1: \documentclass[12pt,preprint]{aastex}

2: 

3: \def\etal{et~al.}

4: % Next 5 lines define \simless and \simgreat: "less than or approximately

5: % equal to" and "greater than or approximately equal to".

6: \newbox\grsign \setbox\grsign=\hbox{$>$} \newdimen\grdimen \grdimen=\ht\grsign

7: \newbox\simlessbox \newbox\simgreatbox

8: \setbox\simgreatbox=\hbox{\raise.5ex\hbox{$>$}\llap

9:      {\lower.5ex\hbox{$\sim$}}}\ht1=\grdimen\dp1=0pt

10: \setbox\simlessbox=\hbox{\raise.5ex\hbox{$<$}\llap

11:      {\lower.5ex\hbox{$\sim$}}}\ht2=\grdimen\dp2=0pt

12: \def\simgreat{\mathrel{\copy\simgreatbox}}

13: \def\simless{\mathrel{\copy\simlessbox}}

14: % Next lines define "approximately proportional to"

15: \newbox\simppropto

16: \setbox\simppropto=\hbox{\raise.5ex\hbox{$\sim$}\llap

17:      {\lower.5ex\hbox{$\propto$}}}\ht2=\grdimen\dp2=0pt

18: \def\simpropto{\mathrel{\copy\simppropto}}

19: 

20: 

21: \shorttitle{EZ\_Ages Algorithm}

22: \shortauthors{Graves \& Schiavon}

23: 

24: \begin{document}

25: 

26: \title{Measuring Ages and Elemental Abundances from Unresolved Stellar

27:   Populations: Fe, Mg, C, N, and Ca}

28: 

29: \author{Genevieve J. Graves\altaffilmark{1} \& Ricardo P. Schiavon\altaffilmark{2}}

30: 

31: \altaffiltext{1}{UCO/Lick Observatory, University of California, Santa

32: 	Cruz, CA 95064}

33: \altaffiltext{2}{Gemini Observatory, 670 N. A'ohoku Place, Hilo, HI 96720}

34: 

35: \begin{abstract}

36: 

37: We present a method for determining mean light-weighted ages and

38: abundances of Fe, Mg, C, N, and Ca, from medium resolution

39: spectroscopy of unresolved stellar populations. The method, pioneered

40: by Schiavon (2007), is implemented in a publicly available code called

41: EZ\_Ages. The method and error estimation are described, and the

42: results tested for accuracy and consistency, by application to

43: integrated spectra of well-known Galactic globular and open

44: clusters. Ages and abundances from integrated light analysis agree

45: with studies of resolved stars to within $\pm0.1$dex for most

46: clusters, and to within $\pm0.2$ dex for nearly all cases. The results

47: are robust to the choice of Lick indices used in the fitting to within

48: $\pm0.1$ dex, except for a few systematic deviations which are clearly

49: categorized. The realism of our error estimates is checked through

50: comparison with detailed Monte Carlo simulations. Finally, we apply

51: EZ\_Ages to the sample of galaxies presented in \citet{tho05} and

52: compare our derived values of age, [Fe/H], and [$\alpha$/Fe] to their

53: analysis. We find that [$\alpha$/Fe] is very consistent between the

54: two analyses, that ages are consistent for old ($Age > 10$ Gyr)

55: populations, but show modest systematic differences at younger ages,

56: and that [Fe/H] is fairly consistent, with small systematic

57: differences related to the age systematics. Overall, EZ\_Ages provides

58: accurate estimates of fundamental parameters from medium resolution

59: spectra of unresolved stellar populations in the old and

60: intermediate-age regime, for the first time allowing quantitative

61: estimates of the abundances of C, N, and Ca in these unresolved

62: systems.

63: 

64: \end{abstract}

65: 

66: \keywords{methods: data analysis, galaxies: abundances, galaxies: star

67: clusters}

68: 

69: \section{Introduction}

70: 

71: The study of stellar populations in galaxies is going through very

72: exciting transformations.  The advent of extensive and high-quality

73: data sets for galaxies in the local and distant universe poses a

74: demand for ever improving stellar population synthesis models.  In

75: this paper, we present a new tool, named EZ\_Ages, that simultaneously

76: address two on-going needs in the field of stellar population

77: synthesis: 1) it extracts accurate ages and abundances of several

78: elements from integrated spectra of stellar populations, and 2) it

79: performs this task in an automatic fashion, which is well-suited for

80: applications to large data sets.

81: 

82: Early stellar population synthesis models for single-burst

83: populations---so called ``single stellar populations'' or SSPs---were

84: based on low resolution stellar spectra.  The models of \citet{wor94}

85: and \citet{vaz96} used optical absorption-line indices with 8--10

86: {\AA} resolution and limited spectral coverage, while \citet{bc93}

87: synthesized model spectra with broad wavelength coverage (ultraviolet

88: through infrared) and resolution of 10--20 {\AA} in the optical.  All

89: models were for solar-abundance ratios only.  Since then, a new

90: generation of SSP models have pushed to higher resolution (2--3 {\AA}

91: for \citealt{bc03} and \citealt{vaz03}) and to variable abundance

92: patterns (\citealt{bor95}; \citealt{wei95}; \citealt{tan98};

93: \citealt{tra00}; \citealt{tho03}; \citealt{mar05}; \citealt{sch07} and

94: \citealt{coe07}).

95: 

96: It is well known that many stellar populations, including the Galactic

97: halo and bulge, Galactic globular clusters, massive early-type

98: galaxies, and the bulges of spiral galaxies, have super-solar

99: [$\alpha$/Fe]. In resolved systems, where abundance analysis of

100: individual stars is possible, it has been known for a while that not

101: all $\alpha$ elements are equally enhanced.  On the other hand, in

102: unresolved systems it has been only recently that evidence started

103: accumulating to the fact that observed stellar populations have

104: abundance patterns which cannot be described by only two parameters

105: (ie.  [Fe/H] and [$\alpha$/Fe]) but rather require more complicated

106: abundance patterns.  In particular, it seems that not all the $\alpha$

107: elements are equally enhanced in early-type galaxies (e.g.,

108: \citealt{vaz97}; \citealt{wor98}; \citealt{tra98}; \citealt{hen99};

109: \citealt{tho03}; \citealt{sch07}).  In the best studied, closest,

110: resolved spheroidal system, the Galactic bulge, there is also recent

111: evidence that not all $\alpha$-elements are equally enhanced

112: \citep{ful05,ful07}.  In particular, \citet{ful07} found that, while

113: Mg is strongly enhanced relative to Fe ([Mg/Fe] $\sim$ +0.3 across a

114: range of [Fe/H]), other $\alpha$ elements such as Si, Ca, Ti, and O

115: tend to follow a disk-like trend, with [X/Fe] decreasing with

116: increasing [Fe/H].  Interestingly, in the case of oxygen---arguably

117: the most important among the $\alpha$ elements---they found [O/Fe]

118: $\approx 0$ at [Fe/H] = 0.  \citet{ben04} show a similar result for

119: Milky Way disk stars in both the thin and thick disk.  \citet{ath03}

120: demonstrates that warm ionized gas in a sample of 7 local early-type

121: galaxies has mean [O/H] $\approx -0.1$, while the stellar populations

122: of the galaxies have super-solar [Mg/H].  Clues to variations of the

123: abundances of other elements, such as C and N, particularly as a

124: function of environment, have been found in a number of studies.  The

125: Lick indices C$_2$4668 and CN$_2$, which are sensitive to C and O (and

126: N, in the case of CN$_2$), have been shown to be weaker in early-type

127: galaxies in dense environments as compared to galaxies in the field,

128: while the strength of the Mg~{\it b} index remains unchanged across

129: environments \citep{san03}.  Finally, a number of studies have found

130: evidence that Ca seems to behave as an Fe-group element in elliptical

131: galaxies, rather than following the trend of other $\alpha$ elements

132: (\citealt{vaz97}; \citealt{wor98}; \citealt{tra98}; \citealt{hen99};

133: \citealt{tho03}).  This result has recently been questioned by

134: \citet{pro05} and \citet{sch07}, who show that the Ca4227 index is

135: strongly affected by CN lines, which may be partially responsible for

136: the low [Ca/Fe] suggested by other authors.  Although

137: $\alpha$-enhanced models are an important improvement over

138: solar-scaled-only, it is fundamental that models with more degrees of

139: freedom be developed, which are able to explore the complexity

140: suggested by the data.  By establishing a more detailed snapshot of

141: the abundance patterns of stars in galaxies, these models can pose

142: fundamental constraints on models of galaxy formation.

143: 

144: \citet{sch07} presents a new set of SSP models which do exactly that.

145: These models explicitly include adjustable abundance patterns for

146: multiple elements, allowing the user to separately vary the abundances

147: of C, N, O, Mg, Ca, Na, Si, Cr, and Ti.  This paper describes an

148: algorithm for searching through the space of variable abundance models

149: to find the abundance mix which best fits the Lick indices measured in

150: a given input stellar population spectrum.  This algorithm has been

151: implemented in an IDL code package, called ``EZ\_Ages'' which is

152: publicly available for use.  Section \ref{seq_grid} describes the

153: sequential grid inversion algorithm.  In \S\ref{compare_clusters}, we

154: present ages and abundances estimated by EZ\_Ages for the Galactic

155: globular clusters NGC 6121, 47 Tuc, NGC 6441, and NGC 6528, as well as

156: the open cluster M67.  These values are compared to results in the

157: literature determined from isochrone fitting of cluster photometry and

158: from high-resolution spectral abundance analysis for individual stars

159: in the clusters.  In \S\ref{compare_galaxies} we apply EZ\_Ages to

160: galaxy data from the literature, contrasting our results with those by

161: other authors.  Section \ref{conc} summarizes our conclusions.

162: 

163: The beta-version of EZ\_Ages is available for download, along with

164: instructions on installation and

165: use.\footnote[1]{www.ucolick.org/$\sim$graves/EZ\_Ages.html}

166: 

167: \section{Sequential Grid Inversion Algorithm}\label{seq_grid}

168: 

169: \subsection{Schiavon Simple Stellar Population Models}

170: 

171: The code EZ\_Ages is designed for use with the single stellar

172: population models described in \citet[hereafter S07]{sch07}.  The

173: details of the model are described in S07; here we briefly discuss the

174: aspects of the model relevant to its practical use.

175: 

176: These models include a choice of two sets of isochrones by the Padova

177: group: one is the solar-scaled isochrones of \citet{gir00} and the

178: other is the $\alpha$-enhanced set of isochrones by \citet{sal00} with

179: average [$\alpha/$Fe$] = +0.42$ and with individual abundance ratios

180: chosen to match metal-poor field stars.\footnote[2]{The following

181: $\alpha$ elements are enhanced as shown: [O/Fe] = $+0.50$, [Ne/Fe] =

182: $+0.29$, [Mg/Fe] = $+0.40$, [Si/Fe] = $+0.30$, [S/Fe] = $+0.33$,

183: [Ca/Fe] = $+0.50$, [Ti/Fe] = $+0.63$, [Ni/Fe] = $+0.02$.  All other

184: abundance-ratios are solar.}  However, the models based on the

185: $\alpha$-enhanced isochrone must be used with caution because

186: \citet{wei06} have shown that, due to a problem with the opacity

187: tables adopted, those isochrones predict temperatures that are

188: slightly too high at both the main-sequence turnoff and the giant

189: branch.  The effect is strongest at solar metallicity, where it

190: results in ages that are slightly too old.  For details, see \S4.3 of

191: S07.  Developing isochrones with variable abundance ratios is an

192: active area of current research (see recent work by \citealt{dot07})

193: and we hope to incorporate a broader range of isochrones into EZ\_Ages

194: in the near future.  

195: 

196: In addition to a choice of isochrones, the S07 models allow the user

197: to select an input abundance pattern.  The abundance ratios that can

198: be varied are: [Mg/Fe], [C/Fe], [N/Fe], [O/Fe], [Ca/Fe], [Na/Fe],

199: [Si/Fe], [Cr/Fe], and [Ti/Fe].  The S07 models output predicted Lick

200: index measurements for a grid of age and [Fe/H] values, based upon the

201: input abundance ratios.  The models span the age range $0.1 \le t \le

202: 15.8$ Gyr, with $-1.3 \le$ [Fe/H] $\le +0.2$ for the solar-scaled

203: isochrone and $-0.8 \le$ [Fe/H] $\le +0.3$ for the $\alpha$-enhanced

204: isochrone, though only models older than 0.8 Gyr are computed with

205: variable abundance ratios.

206: 

207: The Lick indices included in the S07 model are H$\delta_A$,

208: H$\delta_F$, H$\gamma_A$, H$\gamma_F$, H$\beta$, CN$_1$, CN$_2$,

209: Ca4227, G4300, Fe4383, C$_2$4668, Fe5015, Mg$_2$, Mg~{\it b}, Fe5270,

210: and Fe5335, as defined in Table 1 of \citet{wor94} and Table 1 of

211: \citet{wor97}.  The output model indices can then be compared with

212: Lick indices measured in the integrated spectrum of a real stellar

213: population (i.e., a stellar cluster or a galaxy) to determine the

214: best-fitting mean light-weighted age and [Fe/H] for the stellar

215: population, for a given set of model abundance ratios.

216: 

217: The Lick indices included in the S07 model are summarized in Table

218: \ref{indextable}, along with their main abundance sensitivities.

219: These are taken from the works of \citet{ser05} and \citet{kor05}.

220: The sensitivities from \citet{kor05} are taken from their Tables 4 and

221: 5, which give abundance sensitivities for turnoff stars and giant

222: branch stars respectively.  We have chosen to include only those two

223: evolutionary phases in the table, as they dominate the stellar

224: population model spectrum (accounting for $\sim 90$\% of the

225: integrated light), even though all three evolutionary phases are taken

226: into account in the model computations.  In addition to the

227: sensitivities to individual element abundances shown in the table, all

228: indices are sensitive to changes in the total metallicity.  We

229: interpret the effect of [Z/H] on Fe5270 and Fe5335 as being due

230: primarily to changes in Fe because none of the other elements

231: investigated by KMT05 appear to affect these indices.  Table

232: \ref{indextable} shows reasonable agreement between the two different

233: sets of sensitivity tables.

234: 

235: Abundance variations enter the model in two ways, firstly through the

236: choice of isochrone (solar-scaled or $\alpha$-enhanced).  At this

237: stage, the model takes into account the effect of the abundance mix on

238: stellar evolution---specifically, how the chemical mix affects

239: temperatures and luminosities of stars of different masses and at

240: different evolutionary stages.  Admittedly, this is done in an

241: approximate fashion, given that unfortunately we only have isochrones

242: computed for two abundance patterns.  Secondly, the effect of

243: individual elemental abundances on line and continuum opacities in

244: stellar atmospheres is incorporated into model Lick index predictions,

245: using the response functions computed by \citet{kor05}.  At this

246: second stage, it is possible to vary elemental abundances

247: independently; at the first stage, the only choices are

248: [$\alpha$/Fe]~=~0 and $+0.42$, with total metallicity varying.

249: 

250: It should be stated clearly that these models are cast in terms of

251: [Fe/H], and not total metallicity, [Z/H].  This reflects our choice to

252: deal explicitly with quantities that can be inferred from measurements

253: taken in the integrated spectra of galaxies---total metallicity not

254: being one of them, given our current inability to use integrated

255: spectra of stellar populations to constrain the most abundant of all

256: metals, oxygen (see discussion in \S4.4.1 of S07).  Another advantage

257: with casting models in terms of [Fe/H] is that each elemental

258: abundance can be treated separately, so that the effect of its

259: variation can be studied in isolation from every other elemental

260: abundance (at the cost of varying the total metallicity).  In the case

261: of models cast in terms of [Z/H], it is impossible to vary the

262: abundance of a single element, because enhancing one element means

263: decreasing the abundances of all other elements to keep total

264: metallicity constant.  As an example, in models cast in terms of

265: [Z/H], a solar-metallicity ([Z/H]~=~0) $\alpha$-enhanced population

266: has only a slightly higher abundance of $\alpha$ elements than the

267: Sun, while having a much lower iron-abundance.  It is the large iron

268: abundance difference that is responsible for the bulk of the

269: difference between those two hypothetical models, {\it not} the small

270: difference in $\alpha$-element abundances.  For this reason, claims

271: that higher order Balmer lines are strongly affected by

272: $\alpha$-element abundances \citep{tho04} should be taken with caution

273: (see \S4.3 S07 for a thorough discussion of this point).

274: 

275: \subsection{Fiducial Age and [Fe/H]}

276: 

277: In practice, one would like to use absorption line measurements

278: in the integrated spectrum of a given stellar population to determine

279: not only the age and [Fe/H] for a fixed set of abundance ratios,

280: but to find the best set of input abundances as well.  The brute-force

281: method of finding the best-fitting abundances would involve generating

282: a set of models that span the available space of age, [Fe/H], and

283: abundance patterns, and then identifying the model whose predicted

284: Lick indices best match those measured in the spectrum.  However,

285: with nine variable abundance ratios, the search space quickly becomes

286: very large.  If each of those abundance ratios is represented by

287: only four possible values, with 23 model ages and four possible

288: values of [Fe/H], that would mean creating more than 24 million

289: models!

290: 

291: Instead, we choose to perform a directed search for the best model,

292: taking advantage of the fact that various Lick indices are sensitive

293: to only a few, often different, elemental abundances, as indicated in

294: Table~\ref{indextable}.  These sensitivity variations have been

295: modeled by several groups using spectrum synthesis, based on model

296: stellar atmospheres (see recent papers by \citealt{kor05} and

297: \citealt{ser05}).  A more detailed discussion of the motivation behind

298: this method can be found in S07.

299: 

300: EZ\_Ages begins by computing a set of models with the chosen isochrone

301: (either solar-scaled or $\alpha$-enhanced), using solar abundance

302: ratios for the stellar atmospheres.  It then uses a pair of lines that

303: are sensitive to age and [Fe/H], but relatively {\it insensitive} to

304: other elemental abundances---a Balmer line and an Fe line---to

305: determine a fiducial age and [Fe/H] for the stellar population in

306: question.  From Table 1 we can see that H$\beta$ and the iron lines

307: Fe5270 and Fe5335 are good choices for determining the fiducial, as

308: they are mostly insensitive to other element abundances.  In EZ\_Ages,

309: the default choice of indices for calculating fiducial age and [Fe/H]

310: are H$\beta$ and $\langle$Fe$\rangle$, an average of Fe5270 and

311: Fe5335.

312: 

313: The top left panel of Figure \ref{grids} shows a plot of model grids

314: for H$\beta$ and $\langle$Fe$\rangle$.  Dotted lines connect constant-age models

315: (from top to bottom: 1.2, 2.2, 3.5, 7.0, and 14.1 Gyr) while solid

316: lines connect models with the same [Fe/H] (from left to right: -1.3,

317: -0.7, -0.4, 0.0, and +0.2).  The square shows a sample data point for

318: a galaxy with age $\approx 7$ Gyrs and [Fe/H] $\approx -0.2$.  In this

319: figure, it is easy to see which box of the grid encloses the data

320: point in question.  This gives a bounded range in age and [Fe/H] for

321: the data point.  A two-dimensional linear interpolation within the

322: gridbox is then used to convert the data point in index-index space

323: into fiducial values for the age and [Fe/H].

324: 

325: \subsection{Determining Non-Solar Abundance Ratios}  \label{method}

326: 

327: Once a fiducial age and [Fe/H] have been determined from the

328: H$\beta$--$\langle$Fe$\rangle$ model grid, one can then make a similar grid plot

329: with an index which is sensitive to non-solar abundance ratios, for

330: instance a Balmer line plotted against Mg~{\it b}.  Mg~{\it b} is

331: strongly affected by the [Mg/Fe] abundance ratio, at fixed [Fe/H].  If

332: the chosen S07 model is a good match to the data, then the age and

333: [Fe/H] estimated by grid inversion from the H$\beta$--Mg~{\it b} plot

334: should match the fiducial age and [Fe/H] from the H$\beta$--$\langle$Fe$\rangle$

335: plot.  In our example in Figure \ref{grids}, this is clearly not the

336: case; the H$\beta$--Mg~{\it b} plot in the top right panel of Figure

337: \ref{grids} gives a larger value of [Fe/H] than the H$\beta$--$\langle$Fe$\rangle$

338: plot in the top left panel. This is an indication that the example

339: galaxy has a larger [Mg/Fe] ratio than is predicted in a solar-scaled

340: model.  The sequential grid inversion algorithm then increases the

341: input value of [Mg/Fe] and recomputes the S07 model.  Increasing

342: [Mg/Fe] will increase the strength of the Mg~{\it b} index {\it at

343: fixed [Fe/H]}, which will have the effect of sliding the entire set of

344: model grids to the right in the H$\beta$--Mg~{\it b} plot, while

345: keeping them essentially unchanged in the H$\beta$--$\langle$Fe$\rangle$ plane.

346: This~{\it lowers} the value of [Fe/H] that is estimated by grid

347: inversion in the H$\beta$--Mg~{\it b} plot, bringing it into agreement

348: with that obtained in the H$\beta$--$\langle$Fe$\rangle$ diagram.  This can been

349: seen in the lower panels of Figure \ref{grids}, which show

350: H$\beta$--$\langle$Fe$\rangle$ and H$\beta$--Mg~$b$ plots for an S07 model

351: computed with [Mg/Fe]$=+0.3$.  For the data shown in this plot, the

352: age and [Fe/H] values indicated by the model grids are consistent for

353: both metal lines.  In the sequential grid-inversion algorithm

354: implemented in EZ\_Ages, [Mg/Fe] is increased incrementally until the

355: [Fe/H] estimated from the H$\beta$--Mg~{\it b} plot matches the

356: fiducial [Fe/H] from the H$\beta$--$\langle$Fe$\rangle$ plot.  This process is

357: then repeated for other Lick indices and other abundance ratios.

358: 

359: The key to the sequential grid inversion algorithm is to proceed with

360: the abundance fitting in such a way as to only adjust one abundance at

361: a time.  Once fiducial values for the age and [Fe/H] of the system

362: have been determined, the goal is to next fit a Lick index that

363: introduces only one additional abundance.  Mg~{\it b} is a good choice

364: because it is dominated by Mg and Fe (see Table 1), with the latter

365: being determined from the H$\beta$--$\langle$Fe$\rangle$ grid

366: inversion.  By adjusting [Mg/Fe] until the H$\beta$--Mg~{\it b} grid

367: inversion matches the fiducial values, EZ\_Ages can find the

368: best-fitting Mg abundance.

369: 

370: Another index which EZ\_Ages can use to fit a single element is

371: C$_2$4668, which is dominated by C and O.  Oxygen is notoriously

372: difficult to estimate in old (ie. non-star forming) unresolved stellar

373: populations, therefore EZ\_Ages does not try to model [O/Fe] (but see

374: discussion in \S\ref{consist}).  Instead, it allows the user to choose

375: an input oxygen abundance.  For consistency, this should be chosen to

376: match the input isochrone: [O/Fe] = 0.0 for the solar-scaled

377: isochrone, [O/Fe] = +0.5 for the $\alpha$-enhanced isochrone.  With

378: [O/Fe] set by the user, this leaves only C as a variable abundance

379: which contibutes strongly to C$_2$4668.  EZ\_Ages adjusts [C/Fe] so

380: that the [Fe/H] estimated in the H$\beta$--C$_2$4668 plot

381: matches the fiducial.

382: 

383: The only indices which are strongly affected by nitrogen and which

384: might therefore allow a fit of [N/Fe] are the two CN lines, CN$_1$ and

385: CN$_2$, both of which respond strongly to changes in C, N, and O.  As

386: discussed above, [O/Fe] is set by the user.  This leaves C and N as

387: variable abundances contributing to the CN indices.  However, if

388: [C/Fe] has already been determined by fitting C$_2$4668, then the CN

389: indices can be used to fit for [N/Fe] at the calculated value of

390: [C/Fe].  The computed value of [N/Fe] is therefore dependent upon the

391: calculated value of [C/Fe] and a robust error analysis should propagate

392: errors from [C/Fe] into the the errors in [N/Fe].  

393: 

394: The S07 models include one Lick index dominated by Ca and can

395: therefore be used to fit [Ca/Fe]: Ca4227.  Because the blue continuum

396: region used in measuring Ca4227 overlaps with a CN absorption band,

397: this index is also very sensitive to [C/Fe] and [N/Fe], so that these

398: abundance ratios must be fit properly before using Ca4227 to determine

399: [Ca/Fe].  As with [N/Fe], errors in the calculated values of [C/Fe]

400: and [N/Fe] contribute to the error in [Ca/Fe].  Again, this should be

401: taken into consideration in a complete error analysis (see

402: \S\ref{errors}).

403: 

404: The set of available Lick indices and their element abundance

405: sensitivities thus prescribe an order for abundance fitting: [C/Fe],

406: then [N/Fe], then [Ca/Fe].  Note that Mg~{\it b}, the preferred

407: index used to fit [Mg/Fe], does not depend strongly on these

408: abundances, nor do the other abundances depend on it, therefore it

409: can be fit in any order.

410: 

411: Based on Table \ref{indextable}, it is clear that some indices are

412: preferable for fitting: those that are dominated by the elemental

413: abundances that we are using them to estimate and no others.  Because

414: of this, EZ\_Ages has a built-in hierarchy which it uses to choose the

415: indices used in abundance fitting.  Among the Balmer lines, the

416: preferred index is H$\beta$ because it is mildly sensitive to total

417: metallicity, but relatively insensitive to individual abundances.  It

418: is followed by H$\delta_F$ and H$\gamma_F$, which are somewhat

419: sensitive to individual elements, and lastly H$\delta_A$ and

420: H$\gamma_A$, which are broader indices (designed to measure Balmer

421: strengths in populations with a substantial A star component) and

422: therefore include a larger number of other absorption features in the

423: index and pseudo-continuum bandpasses.  

424: 

425: For iron indices, the order of preference is $\langle$Fe$\rangle$ (an

426: average of Fe5270 and Fe5335), Fe5270, Fe5335, Fe4383, and Fe5015.

427: The indices Fe5270 and Fe5335 are dominated by Fe and total

428: metallicity effects, making them relatively clean indicators of

429: [Fe/H].  Of the other two lines, both of which are influenced by

430: element abundance ratios, we prefer Fe4383 over Fe5015 for two

431: reasons: Fe5015 falls on a CCD defect in the globular cluster data

432: used in \S\ref{compare_clusters} to test the output of EZ\_Ages and

433: thus can only be tested for M~67, where it appears to perform

434: similarly well to Fe4383, and the Fe5015 bandpass overlaps the

435: [O\textsc{iii}]$\lambda$5007 emission line from ionized gas sometimes

436: present in galaxies, thus limiting its usefulness in stellar

437: population modeling of galaxies.

438: 

439: For fitting [Mg/Fe], the preferred index is Mg~{\it b} followed by

440: Mg$_2$, which is more sensitive to C than is Mg~{\it b}.  For [C/Fe],

441: it is C$_2$4668 followed by G4300, for reasons discussed in detail in

442: \S\ref{consist}.  For [N/Fe], it is CN$_1$ followed by CN$_2$,

443: although this is fairly arbitrary as there is no strong reason to

444: prefer one CN index over the other.  For [Ca/Fe], the only available

445: index is Ca4227.

446: 

447: It is possible to manually override this order of preference if a

448: given index is unavailable, or to investigate the effects of using

449: other lines in the fit.  Note that EZ\_Ages will only fit an abundance

450: ratio if at least one of the indices needed to estimate it is

451: available.  Also, abundances which depend on other abundance ratios

452: will not be fit if the prerequisite abundance is not available (e.g.,

453: EZ\_Ages will not use a CN index to fit [N/Fe] unless [C/Fe] has been

454: determined from C$_2$4668 or G4300).

455: 

456: Thus far, we have focused only on the dominant abundances affecting

457: each index, however many indices have a weak dependence on other

458: abundances.  Particularly if the most desirable Lick indices are not

459: available due to limited wavelength coverage, bad skyline subtraction,

460: or nebular emission within the galaxy and other lines have to be used

461: (such as H$\delta_F$ in place of H$\beta$ or Fe4383 in place of Fe5270

462: or Fe5335), the computed fiducial values for age and [Fe/H] or other

463: abundance values may change once the abundance fitting is complete,

464: because some of the newly determined abundances may affect the indices

465: used to determine those fiducial values.  To ensure that the final

466: abundance ratio is consistent for all of the Lick indices used in the

467: fit, EZ\_Ages iterates the fitting process up to four times.  It uses

468: the best-fitting abundance pattern from the previous iteration to

469: compute new fiducial values for age and [Fe/H], then repeats the

470: fitting process until further iterations do not improve the fit.

471: 

472: The S07 models account for variations in Lick index strength due to

473: abundance variations in the elements Fe, Mg, C, N, Ca, O, Na, Si, Cr,

474: and Ti.  However, only the first five of these elements can be fit

475: using the algorithm described above.  The elements O, Na, Si, Cr, and

476: Ti do not dominate any of the Lick indices modeled by S07 and

477: therefore cannot be fit in this process.  The EZ\_Ages code allows the

478: user to supply a value for any of these unmodeled elements to use as

479: input to the S07 models.  In the absence of user-supplied values,

480: EZ\_Ages has a set of default abundances for these elements.  By

481: default, it sets [O/Fe] to match the chosen isochrone ([O/Fe] = 0.0

482: for the solar-scaled isochrones, [O/Fe] = $+0.5$ for the

483: $\alpha$-enhanced isochrone).  The default for the $\alpha$-elements

484: Na, Si, and Ti is that they are set to follow Mg, so that [Na/Fe] =

485: [Si/Fe] = [Ti/Fe] = [Mg/Fe].  The iron-peak element Cr is set to

486: follow Fe, so that [Cr/Fe] = 0.0.  All of the modeling reported in

487: this analysis was done using the default settings for the unmodeled

488: elements and with the solar-scaled isochrone.  To see the effect of

489: using the $\alpha$-enhanced isochrone or super-solar [O/Fe] on

490: abundance fitting, see \S4.3 of S07 and Figures 12 and 14 in

491: \citet{gra07}.

492: 

493: \subsection{Error Calculations} \label{errors}

494: 

495: If errors in the Lick index data are provided as input, EZ\_Ages

496: uses them to estimate the uncertainties in the ages, [Fe/H], and

497: abundance ratios calculated by the sequential grid inversion

498: algorithm.  Uncertainties are assumed to be dominated by measurement

499: errors in the line strengths.  Systematic uncertainties in the models

500: are ignored, except that the usual caveat applies: comparisons of {\it

501: relative} ages and abundances between multiple objects are more

502: reliable than absolute age or abundance estimates.

503: 

504: Uncertainties in age and [Fe/H] due to line strength measurement

505: errors are computed by displacing the measured data point by the input

506: index error, then redoing the grid inversion on the displaced data

507: point to determine the range in age or [Fe/H] constrained by 1$\sigma$

508: error in the measured Balmer or Fe line strength.  Because the grid

509: spacing is non-linear (particularly in age), errors in the positive

510: and negative direction are computed separately.  It has been noted by

511: many previous authors that the errors in SSP age and [Fe/H] as

512: inferred from index-index plots are not independent due to the fact

513: that the model grids are not orthogonal.  As can be seen from an

514: inspection of Figure \ref{grids}, age and [Fe/H] errors are correlated

515: in the sense that lower inferred ages will lead to higher inferred

516: [Fe/H], and vice versa.  EZ\_Ages does not explicitly model this

517: aspect of the inferred errors in age and [Fe/H] values, but users

518: should keep this effect in mind when interpreting results from SSP

519: index-index fitting.

520: 

521: Any error in the choice of fiducial age and [Fe/H] will result in

522: errors in the abundance ratios estimated by fitting to incorrect

523: values of the age or [Fe/H].  Thus uncertainties in the fiducial age

524: and [Fe/H] must be propagated through to determine their effect on the

525: abundance ratio estimates.

526: 

527: For example, EZ\_Ages assumes that the uncertainty in [Mg/Fe] is

528: dominated by two sources: the measurement error of the Mg-sensitive

529: Lick index used in the abundance ratio determination (e.g., Mg~{\it

530: b}) and the uncertainty in the fiducial [Fe/H] which EZ\_Ages

531: attempts to match in the H$\beta$--Mg~{\it b} plot.  In the following

532: discussion, we will use $\Delta$Mg~{\it b} and $\Delta$[Fe/H] to

533: represent the measurement error in Mg~{\it b} and the uncertainty in

534: the fiducial [Fe/H], respectively.  The uncertainty in [Mg/Fe] due to

535: Mg~{\it b} measurement error and fiducial uncertainty are denoted in

536: turn by $\Delta$[Mg/Fe]$_{\rm{Mg}b}$ and

537: $\Delta$[Mg/Fe]$_{\rm{fid}}$.

538: 

539: To determine $\Delta$[Mg/Fe]$_{\rm{Mg}b}$, EZ\_Ages recomputes

540: the [Mg/Fe] abundance for Mg~{\it b}$+\Delta$Mg~{\it b} and Mg~{\it

541: b}$-\Delta$Mg~{\it b}, using the larger of these two uncertainties as

542: the estimated $\Delta$[Mg/Fe]$_{\rm{Mg}b}$.  To determine

543: $\Delta$[Mg/Fe]$_{\rm{fid}}$, EZ\_Ages recomputes [Mg/Fe] using

544: the measured value of Mg~{\it b} to fit [Fe/H]$+\Delta$[Fe/H] and

545: [Fe/H]$-\Delta$[Fe/H], again choosing the larger uncertainty for

546: estimating $\Delta$[Mg/Fe]$_{\rm{fid}}$.  These two sources of error

547: are treated as independent and the total $\Delta$[Mg/Fe] is determined

548: by summing in quadrature $\Delta$[Mg/Fe]$_{\rm{Mg}b}$ and

549: $\Delta$[Mg/Fe]$_{\rm{fid}}$.  

550: 

551: The process for computing $\Delta$[C/Fe] is identical to that for

552: [Mg/Fe], substituting a C-sensitive Lick index for Mg~{\it b}.

553: Because [N/Fe] is determined using a CN-sensitive index such as

554: CN$_1$, $\Delta$[N/Fe] depends not only on the CN$_1$ line strength

555: measurement error and $\Delta$[Fe/H], but on $\Delta$[C/Fe] as well.

556: The contributions to $\Delta$[N/Fe] due to $\Delta$CN$_1$ and

557: $\Delta$[Fe/H] (i.e., $\Delta$[N/Fe]$_{\rm{CN1}}$ and

558: $\Delta$[N/Fe]$_{\rm{fid}}$) are computed as above.  An additional

559: contribution due to $\Delta$[C/Fe] is determined by matching the

560: measured CN$_1$ line strength to the fiducial [Fe/H], using models

561: with carbon abundances given by [C/Fe]$+\Delta$[C/Fe] and

562: [C/Fe]$-\Delta$[C/Fe].  These three sources of error are summed in

563: quadrature to obtain $\Delta$[N/Fe].  A similar process is repeated

564: for [Ca/Fe], with the sole difference that errors in [N/Fe] also

565: contribute to the [Ca/Fe] error budget.

566: 

567: To check that this analysis provides a reasonable estimate of the

568: actual statistical errors, we have used a Monte Carlo method to

569: generate a large set of artificial ``repeat measurements'' of the Lick

570: indices for a test stellar population and compared the resulting

571: distribution of ages, abundances, and abundance ratios to the errors

572: calculated by EZ\_Ages.  For our test data, we used Lick index values

573: measured in a stack of $\sim$10,000 red sequence galaxy spectra taken

574: from the Sloan Digital Sky Survey.  As modeled by {\it EZ\_Ages},

575: these correspond to a stellar population with a mean light-weighted

576: age of 9.1 Gyr, with [Fe/H]$=-0.15$, [Mg/Fe]$=+0.17$, [C/Fe]$=+0.17$,

577: [N/Fe]$=+0.07$, and [Ca/Fe]$=0.00$.

578: 

579: We ran four Monte Carlo simulations for different error ranges

580: (effectively different $S/N$ spectra), in which we assumed the errors

581: were 2\%, 5\%, 10\%, and 20\% of the measured index values.  For each

582: run, we simulated 100 repeat measurements of the data by choosing

583: values for each Lick index drawn from Gaussian distributions centered

584: on the ``true'' data values with widths characterised by the assumed

585: 2\%, 5\%, 10\%, and 20\% errors.  This produced a simulation of 100

586: realizations for each of the four assumed error ranges.  {\it

587: EZ\_Ages} was used to determine the age, [Fe/H], and abundance ratios

588: for each simulated measurement.  The distribution of these values for

589: the 100 simulated repeat measurements can be compared with the error

590: estimate made by EZ\_Ages for the original data values and index

591: measurement errors.  

592: 

593: This comparison is shown in Figure \ref{errplot}.  The errors reported

594: by EZ\_Ages are shown on the x-axis, while the y-axis shows the

595: standard deviation $\sigma$ of the distributions of ages and

596: abundances produced in each Monte Carlo simulation.  As discussed

597: earlier in this section, EZ\_Ages calculates separately errors in the

598: positive and negative direction.  There are thus two error values

599: produced by EZ\_Ages plotted against each Monte Carlo $\sigma$.  The

600: solid line shows a one-to-one relation.  The dashed lines outline

601: error estimates that are within $\pm30$\% of the 1$\sigma$ Monte Carlo

602: spread.

603: 

604: For the simulations with 2\%, 5\%, and 10\% index measurement errors,

605: the age and abundance ratio error estimates from EZ\_Ages are all

606: within 30\% of the results of the Monte Carlo simulation with the

607: exception of the error in the H$\gamma_F$ age, which is overestimated

608: by about 0.3 Gyr in the 2\% error case.  Although this is a

609: substantial fraction of the simulated H$\gamma_F$ age error, it is

610: well within the uncertainties in the modeling process, which are of

611: order 1 Gyr in the $\sim9$ Gyr age range corresponding to the test

612: data point \citep{sch07}.  In all other cases, the EZ\_Ages error

613: estimate is an excellent match to the 1$\sigma$ errors produced by the

614: full Monte Carlo realization.  For the simulations with 20\% index

615: measurement errors, EZ\_Ages tends to slightly overestimate the errors

616: as compared to the Monte Carlo simulation, but in all cases is still

617: within 35\% of the simulated error.

618: 

619: The good match shown in this test between the error estimates

620: generated by EZ\_Ages and those produced by a Monte Carlo simulation

621: of the errors suggests that EZ\_Ages does in fact do a reasonable job

622: of estimating the errors in the age, [Fe/H], and abundance ratio

623: values calculated.  This includes the propagation of errors between

624: various stages of the modeling process.  As the data become noisier,

625: EZ\_Ages has a tendency to slightly overestimate the actual

626: uncertainty in the age, [Fe/H], and abundance due to measurement

627: errors.  Figure \ref{errplot} also demonstrates that in order to

628: estimate ages, [Fe/H], and abundance ratios within $\pm$0.1 dex for a

629: typical early type galaxy, Lick indices must be measured with $\pm5$\%

630: accuracy, which requires spectra with $S/N \sim 100$ {\AA}$^{-1}$

631: \citep{car98}.

632: 

633: \subsection{Correlation of Errors in Derived

634:   Parameters}\label{correlated_errs} 

635: 

636: The Monte Carlo error simulation also allows us to investigate the

637: effect of correlated errors in the EZ\_Ages analysis.  Because lines

638: of constant age and [Fe/H] are not orthogonal in

639: H$\beta$-$\langle$Fe$\rangle$ space, errors in either index

640: measurement will result in anti-correlated errors in the inferred age

641: and [Fe/H], as discussed in detail in \citet{tra00}.  For example, an

642: error in H$\beta$ that causes the determined SSP age to be too old

643: will cause the measured [Fe/H] value to be too low, as can be seen

644: from the grids in Figure \ref{grids}.  Likewise, the sequential grid

645: inversion process for fitting abundance ratios depends on the fiducial

646: age and particularly on the fiducial [Fe/H] determination.  This will

647: tend to produce an anti-correlation between [Fe/H] and the measured

648: abundance ratios.

649: 

650: As shown in Figure \ref{errplot}, the simplified error propagation

651: implemented in EZ\_Ages does an excellent job of tracking the total

652: error in each stellar population parameter throughout the fitting

653: process.  However, it does not produce a full error ellipse

654: illustrating correlated errors.

655: 

656: Figure \ref{corr_errs} uses the results of the Monte Carlo error

657: simulation to illustrate the correlated errors that result from

658: sequential abundance fitting.  The 100 simulated data realizations for

659: each of the assumed 2\%, 5\%, 10\%, and 20\% index measurement errors

660: (yellow, red, blue, and green points, respectively) effectively trace

661: out the error ellipses in stellar population parameter space.  Figure

662: \ref{corr_errs}a shows the well-known error anti-correlation between

663: the age and [Fe/H] measurements, as expected from the

664: non-orthogonality of the age and [Fe/H] model grids in

665: H$\beta$-$\langle$Fe$\rangle$ space.

666: 

667: Panels b--e show the abundance ratio determinations for [Mg/Fe],

668: [C/Fe], [N/Fe], and [Ca/Fe] against [Fe/H].  The error ellipses for

669: these data are expected to show an anti-correlation of errors because

670: an over-estimate of [Fe/H] will tend to result in an under-estimate of

671: the other abundances.  For example, if an error in the

672: $\langle$Fe$\rangle$ index causes the fiducial [Fe/H] determination to

673: be too high, the measured abundance ratios for other elements will

674: tend to be too low (at a higher [Fe/H], a lower value of [Mg/Fe] is

675: needed to match the same Mg~$b$ index).  This anti-correlation is

676: observed between [Fe/H] and the abundances [Mg/Fe], [C/Fe], and

677: [Ca/Fe].  Interestingly, there appears to be no correlation in the

678: errors in [Fe/H] and [N/Fe].  Because [N/Fe] is fit using a CN index,

679: it depends strongly on the value of [C/Fe] previously determined by

680: fitting either C$_2$4668 or G4300 and thus we expect a

681: anti-correlation of errors between [C/Fe] and [N/Fe].  This is indeed

682: the case, as illustrated in Figure \ref{corr_errs}f, and may explain

683: the lack of correlation in the errors in [N/Fe] and [Fe/H].  As in the

684: case of [N/Fe], the fitted values of [Ca/Fe] depend on other abundance

685: fits and thus may produce correlated errors.  These are illustrated in

686: Figure \ref{corr_errs}g,h.

687: 

688: The effect of correlated errors can produce spurious trends in a

689: dataset, thus the correlations illustrated here should be kept in mind

690: when interpreting abundance patterns from EZ\_Ages or a similar

691: abundance fitting process.  Of course, the best guard against these

692: effects is to perform abundance fitting only on spectra with $S/N \ge

693: 100$ \AA$^{-1}$, corresponding to 5\% index errors (red points in

694: Figure \ref{corr_errs}) or better.

695: 

696: \section{Testing the Models on Cluster Spectra}\label{compare_clusters}

697: 

698: The validation of stellar population synthesis models depends

699: crucially on their ability to match the integrated properties of

700: simple systems, such as stellar clusters, which can be reasonably

701: approximated by single stellar populations.  In fact, this simple

702: truism disguises the considerable complexity of the task, as attested

703: by the vast literature dedicated to using cluster observations to

704: constrain stellar population synthesis models (S07; \citealt{lee05};

705: \citealt{pro04}; \citealt{sch04a,sch04b}; \citealt{mar03};

706: \citealt{sch02a,sch02b}; \citealt{bea02}; \citealt{vaz01};

707: \citealt{gib99}; \citealt{sch99}; \citealt{bru97}; \citealt{ros94};

708: and references therein).  S07 presented a lengthy discussion of the

709: comparison of his models with integrated data for well-known Galactic

710: open and globular clusters.  Models were computed for a range of ages

711: and metallicities, adopting the known abundance patterns of M~5,

712: 47~Tuc, NGC~6528, and M~67. These were then compared with Lick index

713: measurements taken in the (very high S/N) cluster spectra.  As a

714: result, it was found that predictions for Balmer and metal lines

715: agreed with the data to within 1-$\sigma$ index measurement errors.

716: Also very importantly, the best-matching ages differed by no more than

717: $\sim$1~Gyr from the value obtained from analysis of cluster

718: color-magnitude diagrams (CMDs).  The only exception was M~5, for

719: which the age found was too low, due to the presence of blue

720: horizontal branch (HB) stars, which are not accounted for by the

721: models (see \S\ref{consist} for a further discussion of this point).

722: 

723: In this section we ask the inverse question: how well do the models

724: fare when used to convert a given set of index measurements into ages

725: and elemental abundances?  Again, we rely on data for clusters whose

726: abundances and ages are well constrained from analysis of individual

727: stellar spectra and deep cluster CMDs.  Two important model checks are

728: performed.  We start by submitting the models to the most challenging

729: of the two tests (\S\ref{real}), which consists of checking whether

730: the results obtained with EZ\_Ages match those from detailed resolved

731: analysis of high-resolution spectra and deep photometry of stellar

732: members.  While our {\it goal} is to match ages and the abundances of

733: Fe, Mg, C, N, and Ca to within 0.1 dex our {\it requirement} is that

734: differences be smaller than 0.2 dex.  Moreover, this set of

735: specifications should be met in a quick and non-interactive fashion,

736: suitable for applications to large data sets through use of EZ\_Ages.

737: As a second check (\S\ref{consist}), we verify the internal

738: consistency of EZ\_Ages, by comparing the results obtained for a given

739: set of observations, when using different combinations of Lick indices

740: as age and metal-abundance indicators.

741: 

742: The clusters of choice for this exercise are selected on the basis

743: of availability of good quality age and metal abundance determinations

744: (Table~\ref{abuntable}).  They span a wide range of metallicities,

745: from [Fe/H] $\sim$ --1.2 (NGC~6121) to nearly solar metallicity

746: (M~67 and NGC~6528), and include both old and intermediate-age

747: (M~67) clusters.  Lick indices for these clusters are presented in

748: Table \ref{cluster_indices}.

749: 

750: \subsection{Reality Check}  \label{real}

751: 

752: In Figure~\ref{cluster_grids} cluster data are compared with the S07

753: solar abundance pattern models in a few index-index plots to

754: illustrate that solar-scaled models {\it cannot} match the full set of

755: observed cluster indices.  The indices displayed are the default age

756: and metal-abundance indicators employed by EZ\_Ages.  S07 showed that

757: H$\beta$, $\langle$Fe$\rangle$, Mg~$b$, C$_2$4668, CN$_1$ (or CN$_2$),

758: and Ca4227 are the most reliable, recommended indices for age, [Fe/H],

759: [Mg/Fe], [C/Fe], [N/Fe], and [Ca/Fe] determinations.  This will be

760: explored further in \S\ref{consist}.  Henceforth, we will refer to

761: this set of indices as the {\it standard set}.

762: 

763: The models displayed in Figure~\ref{cluster_grids} were computed for

764: the solar abundance pattern, and the data were measured in the

765: integrated spectra of the above globular clusters from the

766: \citet{sch05} spectral library and that of M~67, obtained by

767: \citet{sch04a}, and revised by S07.  In the upper left panel, model

768: and data are compared in the $\langle$Fe$\rangle$ vs. H$\beta$ plane.

769: S07 showed that these indices are excellent indicators of [Fe/H] and

770: age, respectively.  While the $\langle$Fe$\rangle$ average index is

771: relatively insensitive to the abundances of elements other than Fe,

772: H$\beta$ is mostly sensitive to age, being only mildly sensitive to

773: the overall metallicity of the stellar population.  Therefore, the age

774: and iron abundance of each cluster can be read fairly accurately from

775: the position of the cluster data on the model grid---except for the

776: case of NGC~6121 where the presence of blue HB stars affects the

777: H$\beta$-based age (see discussion in \S\ref{consist}).  The remaining

778: panels show plots of other metal-line indices against H$\beta$, and

779: $\langle$Fe$\rangle$ against H$\delta_F$.  In the metal-line vs.\

780: H$\beta$ plots, one can see that the positions on the model grids

781: where the data for most clusters fall are different from those they

782: occupy on the $\langle$Fe$\rangle$ vs. H$\beta$ plot.  As discussed in

783: \S\ref{method}, this is a clear indication that the abundance pattern

784: of the cluster stars is different from that of the Sun.  We apply the

785: method presented in the previous section to determine the abundance

786: pattern of the clusters.

787: 

788: We ran EZ\_Ages on the cluster data presented in

789: Table~\ref{cluster_indices}. The results are listed in

790: Table~\ref{abuntable}, together with abundance and age determinations

791: from the literature.  The literature values are based on abundance

792: analyses of high-resolution spectra of individual cluster stars and on

793: deep cluster CMDs, respectively.  Differences between our results and

794: those from the literature are displayed in Figure~\ref{comp_lit}.  In

795: that plot, the dashed lines indicate the $\pm0.1$ dex goal stated in

796: the EZ\_Ages specifications discussed in \S\ref{compare_clusters}.

797: Average values from the literature are displayed in black and

798: individual determinations are shown in gray.  For carbon and nitrogen,

799: the values displayed are averages between CN-strong and CN-weak stars,

800: as this best approximates the effect of unresolved spectroscopic

801: observations.

802: 

803: A first inspection of the results shows that most abundance points

804: fall between the two dashed lines, indicating that we are meeting the

805: 0.1-dex goal for most of the abundances and for most clusters.

806: Agreement is in fact particularly good for the abundances of carbon

807: and nitrogen, where the 0.1-dex goal was reached for all clusters for

808: which these elemental abundances are available in the literature.  For

809: magnesium and iron, we meet the 0.1-dex goal for some clusters, and

810: the 0.2-dex requirement for all of them, with one exception: for

811: NGC~6528, we find [Fe/H] = $-0.26$, in reasonably good agreement with

812: the value [Fe/H] = $-0.15$ reported by \citet{zoc04} and \citet{bar04}

813: but in significant disagreement with the value [Fe/H] = $+0.1$

814: reported by \citet{car01}.  For calcium, there is one cluster only for

815: which the 0.2-dex requirement is not met.  For all the others, we meet

816: the 0.1-dex goal.  Given that this is the first systematic attempt at

817: deriving abundances for all these five elements from integrated light,

818: these results can be considered very satisfactory.  Below we discuss

819: the few exceptions for which agreement is not so good.

820: 

821: Although the values of [Fe/H] determined by EZ\_Ages are almost all

822: within the 0.2-dex requirement, they are consistently slightly lower

823: than the literature abundances determined from invidual stars.  The

824: zeropoint for [Fe/H] in the models is based upon the [Fe/H]

825: measurements for the stellar library from which the S07 model is

826: constructed (the S07 models have not been separately calibrated to the

827: Solar spectrum) and the zeropoint is not known to perfect accuracy.  A

828: systematic 0.1-dex underprediction of [Fe/H] by the models is

829: possible.  However, this is not necessarily more uncertain than the

830: zeropoints in the cluster abundance determinations from individual

831: stars.  In the case of NGC~6528, for which multiple literature sources

832: are available, the abundance determinations in the literature show

833: substantial scatter: 0.25 dex in [Fe/H], 0.07 dex in [Mg/Fe], and 0.63

834: in [Ca/Fe].  In this context, it is difficult to address zeropoint

835: issues on the order of 0.1 dex by comparison with existing cluster

836: data.

837: 

838: In general we note that, for some clusters, elemental abundances

839: determined by different groups are in substantial disagreement, which

840: makes the task of interpreting our results a little more complicated.

841: That is especially the case of NGC~6528 mentioned above, for which

842: iron abundances from two groups differ by $\sim$ 0.25 dex, and calcium

843: abundances differ by more than 0.6 dex!  While a definitive statement

844: on our ability to match that cluster's abundances with EZ\_Ages awaits

845: the solution of this discrepancy, we call attention to the fact that

846: agreement with the average values is very good.

847: 

848: EZ\_Ages also performs very well in age determinations.  For all

849: clusters, with the sole exception of NGC~6121, the 0.2-dex requirement

850: is met, and for 47~Tuc and M~67, the 0.1-dex goal is also achieved.

851: For NGC~6121, the age obtained is too young because of the

852: contribution by blue HB stars which tend to boost the strength of

853: Balmer lines in the spectra of old stellar populations, mimicking a

854: younger age (see \S\ref{consist}).  The effect of blue HB stars on

855: Balmer lines has been pointed out in previous studies

856: (\citealt{fre95}; \citealt{mt00}; \citealt{sch04b}; \citealt{tra05};

857: \citealt{sch07}) and it can potentially lead to age underestimates.

858: 

859: The cluster NGC~6441 also contains blue HB stars \citep{ric97}, but

860: EZ\_Ages determines a relatively old age for the cluster when using

861: the H$\beta$ index in the modelling process, which seems to contradict

862: this interpretation.  Unfortunately, we could not find a CMD-based age

863: for this cluster in the literature so we cannot make a statement about

864: how the age result from EZ\_Ages compares to an independent age

865: estimate.  However, when the bluer Balmer line H$\delta_F$ is used in

866: the modelling of NGC~6441, EZ\_Ages finds a younger age than that

867: determined from H$\beta$, suggesting that the influence of the blue HB

868: stars does show up in H$\delta_F$.  As discussed by \citet{sch04b},

869: the signature of blue HB stars is stronger at bluer wavelengths, so

870: that they tend to impact age determinations based on H$\delta$ more

871: strongly than those based on H$\beta$.  This can be clearly seen in

872: the bottom right panel of Figure~\ref{cluster_grids}, where both

873: NGC~6121 and NGC~6441 look {\it younger} in the $\langle$Fe$\rangle$

874: vs.  H$\delta_F$ plot than in that involving H$\beta$.

875: 

876: In NGC~6441, the blue HB stars are greatly outnumbered by red HB

877: stars, unlike NGC~6121 in which the blue HB population is substantial.

878: They therefore have only a modest influence at the wavelength of

879: H$\beta$, but a larger influence at the bluer wavelength of

880: H$\delta_F$, where the blue HB stars contribute a greater fraction of

881: the total light \citep{sch04b}.  We further discuss this point in

882: \S\ref{consist}.

883: 

884: Finally, we note that the values we found for the abundances of C, N,

885: and Ca for NGC~6121 are suspiciously low.  Unfortunately, we could not

886: find literature values for C and N abundances to compare with our

887: estimates.  However, the numbers Table~\ref{abuntable} and

888: Figure~\ref{comp_lit} show that our [Ca/Fe] estimate is too low by

889: about 0.3 dex, which confirms our suspicions that there might be a

890: problem with our procedure for [Fe/H] $\simless -1.0$.  While this

891: clearly deserves further investigation, for the time being {\it we

892: strongly discourage users from applying EZ\_Ages in the [Fe/H]

893: $\simless -1.0$ regime}.

894: 

895: Another check on the accuracy of the stellar population modeling is to

896: compare the index predictions of the best-fitting model to the

897: observed line strengths.  Figure \ref{model_ind} shows the observed

898: values of the Lick indices as measured in the five test cluster

899: spectra, plotted against the predicted index line strengths from the

900: best-fitting S07 model, as determined by EZ\_Ages.  Black and gray

901: symbols show indices that are and are not included in the fitting

902: process, respectively.  The indices used in the fitting process are

903: extremely well reproduced by the best-fitting model.  In addition, the

904: indices that are not used in the fitting process are also reasonably

905: well matched by the model predictions, with some scatter but with no

906: indication of systematic problems in the modeling of individual

907: indices.

908: 

909: \subsection{Consistency Check}  \label{consist}

910: 

911: In this test, we verify how the results vary when different index sets

912: are adopted as inputs to EZ\_Ages.  This is important, because the

913: full set of indices considered by EZ\_Ages are not always available to

914: observers, due to limitations such as those determined by the

915: instrumental setup adopted, or by the redshift of the sample studied.

916: Therefore, it is important to make sure that age and metal abundances

917: obtained from EZ\_Ages do not depend on which absorption-line indices

918: are employed---or, if they do, that the systematics is understood and

919: can be accounted for.  To test for this, we ran EZ\_Ages on multiple

920: combinations of line indices as measured for the clusters NGC~6441 and

921: M~67, and inter-compared the ages and metal abundances obtained.  As a

922: basis for the comparisons, we adopt the results obtained when using

923: the standard set, i.e., H$\beta$, $\langle$Fe$\rangle$, Mg~$b$,

924: C$_2$4668, CN$_1$, and Ca4227.  The indices used in each separate

925: model fit are listed in Table~\ref{diff_ind_tab}.  We substitute each

926: alternative index separately to asses the effect due to that

927: individual index.  We also examine some combinations of indices,

928: specifically a ``balmer'' model, in which we fit using the average of

929: H$\beta$, H$\gamma_F$, and H$\delta_F$, an ``all'' model, in which all

930: lines are included in the fit, and a ``high-z'' model, which simulates

931: the effect of applying EZ\_Ages to higher redshift data, where only

932: the bluer indices ($\lambda < 4400$ \AA) may be available because the

933: indices at longer wavelengths are redshift out of the optical

934: spectrum.  Results from the fitting process are provided in

935: Table~\ref{diff_fit_tab} and displayed in

936: Figures~\ref{vary_fit_NGC6441} and \ref{vary_fit_M67}, where residuals

937: relative to the age and metal abundances obtained with the standard

938: index set are plotted for the various index combinations.

939: 

940: Looking first at NGC~6441, Figure~\ref{vary_fit_NGC6441} shows that,

941: for the vast majority of the index combinations, ages and metal

942: abundances vary by less than $\pm0.1$ dex, indicating that EZ\_Ages

943: and the S07 models have attained an outstanding degree of consistency.

944: There are a few exceptions, though, which occur for the following

945: index combinations: 1) When G4300 is used as a carbon abundance

946: indicator in place of C$_2$4668 (panel c, triangles); 2) When higher

947: order Balmer lines are used as age indicators (panel a); 3) When

948: higher order Balmer lines are used as age indicators and Fe4383 is

949: used as the iron-abundance indicator as in the ``high-z'' model (panel

950: d, triangles) or when the higher order Balmer lines and Fe4383 are

951: averaged in with the other lines as in the ``all'' model (panel d,

952: squares).  In case 1, the G4300-based [C/Fe] and [N/Fe] values differ

953: from those obtained from the standard indices by --0.14 and +0.18,

954: respectively.  In case 2, ages tend to be younger when higher-oder

955: Balmer lines are used in place of H$\beta$, by up to $\sim$ --0.2 dex.

956: Finally, case 3 is similar to case 2, except that [Fe/H] is too high,

957: and consequently all the other abundances are too low.

958: 

959: The difference found in case 1 when replacing C$_2$4668 by the G4300

960: index is due to the different sensitivities of the two indices to the

961: ratio between the abundances of carbon and oxygen (C/O).  The two

962: indices are sensitive to variations of the oxygen abundance because of

963: details of the molecular dissociation equilibrium in the atmospheres

964: of cool stars.  Of all molecules in whose formation carbon and oxygen

965: take part, carbon monoxide (CO) is the hardest one to break, because

966: it has the highest dissociation energy.  Therefore, at the

967: temperatures prevalent in the atmospheres of G and K stars, most

968: available free carbon and oxygen atoms are locked in CO.  As a result,

969: variations in the abundance of oxygen, usually the most abundant of

970: the two species, have a strong influence on the amount of carbon that

971: is free to form other molecules, such as C$_2$ and CH, which are

972: responsible for the vibrational bands measured by the C$_2$4668 and

973: G4300 indices, respectively.  Therefore, an increase in the abundance

974: of oxygen tends to provoke a decrease in the strength of those

975: molecular bands.

976: 

977: While the concentration of the C$_2$ molecule depends quadratically on

978: the abundance of carbon, that of CH depends only linearly on that

979: abundance.  As a result, C$_2$4668 is far more sensitive to carbon

980: than to oxygen, while the the sensitivity of the G4300 index to those

981: two abundances is approximately the same (though for both indices, the

982: sensitivity to carbon and oxygen abundances have opposite signs).

983: Although this makes the G4300 index a more uncertain indicator of

984: carbon abundance, it raises the possibility that a combination of the

985: two indices may be used to constrain the abundance of oxygen.  We

986: verified that possibility by raising the input oxygen abundance in the

987: models from [O/Fe]~=~0 to +0.3.  As a result, the carbon and nitrogen

988: abundances inferred from use of C$_2$4668 and G4300 indices agreed to

989: within 0.05 dex.  Furthermore, the ages and abundances inferred for

990: the O-enhanced models were not substantially different from the

991: results using the [O/Fe]~=~0 model\footnote[3]{The changes in age and

992: abundance determinations from increasing [O/Fe] to $+0.3$ dex are

993: small because only the effect on the atmospheric line strengths are

994: taken into account.  The effect of enhancing [O/Fe] in the isochrone

995: are not included because reliable O-enhanced isochrones have not yet

996: been incorporated into the S07 model.  An O-enhanced isochrone would

997: result in younger age measurements without significantly changing the

998: inferred abundances.  S07 shows that model grids with

999: $\alpha$-enhanced isochrones move along lines of constant [Fe/H] in

1000: index-index plots.  The result is that the inferred ages are different

1001: when using $\alpha$-enhanced isochrones, but that the abundance

1002: determinations are relatively unchanged.}: [N/Fe] increased by $+0.12$

1003: dex, while the ages inferred from H$\gamma_F$ and H$\delta_F$

1004: increased by 0.6 Gyr.  The values of H$\beta$ age, [Fe/H], [Mg/Fe],

1005: and [Ca/Fe] showed negligible change ($< 0.1$ Gyr or $< 0.05$ dex).

1006: Interestingly, the oxygen abundance in NGC~6441 stars ranges from

1007: [O/Fe] = --0.05 to 0.34 \citep{gra06}, with stars in the upper end of

1008: the interval having ``normal'' oxygen abundances, and likely being

1009: more numerous and dominating the cluster light.  Therefore, the value

1010: we obtained from the cluster's integrated light is in good agreement

1011: with the known cluster abundance, which makes it possible that oxygen

1012: abundances might be inferred from a combination of the C$_2$4668 and

1013: G4300 indices.  This will be further investigated in a forthcoming

1014: paper.

1015: 

1016: The dependence in case 2 of the resulting age on the Balmer line

1017: adopted is not surprising, given that NGC~6441 is characterized by the

1018: presence of blue HB stars (\citealt{ric97}, \citealt{bus07}).  Because

1019: these stars have early-F and A spectral types, they tend to strengthen

1020: Balmer lines in the integrated spectra.  That effect is obviously more

1021: important at bluer wavelengths, where the contribution by early-F and

1022: A stars to the integrated light is the greatest.  Therefore,

1023: higher-order Balmer lines tend to be more strongly affected.  As a

1024: result, when models that do not take into account blue HB stars are

1025: compared with the data, they tend to infer systematically younger

1026: ages, and more so according to higher order Balmer lines (see

1027: discussion in S07).  \citet{sch04b} have in fact shown that this

1028: effect can be used to constrain the morphology of the HB of globular

1029: clusters, solely on the basis of Balmer line strengths in their

1030: integrated spectra.  In particular, the ratio between H$\delta_F$ and

1031: H$\beta$ was shown to be very sensitive to the presence of blue HB

1032: stars.

1033: 

1034: Before discussing case 3, where Fe4383 and H$\delta_F$ are used in the

1035: fitting process, let us look at the results of the consistency test

1036: for M~67.  Here again there is agreement to within $\pm0.1$ dex for

1037: any combination of indices used, with the exception of combining

1038: Fe4383 and H$\delta_F$.  The three Balmer lines all give consistent

1039: results for M~67 (panel a), because M~67 does not contain blue HB

1040: stars and because the integrated spectrum of M~67 used in the this

1041: analysis has been carefully constructed to exclude blue stragglers

1042: \citep{sch04a}.  Unfortunately, there is a problem (as yet not

1043: well-understood) with the G4300 measurements for this cluster

1044: \citep{sch04a}, thus we cannot use M~67 to test our hypothesis about

1045: the effect of oxygen abundance on G4300 and C$_2$4668.

1046: 

1047: Indeed, the only inconsistency in the abundance determinations for

1048: M~67 is in the ``high-z'' model, in which Fe4383 and H$\delta_F$ are

1049: used in the fitting process (case 3 from above).  Here, as with

1050: NGC~6441, we not only derive younger ages, but the resulting iron

1051: abundance is higher by $\sim +0.24$ dex.  Consequently, all the other

1052: elemental abundances are found to be somewhat low.  For M~67, we

1053: cannot invoke blue HB stars to explain this discrepancy.

1054: Interestingly, substituting H$\delta_F$ or Fe4383 into the standard

1055: index set {\it individually} (triangles in panel a and squares in

1056: panel b) does not result in large discrepancies from the standard

1057: model.  How then can we understand the dramatic differences seen when

1058: both H$\delta_F$ and Fe4383 are used in the modelling process?

1059: 

1060: Figure \ref{red_blue_grids} shows solar abundance model grids from the

1061: S07 model.  The M~67 data are overplotted as diamonds.  In panel a, we

1062: see that H$\beta$ and $\langle$Fe$\rangle$ produce model grids where

1063: lines of constant age and constant [Fe/H] are nearly perpendicular.

1064: This is one of the reasons this index combination is so useful for age

1065: and [Fe/H] determinations.  When Fe4383 is substituted for

1066: $\langle$Fe$\rangle$ (panel b), the lines of constant [Fe/H] are less

1067: nearly vertical, lessening the diagnostic power of the index-index

1068: grid.  Likewise, when H$\delta_F$ is substituted for H$\beta$ (panel

1069: c), the lines of constant age are less nearly horizontal, again

1070: constraining the age and [Fe/H] determinations.  This is because, in

1071: the blue region, line crowding is higher, making H$\delta_F$ more

1072: sensitive to metallicity.  Moreover, the blue spectral region is more

1073: strongly affected by warm stars than the red, making Fe4383 more

1074: sensitive to age than its redder counterparts.  Notice, however, that

1075: there is still enough spread in the grids in panels b and c that small

1076: errors or zero-point uncertainties in the data or the models do not

1077: result in substatially different age and [Fe/H] measurements for M~67.

1078: 

1079: Now consider panel d.  When H$\delta_F$ and Fe4383 are combined in the

1080: fitting process, the model grid lines are very far from perpendicular

1081: and the model space collapses down toward the familiar age-metallicity

1082: degeneracy.  Here, small errors or zero-point uncertainties result in

1083: substantially different age and [Fe/H] measurements.  Even the slight

1084: metallicity difference seen between panels a and b becomes a

1085: substantial discrepancy in panel d.  This raises warning flags for

1086: using these indices to determine age and [Fe/H].  In particular, we

1087: note that Fe4383 appears to {\it always} give slightly higher values

1088: of [Fe/H] in the data presented both here and in ongoing work by the

1089: authors, suggesting that there is some inaccuracy in the modelling of

1090: this index.  The fact that the discrepancy appears in the case of

1091: M~67, whose abundances are known to be solar-scale, suggests that this

1092: may be a zeropoint problem in the models, rather than a problem in

1093: calculating the abundance sensitivities of this index.  Combined with

1094: the increased degeneracy of the H$\delta_F$-Fe4383 index-index space,

1095: this means that EZ\_Ages should be used cautiously when determining

1096: stellar population parameters in higher redshift objects where

1097: H$\beta$ is not available, and that the need for high $S/N$ data is

1098: particularly acute in this regime.  Even when the problem with Fe4383

1099: is solved, the degeneracy of H$\delta_F$-Fe4383 index-index space

1100: means that small errors or zeropoint uncertainties in data can result

1101: in errors in the determination of age and [Fe/H] from these indices.

1102: This likely applies to all stellar population models, not just those

1103: presented in this work and S07.

1104: 

1105: In summary, we conclude that EZ\_Ages is characterized by a remarkable

1106: degree of consistency, in that ages and metal abundances inferred from

1107: different index sets agree in the vast majority of cases to within

1108: $\pm0.1$ dex.  Two of the inconsistencies we found, namely, that

1109: different Balmer lines indicate different ages, and different

1110: carbon-abundance indicators indicate different carbon abundances, can

1111: in fact be potentially explored to extract even more information from

1112: the integrated spectra of stellar populations.  While blue Balmer

1113: lines were shown to be useful to indicate the presence of blue HB

1114: stars in globular clusters \citep{sch04b}, the carbon index

1115: discrepancy can potentially be used to constrain the abundance of

1116: oxygen.  The latter effect will be further investigated in a

1117: forthcoming paper.  The third inconsistency is due to the fact that

1118: the combined H$\delta_F$-Fe4383 index-index space is substantially

1119: more degenerate than H$\beta$-$\langle$Fe$\rangle$ space, making

1120: accurate age and [Fe/H] determinations from these indices difficult.

1121: This is a matter of concern in integrated studies of globular clusters

1122: \citep{sch04b} and higher redshift galaxies \citep{sch06}, for which

1123: only these bluer lines may be available.

1124: 

1125: \section{Comparison with Abundance Modeling of

1126:   Galaxies}\label{compare_galaxies} 

1127: 

1128: Having performed the reality and consistency checks above, we now turn

1129: to comparisons between EZ\_Ages results and those coming from

1130: application of other models in the literature.  We focus on the models

1131: by \citet[hereafter TMB03]{tho03}, which also make predictions for

1132: variable abundance patterns.  The TMB03 models also attempt to match

1133: non-solar abundance patterns by fitting Lick indices in unresolved

1134: stellar population spectra.  These include both solar-scaled and

1135: $\alpha$-enhanced models, in which the set of $\alpha$ elements N, O,

1136: Mg, Ca, Na, Ne, S, Si, and Ti are slightly enhanced, while the iron

1137: peak elements Cr, Mn, Fe, Co, Ni, Cu, and Zn are significantly

1138: depressed, in order to vary [$\alpha$/Fe] at fixed total metallicity

1139: [Z/H].  In these models, [C/Fe] is fixed at solar.

1140: 

1141: \citet[hereafter T05]{tho05} use the TMB03 models to estimate ages,

1142: [Z/H], and [$\alpha$/Fe] for a sample of 124 early-type galaxies,

1143: based on measurement of the Lick indices H$\beta$, $\langle$Fe$\rangle$, and

1144: Mg~$b$.  To compare the stellar population modeling results of using

1145: EZ\_Ages and the S07 models to those using TMB03, we have run EZ\_Ages

1146: on the sample of 124 galaxies presented in T05, using the values of

1147: H$\beta$, $\langle$Fe$\rangle$, and Mg~$b$ given in their Table 2.  With these

1148: indices, EZ\_Ages can determine the SSP age, [Fe/H], and [Mg/Fe]

1149: values of the sample galaxies and compare these results with those of

1150: the TMB03 models used in T05.

1151: 

1152: Although the details of the model fitting procedure differ, T05

1153: perform a very similar analysis to that of EZ\_Ages, determining age,

1154: [Z/H], and [$\alpha$/Fe] from the same Lick indices that are the

1155: standard set EZ\_Ages uses to calculate age, [Fe/H], and [Mg/Fe].  The

1156: models used in T05 use the same \citet{kor05} index sensitivity

1157: functions as S07 and should therefore show the same index strength

1158: variations when the abundances are modified.  One practical difference

1159: is that the TMB03 models are cast in terms of [Z/H], rather than

1160: [Fe/H].  In order to compare results from the two models, we convert

1161: T05's estimated [Z/H] into [Fe/H] using the conversion proposed by

1162: TMB03: [Fe/H] = [Z/H] $- 0.94$ [$\alpha$/Fe] (TMB03).

1163: 

1164: The TMB03 models assume that all $\alpha$-elements track Mg.  EZ\_Ages

1165: follows a similar convention, by default setting Na, Si, and Ti to

1166: track Mg in the modeling process, while Cr follows Fe.  A significant

1167: advantage of EZ\_Ages and the S07 models is that they also track the

1168: [C/Fe], [N/Fe], and [Ca/Fe] abundance ratios using C, N, and

1169: Ca-sensitive Lick indices, rather than making assumptions about how

1170: these elements vary.  However, these differences should not affect the

1171: comparison of age, [Fe/H], and [$\alpha$/Fe] results presented here,

1172: as H$\beta$, $\langle$Fe$\rangle$, and Mg~$b$ are predominantly

1173: sensitive to age, Fe and Mg only (see Table \ref{indextable}), and are

1174: thus treated the same by both models through the \citet{kor05}

1175: sensitivity functions.

1176: 

1177: The results of fitting the T05 galaxies using EZ\_Ages are shown in

1178: Figure \ref{compare_thomas}.  In each panel, the dashed line shows a

1179: one-to-one relation and the solid line shows the best fitting line to

1180: the data while keeping the one-to-one slope fixed.  The resulting

1181: zeropoint offsets between the model fits are indicated in the upper

1182: left corner of each panel.  In general, there is good agreement

1183: between the results coming from application of the two models to the

1184: same data set.  The scatter is small and the zeropoint offsets are

1185: modest.  The values of [$\alpha$/Fe] agree to within a very small

1186: offset.  The ages estimated in T05 are typically $\sim 25$\% (0.13 dex)

1187: younger than those we find with EZ\_Ages, while the T05 values of

1188: [Fe/H] are $\sim 0.08$ dex higher.

1189: 

1190: In addition to the zeropoint offsets in the age and [Fe/H]

1191: comparisons, the two different analyses do not truly follow a

1192: one-to-one slope for either parameter: the [Fe/H] relation is

1193: significantly steeper than the unity relation, while the age

1194: comparison shows some curvature.  At old ages ($\log$ age $\simgreat

1195: 1.0$ as estimated by EZ\_Ages) the slope is very nearly one-to-one,

1196: and the agreement between the two age estimates is much better, with a

1197: zeropoint offset of 15\% ($-0.07$ dex) between the T05 ages and those

1198: from EZ\_Ages.  At intermediate ages ($0.8 < \log \mbox{age} < 1.0$ as

1199: estimated by EZ\_Ages) the relation is steeper than unity, while at

1200: young ages ($\log \mbox{age} \simless 0.8$) the relation flattens out

1201: again, although at a substantial zeropoint offset.  At intermediate

1202: and young ages, the T05 ages are 33\% (0.17 dex in log age) younger.

1203: The origin of the curvature in the age relation is unclear.  In any

1204: case, age agreement for old galaxies is very good (within 15\%) and is

1205: still within 33\% for all galaxies down to the youngest SSP ages

1206: estimated in the T05 sample.

1207: 

1208: Based on the cluster comparison of \S\ref{real} and Figure

1209: \ref{comp_lit}, the zeropoint uncertainties in the S07 models are

1210: roughly 0.07 dex in $\log$ age, 0.1 dex in [Fe/H], and 0.05 dex in

1211: [Mg/Fe].  The differences between the T05 and the EZ\_Ages results are

1212: therefore within the zeropoint uncertainties of the models, with the

1213: exception of the age estimates for galaxies younger than 10 Gyr.  For

1214: these galaxies, the T05 results are younger by more than twice the

1215: indicated zeropoint uncertainties in the S07 and EZ\_Ages models.  The

1216: best way to resolve this discrepancy is to compare the models with

1217: data for metal-rich, intermediate-age clusters.  While the models by

1218: S07 have been shown in the previous section and in Schiavon (2007) to

1219: match the data for M~67 ( $\sim$ 3.5 Gyr-old and solar metallicity),

1220: more data are clearly needed to better constrain the models in this

1221: crucial age/metallicity regime.

1222: 

1223: The non-unity slope of the [Fe/H] comparison also deserves further

1224: investigation.  While the differences are negligibly small in the

1225: solar metallicity regime, they climb up to 0.2 dex at the

1226: high-metallicity end.  For the most metal-rich galaxies, EZ\_Ages

1227: obtains [Fe/H] $\sim$ 0.2, and T05 find galaxies to be more metal-rich

1228: by $\sim$ 0.2 dex.  As in the case of age determinations, we again

1229: find ourselves lacking data that could help decide between the two

1230: model sets, given the non-existence of integrated spectroscopy for

1231: super-metal-rich clusters.  It would be interesting, however, to

1232: compare results from the two models by confronting them with cluster

1233: data in a regime for which such data are available, such as moderately

1234: metal-poor clusters.

1235: 

1236: In Figure \ref{thomas_slope}, we plot the difference between the

1237: values of [Fe/H] estimated in T05 and by EZ\_Ages as a function of

1238: various other properties of the stellar populations .  The top panel

1239: shows the [Fe/H] differences as a function of SSP age and

1240: $\alpha$-enhancement.  The differences in the [Fe/H] estimates do not

1241: appear to be strongly correlated with either SSP age or

1242: $\alpha$-enhancement.  In the lower panel, we show the differences in

1243: [Fe/H] as a function of the differences in the estimated ages and

1244: $\alpha$-enhancements.  The [Fe/H] differences are correlated with

1245: differences in the age estimates between the two different models.

1246: Furthermore, the correlation is in the direction expected from the

1247: correlated errors in the index-index plots: where T05 obtain

1248: substantially younger ages than EZ\_Ages, they also find higher

1249: [Fe/H].  The difference in the age estimates by the different models

1250: therefore naturally explains the observed differences in [Fe/H], both

1251: the zeropoint shift and the non-unity slope of the [Fe/H] comparison.

1252: Estimates of [$\alpha$/Fe] are substantially less biased by a

1253: zeropoint error in the age.  That is because the Mg~$b$ and

1254: $\langle$Fe$\rangle$ indices have very similar age-dependence, so that

1255: the zero-point uncertainties in the parameters determined by these two

1256: indices ([Mg/H] and [Fe/H]) cancel out.

1257: 

1258: From this analysis, it is clear that the TMB03 models and EZ\_Ages

1259: give fairly consistent results, with a relatively small disagreement

1260: in the age zeropoints of the models which produces modest

1261: disagreements in [Fe/H].  Abundance ratios such as [$\alpha$/Fe] are

1262: much less sensitive to zeropoint offsets.  The comparison of EZ\_Ages

1263: results with ages from CMD fitting of Galactic clusters

1264: (\S\ref{compare_clusters}) indicates that the age zeropoint in

1265: EZ\_Ages is correct to within 0.07 dex in $\log$ age.  The [Z/H] and

1266: [$\alpha$/Fe] results from TMB03 models have been compared with

1267: cluster data \citep{mar03,tho03}, but age estimates have not been

1268: independently tested, therefore it is difficult to say what the

1269: zeropoint uncertainties in the age determinations for these models

1270: should be.  We therefore conclude that EZ\_Ages and the S07 models

1271: give results in good agreement with the TMB03 models (at least in the

1272: high-metallicity regime tested in this analysis of galaxy spectra),

1273: but that at intermediate ages (age $< 10$ Gyr) the TMB03 SSP ages may

1274: underestimate the true SSP ages by as much as 33\% (0.17 dex in $\log$

1275: age).

1276: 

1277: A more general result from this comparison is that the sequential grid

1278: inversion algorithm presented in this paper, combined with the SSP

1279: models of S07, does an excellent job of reproducing [Mg/Fe] abundances

1280: obtained by other groups with other models.  Although the EZ\_Ages

1281: abundance results for [C/Fe], [N/Fe], and [Ca/Fe] in unresolved

1282: stellar populations cannot be compared to other models (since no other

1283: models exist for the comparison), the modeling process for fitting

1284: these other abundances is the same as that used to fit [Mg/Fe] and the

1285: success of this process for [Mg/Fe] bodes well for other abundances.

1286: The test presented in this section, combined with the stringent tests

1287: with cluster data described in \S\ref{compare_clusters}, are

1288: reassuring evidence that this modeling process is reasonable and can

1289: be used to provide a {\it quantitative} assessment of multiple

1290: elemental abundances.

1291: 

1292: \section{Conclusions}\label{conc}

1293: 

1294: In this paper, we have presented a methodology for measuring SSP ages,

1295: [Fe/H], and individual abundance ratios [Mg/Fe], [C/Fe], [N/Fe], and

1296: [Ca/Fe] for unresolved stellar populations.  We do this by exploiting

1297: the different sensitivities of various Lick indices to a variety of

1298: elemental abundances, and the ability of the S07 stellar population

1299: models to accurately model the Lick indices for a wide range of input

1300: abundance patterns.  The algorithm presented here has been implemented

1301: in the IDL code package EZ\_Ages, which is available for download and

1302: general use.

1303: 

1304: We have subjected the modeling process described here to numerous

1305: rigorous tests and comparisons, with the following results:

1306: \begin{list}{}{}

1307: \item[1.] {\it Comparison with Galactic cluster data:}

1308: \begin{list}{}{}

1309: \item[a.] {\it Ages:} EZ\_Ages age estimates reproduce the results of

1310:   cluster CMD fitting to within 0.15 dex for all clusters, with the

1311:   exception of NGC~6121, whose Balmer line strengths are affected by a

1312:   blue HB.  For two out of three of the remaining clusters, the

1313:   EZ\_Ages age is within 0.1 dex of the CMD result.

1314: \item[b.] {\it [Fe/H] and [Mg/Fe]:} EZ\_Ages estimates match the

1315:   results from high-resolution spectroscopy of individual cluster

1316:   members to within 0.1 dex for some of the clusters, and to within

1317:   0.2 dex for all clusters (with the exception of the disagreement

1318:   between our value of [Fe/H] for NGC~6528 and the results of

1319:   \citealt{car01}---we are in $\sim \pm0.1$ dex agreement with the

1320:   results of \citealt{zoc04} and \citealt{bar04} for [Fe/H] for this

1321:   cluster).

1322: \item[c.] {\it [C/Fe] and [N/Fe]:} For the only two clusters where C

1323:   and N abundances are available (47~Tuc and M~67), EZ\_Ages results

1324:   match those from high-resolution spectroscopy of individual cluster

1325:   members to within 0.1 dex.  For NGC~6121, we find suspiciously low

1326:   values, that might indicate a problem at the low metallicity end of

1327:   our models.  Therefore, we strongly caution users against using

1328:   EZ\_Ages to determine these abundances for systems with [Fe/H]

1329:   $\simless -1.0$.

1330: \item[d.] {\it [Ca/Fe]:} EZ\_Ages results are within $\sim 0.1$ dex

1331:   for all clusters except NGC~6121.  The latter is certainly related

1332:   to the problem mentioned above regarding [C/Fe] and [N/Fe] in the

1333:   metal-poor regime, so we do not recommend use of EZ\_Ages for

1334:   [Ca/Fe] determinations for systems with [Fe/H] $\simless -1.0$.

1335: \end{list}

1336: \item[2.] {\it Consistency test performed on the cluster NGC~6441:}

1337:   using different combinations of Lick indices in the fitting process

1338:   yields results that are consistent to within $\pm 0.1$ dex for all

1339:   stellar population parameters except for the following cases:

1340: \begin{list}{}{}

1341: \item[a.] Using G4300 instead of C$_2$4668 to fit [C/Fe] results in

1342:   values of [C/Fe] and [N/Fe] that differ by $-0.14$ and $+0.18$ dex,

1343:   respectively.  This difference is probably due to differences in the

1344:   index responses to O, which are not explored in this work.

1345: \item[b.] Using H$\gamma_F$ or H$\delta_F$ instead of H$\beta$ as an

1346:   age indicator results in younger ages for the cluster NGC~6441 due

1347:   to the effect of the cluster's blue HB.  This discrepancy does not

1348:   exist for M~67, which does not have a blue HB, and whose integrated

1349:   spectrum is constructed to exclude blue HB stars.

1350: \item[c.] Using the bluest Balmer and Fe lines available, H$\delta_F$

1351:   and Fe4383, can result in age and [Fe/H] measurements that are

1352:   substantially different than those determined using the standard set

1353:   of indices, due to the increased degeneracy of the model space for

1354:   these indices.  This poses a problem for high redshift studies of

1355:   stellar populations, where often only the bluest lines are available

1356:   for analysis.

1357: \end{list}

1358: \item[3.] {\it Error analysis test:} EZ\_Ages implements an algorithm

1359:   that attempts to propagate errors through the modeling process in

1360:   an efficient way that takes into account the dependence of abundance

1361:   fitting on other parts of the population fitting process.  Monte

1362:   Carlo simulations show that this simplified error estimation does an

1363:   excellent job of matching the true errors in the population

1364:   parameters due to measurement errors in the Lick indices.

1365: \item[4.] {\it Comparison with \citet{tho05} results:}

1366: \begin{list}{}{}

1367: \item[a.] Results for [$\alpha$/Fe] are consistent between EZ\_Ages

1368:   results and those using TMB03 models.  

1369: \item[b.] Age and [Fe/H] comparisons show little scatter.  There are

1370:   zeropoint offsets of 15\%--33\% (0.07--0.17 dex) in age and 0.08 dex

1371:   in [Fe/H] between the two analyses, and a non-unity slope in the

1372:   [Fe/H] comparison.  The age differences are small for the oldest

1373:   (age $> 10$ Gyr) galaxies and increase for younger galaxies.  The

1374:   [Fe/H] effects seem to be a natural result of the age differences.

1375:   The S07 and EZ\_Ages age zeropoints are shown here to be good to

1376:   within 0.07 dex, while the TMB03 age zeropoints are untested.  This

1377:   suggests that the ages derived in T05 may be too young by up to 33\%

1378:   (0.17 dex) for intermediate age galaxies (age $< 10$ Gyr).

1379: \item[c.] Absolute estimates of abundance ratios are more reliable

1380:   than absolute estimates of age or [Fe/H] because zeropoint

1381:   uncertainties tend to cancel out.

1382: \item[d.] The sequential grid inversion algorithm method is proven to

1383:   work for estimating [Mg/Fe].  Since this method is the same as the

1384:   one used to determine other abundance ratios, those estimates are

1385:   likely to also be reliable, as demonstrated in the comparison with

1386:   cluster data.

1387: \end{list}

1388: \end{list}

1389: 

1390: Overall, EZ\_Ages and the S07 SSP models do an excellent job of

1391: fitting the SSP age, [Fe/H], and abundance ratios [Mg/Fe], [C/Fe],

1392: [N/Fe], and [Ca/Fe] for unresolved stellar populations.  The small

1393: zeropoint uncertainties in age and [Fe/H] estimates illustrated by

1394: comparisons with Galactic cluster data demonstrate that absolute

1395: estimates can be made for these quantities with high $S/N$ data.

1396: Absolute estimates of elemental abundances are robust to the (small)

1397: zeropoint uncertainties.  EZ\_Ages and the S07 models therefore make

1398: it possible, for the first time, to perform a quantitative assessment

1399: of multiple individual elemental abundances from medium-resolution

1400: spectra of unresolved stellar populations.  Furthermore, the abundance

1401: fitting process can be run in an automated way on large data sets.

1402: With these innovations, stellar population analysis is better

1403: positioned than ever before to address the task of unravelling the

1404: star formation histories of stellar systems.

1405: 

1406: \acknowledgements{The authors wish to thank an anonymous referee for

1407:   excellent and thought-provoking comments that improved the quality

1408:   of this work.  They wish to thank Andreas Korn for valuable

1409:   information concerning use of the Korn et al. (2005) sensitivity

1410:   tables.  This work is supposrted by NSF grants AST 00-71198 and AST

1411:   05-07483.  R. P. S. is supported by Gemini Observatory, which is

1412:   operated by the Association of Universities for Research in

1413:   Astronomy, Inc., on behalf of the international Gemini partnership

1414:   of Argentina, Australia, Brazil, Canada, Chile, the United Kingdom,

1415:   and the United States of America.}

1416: 

1417: \begin{thebibliography}{}

1418: \bibitem[Athey(2003)]{ath03} Athey, A. E. 2003, PhD Thesis, University

1419: of Michigan

1420: \bibitem[Barbuy et al.(2004)]{bar04} Barbuy, B., et al. 2004, MmSAI,

1421:   75, 398

1422: \bibitem[Beasley et al.(2002)]{bea02} Beasley, M. A., Hoyle, F. \&

1423: Sharples, R. M. 2002, \mnras, 336, 168

1424: \bibitem[Bensby, Feltzing, \& Lundstr\"om(2004)]{ben04} Bensby, T.,

1425: Feltzing, S., \& Lundstr\"om, I. 2004, \aap, 415, 155

1426: \bibitem[Borges et al.(1995)]{bor95} Borges, A.C., Idiart, T.P., de

1427: Freitas Pacheco, J.A. \& Thevenin, F. 1995, \aj, 110, 2408

1428: \bibitem[Briley et al.(2004)]{bri04} Briley, M. M., Harbeck, D.,

1429:   Smith, G. H., \& Gebel, E. K. 2004, \aj, 127, 1588

1430: \bibitem[Bruzual \& Charlot(1993)]{bc93} Bruzual, G., \& Charlot,

1431: S. 1993, \apj, 405, 538

1432: \bibitem[Bruzual et al.(1997)]{bru97} Bruzual, G., Barbuy, B., Ortolani,

1433: 	S., Bica, E., Cuisinier, F., Lejeune, T., Schiavon, R.P. 1997, \aj,

1434: 	114, 1531

1435: \bibitem[Bruzual \& Charlot(2003)]{bc03} Bruzual, G., \& Charlot,

1436: S. 2003, \mnras, 344, 1000

1437: \bibitem[Busso et al.(2007)]{bus07} Busso, G., Cassisi, S., Piotto, G.,

1438: 	Castellani, M., Romaniello, M., Catelan, M., Djorgovski, S.G.,

1439: 	Recio Blanco, A., Renzini, A., Rich, M.R., Sweigart, A., \&

1440: 	Zoccali, M.  2007, \aap, 105, 119

1441: \bibitem[Cardiel et al.(1998)]{car98} Cardiel, N., Gorgas, J.,

1442:   Cenarro, J., \& Gonz{\'a}lez, J. J. 1998, \aaps, 127, 597

1443: \bibitem[Carretta et al.(2001)]{car01} Carretta, E., Cohen, J. G.,

1444:   Gratton, R. G., \& Behr, B. 2001, \aj, 122, 1469

1445: \bibitem[Carretta et al.(2004)]{car04} Carretta, E., Gratton, R. G.,

1446:   Bragaglia, A., Bonifacio, P., \& Pasquini, L. 2004, \aap, 416, 925

1447: \bibitem[Coelho et al.(2007)]{coe07} Coelho, P., Bruzual, G., Charlot,

1448:   S., Weiss, A., Barbuy, B., \& Ferguson, J. W. 2007, \mnras, 382, 498

1449: \bibitem[Dotter et al.(2007)]{dot07} Dotter, A., Chaboyer, B.,

1450:   Ferguson, J. W., Lee, H.-C., Worthey, G., Jevremovi{\'c}, D., \&

1451:   Baron, E. 2007, \apj, 666, 403

1452: \bibitem[Freitas Pacheco \& Barbuy (1995)]{fre95} Freitas Pacheco, J.A. \&

1453: Barbuy B. 1995, \aap, 302, 718

1454: \bibitem[Fulbright, Rich, \& McWilliam(2005)]{ful05} Fulbright, J. P., Rich, R. M.,

1455: \& McWilliam, A. 2005, Nuclear Physics A, 758, 197

1456: \bibitem[Fulbright et al.(2007)]{ful07} Fulbright, J. P., Rich, R. M.,\&

1457: McWilliam, A.  2007, \apj, 661, 1152

1458: \bibitem[Gibson et al.(1999)]{gib99} Gibson, B.K., Madgwick, D.S., Jones,

1459: 	L.A., Da Costa, G.S. \& Norris, J.E. 1999, \aj, 118, 1268

1460: \bibitem[Girardi et al.(2000)]{gir00} Girardi, L., Bressan, A.,

1461:   Bertelli, G., \& Chiosi, C. 2000, \aaps, 141, 371

1462: \bibitem[Gratton et al.(2006)]{gra06} Gratton, R. G., Lucatello, S.,

1463:   Bragaglia, A., Carretta, E., Momany, Y., Pancino, E., \& Valenti,

1464:   E. 2006, \aap, 455, 271

1465: \bibitem[Graves et al.(2007)]{gra07} Graves, G. J., Faber, S. M.,

1466:   Schiavon, R. P., \& Yan, R. 2007, \apj, submitted

1467:   (astro-ph/0707.1523)

1468: \bibitem[Henry \& Worthey(1999)]{hen99} Henry, R.B.C. \& Worthey,

1469: 	G. 1999, \pasp, 111, 919

1470: \bibitem[Ivans et al.(1999)]{iva99} Ivans, I. I., Sneden, C., Kraft,

1471:   R. P., Suntzeff, N. B., Smith, V. V., Langer, G. E., \& Fulbright,

1472:   J. P. 1999, \aj, 118, 1273

1473: \bibitem[Korn, Maraston, \& Thomas(2005)]{kor05} Korn, A. J.,

1474:   Maraston, C., \& Thomas, D. 2005, /aap, 438, 685

1475: \bibitem[Lee \& Worthey (2005)]{lee05} Lee, H.-c. \& Worthey, G. 2005, \apjs, 160, 176

1476: \bibitem[Maraston(2005)]{mar05} Maraston, C. 2005, \mnras, 362, 799

1477: \bibitem[Maraston \& Thomas (2000)]{mt00} Maraston, C. \& Thomas, D. 2000,

1478: 	\apj 541, 126

1479: \bibitem[Maraston et al.(2003)]{mar03} Maraston, C., Greggio, L.,

1480: Renzini, A., Ortolani, S., Saglia, R.P., Puzia, T.H., \& Kissler-Patig,

1481: M.  2003, \aap, 400, 823

1482: \bibitem[Ortolani et al.(2001)]{ort01} Ortolani, S., Barbuy, B., Bica,

1483:   E., Renzini, A., Zoccali, M., Rich, R. M., \& Cassisi, S. 2001,

1484:   \aap, 376, 878

1485: \bibitem[Prochaska et al.(2005)]{pro05} Prochaska, L.C., Rose, J.A. \&

1486: 	Schiavon, R.P. 2005, \aj, 130, 2666

1487: \bibitem[Proctor et al.(2004)]{pro04} Proctor, R.N., Forbes, D.A.

1488: \& Beasley, M.A. 2004, \mnras, 355 1327

1489: \bibitem[Rich et al.(1997)]{ric97} Rich, R.M., Sosin, C., Djorgovski, S.G.,

1490: 	Piotto, G., King, I.R., Renzini, A., Phinney, E.S., Dorman, B.,

1491: 	Liebert, J., \& Meyla, G.  1997, \apj, 484, L25

1492: \bibitem[Rose (1994)]{ros94} Rose, J.A. 1994, \aj, 107, 206

1493: \bibitem[Salaris \& Weiss(2002)]{sal02} Salaris, M., \& Weiss,

1494:   A. 2002, \aap, 388, 492

1495: \bibitem[Salasnich et al.(2000)]{sal00} Salasnich, B., Girardi, L.,

1496:   Weiss, A., \& Chiosi, C. 2000, \aap, 361, 1023

1497: \bibitem[S{\'a}nchez-Bl{\'a}zquez et al.(2003)]{san03}

1498: S{\'a}nchez-Bl{\'a}zquez, P., Gorgas, J., Cardiel, N., Cenarro, J., \&

1499: Gonz{\'a}lez, J. J. 2003, \apjl, 590, 91

1500: \bibitem[Schiavon(2007)]{sch07} Schiavon, R. P. 2007, \apjs, 171, 146

1501: \bibitem[Schiavon \& Barbuy (1999)]{sch99} Schiavon, R.P. \& Barbuy, B., 

1502: 	1999, \apj, 510, 934

1503: \bibitem[Schiavon et al.(2002a)]{sch02a} Schiavon, R. P., Faber, S. M.,

1504:   Castilho, B.V., \& Rose, J. A. 2002a \apj, 580, 850

1505: \bibitem[Schiavon et al.(2002b)]{sch02b} Schiavon, R. P., Faber, S. M.,

1506:   Rose, J. A., \& Castilho, B. V. 2002b, \apj, 580, 873

1507: \bibitem[Schiavon et al.(2004a)]{sch04a} Schiavon, R.P., Caldwell, N. 

1508: \& Rose, J.A. 2004a, \aj, 127, 1513

1509: \bibitem[Schiavon et al.(2004b)]{sch04b} Schiavon, R.P., Rose, J.A., 

1510: Courteau, S. \& MacArthur, L. 2004b, \apj, 608, L33

1511: \bibitem[Schiavon et al.(2005)]{sch05} Schiavon, R.P., Rose, J.A., 

1512: Courteau, S. \& MacArthur, L. 2005, \apjs, 160, 163

1513: \bibitem[Schiavon et al.(2006)]{sch06} Schiavon, R.P. et al. 2006, \apj,

1514: 	651, L93

1515: \bibitem[Serven, Worthey, \& Briley(2005)]{ser05} Serven, J., Worthey,

1516:   G., \& Briley, M. M. 2005, \apj, 627, 754

1517: \bibitem[Shetrone \& Sanquist(2000)]{she00} Shetrone, M. D., \&

1518:   Sanquist, E. L. 2000, \aj, 120, 1913

1519: \bibitem[Tantalo et al.(1998)]{tan98} Tantalo, R., Chiosi, C. \& Bressan,

1520: 	A. 1998, \aap, 333, 419

1521: \bibitem[Tautvaisiene et al.(2000)]{tau00} Tautvaisiene, G.,

1522:   Edvardsson, B., Tuominen, I., \& Ilyin, I. 2000, \aap, 360, 499

1523: \bibitem[Thomas, Maraston, \& Bender(2003)]{tho03} Thomas, D.,

1524: 	Maraston, C. \& Bender, R., 2003, \mnras, 343, 279

1525: \bibitem[Thomas et al.(2004)]{tho04} Thomas, D., Maraston, C. \& Korn, A.

1526: 	2004, \mnras, 351, 19

1527: \bibitem[Thomas et al.(2005)]{tho05} Thomas, D., Maraston, C., Bender,

1528:   R., \& de Oliveira, C. M. 2005, \apj, 621, 673

1529: \bibitem[Trager et al.(1998)]{tra98} Trager, S. C., Worthey, G.,

1530:   Faber, S. M., Burstein, D., \& Gonz{\'a}lez, J. J. 1998, \apjs, 116, 1

1531: \bibitem[Trager et al.(2000)]{tra00} Trager, S. C., Faber, S. M.,

1532:   Worthey, G., \& Gonz{\'a}lez, J. J. 2000, \aj, 120, 165

1533: \bibitem[Trager et al.(2005)]{tra05} Trager, S.C., Worthey, G., Faber,

1534: 	S.M. \& Dressler, A. 2005, \mnras, 362, 2

1535: \bibitem[Vazdekis et al.(1996)]{vaz96} Vazdekis, A., Casuso, E.,

1536: Peletier, R. F., \& Beckman, J. E. 1996, \apjs, 106, 307

1537: \bibitem[Vazdekis et al.(1997)]{vaz97} Vazdekis, A., Peletier, R.F.,

1538: 	Beckman, J.E. \& Casuso, E. 1997, \apjs, 111, 203

1539: \bibitem[Vazdekis et al.(2001)]{vaz01}  Vazdekis, A., Salaris, M., Arimoto,

1540: 	N. \& Rose, J.A. 2001, \apj, 549, 274

1541: \bibitem[Vakdekis et al.(2003)]{vaz03} Vazdekis, A., Cenarro, A. J.,

1542: Gorgas, J., Cardiel, N., \& Peletier, R. F. 2003, \mnras, 340, 1317

1543: \bibitem[Weiss et al.(1995)]{wei95} Weiss, A., Peletier, R.F. \&

1544: 	Matteucci, F. 1995, \aap, 296, 73

1545: \bibitem[Weiss et al.(2006)]{wei06}Weiss, A., Salaris, M., Ferguson,

1546:   J. W., \& Alexander, D. R. 2006, astro-ph/0605666)

1547: \bibitem[Worthey(1998)]{wor98} Worthey, G. 1998, \pasp, 110, 888

1548: \bibitem[Worthey, Faber, \& Gonz{\'a}lez(1992)]{wor92} Worthey, G.,

1549: Faber, S. M., \& Gonz{\'a}lez, J. J. 1992, \apj, 398, 69

1550: \bibitem[Worthey et al.(1994)]{wor94} Worthey, G., Faber, S. M.,

1551:   Gonz{\' a}lez, J. J., \& Burstein, D. 1994, \apjs, 94, 687

1552: \bibitem[Worthey \& Ottaviani(1997)]{wor97} Worthey, G. \& Ottaviani,

1553:   D. L. 1997, \apjs, 111, 377

1554: \bibitem[Zoccali et al.(2004)]{zoc04} Zoccali, M., et al. 2004, \aap,

1555:   423, 507

1556: \end{thebibliography}

1557: 

1558: \begin{deluxetable}{lcc}

1559: \tabletypesize{\scriptsize}

1560: \tablecaption{Lick Indices in the S07 SSP Models. \label{indextable}}

1561: \tablewidth{0pt}

1562: \tablehead{

1563: \colhead{Index} &

1564: \colhead{KMT Sensitivity\tablenotemark{*}} &

1565: \colhead{SWB Sensitivity\tablenotemark{*}}

1566: }

1567: 

1568: \startdata

1569: \multicolumn{3}{c}{Balmer Indices} \\[3pt]

1570: \tableline \\

1571: H$\delta_A$     &Fe, C$^-$  &Fe, C, V          \\

1572: H$\delta_F$     &Fe, Mg$^+$ &Fe, C, Si         \\

1573: H$\gamma_A$     &C, Fe, Mg  &C, Ti, (Mg)       \\

1574: H$\gamma_F$     &C, Fe      &C, (Si), (Mg)    \\

1575: H$\beta$        &...        &(Ni)             \\

1576: \tableline \\

1577: \multicolumn{3}{c}{Fe Indices} \\[3pt]

1578: \tableline \\

1579: Fe4383          &Fe, Mg, C$^-$ &C, Fe        \\

1580: Fe5015          &Ti, Mg, Fe    &C, Fe, (Ti)  \\

1581: Fe5270          &Fe            &C, (Fe)      \\

1582: Fe5335          &Fe            &Fe, C        \\

1583: \tableline \\

1584: \multicolumn{3}{c}{Mg Indices} \\[3pt]

1585: \tableline \\

1586: Mg$_2$          &Mg, C     &Mg, C, Fe    \\

1587: Mg~{\it b}      &Mg, Fe, C &Mg, Fe, (Cr) \\

1588: \tableline \\

1589: \multicolumn{3}{c}{C, CH, and CN Indices} \\[3pt]

1590: \tableline \\

1591: CN$_1$             &C, N, O      &C, N, O      \\

1592: CN$_2$             &C, N, O      &C, N, O      \\

1593: G4300              &C, O, Fe$^+$ &C, Fe, Ti    \\

1594: C$_2$4668          &C, O         &C, O, (Si)   \\

1595: \tableline \\

1596: \multicolumn{3}{c}{Ca Indices} \\[3pt]

1597: \tableline \\

1598: Ca4227          &Ca, C  &Ca, O, (CN)  

1599: \enddata

1600: \tablecomments{Index sensitivities as given by \citet[KMT]{kor05} and

1601:   \citet[SWB]{ser05}.  Only the top three element sensitivities are

1602:   given, and only those with significance above 1$\sigma$ in the model

1603:   spectra.  For the KMT sensitivities, only those of turn-off

1604:   and giant branch stars are shown, as these two components comprise

1605:   90\% of the light in the S07 models.  For the SWB sensivities, those

1606:   under 2$\sigma$ are shown in parentheses.}

1607: \tablenotetext{*}{In addition to the reported sensitivities for

1608:   individual element abundances, all indices listed here vary with

1609:   total metallicity.}

1610: \tablenotetext{+}{Indicates sensitivities that only appear at high

1611:   metallicity.}

1612: \tablenotetext{-}{Indicates sensitivities that only appear at low

1613:   metallicity.}

1614: 

1615: \end{deluxetable}

1616: 

1617: \begin{deluxetable}{lccccc|cc}

1618: \tabletypesize{\scriptsize}

1619: \tablecaption{Selected Lick Indices for Clusters\label{cluster_indices}}

1620: \tablewidth{0pt}

1621: \tablehead{

1622: \colhead{} &

1623: \colhead{NGC~6121} &

1624: \colhead{47~Tuc} &

1625: \colhead{NGC~6441} &

1626: \colhead{NGC~6528} &

1627: \colhead{M~67} &

1628: \colhead{GC Err$^{*}$} &

1629: \colhead{M~67 Err} 

1630: }

1631: \startdata

1632: H$\delta_F$    & 2.338 & 0.682 & 0.926 & 0.298 & 0.977  &0.117 &0.090\\

1633: H$\gamma_F$    & 1.398 &-0.660 &-0.148 &-1.308 &-0.324  &0.215 &0.050\\

1634: H$\beta$       & 2.280 & 1.602 & 1.804 & 1.572 & 2.246  &0.083 &0.090\\

1635: Fe4383	       & 1.267 & 2.491 & 3.048 & 4.666 & 3.841  &0.267 &0.200\\

1636: Fe5270	       & 1.211 & 1.913 & 2.128 & 2.799 & 2.647  &0.142 &0.090\\

1637: Fe5335	       & 0.984 & 1.685 & 1.785 & 2.417 & 2.332  &0.079 &0.080\\

1638: Mg~{\it b}     & 1.577 & 2.694 & 2.651 & 3.759 & 2.927  &0.148 &0.100\\

1639: Mg$_2$	       & 0.088 & 0.159 & 0.165 & 0.249 & 0.175  &0.015 &0.010\\

1640: G4300	       & 2.496 & 4.656 & 4.001 & 4.819 & 4.740  &0.198 &0.100\\

1641: C$_2$4668      & 0.322 & 1.680 & 1.906 & 4.510 & 4.382  &0.131 &0.100\\

1642: CN$_1$	       &-0.070 & 0.025 & 0.020 & 0.063 &-0.007  &0.010 &0.005\\

1643: CN$_2$	       &-0.047 & 0.050 & 0.047 & 0.093 & 0.018  &0.008 &0.008\\

1644: Ca4227	       & 0.321 & 0.557 & 0.581 & 0.887 & 0.927  &0.041 &0.100\\

1645: \enddata

1646: \tablecomments{All indices are measured in {\AA}, except for CN$_1$ and

1647:   CN$_2$, which are measured in magnitudes, as defined in \citet{wor94}.}

1648: \tablenotetext{*}{Uncertainties in the Lick index measurements for

1649:   the Milky Way globular cluster (GC) data are dominated by the

1650:   zeropoint uncertainties in converting measured indices to the Lick

1651:   system and are thus the same for NGC~6121, 47~Tuc, NGC~6441, and

1652:   NGC~6528. See \citet{sch05} for details.

1653: }

1654: \end{deluxetable}

1655: 

1656: 

1657: \begin{deluxetable}{llccccccc}

1658: \tabletypesize{\scriptsize}

1659: \tablecaption{EZ\_Ages Cluster Abundances Compared to Literature\label{abuntable}}

1660: \tablewidth{0pt}

1661: \tablehead{

1662: \colhead{} &

1663: \colhead{} &

1664: \colhead{Age$^{*}$} &

1665: \colhead{[Fe/H]} &

1666: \colhead{[Mg/Fe]} &

1667: \colhead{[C/Fe]} &

1668: \colhead{[N/Fe]} &

1669: \colhead{[Ca/Fe]} &

1670: \colhead{Ref.}

1671: }

1672: \startdata

1673: NGC 6121  &EZ\_Ages   &$7.6_{-0.7}^{+0.8}$  &$-1.30\pm0.09$ &$+0.36\pm0.11$

1674: &$-0.12\pm0.02$  &$0.15\pm0.15$  &$-0.05\pm0.10$ &\\

1675:           &literature &13$^{\ddag}$ &$-1.18\pm0.00$ &$+0.44\pm0.02$ &\ldots

1676: &\ldots &$+0.26\pm0.02$ &a, b\\

1677: \\

1678: 47 Tuc  &EZ\_Ages   &$13.9_{-3.0}^{+\mbox{max\dag}}$  &$-0.80\pm0.09$ &$+0.28\pm0.05$

1679: &$-0.16\pm0.03$  &$+0.66\pm0.08$  &$+0.08\pm0.03$ &\\

1680:           &literature &12 &$-0.7\pm0.05$ &$+0.4\pm0.1$ 

1681: &$-0.2$/$0.0^{**}$ &$+1.1$/$+0.3^{**}$ &$+0.2\pm0.1$ &c, d, e\\

1682: \\

1683: NGC 6441  &EZ\_Ages   &$11.5_{-1.5}^{+1.9}$  &$-0.64\pm0.07$ &$+0.17\pm0.08$

1684: &$-0.20\pm0.05$  &$+0.54\pm0.13$  &$+0.01\pm0.08$ &\\

1685:           &literature &\ldots &$-0.43\pm0.08$ &$+0.34\pm0.09$ &\ldots &\ldots 

1686: &$+0.03\pm0.04$ &f, g\\

1687: \\

1688: NGC 6528  &EZ\_Ages   &$14.6_{-2.0}^{+\mbox{max\dag}}$  &$-0.26\pm0.06$ &$+0.12\pm0.06$

1689: &$-0.04\pm0.04$  &$+0.31\pm0.08$  &$-0.07\pm0.05$ &\\

1690:           &literature &11 &$-0.15$/$+0.1^{\dag\dag}$

1691: &$+0.07$/$+0.14^{\dag\dag}$ &\ldots &\ldots &$-0.40$/$+0.23^{\ddag}$ &h,

1692: i, j, k\\

1693: \\

1694: M 67  &EZ\_Ages   &$4.0_{-0.58}^{+1.0}$  &$-0.09\pm0.05$ &$0.03\pm0.05$

1695: &$-0.07\pm0.03$  &$-0.01\pm0.05$  &$-0.03\pm0.07$ &\\

1696:           &literature &3.5 &$0.0\pm0.1$ &$0.0\pm0.1$ &$0.0\pm0.1$ 

1697: &$0.0\pm0.1$ &$0.0\pm0.1$ &l, m, n\\

1698: \enddata

1699: \tablenotetext{*}{All reported ages are estimated from H$\beta$, except

1700:   for 47 Tuc, which falls off the model grids in

1701:   H$\beta$-$\langle$Fe$\rangle$ and is therefore calculated using

1702:   H$\gamma_F$.}

1703: \tablenotetext{\dag}{Positive age errors cannot be calculated by EZ\_Ages

1704:   because they exceed the maximum 15.8 Gyr age of the models.}

1705: \tablenotetext{\ddag}{Age determined from relative ages of NGC 6121

1706:   and 47 Tuc in \citet{sal02}, using 47 Tuc age from \citet{sch02b}}

1707: \tablenotetext{**}{CN-strong model / CN-weak model}

1708: \tablenotetext{\dag\dag}{Results differ substantially between authors.

1709:   Reported abundances are \citet{zoc04} and \citet{bar04} /

1710:   \citealt{car01}} 

1711: \tablerefs{(a) \citealt{sal02}, (b) \citealt{iva99}, 

1712:   (c) \citealt{sch02a,sch02b}, (d) \citealt{bri04}, (e) \citealt{car04}, 

1713:   (f), (g) \citet{gra06}, (h) \citealt{ort01}, 

1714:   (i) \citealt{zoc04}, (j) \citealt{bar04}, (k) \citealt{car01}, 

1715:   (l) \citealt{sch04a,sch04b}, (m) \citealt{tau00}, (n) \citealt{she00} }

1716: \end{deluxetable}

1717: 

1718: \begin{deluxetable}{lcccccc}

1719: \tabletypesize{\scriptsize}

1720: \tablecaption{Models Using Different Indices For Fitting\label{diff_ind_tab}}

1721: \tablewidth{0pt}

1722: \tablehead{

1723: \colhead{} &

1724: \multicolumn{6}{c}{Indices used in fit} \\

1725: \cline{2-7} \\

1726: \colhead{Model name} &

1727: \colhead{Balmer} &

1728: \colhead{Fe} &

1729: \colhead{Mg} &

1730: \colhead{C} &

1731: \colhead{N} &

1732: \colhead{Ca}

1733: }

1734: \startdata

1735: Standard &H$\beta$ &$\langle$Fe$\rangle$ &Mg~{\it b} &C$_2$4668 &CN$_1$ &Ca4227 \\

1736: H$\gamma_F$ &{\bf H$\gamma_F$} &$\langle$Fe$\rangle$ &Mg~{\it b} &C$_2$4668 &CN$_1$ &Ca4227 \\

1737: H$\delta_F$ &{\bf H$\delta_F$} &$\langle$Fe$\rangle$ &Mg~{\it b} &C$_2$4668 &CN$_1$ &Ca4227 \\

1738: Balmer &{\bf all} &$\langle$Fe$\rangle$ &Mg~{\it b} &C$_2$4668 &CN$_1$ &Ca4227 \\

1739: Fe4383 &H$\beta$ &{\bf Fe4383} &Mg~{\it b} &C$_2$4668 &CN$_1$ &Ca4227 \\

1740: Fe5015 &H$\beta$ &{\bf Fe5015} &Mg~{\it b} &C$_2$4668 &CN$_1$ &Ca4227 \\

1741: Fe5270 &H$\beta$ &{\bf Fe5270} &Mg~{\it b} &C$_2$4668 &CN$_1$ &Ca4227 \\

1742: Fe5335 &H$\beta$ &{\bf Fe5335} &Mg~{\it b} &C$_2$4668 &CN$_1$ &Ca4227 \\

1743: Mg$_2$ &H$\beta$ &$\langle$Fe$\rangle$ &{\bf Mg$_2$} &C$_2$4668 &CN$_1$ &Ca4227  \\

1744: G4300  &H$\beta$ &$\langle$Fe$\rangle$ &Mg~{\it b} &{\bf G4300} &CN$_1$  &Ca4227 \\

1745: CN$_2$ &H$\beta$ &$\langle$Fe$\rangle$ &Mg~{\it b} &C$_2$4668 &{\bf CN$_2$} &Ca4227 \\

1746: High-z &{\bf H$\delta_F$} &{\bf Fe4383} &... &{\bf G4300} &CN$_1$ &Ca4227 \\

1747: All    &{\bf all} &{\bf all} &{\bf all} &{\bf all} &{\bf all}     &Ca4227 \\

1748: \enddata

1749: \tablecomments{For indices used in the fitting process, bold face

1750:   indicates the index that differs from the standard set.  ``All''

1751:   indicates that all available indices for that element were averaged

1752:   together (i.e., fitting for [C/Fe] was done using an average of

1753:   C$_2$4668 and G4300).  

1754: }

1755: \end{deluxetable}

1756: 

1757: \begin{deluxetable}{lccccccccccccc}

1758: \tabletypesize{\scriptsize}

1759: \tablecaption{EZ\_Ages Results Using Different Indices For Fitting\label{diff_fit_tab}}

1760: \tablewidth{0pt}

1761: \tablehead{

1762: \colhead{} &

1763: \multicolumn{6}{c}{EZ\_Ages results for NGC~6441} &

1764: \colhead{} &

1765: \multicolumn{6}{c}{EZ\_Ages results for M~67} \\

1766: \cline{2-7} \cline{9-14} \\

1767: \colhead{Model} &

1768: \colhead{Age} &

1769: \colhead{[Fe/H]} &

1770: \colhead{[Mg/Fe]} &

1771: \colhead{[C/Fe]} &

1772: \colhead{[N/Fe]} &

1773: \colhead{[Ca/Fe]} &

1774: \colhead{} &

1775: \colhead{Age} & 

1776: \colhead{[Fe/H]} &

1777: \colhead{[Mg/Fe]} &

1778: \colhead{[C/Fe]} &

1779: \colhead{[N/Fe]} &

1780: \colhead{[Ca/Fe]}

1781: }

1782: \startdata

1783: Standard    &11.4 &$-$0.64 &$+$0.17 &$-$0.20 &$+$0.62 &$+$0.04 & &4.0 &$-0.09$ &$+0.03$ &$-0.07$ &$-0.01$ &$-0.03$\\

1784: H$\gamma_F$  &6.9 &$-$0.54 &$+$0.17 &$-$0.22 &$+$0.60 &$+$0.05 & &3.5 &$-0.07$ &$+0.03$ &$-0.06$ &$-0.01$ &$+0.01$\\

1785: H$\delta_F$  &8.2 &$-$0.57 &$+$0.17 &$-$0.22 &$+$0.60 &$+$0.05 & &3.7 &$-0.08$ &$+0.05$ &$-0.06$ &$-0.01$ &$-0.04$\\

1786: Balmer       &8.3 &$-$0.57 &$+$0.17 &$-$0.21 &$+$0.60 &$+$0.05 & &3.7 &$-0.08$ &$+0.03$ &$-0.06$ &$-0.01$ &$-0.01$\\

1787: Fe4383      &11.1 &$-$0.55 &$+$0.08 &$-$0.25 &$+$0.58 &$-$0.04 & &3.8 &$-0.02$ &$-0.03$ &$-0.10$ &$-0.03$ &$-0.09$\\

1788: Fe5015$^{\dag}$&...&...      &...      &...     &...      &... & &4.3 &$-0.17$ &$+0.08$ &$-0.02$ &$-0.04$ &$+0.01$\\

1789: Fe5270      &11.3 &$-$0.62 &$+$0.15 &$-$0.21 &$+$0.60 &$+$0.01 & &3.9 &$-0.10$ &$+0.03$ &$-0.07$ &$-0.01$ &$-0.03$\\

1790: Fe5335      &11.5 &$-$0.66 &$+$0.19 &$-$0.19 &$+$0.64 &$+$0.06 & &4.0 &$-0.09$ &$+0.03$ &$-0.07$ &$-0.01$ &$-0.03$\\

1791: Mg$_2$      &11.4 &$-$0.63 &$+$0.22 &$-$0.20 &$+$0.60 &$+$0.02 & &4.0 &$-0.09$ &$+0.04$ &$-0.07$ &$-0.01$ &$-0.03$\\

1792: G4300$^{*}$ &11.6 &$-$0.65 &$+$0.15 &$-$0.33 &$+$0.79 &$+$0.02 & &...  &...      &...      &...  &...     &...    \\

1793: CN$_2$      &11.4 &$-$0.64 &$+$0.16 &$-$0.20 &$+$0.56 &$+$0.02 & &4.0 &$-0.09$ &$+0.03$ &$-0.07$ &$-0.03$ &$-0.04$\\

1794: High-z$^{*}$&5.7  &$-$0.34 &$+$0.03 &$-$0.33 &$+$0.50 &$-$0.10 & &2.7 &$+0.15$ &...     &$-0.15$ &$-0.04$ &$-0.13$\\

1795: All$^{*\dag}$&7.5 &$-$0.47 &$+$0.12 &$-$0.26 &$+$0.53 &$-$0.02 & &3.7 &$-0.07$ &$+0.03$ &$-0.06$ &$-0.02$ &$-0.01$\\

1796: \enddata

1797: \tablenotetext{*}{There is a problem with G4300 in the M~67 spectrum

1798:   \citep{sch04a}, thus for this cluster no separate model for G4300 is

1799:   computed and the ``high-z'' and ``all'' models are computed with

1800:   C$_2$4668.}

1801: \tablenotetext{\dag}{Fe5015 falls on a bad CCD column in the spectrum

1802:   of NGC~6441, thus for this cluster no separate model for Fe5015 is

1803:   computed and the ``all'' model does not include Fe5015.  }

1804: \end{deluxetable} 

1805: 

1806: 

1807: \clearpage

1808: 

1809: 

1810: \begin{figure}

1811: \epsscale{1}

1812: \plotone{f1.eps}

1813: \caption{Grids produced by the S07 model illustrating the

1814:   effects of super-solar [Mg/Fe].  Solid lines show constant [Fe/H]

1815:   from left to right of -1.3, -0.7, -0.4, 0.0, and +0.2 ([Fe/H] = 0.0

1816:   is shown as a dashed line for reference).  Dotted lines show

1817:   constant age from top to bottom of 1.2, 2.2, 3.5, 7.0, and 14.1 Gyr.

1818:   The square shows an example data point and is the same in both

1819:   panels.  {\it Upper panel:} Models computed with solar-scaled

1820:   abundances.  The Balmer-Fe grid gives fiducial values of $t = 7$ Gyr

1821:   and [Fe/H] = -0.2.  The Balmer-Mg~{\it b} grid shows an [Fe/H] $>

1822:   0.4$ dex higher than the fiducial from the Balmer-Fe grid.  This

1823:   indicates that [Mg/Fe] is super-solar.  {\it Lower panel:}

1824:   Increasing [Mg/Fe] in the models slides the grid to the right toward

1825:   the Balmer-Mg~{\it b} data point, lowering the estimated [Fe/H].

1826:   H$\beta$ is also slightly affected by the abundance change, yielding

1827:   a slightly younger fiducial age.  In the Mg-enhanced model, the

1828:   fiducial [Fe/H] from the Balmer-Fe grid matches the value of [Fe/H]

1829:   estimated from the Balmer-Mg~{\it b} grid, indicating that [Mg/Fe]

1830:   $=+0.3$ is a good fit to the data. }\label{grids}

1831: \end{figure}

1832: 

1833: \clearpage

1834: 

1835: \begin{figure}

1836: \epsscale{1.0}

1837: \plotone{f2.eps}

1838: \caption{ Comparison of the error estimates determined by EZ\_Ages

1839:   with the results of the Monte Carlo error simulation.  Age errors

1840:   are plotted in the left panel and [Fe/H] and abundance errors in the

1841:   right panel.  The solid lines show one-to-one identity relations,

1842:   with a $\pm30$\% spread indicated by dashed lines.  The error

1843:   estimates produced by EZ\_Ages are consistent with those from the

1844:   full Monte Carlo simulations, indicating that EZ\_Ages produces

1845:   reliable error estimates.

1846: }\label{errplot}

1847: \end{figure}

1848: 

1849: \begin{figure}

1850: \epsscale{1.0}

1851: \plotone{f3.eps}

1852: \caption{The distribution of EZ\_Ages results for the Monte-Carlo

1853:   error simulation.  Yellow, red, blue, and green points correspond to

1854:   the realizations of a single data point drawn from the 2\%, 5\%,

1855:   10\%, and 20\% error simulations, respectively.  These trace out the

1856:   error ellipses of the EZ\_Ages fitting analysis and reveal the

1857:   effect of correlated errors. }\label{corr_errs}

1858: \end{figure}

1859: 

1860: \begin{figure}

1861: \epsscale{1.0}

1862: \plotone{f4.eps}

1863: \caption{S07 solar-scaled model grids compared with cluster data.

1864:   Model grids lines are as in Figure \ref{grids}.  Because H$\beta$

1865:   and $\langle$Fe$\rangle$ are relatively insensitive to non-solar

1866:   abundance ratios, the fiducial age and [Fe/H] values for each

1867:   cluster can be read off the model grid in the

1868:   H$\beta$-$\langle$Fe$\rangle$ plot (upper left).  If solar-scaled

1869:   models were a good fit to all clusters, the data would fall in the

1870:   same region of the model space in each index-index plot, as is

1871:   approximately the case with M~67.  The fact that other cluster data

1872:   do not indicates that non-solar abundance patterns are needed to fit

1873:   these clusters.  }\label{cluster_grids}

1874: \end{figure}

1875: 

1876: \begin{figure}

1877: \epsscale{1.0}

1878: \plotone{f5.eps}

1879: \caption{Comparison of EZ\_Ages age ($\log t$), [Fe/H], [Mg/Fe],

1880:   [C/Fe], [N/Fe], and [Ca/Fe] estimates with values from the

1881:   literature.  Error bars indicate the mean observational errors for

1882:   our cluster data.  Not all cluster parameters are available in the

1883:   literature.  For NGC 6528, two sets of abundances are available from

1884:   different groups and these differe substantially (see Table

1885:   \ref{abuntable}).  The individual literature values are plotted in

1886:   gray, with an average of the two literature values plotted in black.

1887:   For 47~Tuc, [C/Fe] and [N/Fe] abundances in the literature are

1888:   computed separately for CN-weak and CN-strong stars.  Here we

1889:   compare the EZ\_Ages results for [C/Fe] and [N/Fe] against the

1890:   average over both sets of stars, as this best simulates their

1891:   combined contribution to the integrated light }\label{comp_lit}

1892: \end{figure}

1893: 

1894: \begin{figure}

1895: \epsscale{1.0}

1896: \plotone{f6.eps}

1897: \caption{Comparison of observed Lick index measurements with the

1898:   predicted index values from the best-fitting S07 model, as

1899:   determined by EZ\_Ages.  All indices included in the S07 models are

1900:   shown for each of the five test clusters.  Measurement errors in the

1901:   observed indices are approximately the size of the plotting symbols.

1902:   Indices shown in black are those used in the abundance fitting

1903:   process; those in gray are not used in the fitting.  Solid lines

1904:   show the one-to-one relation.  Indices not used in the fitting

1905:   process show some discrepancies between model predictions and the

1906:   observed values, but there is no evidence for systematic

1907:   discrepancies in the modeling of any of the

1908:   indices. }\label{model_ind}

1909: \end{figure}

1910: 

1911: \begin{figure}

1912: \epsscale{1.0}

1913: \plotone{f7.eps}

1914: \caption{Variations in age ($\log t$), [Fe/H], [Mg/Fe], [C/Fe],

1915:   [N/Fe], and [Ca/Fe] estimates for NGC~6441 from using different

1916:   indices in the fitting process.  Model names correspond to those in

1917:   Table \ref{diff_ind_tab}.  Error bars indicate the

1918:   observational errors calculated by EZ\_Ages for NGC~6441 using the

1919:   standard set of indices.  (a--c) Results from substituting one

1920:   alternative index into the standard index set (i.e., substituting

1921:   G4300 for C$_2$4668 to fit [C/Fe]), and also for using averages of

1922:   lines (i.e., an average of all Balmer lines instead of H$\beta$).

1923:   All combinations of indices give results consistent within $\pm0.1$

1924:   dex, with the exception of the bluer Balmer lines (panel a), which

1925:   underpredict the cluster age, and G4300 (panel c, triangle), which

1926:   underpredicts [C/Fe] by $-0.13$ dex relative to the fit using

1927:   C$_2$4668.  The underprediction of [C/Fe] results in an

1928:   overprediction of [N/Fe], as expected from the correlated errors

1929:   between [C/Fe] and [N/Fe] (see Figure \ref{corr_errs}).  (d) Results

1930:   for an average of all indices and for fitting when only indices

1931:   blueward of 4400{\AA} are available (simulating higher redshift

1932:   observations).  Fitting with Fe4383 and H$\delta_F$ instead of

1933:   $\langle$Fe$\rangle$ and H$\beta$ (as in the simulated high redshift

1934:   case) results in substantially younger ages and higher [Fe/H].  See

1935:   text \S\ref{consist} for details.  }\label{vary_fit_NGC6441}

1936: \end{figure}

1937: 

1938: \begin{figure}

1939: \epsscale{1.0}

1940: \plotone{f8.eps}

1941: \caption{Variations in age ($\log t$), [Fe/H], [Mg/Fe], [C/Fe],

1942:   [N/Fe], and [Ca/Fe] estimates for M~67 from using different indices

1943:   in the fitting process.  Model names correspond to those in

1944:   Table \ref{diff_ind_tab}.  Error bars indicate the observational

1945:   errors calculated by EZ\_Ages for M~67 using the standard set of

1946:   indices.  (a--c) Results from substituting one alternative index

1947:   into the standard index set (i.e., substituting G4300 for C$_2$4668

1948:   to fit [C/Fe]), and also for using averages of lines (i.e., an

1949:   average of all Balmer lines instead of H$\beta$).  (d) Results for

1950:   an average of all indices and for fitting when only indices blueward

1951:   of 4700{\AA} are available (simulating higher redshift

1952:   observations).  Fitting with Fe4383 and H$\delta_F$ instead of

1953:   $\langle$Fe$\rangle$ and H$\beta$ (as in the simulated high redshift

1954:   case) results in substantially younger ages and higher [Fe/H].  See

1955:   text \S\ref{consist} for details.  }\label{vary_fit_M67}

1956: \end{figure}

1957: 

1958: \begin{figure}

1959: \epsscale{0.9}

1960: \plotone{f9.eps}

1961: \caption{Solar abundance model grids showing the difficulty of

1962:   measuring age and [Fe/H] from Fe4383 and H$\delta_F$.  Lines of

1963:   constant age run nearly horizontally and show (from top to bottom)

1964:   2.5, 2.8, 3.5, 5.0, 7.0, 8.9, and 14.1 Gyr models.  Lines of

1965:   constant [Fe/H] run nearly vertically and show (from left to right)

1966:   $-0.4$, 0.0, and $+0.2$ dex models.  The 3.5 Gyr and solar abundance

1967:   model are shown with dashed lines for reference.  Measured index

1968:   values for M~67 are overplotted as the diamond.  The standard

1969:   indices (panel a) produce model grids with lines of constant age and

1970:   constant [Fe/H] nearly perpendicular.  Line of constant [Fe/H] are

1971:   more sloped from the vertical when using Fe4383 instead of

1972:   $\langle$Fe$\rangle$ (panel b).  Similarly, lines of constant age

1973:   are more sloped from the horizontal when using H$\delta_F$ instead

1974:   of H$\beta$ (panel c).  Substituting one or the other of these

1975:   indices for the standard set causes only small offsets from the

1976:   standard results (see Figure \ref{vary_fit_M67}).  However, the

1977:   combination of Fe4383 and H$\delta_F$ (panel d) results in model

1978:   grids that collapse down on one another, making the results much

1979:   more sensitive to small errors or zeropoint uncertainties in either

1980:   the models or the data.  Because of the near-degeneracy of the

1981:   models in Fe4383-H$\delta_F$ space, the combination of these two

1982:   indices can give results substantially different from the age and

1983:   [Fe/H] measured using the standard index set.

1984:   }\label{red_blue_grids}

1985: \end{figure}

1986: 

1987: \begin{figure}

1988: \epsscale{1.0}

1989: \plotone{f10.eps}

1990: \caption{Comparison of $\log$ Age, [Fe/H], and [$\alpha$/Fe] results

1991:   from T05 and from EZ\_Ages.  The galaxy sample is that of T05, and

1992:   the stellar population parameters are from SSP fits to H$\beta$,

1993:   $\langle$Fe$\rangle$, and Mg~$b$ for each set of models.  T05 models

1994:   are cast in terms of [Z/H], which we compare to EZ\_Ages values of

1995:   [Fe/H] using the conversion given in TMB03 for their

1996:   $\alpha$-enhanced mixture: [Fe/H] = [Z/H] - 0.94 [$\alpha$/Fe].  We

1997:   compare their values of [$\alpha$/Fe] to the EZ\_Ages results for

1998:   [Mg/Fe], since T05 use the Mg~$b$ line as their $\alpha$-enhancement

1999:   indicator.  Dashed lines show the one-to-one relation.  Solid lines

2000:   show the best fit one-to-one slope allowing for a zeropoint offset.

2001:   The size and direction of the zeropoint offset are indicated in the

2002:   top left of each panel.  Error bars in the lower right corner of

2003:   each panel indicate the median observational errors from T05.

2004:   Results for [$\alpha$/Fe] are very consistent between the two

2005:   models.  T05 find ages which are younger by $\sim 35$\% and [Fe/H]

2006:   values which are higher by $\sim 0.08$ dex than the EZ\_Ages

2007:   results.  Also, the slope of the [Fe/H] comparison differs somewhat

2008:   from the one-to-one relation.  These zeropoint differences, as well

2009:   as the slope difference in [Fe/H], can be explained by differences

2010:   in the model ages, which affect the [Fe/H] estimates, but have

2011:   relatively little effect on [$\alpha$/Fe].  The age zeropoint

2012:   difference between the models is somewhat larger than the zeropoint

2013:   uncertainty in the S07 models (the age zeropoint of the models used

2014:   in T05 is uncalibrated), suggesting that the T05 analysis may

2015:   slightly underestimate the age of the sample galaxies.  See text for

2016:   details.  }\label{compare_thomas}

2017: \end{figure}

2018: 

2019: \begin{figure}

2020: \epsscale{1.0}

2021: \plotone{f11.eps}

2022: \caption{{\it Top:} Difference between [Fe/H] estimates for the T05

2023:   galaxy sample from EZ\_Ages and from T05, as a function of stellar

2024:   population age and [Mg/Fe].  {\it Bottom:} Difference in [Fe/H] as a

2025:   function of differences in stellar population age and [$\alpha$/Fe].

2026:   The differences between T05 and EZ\_Ages estimates of [Fe/H] are

2027:   most strongly correlated with differences in the age results between

2028:   the two models (lower left panel).  This suggests that the non-unity

2029:   slope of the [Fe/H] comparison relation in Figure

2030:   \ref{compare_thomas} is due to differences in the age estimates and

2031:   arises from correlated errors (see discussion in \S\ref{errors}).

2032:   The non-orthogonality of model grids in the index-index diagrams (as

2033:   in Figure \ref{grids}) results in correlated errors such that

2034:   underestimating the age of a population will result in

2035:   overestimating the corresponding value of [Fe/H].

2036:   }\label{thomas_slope}

2037: \end{figure}

2038: 

2039: 

2040: \end{document}

2041: