Annu. Rev. Astron. Astrophys. 2006. 44: xxx-xxx Copyright © 2006 by Annual Reviews. All rights reserved

3. ELLIPTICAL GALAXIES IN THE LOCAL UNIVERSE

Observations at high redshift are certainly the most direct way to look at the forming galaxies, and a great observational effort is currently being made in this direction. Yet, high-redshift galaxies are very faint, and only few of their global properties can now be measured. Nearby galaxies can instead be studied in far greater detail, and their fossil evidence can provide a view of galaxy formation and evolution that is fully complementary to that given by high-redshift observations. By fossil evidence one refers to those observables that are not related to ongoing, active star formation, and which are instead the result of the integrated past star formation history. At first studies attempted to estimate ages and metallicities of the dominant stellar populations on a galaxy-by-galaxy basis. But the tools used were still quite rudimentary, being based on largely incomplete libraries of stellar spectra and evolutionary sequences. Hence, through the 1980s progress was relatively slow, and opinions could widely diverge as to whether ellipticals were dominated by old stellar populations -as old as galactic globular clusters- or by intermediate age ones, several billion years younger than globulars (see e.g. O'Connell 1986, Renzini 1986) -with much of the diverging interpretations being a result of the age-metallicity degeneracy. From the beginning of the 1990s progress has been constantly accelerating, and much of this review concentrates on the developments that took place over the past 15 years.

3.1. Color-Magnitude Relation, Fundamental-Plane and Line-Indices

3.1.1 THE COLOR-MAGNITUDE AND COLOR- RELATIONS   That elliptical galaxies follow a tight color-magnitude (C-M) relation was first recognized by Baum (1959), and in a massive exploration Visvanathan & Sandage (1977) and Sandage & Visvanathan (1978a, b) established the universality of this relation with what continues to be the culmination of ETG studies in the pre-CCD era. The C-M relation looked the same in all nine studied clusters, and much the same in the field as well, though with larger dispersion (at least in part due to larger distance errors). The focus was on the possible use of the C-M relation as a distance indicator; however, Sandage & Visvanathan documented the tightness of the relation and noted that it implies the stellar content of the galaxies to be very uniform. They also estimated that both S0's and ellipticals had to be evolving passively since at least ~ 1 Gyr ago. Figure 3 shows a modern rendition for the C-M plot for the Coma cluster galaxies, showing how tight it is, as well as how closely both S0's and ellipticals follow the same relation, as indeed Sandage & Visvanathan had anticipated.

Figure 3. The (U - V) - MV color-magnitude relation for galaxies of the various morphological types that are spectroscopic members of the Coma cluster (Bower et al. 1999).

In a major breakthrough in galaxy dating, Bower, Lucey, & Ellis (1992), rather than trying to age-date galaxies one by one, were able to set tight age constraints on all ETGs in Virgo and Coma at once. Noting the remarkable homogeneity of ETGs in these clusters, they estimated the intrinsic color scatter in the color- relation (see Figure 4) to be (U - V) 0.04 mag, where is the central stellar velocity dispersion of these galaxies. They further argued that - if due entirely to an age dispersion t = (t_H - t_F), such color scatter should be equal to the time scatter in formation epochs, times (U - V) / t, i.e.:

(1)

where t_H is the age of the universe at z = 0, and galaxies are assumed to form before a lookback time t_F. Bower and colleagues introduced the parameter , such that (t_H - t_F) is the fraction of the available time during which galaxies actually form. Thus, for = 1 galaxy formation is uniformly distributed between t ~ 0 and t = t_H - t_F , whereas for < 1 it is more and more synchronized, i.e., restricted to the fraction of time interval t_H - t_F. Adopting (U - V) / t from the models of Bruzual (1983), they derived t_H - t_F < 2 Gyr for = 1 and t_H - t_F < 8 Gyr for = 0.1, corresponding respectively to formation redshifts z_F 2.8 and 1.1 for their adopted cosmology (t_H = 15 Gyr, q_o = 0.5). For the concordance cosmology, the same age constraints imply z_F 3.3 and 0.8, respectively. A value = 0.1 implies an extreme synchronization, with all Virgo and Coma galaxies forming their stars within less than 1 Gyr when the universe had half its present age, which seems rather implausible. Bower and colleagues concluded that ellipticals in clusters formed the bulk of their stars at z 2, and later additions should not provide more than ~ 10% of their present luminosity. Making minimal use of stellar population models, this approach provided for the first time a robust demonstration that cluster ellipticals are made of very old stars, with the bulk of them having formed at z 2.

Figure 4. The relation between the (U - V) color and the central velocity dispersion () for early-type galaxies in the Virgo (open symbols) and Coma (filled symbols) clusters. Red circles represent ellipticals, blue triangles represent S0's. From Bower, Lucey & Ellis (1992).

3.1.2 THE FUNDAMENTAL PLANE   Three key observables of elliptical galaxies, namely the effective radius R_e, the central velocity dispersion , and the luminosity L (or equivalently the effective surface brightness I_e = L / 2 R_e²) relate their structural/dynamical status to their stellar content. Indeed, elliptical galaxies are not randomly distributed within the 3D space (R_e, , I_e), but rather cluster close to a plane, thus known as the fundamental plane (FP), with R_e ^a I_e^b (Dressler et al. 1987; Djorgovski & Davis 1987), where the exponents a and b depend on the specific band used for measuring the luminosity. The projection of the FP over the (R_e, I_e) coordinate plane generates the Kormendy relation (Kormendy 1977), whereas a projection over the (, L = 2 R_e² I_e) plane generates the Faber-Jackson relation (Faber & Jackson 1976). At a time when testing the _M = 1 standard cosmology had high priority, the FP was first used to estimate distances, in order to map deviations from the local Hubble flow and construct the gravitational potential on large scales. Its use to infer the properties of the stellar content of galaxies, and set constraints on their formation, came later. Yet, by relating the luminosity to the structural-dynamical parameters of a galaxy, the FP offers a precious tool to gather information on the ages and metallicities of galaxies, at low as well as at high redshifts.

The mere existence of a FP implies that ellipticals (a) are virialised systems, (b) have self-similar (homologous) structures, or their structures (e.g., the shape of the mass distribution) vary in a systematic fashion along the plane, and (c) contain stellar populations which must fulfill tight age and metallicity constraints. Here we concentrate on this latter aspect.

To better appreciate the physical implications of the FP, Bender, Burnstein, & Faber (1992) introduced an orthogonal coordinate system (₁, ₂, ₃), in which each new variable is a linear combination of log ², log R_e and log I_e. The transformation corresponds to a rotation of the coordinate system such that in the (₁, ₃) projection the FP is seen almost perfectly edge-on. Moreover, if structural homology holds all along the plane, then log M / L = 3^1/2 ₃ + const. If is (almost) unaffected by the dark matter distribution (as currently understood, Rix et al. 1997), then ₃ provides a measure of the stellar M / L ratio, and ₁ log (² R_e) log M a measure of the stellar mass. Bender and colleagues showed that in Virgo and Coma the FP is remarkably "thin", with a 1- dispersion perpendicular to the plane of only (₃) 0.05, corresponding to a dispersion in the M / L ratio 10% at any position along the plane. Moreover, the FP itself is "tilted", with the M / L ratio apparently increasing by a factor ~ 3 along the plane, while the mass is increasing by a factor ~ 100. Note that the tilt does not imply a departure from virialization, but rather a systematic trend of the stellar content with galaxy mass, possibly coupled with a systematic departure from structural homology (e.g., Bender, Burstein & Faber 1992, Ciotti 1997, Busarello et al. 1997).

The narrowness of the FP, coupled to the relatively large tilt ( ₃ / (₃) 0.35/0.05 = 7) requires some sort of fine tuning, which is perhaps the most intriguing property of the FP (Renzini & Ciotti 1993). Although unable to identify one specific origin for the FP tilt, Renzini & Ciotti argued that the small scatter perpendicular to the FP implied a small age dispersion ( 15%) and high formation redshift, fully consistent with the Bower, Lucey & Ellis (1992) argument based on the narrowness of the C-M and color- relations.

The remarkable properties of the FP for the Virgo and Coma clusters were soon shown to be shared by all studied clusters in the local universe. Jørgensen, Franx, & Kjærgaard (1996) constructed the FP for 230 ETGs in 10 clusters (including Coma), showing that the FP tilt and scatter are just about the same in all local clusters, thus strengthening the case for the high formation redshift of cluster ETGs being universal. However, Worthey, Trager, & Faber (1995) countered that the thinness of the FP, C-M, and color- relations could be preserved, even with a large age spread, provided age and metallicity are anticorrelated (with old galaxies being metal poor and young ones being metal rich). This is indeed what Worthey and colleagues reported from their line-indices analysis (see below), indicating a factor of ~ 6 for the range in age balanced by a factor ~ 10 in metallicity (from solar to ~ 10 times solar). If so, then the FP should be thicker in the near infrared, because the compensating effect of metallicity would be much lower at longer wavelength, thus unmasking the full effect of a large age spread (Pahre, Djorgovski, & De Carvalho 1995). But Pahre and colleagues found the scatter of the FP K-band to be the same as in the optical. In addition, its slope implied a sizable variation of M / L_K M^0.16 along the FP, somewhat flatter than in the optical (M / L_V M^0.23), still far from the M / L_K ~ const. predicted by Worthey et al. (1995).

These conclusions were further documented and reinforced by Pahre, Djorgovski, & De Carvalho (1998), Scodeggio et al. (1998), Mobasher et al. (1999), and Pahre, De Carvalho, & Djorgovski (1998), who finally concluded that the origin of the FP tilt defies a simple explanation, but is likely the result of combined age and metallicity trends along the plane (with the most metal rich galaxies being actually the oldest), plus an unidentified systematic deviation from structural homology. Several possibilities for the homology breaking have been proposed and investigated, such as variation in stellar and/or dark matter content and/or distribution, anisotropy, and rotational support (e.g., Ciotti, Lanzoni, & Renzini 1996, Prugniel & Simien 1996, Ciotti & Lanzoni 1997). Recently, Trujillo, Burkert, & Bell (2004) argued that one fourth of the tilt is due to stellar population (i.e., a combination of metallicity and age), and three quarters of it to structural nonhomology in the distribution of the visible matter.

Of special interest is the comparison of the FP in clusters and in the field, because one expects all formation processes to be faster in high density peaks of the matter distribution. This was tested by Bernardi et al. (2003b, 2006) with a sample of ~ 40,000 SDSS morphology- and color-selected ETGs spanning a wide range of environmental conditions, from dense cluster cores to very low densities. Bernardi and colleagues found very small, but detectable differences in the FP zero point; the average surface brightness is ~ 0.08 mag brighter at the lowest density extreme compared to the opposite extreme. As the sample galaxies are distributed in redshift up to z ~ 0.3, they used the observed lookback time to empirically determine the time derivative of the surface brightness (hence in a model-independent fashion) and estimated that the 0.08 mag difference in surface brightness implies an age difference of ~ 1 Gyr, and therefore that galaxies in low density environments are ~ 1 Gyr younger compared to those in cluster cores.

3.1.3 THE LINE-STRENGTH DIAGNOSTICS   Optical spectra of ETGs present a number of absorption features whose strength must depend on the distributions of stellar ages, metallicities and abundance ratios, and therefore may give insight over such distributions. To exploit this opportunity, Burstein et al. (1984) introduced a set of indices now known as the Lick/IDS system, and started taking measurements for a number of galaxies. The most widely used indices have been the Mg₂ (or Mgb), <Fe>, and the H indices, measuring respectively the strength of MgH+MgI at 5156-5197Å, the average of two FeI lines at 5248 and 5315Å, and of H.

A first important result was the discovery that theoretical models based on solar abundance ratios adequately describe the combinations of the values of the <Fe> and Mg₂ indices in low-luminosity ETGs, but fail for bright galaxies (Peletier 1989, Gorgas, Efstathiou, & Aragón-Salamanca 1990, Faber, Worthey, & Gonzales 1992, Worthey, Faber, & Gonzales 1992, Davies, Sadler, & Peletier 1993, Jørgensen 1997). This implies either that population synthesis models suffered from some inadequacy at high metallicity (possibly due to incomplete stellar libraries), or that massive ellipticals were genuinely enriched in magnesium relative to iron, not unlike the halo stars of the Milky Way (e.g., Wheeler, Sneden, & Truran 1989). As for the Milky Way halo, such an -element overabundance may signal a prompt enrichment in heavy elements from Type II supernovae, with the short star-formation timescale having prevented most Type Ia supernovae from contributing their iron while star formation was still active. Yet, a star-formation timescale decreasing with increasing mass was contrary to the expectations of galactic wind/monolithic models (e.g., Arimoto & Yoshii 1987), where the star formation timescale increases with the depth of the potential well (Faber, Worthey & Gonzales 1992). However, as noted by Thomas (1999), the contemporary semi-analytical models did not predict any -element enhancement at all, no matter whether in low- or high-mass ETGs. Indeed, Thomas, Greggio, & Bender (1999) argued that the -enhancement, if real, was also at variance with a scenario in which massive ellipticals form by merging spirals, and required instead that star formation was completed in less than ~ 1 Gyr. Therefore, assessing whether the -enhancement was real, and in that case measuring it, had potentially far reaching implications for the formation of ETGs.

Two limitations had to be overcome in order to reach a credible interpretation of the <Fe> - Mgb plots: (a) existing synthetic models for the Lick/IDS indices were based on stellar libraries with fixed [/Fe] (Worthey 1994, Buzzoni 1995), and (b) an empirical verification of the reality of the -enhancement was lacking. In an attempt to overcome the first limitation, Greggio (1997) developed a scaling algorithm that allowed one to use existing models with solar abundance ratios to estimate the Mg overabundance, and she concluded that an enhancement up to [Mg/Fe] +0.4 was required for the nuclei of the most massive ellipticals (see also Weiss, Peletier, & Matteucci 1995). She also concluded that a closed-box model for chemical evolution failed to explain the very high values of the Mg₂ index of these galaxies. Indeed, the numerous metal-poor stars predicted by the model would obliterate the Mg₂ feature, hence the nuclei of ellipticals had to lack substantial numbers of stars more metal poor than ~ 0.5Z. Besides, very old ages ( 10 Gyr) and -enhancement were jointly required to account for galaxies with strong Mg₂. Eventually, Thomas, Maraston, & Bender (2003) produced a full set of synthetic models with variable [/Fe], and Maraston et al. (2003) compared such models to the indices of ETGs and of metal-rich globular clusters of the Galactic bulge, for which the -enhancement has been demonstarted on a star-by-star basis by high resolution spectroscopy. The result is displayed in Figure 5, showing that indeed the new models indicate for the bulge globulars an enhancement of [/Fe] ~ +0.3, in agreement with the stellar spectroscopy results, and similar to that indicated for massive ETGs.

Figure 5. The <Fe> index versus the Mgb index for a sample of halo and bulge globular clusters, the bulge integrated light in Baade's Window, and for elliptical galaxies from various sources. Overimposed are synthetic model indices (from Thomas, Maraston & Bender 2003) with solar metallicity ([Z/H] = 0), various -enhancements as indicated, and an age of 12 Gyr (black solid lines). The cyan grid shows a set of simple stellar population models (from Maraston 1998) with solar abundance ratios, metallicities from [Fe/H] = -2.25 to +0.67 (bottom to top), and ages from 3 to 15 Gyr (left to right). The blue grid offers an example of the so-called age-metallicity degeneracy. From Maraston et al. (2003).

Other widely used diagnostic diagrams involved the H index along with <Fe> and Mg₂ or Mgb. The Balmer lines had been suggested as good age indicators (e.g., O'Connell 1980; Dressler & Gunn 1983), an expectation that was confirmed by the set of synthetic models constructed by Worthey (1994) with the aim of breaking the age-metallicity degeneracy that affects the broad-band colors of galaxies. Worthey's models were applied by Jørgensen (1999) to a sample of 115 ETGs in the Coma cluster, and by Trager et al. (2000) to a sample of 40 ETGs biased toward low-density environments, augmented by 22 ETGs in the Fornax cluster from Kuntschner & Davies (1998), which showed systematically lower H indices. From these samples, and using the H - Mgb and H - <Fe> plots from Worthey's models, both Jørgensen and Trager and colleagues concluded that ages ranged from a few to almost 20 Gyr, but age and metallicity were anticorrelated in such a way that the Mgb - , C-M, and FP relations may be kept very tight. Moreover, there was a tendency for ETGs in the field to appear younger than those in clusters. Yet, Trager and colleagues cautioned that H is most sensitive to even low levels of recent star formation, and suggested that the bulk of stars in ETGs may well be old, but a small "frosting" of younger stars drives some galaxies toward areas in the H - Mgb and H - <Fe> plots with younger SSP ages. Finally, for the origin of the -enhancement Trager and colleagues favored a tight correlation of the IMF with , in the sense of more massive galaxies having a flatter IMF, hence more Type II supernovae. However, with a flatter IMF more massive galaxies would evolve faster in luminosity with increasing redshift, compared to less massive galaxies, which appears to be at variance with the observations (see below).

These conclusions had the merit of promoting further debates. Maraston & Thomas (2000) argued that even a small old, metal-poor component with a blue horizontal branch (like in galactic globulars) would increase the H index thus making galaxies look significantly younger than they are. Even more embarrassing for the use of the H - Mgb and H - <Fe> plots is that a perverse circulation of the errors automatically generates an apparent anticorrelation of age and metallicity, even where it does not exist. For example, if H is overestimated by observational errors, then age is underestimated, which in turn would reduce Mgb below the observed value unless the younger age is balanced by an artificial increase of metallicity. Trager and colleagues were fully aware of the problem, and concluded that only data with very small errors could safely be used. Kuntschner et al. (2001) investigated the effect by means of Monte Carlo simulations, and indeed showed that much of the apparent age-metallicity anticorrelation is a mere result of the tight correlation of their errors. They concluded that only a few outliers among the 72 ETGs in their study are likely to have few-billion-year-old luminosity-weighted ages, and these were typically galaxies in the field or loose groups, whereas a uniformly old age was derived for the vast majority of the studied galaxies. Moreover, younger ages were more frequently indicated for S0 galaxies (Kuntschner & Davies 1998). Nevertheless, in a cluster with very tight C-M and FP relations such as Coma, a large age spread at all magnitudes was found for a sample of 247 cluster members (Poggianti et al. 2001), and a sizable age-metallicity anticorrelation was also found for a large sample of SDSS galaxies (Bernardi et al. 2005).

As already alluded to, the main pitfall of the procedure is that the various indices depend on all three population parameters one is seeking to estimate: thus H is primarily sensitive to age, but also to [Fe/H] and [/Fe], <Fe> is sensitive to [Fe/H], but also to age and [Mg/Fe]; etc. Thus, the resulting errors in age, [Fe/H] and [Mg/Fe] are all tightly correlated, and one is left with the suspicion that apparent correlations or anticorrelations may be an artifact of the procedure, rather than reflecting the real properties of galaxies. In an effort to circumvent these difficulties, Thomas et al. (2005) renounced to trust the results galaxy by galaxy. They rather looked at patterns in the various index-index plots and compared them to mock galaxy samples generated via Monte Carlo simulations that fully incorporated the circulation of the errors. The real result was not a set of ages and metallicities assigned to individual galaxies, but rather age and metallicity trends with velocity dispersion, mass and environments. Having analyzed a sample of 124 ETGs in high- and low-density environments, Thomas and colleagues reached the following conclusions: (a) all three parameters -age, metallicity and [/Fe]- correlate strongly with , and, on average, follow the relations:

	(2)
	(3)
	(4)

where quantities in brackets/not in brackets refer to low-density/high-density environments, respectively. (b) For ETGs less massive than ~ 10¹⁰ M there is evidence for the presence of intermediate-age stellar populations with near-solar Mg/Fe. Instead, massive galaxies ( 10¹¹ M) appear dominated by old stellar populations, whereas at intermediate masses the strength of H requires either some intermediate age component or a blue horizontal branch (HB) contribution. (c) By and large this picture applies to both cluster and field ETGs, with cluster galaxies having experienced the bulk of their star formation between z ~ 5 and 2, and this activity appears to have been delayed by ~ 2 Gyr in the lowest density environments, i.e., between z ~ 2 and ~ 1. Figure 6 qualitatively summarizes this scenario, in which the duration of star formation activity decreases with increasing mass (as required by the [Mg/Fe] trend with ), and extends to younger ages for decreasing mass (as forced by the H - relation). Note that the smooth star-formation histories in this figure should be regarded as probability distributions, rather than as the actual history of individual galaxies, where star formation may indeed take place in a series of bursts. Qualitatively similar conclusions were reached by Nelan et al. (2005), from a study of ~ 4000 red-sequence galaxies in ~ 90 clusters as part of the National Astronomical Observatory Fundamental Plane Survey. Assuming the most massive galaxies ( ~ 400 km s^-1) to be 13 Gyr old, they derived an age of only 5.5 Gyr for less massive galaxies ( ~ 100 km s^-1). Note that the age- scaling of Thomas and colleagues would have given a much older age (~ 9.5 Gyr). Taken together, Equations 2 and 3 imply a trend of M / L_V by a factor ~ 1.8 along the FP (from = 100 to 350 km s^-1), thus accounting for almost two thirds of the FP tilt.

Figure 6. The scenario proposed by Thomas et al. (2005) for the average star formation history of early-type galaxies of different masses, from 5 × 109 M up to 1012 M, corresponding to 100 to ~ 320 km s-1, for the highest and lowest environmental densities, respectively, in the upper and lower panel.

As extensively discussed by Thomas et al. (2005), one residual concern comes from the possibility that part of the H strength may be due to blue HB stars. Besides a blue HB contribution by low-metallicity stars (especially in less massive galaxies), blue HB stars may also be produced by old, metal-rich populations, and appear to be responsible for the UV upturn in the spectrum of local ETGs (Brown et al. 2000, Greggio & Renzini 1990). In the Thomas et al. sample, some S0 outliers with strong H and strong metal lines would require very young ages and extremely high metallicity (up to ~ 10 times solar), and may better be accounted for by an old, metal-rich population with a well-developed blue HB.

The Mg₂ - relation has also been used to quantify environmental differences in the stellar population content. The cluster/field difference turns out to be small, with Mg₂ ~ 0.007 mag, corresponding to ~ 1 Gyr difference - field galaxies being younger - within a sample including ~ 900 ETGs (Bernardi et al. 1998), though no statistically significant environmental dependence of both Mg₂ and H was detected within a sample of ~ 9,000 ETGs from the SDSS (Bernardi et al. 2003a). Still from SDSS, coadding thousands of ETG spectra in various luminosity and environment bins, Eisenstein et al. (2003) detect clear trends with the environment thanks to the resulting exquisite S/N, but the differences are very small, and Eisenstein and colleagues refrain from interpreting them in terms of age/metallicity differences.

These results from the analysis of the Lick/IDS indices, including large trends of age with , or even large age-metallicity anticorrelations, have yet to be proven consistent with the FP and C-M relations of the same galaxies as established specifically for the studied clusters. Feeding the values of the indices, the synthetic models return ages, metallicities, and -enhancements. But along with them the same models also give the colors and the stellar M / L ratio of each galaxy in the various bands, hence allowing one to construct implied FP and C-M relations. It would be reassuring for the soundness of the whole procedure if such relations were found to be consistent with the observed ones. To our knowledge, this sanity check has not been attempted yet. The mentioned trends and correlations, if real, would also have profound implications for the evolution of the FP and C-M relations with redshift, an opportunity that will be exploited below.

3.2. Ellipticals Versus Spiral Bulges

The bulges of spiral galaxies are distinguished in "true bulges", typically hosted by S0-Sb galaxies, and "pseudobulges" usually (but not exclusively) in later-type galaxies (Kormendy & Kennicutt 2004). True (classical) bulges have long been known as similar to ellipticals of comparable luminosity, in both structure, line strengths and colors (e.g., Bender, Burnstein & Faber 1992, Jablonka, Martin, & Arimoto 1996, Renzini 1999, and references therein). Peletier et al. (1999) were able to quantify this similarity using Hubble Space Telescope (HST) WFPC2 (Wide Field Planetary Camera 2) and NICMOS (Near Infrared Camera and Multiobject Spectrometer) observations, and concluded that most (true) bulges in their sample of 20 spirals (including only 3 galaxies later than Sb) had optical and optical-IR colors similar to those of Coma ellipticals. Hence, like in Coma ellipticals their stellar populations formed at z 3, even if most of the galaxies in their sample are in small groups or in the field. More recently, Falcon-Barroso, Peletier, & Balcells (2002) measured the central velocity dispersion for the same sample observed by Peletier and colleagues, and constructed the FP for these bulges, showing that bulges in this sample tightly follow the same FP relation as cluster ellipticals, and therefore had to form their stars at nearly the same epoch. The similarity of true bulges and ellipticals includes the tendency of less massive objects to have experienced recent star formation, as indicated by their location in the Mg₂ - diagram in Figure 7. These similarities between true (classical) bulges and ellipticals suggest a similar origin, possibly in merger-driven starbursts at high redshifts. Pseudobulges, instead, are more likely to have originated via secular evolution of disks driven by bars and other deviations from axial symmetry, as extensively discussed and documented by Kormendy & Kennicutt (2004). Several of the objects in the Prugniel and colleagues sample in Figure 7 are likely to belong to the pseudobulge group. From the Lick/IDS indices of a sample of bulges, Thomas and Davies (2006) argue that the same scenario depicted in Figure 6 for ETGs, applies to bulges as well, the main difference being that bulges are on average less massive, hence on average younger than ETGs.

Figure 7. The Mg2 - relation for the spiral bulges studied by Falcon-Barroso et al. (2002), labelled "our bulges" in the insert, are compared to bulges in other samples (B92 = Bender, Burnstein & Faber 1992, P01 = Prugniel, Maubon & Simien 2001, JMA96 = Jablonka, Martin & Arimoto 1996). The solid lines are the bet fits to the corresponding data, while the dashed line show the average relation for cluster ellipticals from Jørgensen, Franx & Kjærgaard (1996).

Looking near to us, HST and ground based photometry of individual stars in the Galactic bulge have shown that they are older than at least 10 Gyr, with no detectable trace of an intermediate age component (Ortolani et al. 1995, Kuijken & Rich 2002, Zoccali et al. 2003). HST/NICMOS photometry of stars in the bulge of M31 has also shown that their H-band luminosity function is virtually identical to that of the Galactic bulge, and by inference should have nearly identical ages (Stephens et al. 2003). These two bulges belong to spirals in a rather small group, and yet appear to have formed their stars at an epoch corresponding to z 2, not unlike most ellipticals.

3.3. Summary of the Low-Redshift (Fossil) Evidence

The main observational constraints on the epoch of formation of the stellar populations of ETGs in the near universe can be summarized as follows:

The C-M, color- and FP relations for ETGs in clusters indicate that the bulk of stars in these galaxies formed at z 2-3.

The same relations for the field ETGs suggest that star formation in low density environments was delayed by ~ 1-2 Gyr.

The more massive galaxies appear to be enhanced in Mg relative to iron, which indicates that the duration of the star-formation phase decreases with increasing galaxy mass, having been shorter than ~ 1 Gyr in the most massive galaxies.

Interpretations of the Lick/IDS indices remains partly controversial, with either an age-metallicity anticorrelation, or an increase of both age and metallicity with increasing .

These trends are qualitatively illustrated in Figure 6, showing that the higher the final mass of the system, the sooner star formation starts and more promptly subsides, in an apparently "antihierarchical" fashion. A trend in which the stellar population age and metallicity are tightly correlated to the depth of the potential well (as measured by ) argues for star formation, metal enrichment, supernova feedback, merging, and violent relaxation having been all concomitant processes rather than having taken place sequentially.

The fossil evidence illustrated so far is in qualitative agreement with complementary evidence at low as well as high redshift, now relative to star-forming galaxies as opposed to quiescent ones. At low z, Gavazzi (1993) and Gavazzi, Pierini, & Boselli (1996) showed that in local (disk) galaxies the specific star-formation rate anticorrelates with galaxy mass, a trend that can well be extended to include fully quiescent ellipticals. On this basis, Gavazzi and collaborators emphasized that mass is the primary parameter controlling the star-formation history of galaxies, with a sharp transition at L_H 2 × 10¹⁰ L (corresponding to ~ 2 × 10¹⁰ M) between late-type, star-forming galaxies and mostly passive, early-type galaxies (Scodeggio et al. 2002). This transition mass has then been precisely located at ~ 3 × 10¹⁰ M with the thorough analysis of the SDSS database (Kauffmann et al. 2003). In parallel, high redshift observations have shown that the near-IR luminosity (i.e., mass) of galaxies undergoing rapid star formation has declined monotonically from z ~ 1 to the present, a trend for which Cowie et al. (1996) coined the term down-sizing. This is becoming a new paradigm for galaxy formation, as the anticorrelation of the specific star-formation rate with mass is now recognized to persist well beyond z ~ 2 (e.g., Juneau et al. 2005, Feulner et al. 2005).