) relation than the relation usingthe spectroscopically derived [Fe/H] values. Metallicity Distributions
In Figure 11 we demonstrate the effect of our newcolor–metallicity relations on the derived metallicity dis-tributions. In the left panel we compare the NGVS(yellow) and ACSVCS (green) color distributions. ForNGVS we only show clusters within R gal < . R gal ∼
13 kpc and we see bimodality clearlyin the color distribution for that sample. Meanwhile, theNGVS sample extends more than twice as far out andbimodality gets completely washed out in its color dis-tribution.In the middle panel we compare the spectroscopically-derived metallicity distribution (grey) with the metallic-ity distributions derived from the ACSVCS and NGVSsamples using their respective color–metallicity relationsfor two galactocentric radius cut-offs: R gal < . R gal < . R gal = 10 . [ F e / H ] Observed
Observed
Peng+ 20060.6 0.8 1.0 1.2 1.4 1.6( g z ) NGVS2.01.51.00.50.00.51.0 [ F e / H ] Synthetic g z ) ACSVCS2.01.51.00.50.00.51.0
Synthetic
Figure 8. (Top-left): Color–metallicity relation using observed NGVS g − z colorsfor the 172 GCs that are in both the spectroscopicsample and NGVS. (Bottom-left): Same as top-left with synthetic colors for all 177 GCs in the spectroscopic sample. (Top-right): Color–metallicity relation using observed ACSVCS colors for the 37 GCs that are in both that and the spectroscopic sample. (Bottom-right):Synthetic color–metallicity relation in the ACSVCS bands for all 177 GCs in the spectroscopic sample. In each panel we show the best-fitline and 100 samples drawn from the posterior distribution by fitting the corresponding data points with a linear model (see text fordetails). In the right panels we show the Peng et al. (2006) relation (dashed green). The regression algorithm detects outliers in the datawhich are shown in each plot by the red circles.
10 5 0 5 10NGVS (obs) [Fe/H]/0.000.050.100.150.20 = 0.25 = 2.62 10 5 0 5 10ACS (obs) [Fe/H]/0.000.050.100.150.20 = 0.10 = 2.14 10 5 0 5 10NGVS (syn) [Fe/H]/0.000.050.100.150.20 = 0.25 = 2.69 10 5 0 5 10ACS (syn) [Fe/H]/0.000.050.100.150.20 = 0.25 = 2.70
Figure 9.
Normalized histograms of the residuals between the observed [Fe/H] values and the values predicted by the best-fit color–metallicity relations divided by the observed [Fe/H] uncertainties. We have indicated the mean offset, µ , and standard deviation, σ for thedistribution of residuals. A Gaussian distribution with σ = 1 is also shown. g z ) ACS2.52.01.51.00.50.00.51.0 [ F e / H ] Peng+ 2006M87 GCsMW GC (alf)MW GC (lit)
Figure 10.
We show synthetic ACS g − z color versus metallicity for the M87 clusters (black) and the Milky Way GCs. The inclusion ofthe MW GCs in the Peng et al. (2006) analysis explains much of the discrepancy between our color–metallicity relations. g z ) R < . K p c NGVSACSVCS 1.5 1.0 0.5 0.0 0.5[Fe/H]0.00.20.40.60.81.0 From spectroscopyNGVS + ACSVCSNGVS + ACSVCS (R gal < 10.5) 2.0 1.5 1.0 0.5 0.0 0.5[Fe/H]0.00.20.40.60.81.0 Peng+ 2006
Figure 11. (Left): Distributions of the ( g − z ) colors from NGVS (yellow) and ACSVCS (green). The ACSVCS sample is redder and moremetal-rich than the NGVS sample on average because it is drawn from a more central region of the galaxy. We limited the NGVS sampleto objects within R gal < . R gal < . R gal < . B ( g - z ) Blue GCsRed GCs 212019181716 M B [ F e / H ] Villaume+ 2019Peng+ 2006
Figure 12. (Left) Mean values of the blue and red GC colors as a function of host galaxy luminosity in seven bins of host galaxymagnitude (see Peng et al. 2006, for details). (Right) Mean metallicities of the blue and red GC populations using the color–metallicityrelation determined in this work (solid lines) and the best-fit lines from Peng et al. (2006) (dashed lines). The different color–color metallicityestablished in this work propagates to a dramatically different metal-poor relation. ities. We derived uncertainties for the metallicity valuesby doing Monte Carlo sampling of the color–metallicityrelation using the color uncertainties.We show the linear fit to the new relations in the solidlines. We show the relations Peng et al. (2006) deter-mined as dashed lines. As we would expect from theprevious results, the new relation between host galaxyluminosity and mean metallicity for the metal-rich GCsis similar to the Peng et al. (2006) result but the relationfor the metal-poor GCs is shallower and more metal richthan the Peng et al. (2006) result. DISCUSSION
Which Metallicity is it Anyway?
The difference between our and the Peng et al. (2006)color–metallicity relationship is substantial for the blueGCs. We can understand this difference by examiningthe origin of the [Fe/H] values Peng et al. (2006) used intheir analysis. First, the Milky Way GCs make up themajority of the blue GCs used in the Peng et al. (2006)sample. We demonstrated in Figure 6 that the MilkyWay GCs are more metal-poor than the M87 GCs. In2Figure 10 we show that, using both literature [Fe/H] val-ues and [Fe/H] values derived from full-spectrum fitting,the Milky Way GCs have a different color–metallicity re-lation than the M87 GCs. The closeness of the Peng et al.(2006) relation in the blue to the Milky Way GC relationis highly suggestive that the presence of the Milky WayGCs is driving and biasing the relation in the blue forPeng et al. (2006).Second, we show in Figure 6 that even though the GCsin our sample and the Cohen et al. (1998) sample spana similar color range, the Cohen et al. (1998) metallicityvalues are systemically lower than the metallicities wederive. There are no GCs that are shared between theCohen et al. (1998) sample and our sample but we canunderstand the differences between the two by bearing inmind two related facts: the fitting functions that underliethe Worthey et al. (1994) models are not well-calibratedat high metallicities and the Cohen et al. (1998) metal-licities are placed on the Zinn (1985) metallicity scalewhich is set by Milky Way GCs.The former was discussed in Cohen et al. (2003) as aserious concern. Cohen et al. (2003) redid the [Fe/H] de-terminations of the M87 GCs from Cohen et al. (1998)by extrapolating the models to higher metallicity by as-suming that the indices are on the damping part of thecurve of growth. This affected five M87 GCs in their sam-ple. We are, to be clear, using the Cohen et al. (1998)metallicities in this work as Peng et al. (2006) did.For the latter, Cohen et al. (1998) noted that fromtheir qualitative analysis of the line indices of both theMilky Way and M87 GCs, the M87 GCs have a metal-richtail that extends to significantly higher metallicities thanthe Milky Way GCs, which we confirm. The relationthey use to scale the Worthey et al. (1994) models to theMilky Way GCs is [Fe / H] Z = 0 . × [Fe / H] W − . α elements(Tripicco and Bell 1995).We also find that the color–metallicity relation differsbetween the Milky Way and M87, especially near the blueend ( g − z (cid:46) . − .
5) and found thatthe metal-poor M87 GCs would have to be about 4 Gyryounger than the Milky Way GCs to explain the colordifference. We also cannot rule out the possible effectsthat α elements or the morphology of the blue horizontalbranch have on the color. Bimodality
Bimodality of GC systems has been the dominantparadigm in which extragalactic GC studies have been conducted over the past 30 years. In this paper we de-fer quantitative analysis of the subpopulations of the GCsystem around M87 to a forthcoming paper on the sub-ject. This is to more appropriately address the com-plexities around the topic that have been raised recently.Even with just the M87 system, consensus has yet to bereached on the number of subpopulations that make upthe system (e.g., Strader et al. 2011; Agnello et al. 2014;Oldham and Auger 2016). With that being said, thereare still some things worth pointing out.First, Cohen et al. (1998) detected bimodality in M87only after excluding the metal-rich tail from their anal-ysis. Usher et al. (2012) speculated that the lack of con-vincing proof for bimodality from Cohen et al. (1998)was a result of their typically bright targets. Since Co-hen et al. (1998) we have become aware of the blue-tilt phenomenon as well as liminal objects like ultra-compact dwarfs that could contaminate populations ofbright canonical GCs (e.g., Usher et al. 2012; Villaumeet al. 2017b).Second, we take advantage of obtaining color–metallicity relations using both the NGVS and ACSVCSdatasets by converting both into metallicity and combin-ing the data sets. The ACSVCS data probe the very in-ner region of the M87 GC system while the NGVS dataextends further out. We see that the color-convertedMDF is consistent with the spectroscopically determinedMDF. Furthermore, bimodality can be seen visually fromthe MDFs, especially when only GCs within R gal < . z ∼ Implications for GC and Galaxy Formation
We have derived a new galaxy luminosity–GC metal-licity relation separately for the blue and red GCs in theVirgo galaxies included in Peng et al. (2006) (Figure 12).The difference in our new color–metallicity relation istwo-fold: the metal-poor GCs now correlate with galaxyluminosity less strongly than previously measured, andthe metal-poor GCs are more metal-rich than what Penget al. (2006) determined.Larsen et al. (2001) were the first to assess the relation-ship between GC subpopulation metallicity and galaxy lu-minosity with a homogeneously acquired sample. ThenStrader et al. (2004) combined elliptical galaxy data froma variety of sources (Larsen et al. 2001; Kundu and Whit-more 2001a,b) with data from spiral galaxies (Harris1996; Barmby et al. 2000) to look at just the metal-poorGCs. Most recently Peng et al. (2006) determined thisrelationship for the Virgo Cluster galaxies. Like Larsenet al. (2001), Peng et al. (2006) found shallower slopesfor the metal-poor GCs relative to the metal-rich GCs.There is remarkable similarity between the slopes thatLarsen et al. (2001), Strader et al. (2004), and Peng et al.(2006) found for the metal-poor GCs.We already know that the difference between our re-lation and the relation from Peng et al. (2006) is due tothe color–metallicity relation. What about the differencewith Strader et al. (2004)? Strader et al. (2004) used theBarmby et al. (2000) color–metallicity relation based ona sample of M31 GCs. Barmby et al. (2000) noted thatthe M31 color–metallicity relation is similar to the MilkyWay relation. This raises the likelihood that it is not anappropriate way to convert colors to metallicities for theearly-type galaxies included in the Strader et al. (2004)sample. The similarity in slopes between Strader et al.(2004) and Peng et al. (2006) might be an artifact of thesimilar source of their respective color–metallicity rela-tions.To explain the correlation between galaxy luminosityand blue GC metallicity Strader et al. (2005) and Brodie and Strader (2006) invoked the concept of “biasing”, alsointroduced in the context of Milky Way stellar halo as-sembly by Robertson et al. (2005). In short, the pro-genitor satellites that now constitute the stellar halosof massive galaxies were more metal-rich, at fixed mass,than present day satellites. In the light of the new, muchweaker correlation, this needs to be reassessed. The newcorrelation could indicate that biasing is not as strongas an effect as once thought. Put another way, the newcorrelation suggests that metal-poor GCs formed irre-spective of their host galaxies.The change in metallicity intercept for the metal-poorGCs on this relation has implications for their formationepoch. Forbes et al. (2015) evolved the galaxy mass–GCmetallicity relation through redshift to determine bulkages of the GCs belonging to the galaxies in the SLUGGSsurvey (e.g., Usher et al. 2012). In their model, highermetallicities indicate younger ages and/or more massivehosts. The slopes of their metal-poor and metal-rich re-lations are not totally consistent with what we present inthis work, but the intercepts are roughly similar. Follow-ing the logic of Forbes et al. (2015), the nearly constantslope we find for the metal-poor GCs as a function ofgalaxy luminosity indicates that the metal-poor GCs inthe Virgo Cluster formed at nearly the same time. Thecorrelation between the metal-rich GC [Fe/H] values andhost galaxy luminosity indicates that the metal-rich GCsaround the giant galaxies formed more recently than themetal-rich GCs around the dwarf galaxies. The increasein metallicity for the metal-poor GCs may also help easethe tension between simulated and observational resultsas discussed in the Introduction, if it indicates that theGCs formed in more massive satellites.It is important to note the crucial underlying caveatof Figure 12 – that the color–metallicity relation we de-veloped for M87 is applicable to the other Virgo Clus-ter galaxies in the Peng et al. (2006) analysis. This isprobably not a good assumption, particularly in lightof the Powalka et al. (2016) results which showed thatcolor–color relations in the NGVS sample depend on en-vironment, with colors on the whole becoming bluer withincreased radial distance from M87 and that GCs > SUMMARY • We have fitted a spectroscopic sample of GCsaround M87 with full-spectrum SPS models andobtained [Fe/H] for 177 GCs. We demonstratethat the metallicity values we derive are systemati-cally higher-metallicity than previous spectroscopicstudies. We attribute this difference to the limita-tions of the previously-used Worthey et al. (1994)SPS models and because the previously determinedmetallicity values were scaled to match the MilkyWay GCs, which are, as a whole, lower in metallic-ity than the M87 GCs. • Using synthetic photometry from flux-calibratedstellar population models we determine a trans-4 formation between the NGVS and ACSVCS pho-tometric systems: ( g − z ) ACSVCS = 1 . g − z ) NGVS − . • We derived new color–metallicity relations usingboth NGVS and ACSVCS g − z colors. OurACSVCS color–metallicity relation differs signifi-cantly for the blue GCs from the previously pub-lished color–metallicity relation using the ACS fil-ters. This is because we find the relation for theMilky Way GCs to be significantly different thanthe relation for the M87 GCs. We discuss the nec-essary age difference needed to explain this result,but previous work in colors of Virgo Cluster GCssuggested that there is some environmental effecton chemical abundance patterns. • While we advocate that color–metallicity relationsbe confirmed with spectroscopic follow-up for in-dividual galaxies, we assume that in this respectthe Virgo cluster galaxies are similar to one an-other and as a result we find a shallower galaxyluminosity-GC metallicity relation for the blue GCsthan previous studies. This could either indicatethat progenitor satellites were less massive thanpreviously thought, or the properties of metal-poorGCs are not as dependent on their present-day hostgalaxy as metal-rich GCs.REFERENCES
A. Pillepich, D. Nelson, L. Hernquist, V. Springel, R. Pakmor,P. Torrey, R. Weinberger, S. Genel, J. P. Naiman,F. Marinacci, et al., MNRAS , 648 (2018), 1707.03406.D. A. Forbes, N. Pastorello, A. J. Romanowsky, C. Usher, J. P.Brodie, and J. Strader, MNRAS , 1045 (2015), 1506.06820.J. P. Brodie and J. Strader, Annual Review of Astronomy andAstrophysics , 193 (2006), astro-ph/0602601.S. van den Bergh, Annual Review of Astronomy and Astrophysics , 217 (1975).J. P. Brodie and J. P. Huchra, ApJ , 157 (1991).K. Gebhardt and M. Kissler-Patig, AJ , 1526 (1999),astro-ph/9906499.A. Kundu and B. C. Whitmore, AJ , 2950 (2001a),astro-ph/0103021.S. S. Larsen, J. P. Brodie, J. P. Huchra, D. A. Forbes, and C. J.Grillmair, AJ , 2974 (2001), astro-ph/0102374.P. Cˆot´e, M. J. West, and R. O. Marzke, ApJ , 853 (2002),astro-ph/0111388.J. Strader, J. P. Brodie, A. J. Cenarro, M. A. Beasley, and D. A.Forbes, AJ , 1315 (2005), astro-ph/0506289.K. L. Rhode, S. E. Zepf, and M. R. Santos, ApJ , L21 (2005),astro-ph/0507551.H. Li and O. Y. Gnedin, ApJ , 10 (2014), 1405.0763.W. E. Harris, S. M. Ciccone, G. M. Eadie, O. Y. Gnedin,D. Geisler, B. Rothberg, and J. Bailin, ApJ , 101 (2017),1612.08089.W. E. Harris, B. C. Whitmore, D. Karakla, W. Oko´n, W. A.Baum, D. A. Hanes, and J. J. Kavelaars, ApJ , 90 (2006),astro-ph/0508195.E. W. Peng, A. Jord´an, P. Cˆot´e, J. P. Blakeslee, L. Ferrarese,S. Mei, M. J. West, D. Merritt, M. Milosavljevi´c, and J. L.Tonry, ApJ , 95 (2006), astro-ph/0509654.A. Jord´an, J. P. Blakeslee, E. W. Peng, S. Mei, P. Cˆot´e,L. Ferrarese, J. L. Tonry, D. Merritt, M. Milosavljevi´c, andM. J. West, The Astrophysical Journal Supplement Series ,509 (2004), astro-ph/0406433.J. G. Cohen, J. P. Blakeslee, and A. Ryzhov, ApJ , 808(1998), astro-ph/9709192.J. G. Cohen, J. P. Blakeslee, and P. Cˆot´e, ApJ , 866 (2003),astro-ph/0304333. S.-J. Yoon, S. K. Yi, and Y.-W. Lee, Science , 1129 (2006),astro-ph/0601526.S.-Y. Lee, C. Chung, and S.-J. Yoon, The Astrophysical JournalSupplement Series , 2 (2019), 1811.00018.A. Alves-Brito, G. K. T. Hau, D. A. Forbes, L. R. Spitler,J. Strader, J. P. Brodie, and K. L. Rhode, MNRAS , 1823(2011), 1107.0757.C. Usher, D. A. Forbes, J. P. Brodie, C. Foster, L. R. Spitler,J. A. Arnold, A. J. Romanowsky, J. Strader, and V. Pota,MNRAS , 1475 (2012), 1207.6402.J. P. Brodie, C. Usher, C. Conroy, J. Strader, J. A. Arnold, D. A.Forbes, and A. J. Romanowsky, ApJ , L33 (2012),1209.5390.C. Conroy, G. J. Graves, and P. G. van Dokkum, ApJ , 33(2014), 1303.6629.J. Choi, C. Conroy, J. Moustakas, G. J. Graves, B. P. Holden,M. Brodwin, M. J. I. Brown, and P. G. van Dokkum, ApJ ,95 (2014), 1403.4932.C. Conroy, A. Villaume, P. G. van Dokkum, and K. Lind, ApJ , 139 (2018).J. Strader, A. J. Romanowsky, J. P. Brodie, L. R. Spitler, M. A.Beasley, J. A. Arnold, N. Tamura, R. M. Sharples, andN. Arimoto, The Astrophysical Journal Supplement Series ,33 (2011), 1110.2778.J. C. Mihos, P. Harding, J. J. Feldmeier, C. Rudick,S. Janowiecki, H. Morrison, C. Slater, and A. Watkins, ApJ , 16 (2017), 1611.04435.L. J. Oldham and M. W. Auger, MNRAS , 820 (2016).A. Jord´an, E. W. Peng, J. P. Blakeslee, P. Cˆot´e, S. Eyheramendy,L. Ferrarese, S. Mei, J. L. Tonry, and M. J. West, TheAstrophysical Journal Supplement Series , 54 (2009).A. Villaume, C. Conroy, B. Johnson, J. Rayner, A. W. Mann,and P. van Dokkum, The Astrophysical Journal SupplementSeries , 23 (2017a).P. S´anchez-Bl´azquez, R. F. Peletier, J. Jim´enez- Vicente,N. Cardiel, A. J. Cenarro, J. Falc´on-Barroso, J. Gorgas,S. Selam, and A. Vazdekis, MNRAS , 703 (2006),astro-ph/0607009.A. W. Mann, G. A. Feiden, E. Gaidos, T. Boyajian, and K. vonBraun, ApJ , 64 (2015), 1501.01635.R. P. Schiavon, J. A. Rose, S. Courteau, and L. A. MacArthur,The Astrophysical Journal Supplement Series , 163 (2005),astro-ph/0504313.J. C. Roediger, S. Courteau, G. Graves, and R. P. Schiavon, TheAstrophysical Journal Supplement Series , 10 (2014),1310.3275.P. S´anchez-Bl´azquez, P. Ocvirk, B. K. Gibson, I. P´erez, and R. F.Peletier, MNRAS , 709 (2011), 1103.3796.S. M. Faber, E. D. Friel, D. Burstein, and C. M. Gaskell, TheAstrophysical Journal Supplement Series , 711 (1985).D. Burstein, S. M. Faber, and J. J. Gonzalez, AJ , 1130 (1986).G. Worthey, S. M. Faber, J. J. Gonzalez, and D. Burstein, TheAstrophysical Journal Supplement Series , 687 (1994).J. G. Cohen and A. Ryzhov, ApJ , 230 (1997),astro-ph/9704051.D. A. Hanes, P. Cˆot´e, T. J. Bridges, D. E. McLaughlin,D. Geisler, G. L. H. Harris, J. E. Hesser, and M. G. Lee, ApJ , 812 (2001), astro-ph/0106004.S. E. Strom, J. C. Forte, W. E. Harris, K. M. Strom, D. C. Wells,and M. G. Smith, ApJ , 416 (1981).E. L. Fitzpatrick, Publications of the Astronomical Society of thePacific , 63 (1999), astro-ph/9809387.D. J. Schlegel, D. P. Finkbeiner, and M. Davis, ApJ , 525(1998), astro-ph/9710327.L. Ferrarese, P. Cˆot´e, J.-C. Cuilland re, S. D. J. Gwyn, E. W.Peng, L. A. MacArthur, P.-A. Duc, A. Boselli, S. Mei,T. Erben, et al., The Astrophysical Journal Supplement Series , 4 (2012).D. W. Hogg, J. Bovy, and D. Lang, ArXiv e-printsarXiv:1008.4686 (2010), 1008.4686.W. E. Harris, AJ , 1487 (1996).R. Zinn, ApJ , 424 (1985).J. Choi, A. Dotter, C. Conroy, M. Cantiello, B. Paxton, andB. D. Johnson, ApJ , 102 (2016), 1604.08592.M. J. Tripicco and R. A. Bell, AJ , 3035 (1995).A. Agnello, N. W. Evans, A. J. Romanowsky, and J. P. Brodie,MNRAS , 3299 (2014), 1401.4461. A. Villaume, J. Brodie, C. Conroy, A. J. Romanowsky, andP. van Dokkum, ApJ , L14 (2017b).A. J. Romanowsky, J. Strader, J. P. Brodie, J. C. Mihos, L. R.Spitler, D. A. Forbes, C. Foster, and J. A. Arnold, ApJ , 29(2012), 1112.3959.J. Pfeffer, J. M. D. Kruijssen, R. A. Crain, and N. Bastian,MNRAS , 4309 (2018).N. Choksi, O. Y. Gnedin, and H. Li, MNRAS , 2343 (2018),1801.03515.J. Strader, J. P. Brodie, and D. A. Forbes, AJ , 3431 (2004),astro-ph/0403160.A. Kundu and B. C. Whitmore, AJ , 1251 (2001b),astro-ph/0105198.P. Barmby, J. P. Huchra, J. P. Brodie, D. A. Forbes, L. L.Schroder, and C. J. Grillmair, AJ , 727 (2000),astro-ph/9911152.B. Robertson, J. S. Bullock, A. S. Font, K. V. Johnston, andL. Hernquist, ApJ , 872 (2005), astro-ph/0501398. M. Powalka, T. H. Puzia, A. Lan¸con, E. W. Peng, F. Sch¨onebeck,K. Alamo-Mart´ınez, S. ´Angel, J. P. Blakeslee, P. Cˆot´e, J.-C.Cuilland re, et al., ApJ , L5 (2016), 1608.08628.F. P´erez and B. E. Granger, Computing in Science andEngineering , 21 (2007), ISSN 1521-9615, URL http://ipython.org . SciPy: Open source scientific tools for python (2001), URL .S. Van Der Walt, S. C. Colbert, and G. Varoquaux, Computing inScience & Engineering , 22 (2011).J. D. Hunter, Computing In Science & Engineering , 90 (2007).M. B. Taylor, in Astronomical Data Analysis Software andSystems XIV , edited by P. Shopbell, M. Britton, and R. Ebert(2005), vol. 347 of
Astronomical Society of the PacificConference Series , p. 29., p. 29.