3 Selected results

The data set provides an immense amount of information. Our purpose here is to draw attention to selected empirical facts that do, in our opinion, deserve further investigation either observationally or theoretically. We also identify individual stars of special interest.

The use of near-IR molecular indices for classification purposes is discussed by Alvarez et al. (2000). Ways to include the data in population synthesis calculations and first results are presented in Lançon (1999), Mouhcine & Lançon (1999a,b, 2000) and Lançon et al. (1999). The construction of a practical library of suitably averaged empirical LPV spectra and a discussion of the appropriate temperature scale will be presented in more detail in a forthcoming paper (Mouhcine et al., in preparation). First comparisons of the empirical data with the O-rich LPV model spectra described in Bessell et al. (1991, 1996) and Hofmann et al. (1998), and with the carbon star models of the Vienna group (Loidl et al. 2000) have been made, and will be reported in detail elsewhere.

3.1 Line and band identifications

The optical spectra of late-type O-rich stars are shaped by TiO and VO bands. Identifications are given in Phillips (1969), Turnshek et al. (1985), Brett (1989, 1990) and Valenti et al. (1998), and are shown in Fig. 4. The Ca II triplet (8498, 8542 and 8662 Å) is seen in stars of early M spectral type (Figs.B1, B2, and B9, but disappears when TiO absorption takes over for later types and in most LPV spectra of the present sample. Identifications for features between 1 and 1.3 $\mu$m are given by Joyce et al. (1998). The H₂O band around 1.15 $\mu$m is easily confused with the CN band present in red supergiants, and is heavily blended with TiO and VO bands in the coolest LPVs. For easier comparison with the data of Joyce et al. (1998), Fig. 5 provides an enlargement of this region for a variety of cool stars, on the wavenumber scale adopted by these authors. At longer wavelengths, CO and H₂O are the dominant molecules but OH and, mainly in supergiants, CN absorption also occur. Origlia et al. (1993, and references therein) identify and model the main molecular and atomic features found between 1.4 and 1.73 $\mu$m. Identifications for the K atmospheric window are given by Kleinman & Hall (1986) and Wallace & Hinkle (1997).

The dominant molecules vary from star to star, even when the sample is restricted to O-rich objects. An illustration is given in Fig.6, which shows that OH lines can be more prominent that CO lines in the H window.

Figure 5: Main band identifications in the region between 1.02 and 1.28 $\mu$m (Joyce et al. 1998). From top to bottom: the cool O-rich Mira RCha, the warm O-rich Mira SCar, the C-rich Mira RUPup, and the red supergiants HD101712 and IRC-20427. Relative flux units are arbitrary

Band identifications for C-rich stars will be given by Loidl et al. (2000).

   
3.2 Giants, supergiants and O-rich LPVs

The present data (Figs.B1, B2, B3) confirm expected effects of temperature ( $T_{\rm eff}$) and surface gravity (g) among static giants and supergiants (see e.g. Johnson & Méndez 1970; Bessell et al. 1989a). For instance, CO bands and metal lines (e.g. around 1.3 $\mu$m and between 2.2 and 2.3 $\mu$m) deepen with decreasing $T_{\rm eff}$ and g. Among O-rich stars, only supergiants display prominant near-IR CN absorption (in particular at 1.1 $\mu$m).

Red supergiants have lower effective temperatures than giants of the same spectral type (e.g. Flower et al. 1977; Dyck et al. 1996). This still controversial statement (Bessell 1998) is supported by our data if we take the optical features seen at low resolution as an indication of the spectral type: at similar optical features, supergiants have redder energy distributions than giants (see spectra and Fig. 7). The apparent shift of the maximum of the energy distribution to longer wavelengths may be due in part to CN absorption, in particular between 0.8 and 1.05 $\mu$m; the extinction vector in Fig. 7 indicates a possible partial contribution of reddening, due to circumstellar or interstellar extinction towards the low galactic latitude supergiants.

Figure 6: Relative importance of CO and OH in the H window, in the Miras SV Lib (left) and S Car (right). The dotted lines identify the main 12CO bandheads. Vertical marks identify OH lines. Throughout the window, the 2-0 P series (lower long marks) and 3-1 P series (higher long marks) are seen; the 3-1 Q (1.51$\mu$m, short marks) and R series (1.46 $\mu$m, short marks) are also shown. The curved shape of the spectra is due to H2O absorption

The spectra of LPVs with small visual amplitudes ($\leq 1.3$) are similar to those of static stars. They can be sorted into the giant or supergiant categories using the two-colour diagram shown in Fig. 7. The GCVS supergiant classification of the variables EV Car (SRb), CL Car (SRc), UY Sct (SR) and V774 Sgr (Lb) is confirmed. ET Vir (SRb, $\delta V=0.2$) obviously joins the giants (Fig. B1), and SY Vel (SRb, $\delta V=1.3$, Fig. B6) shows very little difference with the M 5 giant BS4267 (Fig. B1). KV Car (SRb, $\delta V=1.3$) significantly varied between our two observations; although they show relatively strong CO bands, this star's spectra fit into the spectral sequence of giants.

Strong H₂O bands are found to be the characteristic feature of variable M stars. Inside subsamples of M type stars with similar optical spectra and global energy distributions, very different near-IR spectra are found (Fig.8), strong H₂O bands always implying large amplitude variability ( $\delta V >1.7$ in our sample). None of the (quasi-)static M giants of the sample show impressive water vapour bands. This property has been qualitatively reproduced by first generations of models (Bessell et al. 1989a,b, 1996). As a result of pulsation, LPVs develop extended atmospheres with thicknesses comparable to the effective stellar radius (Wood 1979; Bowen 1988; Feuchtinger et al. 1993). Dense cool layers can exist, allowing H₂O and, at later stages, even dust to form. According to the pulsating star models used in Bessell et al. (1996) and Hofmann et al. (1998; relevant data kindly provided by M.Scholz), observations at the wavelengths of water vapour bands sample regions typically a factor of 1.4 - 2 further out in the atmosphere than observations in the quasi-continuum windows (e.g. at 2.2 or 1.04 $\mu$m). The shape of the water bands depends on details of the instantaneous atmospheric structure, which leaves much freedom to explain the variety of observed shapes and depths but makes it difficult to determine the cause for any particular empirical H₂O feature. Some LPV spectra can be relatively well reproduced by existing models, but for the same star the fits at other pulsation phases is often poor, casting doubt on the overall model reliability. Spectra with strong water vapour bands but otherwise warm energy distributions, such as S Car, have not as yet been reproduced by models (Mouhcine & Lançon 1999a; Matsuura et al. 1999).

Figure 7: The temperature sensitive broad band colour $V^{\prime }-K$ (cf. Sect. 2.4) plotted against the optical, spectral type sensitive index S1S3 (defined as $S_{1/3,{\rm Sp}}$ in Fluks et al. 1994). Plus symbols represent both semi-regular and Mira-type O-rich LPVs; stars, circles and squares, respectively, are static giants, red supergiants and C-rich LPVs. Horizontal arrows identify the small amplitude variable KV Car (see text)

Additional systematic differences between stars of various luminosity classes and variability types are described in Alvarez et al. (2000). While segregation tools are already available at the resolution of the present data, it is clear that higher resolution data would be helpful, in particular in distinguishing between the various sources of opacity between 0.8 and 1.2$\mu$m.

   
3.3 Emission lines

Many miras are already classified as emission line stars in the GCVS on the basis of the Balmer line emission seen in the blue part of the spectrum. The most conspicuous near-IR emission line is Paschen$\beta$, which is located in a relatively clean spectral range, be it with respect to intrinsic stellar absorption bands or to telluric absorption. Equivalent widths of up to 6 Å are observed in our sample (RS Hor, March 1996, Fig. B13). Higher order lines of the Paschen series are fainter than P$\beta$and are located in stellar molecular bands of TiO and VO. Brackett$\gamma$ is more difficult to recover and P$\alpha$ completely lost because of telluric H₂O absorption (cf. Fig. 2).

Line emission in Miras is a transient phenomenon: there is no indication in our data that any star displays emission lines throughout its pulsation cycle, a fact already known from optical studies (e.g. Merrill 1940). GCVS phases are available for 25 stars in the sample; we have been able to replace them with more recent phase information only in a few cases of particular interest. Despite the uncertainties on the GCVS phases, the phase distribution of stars with emission lines shows a significant excess at phases 0.75-1 and a corresponding significant dip at phases 0.2-0.45, when compared to a uniform distribution. Models that incorporate shock formation and attenuation in LPV atmospheres predict that these shocks are strongest during the rising part of the light curve or around maximum light (Bessell et al. 1996). With the necessary caution due to limited phase information, the present data are consistent with a physical association between such shocks and the line emission (however, see e.g. Gillet 1988, for an alternative interpretation).

   
3.4 Metallicity

Only few theoretical studies exist of the effect of metallicity, Z, on near-IR spectra of luminous red stars with otherwise identical global properties (e.g. $T_{\rm eff}$, g, atmospheric extension, microturbulent velocity). The concept of metallicity itself is controversial, since several dredge-up episodes are likely to have modified the abundance ratios as compared to solar values.

Figure 8: The M 5 giant BS4267 and the Mira-type LPV S Car (July 1996), two stars with similar optical spectra and large scale energy distribution. Note that water vapour absorption is much more prominent in the variable star S Car. Here and in subsequent spectral plots the dashed areas are those of very strong telluric absorption (season dependent, cf. Fig. 2), in which the signal-to-noise ratio after correction generally remains unacceptably low

Figure 9: The LMC star HV2360 (smoothed) and the two galactic, high radial velocity Miras SLib (May 1996) and SCar (July 1996)

Bessell et al. (1989a) studied the effect of metallicity on low resolution spectra of static stars. Their model spectra show that the effects of Z, $T_{\rm eff}$ and g on spectrophotometric indices are highly degenerate (Lançon & Scholz, ongoing study).

Origlia et al. (1997), based on higher spectral resolution radiative transfer calculations and on static giant stellar structures with $T_{\rm eff}$ $\geq 3000$ K, suggest that the equivalent width of the SiI line at 1.59 $\mu$m provides a relatively robust metallicity scale. The authors note, however, that the line is contaminated by OH in cooler stars (see also Fig. 6). The Al doublet at 1.316$\mu$m (Joyce et al. 1998), or metal features in the K band (Na 2.20 $\mu$m, Ca 2.26 $\mu$m) may provide alternatives. The calibration of all the potential metallicity indicators remains challenging even for static giants, although progress is being made (Ramírez et al. 2000). In LPVs, the strong dependence of the many weak molecular lines on details of the instantaneous stellar structure affects the pseudo-continuum and makes the interpretation of metal lines even more complex.

Model independent insight into the effect of metallicity on LPV spectra can in principle be obtained by comparing stars known or suspected to belong to populations of established metallicities. LMC, SMC and thick disk or halo objects are expected to be metal deficient. S Car and S Lib, whose radial velocities approach 300km s^-1 (Feast 1963), are both commonly associated with Population II or the thick disk. Figure 9 shows that very similar spectra can be found among the halo/thick disk and the LMC/SMC subsamples. Note, however, that HV2360 is a supergiant with mass $\sim$8 $M_{\odot}$ and $M_{\rm bol} \sim -8.15$ (Wood et al. 1983) while SCar and SLib presumably have $M < M_{\odot}$ and $M_{\rm bol} \sim -4$ (from the $M_{\rm bol}-\log P$ relation of Hughes & Wood 1990). Relatively strong CN bandheads at 0.77, 0.92 and 1.1 Å in HV2360 and the other LMC and SMC supergiants (Fig. B26) is consistent with the expectation that hot-bottom burning nucleosynthesis in these stars is converting carbon (probably dredged-up) into nitrogen. The number of matching pairs of spectra in the sample is too small to draw any final conclusion from these comparisons.

Figure 10: Potential metallicity indicators. Left: an index measuring the absorption in the short wavelength wing of the 1.9 $\mu$m H2O feature (in relative magnitudes), versus a measure of the equivalent width of the CO (2-0) bandhead at 2.29 $\mu$m (arbitrary linear wavelength unit). Right: the ratio H2O/CO of these absorption measures versus the equivalent width of the Ca line blend at 2.26 $\mu$m (arbitrary linear wavelength unit). The symbols are as in Fig. 7, except that semi-regular variables are shown as triangles (SRb, SRc) and diamonds (SRa). Arrows identify 4 of the stars with weak Ca and Na line blends: S Car (pointing up), BD Hya (left), RY Cra (down) and RS Hor (right). The strengths of all features vary considerably. The coordinates of the two spectra of S Lib are (11.2, 0.18) and (9.9, 0.16) in the first graph, (31.4, 0.02) and (31.6, 0.02) in the second

Figure 11: Comparison between dereddened Bulge star spectra and spectra of Galaxy field stars (cf.Table 7)

As a further step, we may use the metal lines mentioned earlier to attempt to identify metal-poor stars in the sample and then look for systematic spectral differences between these and the other objects. First, we note that the Na2.20$\mu$m and Ca2.26$\mu$m line blends of the large radial velocity objects SCar and SLib are among the weakest of our sample. Taking weak Ca and Na lines as a tentative criterion for low metallicity, one is led to select the Magellanic Cloud stars as well as spectra of the SRa variables T Cen and BD Hya and the Miras RS Hor, RY CrA, SV Tel. Of these, T Cen has been previously associated with the halo or the thick disk by Alvarez et al. (1997; table in Alvarez 1997). Eye inspection (confirmed by colour measurements) shows that this list contains the hottest objects of the sample, illustrating the degeneracy between the effects of temperature and metallicity. All selected objects have periods below 225 days; reciprocally however, some short period SR variables fail to join the list, e.g. S Phe or SY Vel. The amplitudes of the selected objects range from 1.8 to 5.5 magnitudes in V, so that amplitude provides no further directly useful information for metallicity selection.

When concentrating on the spectra with weak metal lines, eye inspection suggests a systematically high H₂O/CO absorption ratio. The trend is illustrated in Fig. 10. Despite their relatively blue colours, the stars selected tend to have strong water vapour bands. It is tempting to relate the high H₂O/CO absorption ratios to low metallicity, as the abundance of CO depends on the square of the heavy element abundance. However, as already mentioned, apparently weak metal lines can result from a depressed pseudo-continuum due to many enhanced molecular lines, and we may simply have selected relatively hot stars with particularly strong H₂O absorption. Only high resolution spectroscopy can solve this degeneracy.

Among the spectra with small Ca equivalent widths, those of S Lib stand out with relatively strong CO bands, a potential signature of a high C/O ratio (see discussion about SVLib in Appendix A) or high luminosity. We note that SLib is one of the few stars for which the automatic distance assignments of Alvarez et al. (1997), based respectively on optical and near-IR data, gave deviant results.

 

 

Table 7:

Bulge spectrum	Reddening correction	Similar field star	Figures
	E(B-V)
NGC6522n435 (May 1996)	1.6	X Men (Mar. 1996)	B25, B19
NGC6522n426 (May 1996)	0.4	RS Lib (Jun. 1995)	B25, B18
NGC6522n205 (May 1996)	0.0	AFGL 1686 (Apr. 1995)	B25, B24
SGRIn11 (Jul. 1996)	1.0	RS Hya (Feb. 1995)	B25, B15
SGRIn11 (Jul. 1996)	0.5	WW Sco (Jun. 1995)	B25, B22

The relatively long period Bulge stars included in the sample (Fig. B25) are expected to be metal-rich. As discussed in Alvarez et al. (2000), the Bulge spectra are heavily reddened but are unlikely to be contaminated significantly by circumstellar dust emission. Once corrected for extinction, they generally remain redder than optically visible late-type AGB stars of the solar neighbourhood. Wood & Bessell (1983) interpreted this as the consequence of the lower effective temperatures of high metallicity AGB evolutionary tracks. Note, however, that dereddened Bulge spectra can be matched by the spectra of very late type Galaxy field LPVs (near minimum) very well (Fig.11).

Perhaps we may use similarity to dereddened Bulge spectra as a selection criterion for metal-rich field stars. Some matching pairs are listed in Table7. Obviously, using the Bulge stars of our sample as templates leads us to select only stars of the latest spectral types as potentially metal-rich stars.

   
3.5 Variations

Figures B6-B31, each of which displays multiple spectra for the same star, show how spectra can vary with phase through the pulsation cycle. Note that the large V band amplitudes of many Miras are the result of variations of approximately one magnitude in bolometric light, combined with large changes in the bolometric correction BC_V. For S Car ( $\delta V=6.6$), RS Hor ( $\delta V=5.2$), RZ Car ( $\delta V=6.2$) and RS Hya ( $\delta V=5.2$) we find, respectively, $\delta BC_{V} \geq$3.0, 3.7, 3.0 and 3.4. The spectra plotted per object in this paper illustrate the corresponding variations in colours and spectral signatures.

Alvarez & Plez (1998) illustrated clearly that optical signatures like the bands of TiO and VO both vary with phase but do not reach maximum depth at the same time. Similar variations and phase shifts are observed in the infrared. It has already been mentioned in Sect. 3.3 that the HI emission lines are strongest around or somewhat before maximum light.

An interesting new result suggested by the spectra is that CO and metal line absorption may be systematically reduced for a short time interval during the rising part of the light curve, around phase 0.7 (this is perhaps related to the line blurring and veiling noted in the blue spectra of Miras - see Merrill 1940). Searching for individual spectra with apparently reduced CO with respect to all other observations of the same star, we find: the June 1995 spectrum of RZ Car (GCVS phase 0.7, phase $\sim$0.8 according to the AAVSO bulletin, Mattei 1999; Fig.B8), the July 1996 spectrum of X Men (GCVS phase 0.7; Fig. B19), the June 1995 observation of R Phe (GCVS phase 0.62; Fig. B20), the June 1995 spectrum of S Car (GCVS phase 0.8, phase 0.7 - 0.75 according to the AAVSO bulletin and to AFOEV data (AFOEV 1999); Fig. B7) and the May 1996 observation of R Cha (GCVS phase 0.58, $\sim$0.8 according to the AAVSO bulletin 1999, Fig. B10). No spectrum with reduced CO was found at phases larger than 0.9 or smaller than 0.6. No spectrum with significantly enhanced CO (as compared to other phases of the same star) was found around phase 0.7. This phenomenon suggests that the atmosphere (or at least the relatively deep region of the photosphere responsible for the aspect of the CO bands and metal lines) goes through a particularly compact stage while luminosity increases, which is consistent with predictions from pulsation models (e.g. Bessell et al. 1996) in which a shock wave is starting to emerge from the deep layers around phase 0.7. At low resolution, it is not possible to investigate more subtle potential causes for the apparent weakening of the CO bands, such as line doubling or the superimposition of CO emission in the strongest lines (as discussed by Aringer et al. 1999, for SiO bands, or seen in UEqu, Fig. B24 and Appendix A).

Another example of delayed variation cycles is found in the 1 - 1.3 $\mu$m range. Using the fluxes in the pseudo-continuum windows at 1.04 and 1.23 $\mu$m, one can identify a few spectra with particularly blue pseudo-continua in this range but otherwise cool energy distributions and clear VO absorption: RS Hor in Jan.96 (Fig. B13), SV Tel in June95 (Fig. B23), RS Hya in Apr.96 (Fig. B15). All these spectra display HI emission, indicating that they were observed just before or around maximum light. Somewhat later observations in the same pulsation cycle (RSHor in Mar. 96, RSHya in May 96) show similar near-IR spectra, warmer optical spectra and no VO absorption. Again, this points to the existent of a brief moment shortly before maximum light during which the spectral properties originating in deeper layers indicate higher temperatures than the molecular features originating further out.

Pulsating Mira models without heavy mass loss (e.g. Bessell et al. 1996) predict non-periodic behaviour of the upper atmospheric regions so that there are cycle to cycle variations in these regions and in the location and strength of the shocks that run through them. Such changes lead to obvious spectral variations from cycle to cycle. As examples, we may cite:
-- RS Hya (Fig. B15): the June 1995 spectrum has been obtained in an earlier cycle than the others (1.05 cycles before the May 1996 spectrum) and clearly displays much weaker H₂O absorption;
-- RZ Car (Fig. B8): deeper H₂O absorption in the March 1996 observation than in the June 1995 spectrum, observed exactly one cycle earlier.

As expected, the most obvious cycle to cycle variations concern the H₂O bands as these form in the upper cool layers whose structure is sensitive to details of the history of successive shock passages.

Figure 12: Optical light curve of S Car. The AAVSO data plotted was kindly provided by J.A.Mattei in numerical form (we also acknowledge that AFOEV made similar data avalable to us). The acquisition dates of the spectra of Fig. B7 are indicated

The well sampled light curve available for S Car (Fig. 12) allows us to highlight the intimate link between irregularities of the luminosity evolution and cycle-to-cycle variations in the spectra. The February 1995 and December 95 spectra, at the top of Fig. B7, were taken close to maximum light. A higher (optical) luminosity was reached in December than in February; the December data has weaker H₂O features than its February counterpart. The July 96 and March 96 spectra, at the bottom of Fig. B7, correspond to minimum light. The March minimum had a lower luminosity than the July minimum; it shows very deep and broad H₂O absorption and a clear VO band at 1.05 $\mu$m that is absent in the July data. At a given phase, the strength of the molecular features and the current (optical) luminosity are clearly anticorrelated for this irregular Mira variable.

   
3.6 C stars

The spectra of carbon stars have little in common with those of oxygen-rich giants apart from CO absorption and a red energy distribution. Our sample (Figs. B28-B31) is relatively small. However, several general comments can be made.

As compared to O-rich stars, the spectra of C-rich LPVs tend to vary little with phase. O-rich LPVs with visual amplitudes of about 1.8 show significant changes, whereas the available C stars with similar amplitudes don't. The molecules that shape the near-IR spectrum of C stars, essentially CO, CN and C₂, are formed deeper in the atmosphere than their counterparts in O-rich stars, TiO and H₂O (Loidl et al. 2000); the influence of dynamics in deep layers is smaller. Stronger variations with phase are found at longer, mid-IR wavelengths, where the main molecular bands are due to molecules that form at lower temperatures, such as C₂H₂, HCN and C₃ (e.g. Hron et al. 1998). Note that the S/C star BH Cru and the cool, large amplitude ( $\delta V=6.2$ mag) variable R Lep do exhibit significant near-IR variations (see Appendix A regarding peculiarities of RLep).

Since the spectra of low amplitude C-rich LPVs are not very sensitive to pulsation, it is meaningful to compare them to theoretical spectra of static stars. A detailed comparison will be discussed by Loidl et al. (2000). Preliminary results are promising: the comparisons show that molecular line lists and their incorporation into the radiative transfer codes have made very significant progress in recent years and may now be satisfactory. The models indicate that the C/O ratio is a dominant factor in explaining star to star differences, and in particular the strength of the C₂ bandhead at 1.77$\mu$m. A C/O ratio of at least 1.4 is necessary to reproduce the bandheads observed in Y Hya or S Cen. Empirically, carbon stars can easily be ordered into a sequence according the ratio of the strengths of the C₂ bandhead at 1.77$\mu$m and various CO bandheads. The CO bands of the S/C star BH Cru are as strong as those of the coolest supergiant stars of our sample. Correspondingly, C₂ absorption in this object is weak.

An interesting feature of the C-rich spectra is the apparent absence of SiI absorption at 1.59$\mu$m. This empirical fact may simply be the result of blends with the many CN and C₂ lines, although the strong CO bandheads in the same spectral range don't seem to be affected. Alternatively, Si might be bound preferentially into silicon carbides. Within current theories for the formation of SiC grains (e.g. Kozasa et al. 1996) the condensation temperatures are low compared to the temperature required to populate the lower level of the SiI line, making this particular depletion mechanism unlikely; but silicon carbide formation theories are not yet final. OH lines are also absent in carbon star spectra, as expected if all oxygen is locked up in CO.