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

4. ELLIPTICAL GALAXIES AT HIGH REDSHIFT

Perhaps the best way of breaking the age-metallicity degeneracy is by looking back in time, studying galaxies at higher and higher redshifts. In the 1990s this was attempted first with 4m-class telescopes, and later, with impressive success, with 8-10m-class telescopes and HST. Studies first focused on cluster ellipticals, and their extension to field galaxies followed with some delay. Thus, the evolution with redshift of various galaxy properties were thoroughly investigated, such as the C-M and Kormendy relations, the luminosity and mass functions, and the FP. Various ongoing surveys are designed to map the evolution with redshift and local environment of all these properties, along with the number density of these galaxies.

4.1. Cluster Ellipticals Up to z ~ 1

4.1.1 THE COLOR-MAGNITUDE RELATION   With the identification of clusters at higher and higher redshifts, from the mid-1990s it became possible to construct their C-M relation, hence to directly assess the rate of evolution of cluster ETGs. Pioneering studies showed a clearly recognizable red sequence in high-redshift clusters, and gave hints that the color evolution up to z ~ 1 was broadly consistent with pure passive evolution of the galaxies formed at high redshift (Dressler & Gunn 1990; Aragón-Salamanca et al. 1993; Rakos & Schombert 1995). Subsequent studies fully confirmed these early hints and provided accurate estimates for the formation redshift of the bulk of stars in cluster ellipticals. Thus, replicating the Bower, Lucey & Ellis (1992) procedure for a sample of morphologically-selected ETGs in clusters at z ~ 0.5, Ellis et al. (1997) were able to conclude that most of the star formation in ellipticals in dense clusters was completed 5-6 Gyr earlier than the cosmic time at which they are observed, i.e., at z 3. Extending these studies to clusters up to z ~ 0.9, Stanford, Eisenhardt, and Dickinson (1998) showed that pure passive evolution continues all the way to such higher redshift, while the dispersion of the C-M relation remains as small as it is in Virgo and Coma (see Figure 8). Thus, Stanford and colleagues concluded that cluster ellipticals formed the bulk of their stars at z 3, with the small color dispersion arguing for highly synchronized star-formation histories among galaxies within each cluster, and from one cluster to another. These conclusions were reinforced by several other investigators, e.g., Gladders et al. (1998), Kodama et al. (1998), Nelson et al. (2001), De Lucia et al. (2004), and by van Dokkum et al. (2000), who also cautioned about the "progenitor bias" (see below).

Figure 8. The color evolution of early-type galaxies in clusters out to z 0.9 (Stanford, Eisenhardt & Dickinson 1998, Dickinson 1997). The "blue" band is tuned for each cluster to approximately sample the rest frame U-band, whereas the K band is always in the observed frame. Top panel: the redshift evolution of the zero point of "blue"-K color-magnitude (C-M) relation relative to the Coma cluster. A purely passive evolution model is also shown, assuming a formation redshift zF = 5. Middle panel: the intrinsic color scatter, having removed the mean slope of the C-M relation in each cluster and the contribution of photometric errors. The intrinsic scatter of Coma galaxies is shown for reference. Bottom panel: the redshift evolution of the slope of the ("blue"-K) - K C-M relation, relative to the slope of the corresponding relation for galaxies in Coma.

The evolution of the C-M relation was then traced beyond z = 1 thanks to the discovery of higher redshift clusters, primarily by the Rosat Deep Cluster Survey (RDCS, Rosati et al. 1998). Using deep HST/ACS (Advanced Camera for Surveys) i- and z-band images, Blakeslee et al. (2003) found a tight red sequence for morphologically selected ETGs in a z = 1.24 cluster, implying typical ages of ~ 3 Gyr, and formation redshift z_F 3. This was further confirmed by its infrared C-M relation (Lidman et al. 2004). However, some clusters in the range 0.78 < z < 1.27 appear to have a larger color scatter than others, again with ellipticals in those with tight C-M relation having virtually completed their star formation at z 3 (Holden et al. 2004). Presently, the highest redshift clusters known to be dominated by old, massive ETGs are at z 1.4 (Stanford et al. 2005, Mullis et al. 2005).

The redshift evolution of the color of the red sequence in clusters proved to be a very powerful tool in disentangling ambiguities that are difficult to eliminate based only on z ~ 0 observations. From the color evolution of the red sequence in two Abell clusters at z ~ 0.2 and ~ 0.4, Kodama & Arimoto (1997) were able to break the age-metallicity degeneracy plaguing most of the global observables of local ellipticals. In principle, because colors depend both on age and metallicity, the slope of the C-M relation could equally well be reproduced with either age or metallicity increasing with increasing luminosity (or ). However, if age were the dominant effect, then the C-M relation would steepen with lookback time (redshift), as the color of the young galaxies would get more rapidly bluer compared to that of the old galaxies. Instead, the slope of the relation remains nearly the same (see Figure 8). Actually, the Kodama & Arimoto argument can be applied also to the color dispersion within a cluster, demonstrating that the tightness of the C-M (and FP) relation in low-z clusters cannot be due to a conspiracy of age and metallicity being anticorrelated (as advocated e.g., by Worthey, Trager & Faber 1995). If so, the color dispersion would rapidly increase with redshift, contrary to what is seen in clusters up to z ~ 1 (see Figure 8).

4.1.2 THE LUMINOSITY FUNCTION   A cross check of the high formation redshift of ETGs can be provided by looking at their luminosity in distant clusters. If ETGs evolve passively, following a pure luminosity evolution, then their luminosities should increase with increasing redshift by an amount that depends on the formation redshift and on the slope of the IMF.

Initial attempts to detect the expected brightening of the characteristic luminosity (M^*) of the luminosity function (LF) were inconclusive, as Barger et al. (1998) failed to detect any appreciable change between clusters at z = 0.31 and z = 0.56, possibly owing to the small redshift baseline. On the other hand, comparing the LF of z ~ 0 clusters to the LF of a sample of 8 clusters at 0.40 < z < 0.48 Barrientos & Lilly (2003) found a brightening of the characteristic luminosity M^* consistent with passive evolution and high formation redshift, also in agreement with the (U - V) color evolution of the red sequence. In a major cluster survey, De Propris et al. (1999) explored the evolution of the observed K-band LF in 38 clusters with 0.1 < z < 1, and compared the results to the Coma LF. With this much larger redshift baseline, De Propris et al. found the trend of K^* with redshift to be consistent with passive evolution and z_F 2. They pointed out the agreement with the results based on the color evolution of the red sequence galaxies, but emphasized that this behavior of the LF implies that "not only their stellar population formed at high redshift, but that the assembly of the galaxies themselves was largely complete by z ~ 1". Kodama & Bower (2003) and Toft, Soucail, & Hjorth (2003) came to the same conclusions by studying the K-band LF of (respectively two and one) clusters at z ~ 1. Breaking the z = 1 barrier, Toft et al. (2004) constructed a very deep K-band LF of a rich RDCS cluster at z = 1.237, and concluded that the most massive ellipticals that dominate the top end of the LF were already in place in this cluster. They compared the cluster K-band LF (corresponding to the rest-frame z-band LF) to the z-band LF of local clusters (Popesso et al. 2005) and derived a brightening by ~ 1.4 mag in the rest-frame z-band characteristic magnitude, indeed as expected from pure passive evolution. Toft and colleagues also found a substantial deficit of fainter ETGs, which could be seen as a manifestation of the down-sizing effect in a high redshift cluster, a hint of which was also noticed in other clusters at z ~ 0.8 (De Lucia et al. 2004).

However, a more complete study of 3 rich clusters at 1.1 z 1.3, including the z = 1.237 cluster studied by Toft et al. (2004), did not produce evidence of a down-sizing effect, down to at least 4 magnitudes below K^* (Strazzullo et al. 2006). This study confirmed the brightening of M_z^* and M_K^* by ~ 1.3 mag, consistent with passive evolution of a population that formed at z 2, and showed that the massive galaxies were already fully assembled at z ~ 1.2, at least in the central regions of the 3 clusters.

4.1.3 THE KORMENDY RELATION   An alternative way of detecting the expected brightening of old stellar populations at high redshift is by tracing the evolution of the Kormendy relation, which became relatively easy only after the full image quality of HST was restored. Thus, from HST data, the ETGs in a cluster at z = 0.41 were found brighter by M_K = 0.36 ± 0.14 mag (Pahre, Djorgovski, & de Carvalho 1996) or by 0.64 ± 0.3 mag (Barrientos, Schade, & Lopez-Cruz 1996) with respect to local galaxies, consistent (within such large errors) with passive evolution of an old, single-burst stellar population. Schade et al. (1996) using excellent-seeing CFHT (Canada-France-Hawaii Telescope) imaging data for 3 clusters at z = 0.23, 0.43 and 0.55 detected a progressive brightening in galaxy luminosity at a fixed effective radius that once more was estimated to be consistent with passive evolution and formation at high redshift. No differential evolution with respect to ETGs in the cluster surrounding fields was detected.

Turning to HST data, a systematic brightening in the Kormendy relation, again consistent with passive evolution and high formation redshift, was found by several other groups, eventually reaching redshifts ~ 1 (see Schade, Barrientos, & Lopez-Cruz 1997, Barger et al. 1999, Ziegler et al. 1999, Holden et al. 2005a, Pasquali et al. 2006).

4.1.4 THE FUNDAMENTAL PLANE   Besides the high spatial resolution, constructing the FP of high redshift cluster (and field) galaxies requires moderately high-resolution spectroscopy (to get ), hence a telescope with large collective area. With one exception, for a few years this was monopolized by the Keck Telescope, and FP studies of high-z ellipticals first flourished at this observatory. In a crescendo toward higher and higher redshifts, the FP was constructed for clusters at z = 0.39 (van Dokkum & Franx 1996), z = 0.58 (Kelson et al. 1997), z = 0.83 (van Dokkum et al. 1998, Wuyts et al. 2004), and finally at z = 1.25 and 1.27 (Holden et al. 2005b, van Dokkum & Stanford 2003). The early exception was the heroic study of two clusters at z = 0.375 using the 4m-class telescopes at ESO (NTT) and Calar Alto (Bender, Saglia, & Ziegler 1996, Bender, Ziegler, & Bruzual 1996, Bender et al. 1998).

The redshift evolution of the FP depends on a variety of factors. For passive evolution the FP shifts by amounts that depend on a combination of IMF slope, formation redshift, and cosmological parameters. A systematic trend of the IMF slope with galaxy mass would cause the FP to rotate with increasing redshift (Renzini & Ciotti 1993), as it would do for a similar trend in galaxy age. An age dispersion ( t) would cause the scatter perpendicular to the FP to increase with redshift, as, for fixed t, t / t increases for increasing redshift, i.e. decreasing galaxy age (t). Clearly, the behavior of the FP with redshift can give a wealth of precious information on the formation of ellipticals, their stellar populations, and to some extent on cosmology also (Bender et al. 1998).

All the quoted FP studies of high-z clusters conclude that the FP actually shifts nearly parallel to itself by an amount that increases with redshift and is consistent with the passive evolution of stellar populations that formed at high redshifts. The FP shifts imply a decrease of the rest-frame M / L ratio log M / L_B -0.46z (van Dokkum & Stanford 2003), but - as emphasized above - the formation redshift one can derive from it depends on both cosmology and the IMF. Figure 9 illustrates the dependence on the adopted cosmology. Note that for the "old standard" cosmology (_M = 1) galaxies would be older than the universe and therefore the observed FP evolution can effectively rule out this option (see also Bender et al. 1998).

Figure 9. The data points show the redshift evolution of the M / LB ratio of cluster elliptical galaxies as inferred from the shifts of the fundamental plane. The lines refer to the evolution of the M / LB ratio for stellar populations with a Salpeter initial mass function (IMF) (s = 2.35) and formation redshifts as indicated in the left -and middle- panel. The comparison is made for three different cosmologies. The dotted line in the middle panel shows a model with zF=∞ and a steep IMF (s = 3.35). From van Dokkum et al. 1998).

Figure 10 shows the effect of the IMF slope and formation redshift on the expected evolution of the M / L_B ratio. The redshift range z = 0 to 1.5 only probes the IMF between ~ 1 and ~ 1.4 M, which correspond to the masses at the main sequence turnoff M_TO of oldest populations at z = 0 (age ~ 13 Gyr, M_TO M) and of those at z = 1.5 (age ~ 4.5 Gyr and M_TO 1.4 M). Together, Figure 9 and 10 illustrate the formation redshift/IMF/cosmology degeneracies in the FP diagnostics. However, the cosmological parameters are now fixed by other observational evidences, whereas the formation redshift as determined by the color evolution of the cluster red sequence is independent of the IMF slope. In summary, in the frame of the current concordance cosmology, when combining the color and FP evolution of cluster ellipticals one can conclude that the best evidence indicates a formation redshift z_F ~ 3 and a Salpeter IMF slope in the pertinent stellar mass range (1 < M < 1.4 M).

Figure 10. The evolution with redshift of the M* / LB ratio of simple stellar populations of solar metallicity and various initial mass function slopes (dN M-s dM) and formation redshifts, as indicated. The curves are normalized to their value at z = 0. Concordance cosmology (m = 0.3, = 0.7, Ho = 70) is adopted. The data points (from van Dokkum & Stanford 2003) refer to the shifts in the fundamental plane locations for clusters at various redshifts. Note that for such high formation redshifts the stellar mass at the main sequence turnoff is ~ 1.4 M at z = 1.5 and ~ 1.0 M at z = 0, as indicated by the arrows. Adapted from Renzini (2005).

The fact that the cluster FP does not appreciably rotate with increasing redshift is documented down to ~ 100 km s^-1 for clusters out to z ~ 0.3-0.4 (lookback time ~ 4 Gyr) (Kelson et al. 2000, van Dokkum & Franx 1996) and down to ~ 150 km s^-1 out to z ~ 0.8 (lookback time ~ 7 Gyr) (Wuyts et al. 2004). If the large age trends with derived in some of the studies using the Lick/IDS indices were real (see Section 3.1), this would result in a very large rotation of the FP in these clusters. For example, if age were to increase along the FP from 5.5 Gyr to 13 Gyr (at z = 0, and for = 100 and 320 km s^-1, respectively, see Nelan et al. 2005), then at a lookback time of 4 Gyr (z ~ 0.4) the younger population would have brightened by M_B ~ 1.33 mag, and the older one only by ~ 0.46 mag (using models from Maraston 2005), which results in a FP rotation of ~ 0.9 mag in surface brightness. Alternatively, the much shallower age- relation derived by Thomas et al. (2005) implies an age increase from ~ 9.5 Gyr to ~ 11.5 Gyr for increasing from 180 to 350 km s^-1, which implies a rotation of the FP by ~ 0.36 mag in surface brightness by z = 0.8, which is still consistent with the hint that in fact there may be a small FP rotation in a cluster at this redshift (Wuyts et al. 2004).

The scatter about the FP of clusters also remains virtually unchanged with increasing redshift, however some of the claimed age-metallicity anticorrelations derived from the Lick/IDS indices would result in a dramatic increase of the scatter with redshift, causing the FP itself to rapidly blur away.

4.1.5 THE LINE INDICES   The intermediate resolution spectra used for constructing the FP of distant cluster ETGs, were also used to measure age-sensitive line indices that can provide further constraints on the formation epoch. Thus, Bender, Ziegler, & Bruzual (1996) and Ziegler & Bender (1997) measured the Mgb index of 16 ETGs in their two clusters at z = 0.375, and found that the Mgb - relation was shifted toward lower values of the index. From such differences in Mgb index, Ziegler & Bender inferred that the age of the z = 0.375 galaxies is about two thirds of that of ETGs in Coma and Virgo. Therefore, t(z = 0) - t_F = 1.5 × [t(z = 0.375) - t_F], where t is the cosmic time and t_F the cosmic time when the local and distant cluster ETGs formed (which is assumed to be the same). Adopting the t - z relation for the concordance cosmology, one derives t_F ~ 1.7 Gyr, corresponding to z_F 3.

From the strength of the Balmer absorption lines (H and H) as age indicators, Kelson et al. (2001) used data for several clusters up to z = 0.83 and were able to set a lower limit to the formation redshift z_F 2.5, consistent with the above result from the Mgb index.

In summary, the study of the stellar populations in ETGs belonging to distant clusters up to z ~ 1.3 have unambiguously shown that these objects have evolved passively from at least z ~ 2-3. This came from the color, line strength, and luminosity evolution. Moreover, the brightest cluster members at z ~ 1-1.3 and the characteristic luminosity of the LF appear to be brighter than their local counterpart by an amount that is fully consistent with pure passive evolution, indicating that these galaxies were already fully assembled at this high redshift. This may not have been the case for less massive galaxies, as their counts may be affected by incompleteness.

4.1.6 CAVEATS   Although it is well established that ETGs in distant clusters are progenitors to their local analogs and formed at high redshift, some caveats are nevertheless in order. First, as frequently emphasized, the evidence summarized above only proves that at least some cluster galaxies evolved passively from z 1 to the present, but other local ETGs may have z ~ 1 progenitors that would not qualify as ETGs at that redshift, either morphologically or photometrically. This "progenitor bias" (e.g., van Dokkum & Franx 1996) would therefore prevent us from identifying all the z ~ 1 progenitors of local cluster ETGs, some of which may well be still star forming. Second, the slope of the FP is progressively less accurately constrained in higher and higher redshift clusters, because the central velocity dispersion has been measured only for very few cluster members (generally the brightest ones). Third, it is always worth recalling that all luminosity-weighted ages tend to be biased toward lower values by even minor late episodes of star formation. Last, stellar population dating alone only shows when stars were formed, not when the galaxy itself was assembled and reached its observed mass.

4.2. Field versus Cluster Ellipticals up to z ~ 1

In the local universe field ellipticals show small, yet detectable differences compared to their cluster counterpart, being possibly ~ 1 Gyr younger, on average. This t difference, if real, should magnify in relative terms and become more readily apparent when moving to high redshift ( t / t is increasing). Using the color evolution of the red sequence and the shift of the FP with redshift, progress in investigating high-z ETGs in the field has been dramatic in recent years, along with the cluster versus field comparison.

Schade et al. (1999) selected ETGs by morphology from the Canada-France Redshift Survey (CFRS, Lilly et al. 1995a, b) and Low Dispersion Survey Spectrograph (LDSS) redshift survey (Ellis et al. 1996), and constructed the rest-frame (U - V) C-M relation for field ETGs in the 0.2 < z < 1.0 range. They found that the C-M relation becomes progressively bluer with redshift, with (U - V) -0.68 ± 0.11 at z = 0.92 with respect to the relation in Coma, accompanied by a brightening by ~ 1 mag in the rest-frame B band, as derived from the Kormendy relation. To be consistent with the color evolution, this brightening should have been much larger than observed if the color evolution were due entirely to the passive evolution of stellar populations formed at high z. Thus, Schade and colleagues reconciled color and luminosity evolution by invoking a residual amount of star formation (adding only ~ 2.5% of the stellar mass from z = 1 to the present), yet enough to produce the observed fast color evolution. Support for such an interpretation comes from about one third of the galaxies exhibiting weak [OII] emission, which indicates that low-level star formation is indeed fairly widespread. The rate of luminosity evolution was found to be identical to that of cluster ellipticals at the same redshifts, hence no major environmental effect was detected besides the mentioned low level of star formation and a color dispersion slightly broader than in clusters at the same redshift.

With COMBO-17, the major imaging survey project undertaken with the ESO/MPG 2.2m telescope, Wolf et al. (2003) secured deep optical imaging in 17 broad and intermediate bands over a total 0.78 square degree area, from which Bell et al. (2004b) derived photometric redshifts accurate to within z ~ 0.03. The bimodality of the C-M relation, so evident at z ~ 0 (e.g., Baldry et al. 2004), clearly persists all the way to z ~ 1.1 in the COMBO-17 data, and this allowed Bell and colleagues to isolate ~ 5,000 "red sequence" ETGs down to R < 24. As mentioned above, ~ 85% of such color-selected galaxies appear also morphologically early-type on the ACS images of the GEMS (Galaxy Evolution from Morphology and SED) survey (Rix et al. 2004, Bell et al. 2004a). The rest-frame (U - V) color of ETGs in the COMBO-17 survey evolves by a much smaller amount than that reported by Schade et al. (1999) for the morphologically-selected ETGs, i.e., by only ~ 0.4 mag between z = 0 and 1, as expected for an old stellar population that formed at high redshift (z_F 2). This color evolution is also in agreement with the ~ 1.3 mag brightening of the characteristic luminosity M_B^* in the Schechter fit to the observed LF. Thus, when comparing only the color and M_B^* evolution, the field ETGs in the COMBO-17 sample seem to evolve in much the same fashion as their cluster counterparts. Using COMBO-17 data and GEMS HST/ACS imaging, McIntosh et al. (2005) studied a sample of 728 morphology- and color-selected ETGs, finding that up to z ~ 1 the Kormendy relation evolves in a manner that is consistent with the pure passive evolution of ancient stellar populations.

From deep Subaru/Suprime-Cam imaging over a 1.2 deg² field [covered by the Subaru-XMM Deep Survey (SXDS)], Kodama et al. (2004) selected ETGs for having the R - z and i - z colors in a narrow range as expected for passively evolving galaxies in the range 0.9 z 1.1, and the sampled population included both field and cluster ETGs. They found a deficit of red galaxies in the C-M sequence ~ 2 mag fainter than the characteristic magnitude (corresponding to stellar masses below ~ 10¹⁰ M). Less massive galaxies appear to be still actively star-forming, while above ~ 8 × 10¹⁰ M galaxies are predominantly passively evolving. This was interpreted as evidence for down-sizing in galaxy formation (à la Cowie et al. 1996), with massive galaxies having experienced most of their star formation at early times and being passive by z ~ 1, and many among the less massive galaxies experience extended star formation histories.

In a comprehensive study also based on deep, wide-field imaging with the Suprime-Cam at the Subaru Telescope, Tanaka et al. (2005) obtained photometric redshifts based on four or five optical bands and constructed the C-M relations for the two clusters (at z = 0.55 and 0.83) included in the field, and for their extended environment. Tanaka and colleagues further distinguished galaxies in the cluster environment as belonging either to recognized "groups", or otherwise to the "field". The results are shown in Figure 11, where the various plots allow one to visually explore trends with redshift for given environment, or with environment for given redshift. The red sequence appears already in place in the "field" in the highest redshift sample, but no clear color bimodality is apparent. (Note that COMBO-17 does find bimodality at this redshift, possibly due to its photometric redshifts based on many more bands being more accurate.) The color bimodality is instead clearly recognizable in the z = 0.55 "field" sample. At higher environmental densities (labelled "group") the C-M relation of the red sequence is clearly recognizable already in the z = 0.83 sample, and even more so in the "cluster" sample. Tanaka and colleagues argue that the bright and the faint ends of the red sequence are populated at a different pace in all three environments; the more massive red galaxies are assembled first, i.e., the C-M relation grows from the bright end to the faint end in all three environments (not in the opposite way, as one may naively expect in a hierarchical scenario), which is interpreted in terms of down-sizing. Note also that the faint end appears to be well in place in "clusters" at z = 0 while virtually still lacking in the field (see also Popesso et al. 2005). Using HST/WFPC2 imaging over a ~ 30 arcmin² field including the same z = 0.83 cluster, Koo et al. (2005) were able to measure the rest-frame (U - B) colors of the sole bulge component of 92 galaxies with M_B < -19.5, part in the cluster itself, part in the surrounding field. Their very red color does not show any environmental dependence, suggesting similarly old ages and high formation redshifts.

Figure 11. The C-M relations of two high redshift clusters and their surrounding fields at progressive lower density (labelled "group" and "field") are compared to their local counterparts from the Sloan Digital Sky Survey. The red line marks the adopted separation between the red sequence galaxies, and the blue, star-forming galaxies. In the cluster panels the blue dashed lines show the approximate location of galaxies with stellar mass of 1010 (right) and 1011 M (left). From Tanaka et al. (2005).

The public delivery of the Hubble Deep Field data (HDF, Williams et al. 1996) spurred several studies of field ETGs. Thus, Fasano et al. (1998) applied the Kormendy relation to a sample of morphologically selected ETGs in HDF-North and estimated an increase of the surface brightness at a fixed effective radius that was consistent with a high formation redshift (z_F ~ 5), according to the galaxy models by Bressan, Chiosi & Fagotto (1994) and Tantalo et al. (1996).

Although the LF and the C-M and Kormendy relations had already given useful indications on the analogies and differences between cluster and field ETGs, major progress came with the study of the differential evolution (field versus cluster) of the FP with redshift. In early studies, no field/cluster difference had clearly emerged at z = 0.3 (Treu et al. 1999), z ~ 0.4 (Treu et al. 2001), z = 0.55 (van Dokkum et al 2001), and z = 0.66 (Treu et al. 2002). But already at these modest redshifts there were hints that the brightest, most massive ETGs in the field closely follow the FP evolution of their cluster counterparts, while less massive ETGs (especially S0's) appear to evolve slightly faster, and hence look younger. This was more accurately quantified for morphologically-selected ETGs up to z ~ 1 in the HDF-North by van Dokkum & Ellis (2003), showing a field versus cluster difference ln M / L_B = -0.14 ± 0.13 in the FP. This implies that field ETGs are on average younger by only 16% ± 15% at < z> = 0.88. Van Dokkum & Ellis also inferred that the bulk of stellar mass in the observed ETGs must have formed at z 2 even in the field, with only minor star formation at lower redshifts. Then, moving to the wider GOODS (Great Observatories Origin Deep Survey)-South field (Giavalisco et al. 2004a), van der Wel et al. (2004, 2005a) constructed the FP for a total of 33 color and morphology-selected ETGs at 0.60 < z < 1.15, using intermediate resolution spectra taken at the ESO Very Large Telescope (VLT). They also found the most massive galaxies (M_* > 2 × 10¹¹ M) to behave much like their cluster analogs at the same redshifts, while less massive galaxies appeared to be substantially younger. Moreover, all these studies noted the higher proportion of weak [OII] emitters among the field ETGs (~ 20%) compared to their cluster counterparts, as well as the higher proportion of galaxies with strong Balmer lines (the K+A Galaxies).

The main limitations of all these early studies was in the small number of objects observed at each redshift, which must go a long way toward accounting for occasional discrepancies in the results. In a major effort to overcome this limitation, Treu et al. (2005a, b) obtained high-resolution spectra at the Keck telescope for 163 morphologically-selected ETGs in the GOODS-North field, which were distributed over the redshift range 0.2 < z < 1.2. The main results of this study are displayed in Figure 12, showing that the most massive ellipticals in the field do not differ appreciably from their cluster analogs in having luminosity-weighted ages implying z_F 3. However, the lower the mass the larger the dispersion in the M / L_B ratio, with a definite trend toward lower values with decreasing mass, implying lower and lower formation redshift. This demonstrates that completion of star formation in field galaxies proceeds from the most massive to the less massive ones, as is indeed expected from the down-sizing effect (Cowie et al. 1996, Kodama et al. 2004) and is consistent with the scenario shown in Figure 6. This systematic trend in the M / L ratio with galaxy mass results in a "rotation" of the FP with increasing redshift, as less massive galaxies evolve faster in luminosity compared to the more massive, older ones (but one should beware of possible Malmquist bias). This result confirms early hints for a modest rotation of the FP of field ETGs with redshift, as also does a study of the FP of 15 ETGs at 0.9 z 1.3 by di Serego Alighieri et al. (2005), a sample drawn from the K20 survey (Cimatti et al. 2002b). Figure 13 shows the FP for the combined Treu et al. (2005b) and di Serego Alighieri et al. (2005) samples of ETGs with < z> = 1.1, where the rotation with respect to the Coma FP is apparent. Note that a similar FP rotation in two clusters at z = 0.83 and 0.89 has been recently unambiguously detected by Jørgensen et al. (2006), having extended the measurements below ~ 100 km s^-1. Somewhat at variance with the FP studies reported above was the Deep Groth Strip Survey result (Gebhardt et al. 2003), in which no difference in the slope was found up to z ~ 1 compared to the local FP (hence no down-sizing), which was coupled to a much faster luminosity evolution compared to all other results. Treu et al. (2005b) discuss the possible origins of the discrepancy, and attribute it to a combination of selection bias, small number statistics, and relatively low S/N spectra.

Figure 12. The offset log M / LB from the fundamental plane of cluster ellipticals at z ~ 0 for the early-type galaxies in the GOODS-North field (from Treu et al. 2005b). Different symbols are used for early-type (E+S0) galaxies and bulges in late-type (Sa+Sb) galaxies, as well as for the various stellar mass ranges as indicated. The dotted lines are labelled by various formation redshifts.

Figure 13. An edge-on view of the fundamental plane for field ETGs at z ~ 1.1 from di Serego Alighieri et al. (2005, red squares and black circles) and Treu et al. (2005b, green squares). Open triangles refer to the Coma ellipticals from Jørgensen, Franx, & Kjærgaard (1995) and the dashed line is a best fit to the data. The dotted line is shifted parallel to the dashed line by an amount in surface brightness corresponding to the observed shift of the fundamental plane of galaxy clusters (i.e., M / LB = -0.46z, van Dokkum & Stanford 2003). The effective surface brightness in the B band (µeB) is in magnitudes per arcsec2.

From population synthesis models one expects the rate of evolution of the M / L ratio to be slower at longer wavelengths compared to the B band, because it is less affected by the main sequence turnoff moving to cooler temperatures with increasing age. This expectation was qualitatively confirmed by van der Wel et al. (2005b), who found ln(M / L_B) = -(1.46 ± 0.09)z and ln (M / L_K) = -(1.18 ± 0.1)z, which appears to be in agreement with the prediction of some models (Maraston 2005), but not of others (Bruzual & Charlot 2003), possibly owing to the different treatment of the AGB contribution.

Instead of measuring central velocity dispersions directly, Kochanek et al. (2000) and Rusing and Kochanek (2005) estimated them from the lens geometry for a sample of (field) lensing ETGs at 0.2 z 1. The resulting FP shifts appear to be similar to those of cluster ETGs, although with more scatter, and indicate <z_F> 1.5 for the bulk of the stars in the lensing galaxies.

In summary, like at low redshifts, also at z ~ 1 there appear to be small detectable differences between ETGs in high- and low-density regions, but such differences are more evident for faint/low-mass galaxies than for the bright ones.

4.3. Ellipticals Beyond z ~ 1.3

Up to z ~ 1.3 the strongest features in the optical spectrum of ETGs are the CaII H&K lines and the 4000 Å break. But at higher redshifts these features first become contaminated by OH atmospheric lines, and then move to the near-IR, out of reach of CCD detectors. The lack of efficient near-IR multi-object spectrographs (even in just the J band) has greatly delayed the mapping of the ETG population beyond z ~ 1.3. Thus, for almost a decade the most distant spectroscopically confirmed old spheroid was an object at z = 1.55 selected for being a radiogalaxy (Dunlop et al. 1996, Spinrad et al. 1997). The spectral features that made the identification possible included a set of FeII, MgII and MgI lines in the rest-frame near-UV, in the range of ~ 2580-2850 Å, which is typical of F-type stars. The UV Fe-Mg feature offers at once both the opportunity to measure the redshift, and to age-date the galaxy, because it appears only in populations that have been passively evolving since at least a few 10⁸ years. It has been also used to age-date local ETGs (e.g., Buson et al. 2004). Thus, using this feature, Spinrad and colleagues inferred an age of ~ 3.5 Gyr, implying z_F > 5 (even in modern cosmology).

This record for the highest redshift ETG was eventually broken by Glazebrook et al. (2004) and Cimatti et al. (2004), using the same features in the rest-frame UV (see Figure 14). They reported the discovery of, respectively, five passively evolving galaxies at 1.57 z 1.85, and four other such objects at 1.6 z 1.9. All being brighter than K = 20, these galaxies are quite massive (M 10¹¹ M), and hence would rank among the most massive galaxies even in the local universe. This suggests that they were (almost) fully assembled already at this early epoch, and having been passive since at least ~ 1.1 Gyr had to form at redshift 2.7. The four objects found by Cimatti and colleagues are included in the GOODS-South field, and the GOODS deep HST/ACS imaging showed that two objects are definitely elliptical galaxies, and the two others are likely to be S0's.

Figure 14. The rest-frame coadded spectrum of the four passively evolving galaxies at 1.6 < z < 1.9 with the identification of the main spectral features (blue spectrum). The spectrum from Bruzual & Charlot (2003) for a 1-Gyr-old SSP (simple stellar population) model of solar metallicity is also shown (red spectrum). (From Cimatti et al. 2004).

Though breaking the old redshift record was certainly an exciting result, perhaps far more important was the discovery that the surface density of z > 1.5 ETGs is indeed much higher than one would have expected from just the single object found by Dunlop et al. (1996). Indeed, this galaxy was selected from a catalog of radiogalaxies covering a major fraction of the whole sky, whereas the nine galaxies in the Cimatti and colleagues and Glazebrook and colleagues samples come from a combined area of only 62 arcmin².

Further identifications of very high redshift ETGs used this UV feature: McCarthy et al. (2004) reported the discovery of 20 ETGs with 1.3 z 2.15 and K < 20 as part of the Gemini Deep Deep Survey (GDDS) (including the 5 galaxies from Glazebrook et al. 2004). Within the ~ 11 arcmin² area of the Hubble Ultra Deep Field (HUDF, Beckwith et al. in preparation), Daddi et al. (2005b) identified 7 ETGs with 1.39 z 2.5 using their HST/ACS grism spectra. For all these objects the stellar mass derived from the spectral energy distribution (SED, typically extending from the B to the K band) is in excess of ~ 10¹¹ M. Over the same field, Yan et al. (2004) identify 17 objects with photometric redshifts between 1.6 and 2.9, whose SED can be best fit by a dominant ~ 2 Gyr old stellar population, superimposed to a low level of ongoing star formation.

Rather than digging deep into small fields, Saracco et al. (2005) searched for bright high-z ETGs over the ~ 160 arcmin² field of the MUNICS survey (Drory et al. 2001), and selected objects with R - K > 5.3 and K < 18.5 for spectroscopic follow up with the low-resolution, near-IR spectrograph on the TNG 3.5m telescope. They identified 7 ETGs at 1.3 z 1.7, all with mass well in excess of 10¹¹ M.

Altogether, these are virtually all the spectroscopically confirmed ETGs at z > 1.3 known to date. Given the long integration time needed to get spectroscopic redshifts, it became clear that an effective criterion was indispensable to select high-z ETG candidates. To this end, Daddi et al. (2004) introduced a robust criterion based on the B - z and z - K colors, that very effectively selects galaxies at 1.4 z 2.5 (the so-called BzKs), and among them separates the star forming BzKs with BzK (z - K)_AB - (B - z)_AB -0.2, from the passive ones, with BzK < -0.2 and (z - K)_AB > 2.5. The criterion is primarily an empirical one, based on the spectroscopic redshifts from the K20 survey (Cimatti et al. 2002b) and other publicly available data sets. However, Daddi and colleagues showed that synthetic stellar populations of the two kinds (i.e., star forming and passive) do indeed occupy the corresponding areas in this plot, when redshifted to 1.4 < z < 2.5. An application of the method to a 320 arcmin² field is shown in Figure 15 (Kong et al. 2006). In this latter study, it is estimated that the space density of massive and passive BzKs (with K < 20, stellar mass 10¹¹ M and <z> 1.7) is 20 ± 7% that of z = 0 ETGs within the same mass limit. Then there appears to be a sharp drop of passive BzKs beyond z = 2, which in part may be due to the available B-band data being not deep enough (Reddy et al. 2005)

Figure 15. The BzK plot introduced by Daddi et al. (2004) is here shown for objects to a limiting magnitude KVega = 20 from a 320 arcmin2 field (from Kong et al. 2006). Black dots refer to galaxies at z < ~ 1.4, blue dots refer to starforming galaxies at ~ 1.4 < z < ~ 2.5, orange dots refer to passively evolving galaxies at ~ 1.4 < z < ~ 2.5. Green dots are Galactic stars in the same field and purple stars are local stellar standards.

Further candidate ETGs with masses up to a few 10¹¹ M have been identified at even higher redshifts, such as six objects within the HUDF at a (photometric) redshift > 2.8 (Chen & Marzke 2004), where both the redshift and the old age are inferred from the observed break between the J (F110W) and the H (F160W) band being interpreted as the 4000 Å break. One of these objects is undetected in the deep GOODS optical data, but is prominent in the GOODS Spitzer/IRAC 3.5-µm images (M. Dickinson et al., in preparation). Thus its SED shows two breaks, one between the z and the J band, and one between the K and the 3.5 µm IRAC band. Identifying them respectively with the Lyman and Balmer breaks, the object would be placed at z ~ 6.5, it would be passively evolving with z_F > 9, and would have the uncomfortably large mass of a few 10¹¹ M (Mobasher et al. 2005). Lower redshift alternatives give much worse fits to the data, whereas the use of models with strong AGB contribution (Maraston 2005) results in a somewhat less extreme mass and formation redshift.

4.4. Evolution of the Number Density of ETGs to z ~ 1 and Beyond

The studies illustrated so far have shown that ETGs exist up to z ~ 1, both in clusters and in the field, and are dominated by old stellar populations that formed at z 2-3. Moreover, a handful of ETGs has also been identified (over small fields) well beyond z ~ 1. Some of these ETGs appear to be as massive as the most massive ETGs in the local universe, demonstrating that at least some very massive ETGs are already fully assembled at z 1. However, the expectation is for the number of ETGs to start dropping at some redshift, when indeed entering into the star formation phase of these galaxies, or when they were not fully assembled yet. Therefore, what remained to be mapped by direct observations was the evolution with redshift of the comoving number density of ETGs, and to do so as a function of mass and environment while covering wide enough areas of the sky in order to reduce the bias from cosmic variance. Only in this way one could really overcome the so-called progenitor bias. Because deep and wide surveys require so much telescope time, progress has been slow. Cosmic variance may still be responsible for the apparent discrepancies between galaxy counts from different surveys, but occasionally the interpretation itself of the counts may be prone to ambiguities.

One of the main results of the CFRS was that the number density of red galaxies shows very little evolution over the redshift range 0 < z < 1 (Lilly et al. 1995b). Following this study, in an attempt to map the number evolution of ETGs all the way to z ~ 1 , Kauffmann, Charlot, & White (1996) extracted 90 color-selected ETGs without [OII] emission from the CFRS redshift catalog. They used a V / V_max test and concluded that at z = 1 only ~ 1/3 of bright ETGs had already assembled or had the colors expected for old, pure passively evolving galaxies. However, Im et al. (1996) identified ~ 360 ETGs morphologically selected on archival HST images, and also conducted the V / V_max test using photometric redshifts, finding no appreciable number density evolution up to z ~ 1 and a brightening consistent with passive evolution. The V / V_max test was repeated - again using the CFRS sample - by Totani & Yoshii (1998) who concluded that there was no evolution in the number density up to z ~ 0.8, and ascribed the apparent drop at z > 0.8 to a color selection bias. No evolution of the space density of morphologically-selected ETGs up to z ~ 1 was found by Schade et al. (1999) too, who used the HST imaging of the CFRS and LDSS redshift surveys.

Several attempts to trace the evolution of the number density of morphologically-selected ETGs to the highest possible redshifts were made using HDF data. In some of these studies very little, if any, change in the space density of ETGs was found up to z ~ 1 (e.g., Driver et al. 1998, Franceschini et al. 1998, Im et al. 1999), or up to even higher redshifts when combining the HDF optical data with very deep near-IR data (Benitez et al 1999, Broadhurst & Bouwens 2000). These latter authors emphasized that without deep near-IR data many high-z ETGs are bound to remain undetected, and that spectroscopic incompleteness beyond z ~ 0.8 is partly responsible for some of the previous discrepancies. Beyond z ~ 1 a drop in the space density of ETGs was detected in several studies including the HDF (Zepf 1997, Franceschini et al. 1998, Barger et al. 1999, Rodighiero, Franceschini, & Fasano 2001, Stanford et al. 2004). In particular, Stanford and colleagues applied the V / V_max test to a sample of 34 ETGs from the HDF-North that includes deep NICMOS imaging in the H band (F160W), and concluded for a real drop at z > 1, but advocated the necessity to explore much wider fields in order to improve the statistics and cope with cosmic variance. Finally, the HDF-South optical data were complemented by ultra-deep JHK imaging at the VLT (the FIRES survey, Labbé et al. 2003, Franx et al. 2003), revealing a population of near-IR galaxies with very red colors (J - K > 2.3), called distant red galaxies (DRG), a fraction of which may be ETGs at very high redshifts (see below). Compared to HDF-North, its Southern equivalent appeared to be much richer in very red galaxies, e.g., including 7 objects with (V - H)_AB > 3 and H_AB < 25 while HDF-North has only one. Clearly, exploring much wider fields compared to HDF's ~ 5 arcmin² field was imperative in order to make any significant progress.

Passive ETGs formed at very high redshift (e.g., z > 3) would indeed have very red colors at z 1, and thus they should be found among the so-called extremely red objects (ERO), a class defined for having R - K > 5 (or similar color cut), and whose characteristics and relation to ETGs have been thoroughly reviewed by McCarthy (2004). Using a much shallower sample than that from the HDF, but one that covers an area ~ 140 times wider than it, Daddi et al. (2000) and Firth et al. (2002) showed that EROs are much more abundant than previously found in smaller fields and are much more strongly clustered than generic galaxies to the same limiting magnitude K ~ 19. This made them likely candidates for high-z ETGs, and assuming that ~ 70% of EROs are indeed ETGs at z > 1, Daddi, Cimatti & Renzini (2000) concluded that most field ellipticals were fully assembled by z ~ 1. However, Cimatti et al. (2002a) actually found that out of the 30 EROs with secure redshifts and K < 19.2, only 50% are passively evolving objects and these are distributed in the redshift interval 0.8 z 1.3, while the other 50% is made by highly-reddened, actively star-forming galaxies. Interestingly, precisely 50% among a sample of 129 EROs with K < 20.2 have been detected at 24 µm with Spitzer/MIPS, reinforcing the conclusion that up to one half of EROs are likely to be passive precursors to ETGs (Yan et al. 2004). In fact, the fraction of passive EROs decreases to ~ 35% on a spectroscopic complete sample to K = 20 (Cimatti et al. 2003). Nevertheless, the number density of passive EROs appeared to be broadly consistent with no density evolution of ETGs up to z ~ 1, or a modest decrease.

With the COMBO-17 survey Bell et al. (2004b) went a long way toward coping with cosmic variance. With their 5,000 color-selected ETGs up to photometric redshift z ~ 1.1, Bell and colleagues were able to construct their rest-frame B-band luminosity functions in nine redshift bins (0.2 < z < 1.1), and derived the best fit Schechter parameters for them using a fixed value of the faint-end slope, = -0.6. They found that the characteristic luminosity M_B^* brightens by ~ 1.0 mag between z = 0.25 and 1.05, consistent with passive evolution within the errors, and also with the brightening expected from the FP shift ( log M / L_B = -0.46 z), which predicts ~ 0.9 mag. At the same time, the normalization factor ^* drops by a factor of ~ 4, but much of the drop is in the highest redshift bins which may be affected by incompleteness. More robust than either ^* or L^* separately, is their product ^* L_B^* which is proportional to the B-band luminosity density, and this is found to be nearly constant up to z ~ 0.8. This is at variance with a pure passive-evolution scenario, that would have predicted an increase by a factor of ~ 2. Thus, the color of the COMBO-17 red sequence follows nicely the expectation from passive evolution (cf. section 4.2), but the number density of red sequence galaxies does not, and Bell and colleagues concluded that the stellar mass in red sequence galaxies has nearly doubled since z ~ 1.

In a major observational effort at the Keck telescope, Faber et al. (2006, DEEP2 project) secured spectroscopic redshifts for ~ 11,000 galaxies with R < 24.1, and also reanalyzed the COMBO-17 data, finding separate best fit Schechter parameters in various redshift bins up to z ~ 1.1. Faber and colleagues emphasize that ^* and M^* are partly degenerate in these fits, for which the faint-end slope was fixed at = -0.5. Thus, between z = 0.3 and 1.1, M^* brightens by ~ 0.47 mag and ^* drops by a factor of ~ 2.5 for the DEEP2 data, and respectively up by ~ 0.95 mag and down by a factor of ~ 4 for the COMBO-17 data. Once more, much of the ^* drops are confined to the last redshift bin, and emphasis is placed on both DEEP2 and COMBO-17 confirming that the B-band luminosity density is nearly constant up to z ~ 0.8, along with the implication that the mass density in ETGs has increased, presumably by a factor of ~ 2, as estimated by Bell et al. (2004b). Extending the analysis from z = 0.3 to z = 0 (using SDSS data), Faber an colleagues find M^* ~ 1.3 mag and a drop in ^* by a factor of ~ 4 between z = 0 and 1.1, but caution that much of these changes occur between the z ~ 0 survey and their first bin (at z = 0.3) at one end, and in the last redshift bin at the other end, where the data are said to be the weakest.

Both in COMBO-17 and DEEP2 the shape of the Schechter function for the red-sequence galaxies is assumed constant with redshift. As such, by construction this assumption virtually excludes down-sizing, for which ubiquitous indications have emerged both at low as well as high redshift. Indeed, as alluded in Kong et al. (2006) and documented by Cimatti, Daddi & Renzini (2006), COMBO-17 and DEEP2 results can also be read in a different way. Figure 16 shows the evolution of the rest frame B-band LF from COMBO-17, with the continuous line being the local LF for red-sequence galaxies from Baldry et al. (2004). The local LF has been shifted according to the brightening derived from the empirical FP shift with redshift for cluster ETGs (i.e., by M_B = -1.15 z, coming from (M / L_B) = -0.46z (van Dokkum & Stanford 2003), taken as the empirical template for passive evolution. From Figure 16 it is apparent that the brightest part of the LF is fully consistent with pure passive evolution of the most massive galaxies, whereas the fainter part of the LF (below ~ L^*) is progressively depopulated with increasing redshift, an effect that only in minor part could be attributed to incompleteness. Therefore, from these data it appears that virtually all the most massive ETGs have already joined the red sequence by z ~ 1, whereas less massive galaxies join it later. This is what one would expect from the down-sizing scenario, as exemplified e.g., in Figure 6, as if down-sizing was not limited to stellar ages (stars in massive galaxies are older), but it would work for the assembly itself, with massive galaxies being the first to be assembled to their full size. Being more directly connected to the evolution of dark matter halos, an apparent antihierarchical assembly of galaxies may provide a more fundamental test of the CDM scenario than the mere down-sizing in star formation.

Figure 16. The evolution of the rest-frame B-band luminosity function of early-type (red sequence) galaxies from COMBO-17 (Bell et al. 2004) is compared to to the local luminosity function (solid line) from Baldry et al. (2004). The local LF has been shifted in magnitude as indicated in each panel, which corresponds to pure passive evolution as empirically derived from the FP shift of cluster of galaxies ( M / LB = -0.46z, van Dokkum & Stanford 2003). Note that there appears to be no number density evolution at the bright end (i.e., for most massive galaxies), whereas at fainter magnitudes there is a substantial decline with redshift of the number density of ETGs. (From Cimatti, Daddi & Renzini 2006.)

The slow evolution with redshift of the number density of spectrum-selected bright ETGs was also one of the main results of the K20 survey (Pozzetti et al. 2003), and more recently of the VLT VIMOS Deep Survey (VVDS) where the rest-frame B-band LF of ETGs to I < 24 is found to be broadly consistent with passive evolution up to z ~ 1, with the number density of bright ETGs decreasing by ~ 40% between z = 0.3 and 1.1 (Zucca et al. 2006).

Quite the same scenario, in which the most massive ETGs are already in place at z ~ 1 while less massive ones appear later, emerges from the study of the evolution of the stellar mass function for morphologically-selected ETGs in the GOODS fields (Bundy, Ellis & Conselice 2005, Caputi et al. 2006, Franceschini et al. 2006), and especially from the thorough re-analysis by Bundy et al. (2006) of the color-selected ETGs from the DEEP2 survey. No such effect for morphologically-selected ETGs in the GOODS-South field is mentioned in a recent study by Ferreras et al. (2005), who report instead a steep decrease in their number density with redshift. It seems fair to conclude that to fully prove (or disprove) down-sizing in mass assembly, and precisely quantify the effect, one needs to explore the luminosity and mass functions to deeper limits than reached so far, while extending the search to wider areas is needed to overcome cosmic variance. An endeavour of this size requires an unprecedented amount of observing time at virtually all major facilities, both in space and on the ground. The COSMOS project covering 2 square degrees (Scoville 2005) is deliberately targeted to this end, providing public multiwavelength data extending from X rays to radio wavelengths that will allow astronomers to map the evolution of galaxies and AGNs in their large scale structure context, and to derive photometric and spectroscopic redshifts (Lilly 2005) to substantially fainter limits than reached so far. Thus, it will be possible to directly assess the interplay between AGN activity, star-formation onset and quenching, merging, mass growth, and morphological differentiation over the largest presently possible scale, hence promising substantial, perhaps definitive progress in mapping the evolution of ETGs and their progenitors.

It is worth emphasizing that even a low-level of ongoing star formation can make galaxies drop out of our ETG samples, and that theoretical models do not make solid predictions on when star formation is going to cease in a dark matter halo. Therefore, in a broader perspective, what is perhaps more fundamental than the early-type/late-type distinction, is the stellar mass of galaxies and the evolution of the galaxy mass function, irrespective of galaxy type. Several ongoing studies are moving in this directions (e.g., Fontana et al. 2004, Drory et al. 2004, 2005, Bundy, Ellis, & Conselice 2005, Gabasch et al. 2006, Conselice, Blackburne, & Papovich 2005, etc.), but extending the discussion to the general mass assembly of galaxies goes beyond the scope of this review.