The use of our spectra to study the absorption line is hampered by the low resolution, both because faint lines are below the detection threshold and because many lines are blended together. Nevertheless, many features can be easily seen and measured with great accuracy, even if in some cases their identification may not be unique. For the following analysis we have only used the spectra in H and K because a series of facts make our J spectra less useful to study the features: they have the lowest resolution, their SNR is lower than that in the other bands, the features are intrinsically less prominent, and the star classification in this band is still in its infancy (see, for example, Wallace et al., 2000).
To further increase the SNR we decided to use the average of the spectra of E, S0 and Sa because they show no significant difference in the absorption lines, meaning that temperature, luminosity and metallicity of the dominant stellar populations are similar. We will indicate with ES the resulting average spectrum.
We identified the lines (see Table 3) by comparing the features in our spectra to those identified in the late type stars. The literature of reference is now rich enough (e.g., Origlia et al., 1993; Oliva et al., 1995, 1998; Meyer et al, 1998, hereafter M98; Kleinmann & Hall, 1986, hereafter KH86; Vanzi & Rieke 1997; Wallace et al, 1996, 1997) to make the identifications reliable, but the dominant contribution to some features due to blending of many lines is sometime uncertain. In these cases, the synthetic spectra constructed specifically for line identification in the H band by Origlia et al. (1993) were of particular help, while in the K band the high-resolution spectra by Wallace et al. (1996) were taken into particular account.
The first useful results can be derived by qualitatively comparing the H- and K-band parts of the ES spectrum to various stellar spectra libraries and to the results of the spectrophotometric models. This comparison is shown in Figures 10 and 11 where all the spectra were normalized to a fitted continuum and reduced to the same resolution of the observed data.
Figure 10. Line identifications and comparison of the observed H-band spectrum of the early-type galaxies (thick line) normalized to its continuum to several spectral libraries (thin lines) reduced to the same resolution. Panel (a): the galaxy spectrum is compared to a K5-giant star from the libraries by M98, showing a very good agreement. Panel (b): the spectra of a K5 (solid) and a MO giant (dotted) from Pickles (1998) are overplotted to the galaxy data. Even if the two stars are similar, their spectra in this library are very different. Panel (c): the SSP model from BC99 at 10 Gyr with empirical stellar libraries (see previous section and Figure 9) is compared to the data.
Figure 11. As the previous figure, but in the K band. The K5III star spectrum in panel (a) is from KH86.
In panel (a) of Figure 10 a K5III star from M98 is overplotted to the data. The agreement is very close and basically all the features are well reproduced: as expected, cold giant stars dominate the emission at 1.6 µm and the observed spectra can be reproduced even with a single star spectrum Analogous excellent agreement is obtained using spectra from the Wallance & Hinkle (1996) library (not shown) which are limited to a narrower wavelength range.
The following panel shows the spectrum of two stars, a K5III and a M0III, from the more recent stellar libraries by Pickles (1998), library used in several spectrophotometric models. The former star, the same as panel (a), shows a spectrum similar to our data, even if the agreement is not as accurate as for the KH86 spectrum. On the contrary, the M0III star shows a very different behavior, as also other stars of similar spectroscopic and luminosity class. These differences are not explained by the small temperature difference between the K5 and M0 stars (about 160 °K, M98) and such a behavior is not seen in other libraries, as M98 or Wallance & Hinkle (1996). The differences are probably due to the use of several different sources in the derivation of the Pickles (1998) library. In particular Pickles bases the near-IR part of its spectral on the the Lançon & Rocca-Volmerange (1996) library because it covers a wide wavelength range, between 1.43 and 2.5 µm, with accurate flux calibration. Unfortunately, some features of the spectra of this library seem very variable from one spectral type to the next ones and very different from other libraries (as M98 and KH86) and from our spectra. Therefore they probably are observational artifacts. As a consequence, any spectrophotometric model based on this library will not be able to accurately reproduce the features in near-IR galaxy spectra. This is shown in the following (c) panel: here the observed spectrum is compared to the BC99 SSP galaxy model at the age of the best fit in Figure 8. This model is the only one in our sample having a resolution in the near-IR high enough to allow a meaningful comparison of the spectral features. Even if the general shape of the continuum is well reproduced, as shown in Figure 8, the agreement with the spectra features is generally poor: as an example, the deep absorption lines predicted by the models at 1.665 and 1.705 µm are not present in the observed data.
Same conclusions follow from the spectra in the K band shown in Figure 11: the K5III star from KH86 in panel (a) shows a very good agreement with the ES spectrum. Only the CO bands at 2.29 and 2.32 µm are deeper in the star spectrum than in the ES one, because of either metallicity effects, or the contribution of stars warmer than K5 or on the main sequence, or differences in the fitted continuum level above 2.28 µm. Analogous excellent agreement is obtained with the K5III star spectrum from the Wallance & Hinkle (1996) library (not shown).
In panel (b) and (c) the Pickles (1998) and BC99 spectra are shown: again, some features are not reproduced. For example, the prominent NaI feature at 2.21 µm is almost completely absent from the M0III star and from the model spectra. Also the CaI feature at 2.26 µm appears deeper in the observed data than in the model spectrum. Given that the general shape of the continuum is well reproduced by the model, (see the previous section) the problem must be in the used stellar library.
To summarize, libraries of stellar spectra exist which reproduce the observed galaxy spectra, while others show very different features. Spectrophotometric models based on the second set of spectra might not be able to reproduce the observed features. Our spectra will make it possible to tune the models, especially when new and more extended libraries of near-IR stellar spectra will be available (see, for example, Ivanov et al, 1999; Mouhcine & Lançon, 1999).