3 Spectral classification

3.1 TiO bands in M-type and late K-type giants

M-type and late K-type stars show prominent TiO bands in the near IR (cf. Fig. 1). It is well established that the strengths of these bands correlate tightly with spectral type (e.g. Keenan & Hynek 1945; Sharpless 1956; Ramsey 1981; Kenyon & Fernández-Castro 1987; Schild et al. 1992). Kenyon & Fernández-Castro (1987) and Schulte-Ladbeck (1988) succesfully classified symbiotic spectra using these bands. We basically apply their methods to a larger sample of targets. For our classification we use the depths of the heads of the strong TiO bands at $\lambda\lambda$7054, 7589, 8432, 8859, and 9209.

  

Figure 1: Typical near IR spectra of s-type symbiotic systems containing an M-giant. The uppermost curve shows the spectrum of a system with a late M-giant (AS221, M7.5); the middle displays the M3.5 spectrum of HK Sco; the bottom spectrum is an early M-type (Hen1342, M0). The indicated TiO band heads (dotted lines) are used in the analysis. For clarity we left a gap at the location of the strongest absorptions from the telluric A band ($\lambda$7593) in order to illustrate the spectral characteristics of the TiO $\lambda$7589 band

We determined the spectral types by comparing the strengths of the TiO band heads of the target stars to those of the spectral standards. We either carried out the comparison "by eye'', or we judged the ratio of the two spectra. Figure 2 shows an example. In this procedure we payed particular attention to the different spectral resolution and spectral coverage of our spectra. Especially the spectral types derived from the low resolution ($\Delta\lambda=15$ Å) spectra taken during the run at OHP on JD 47754 should be considered with caution, as numerous weak emission lines may veil the molecular absorptions. Fortunately, the majority of these observations can be checked with higher resolution spectra, which were taken two days later.

  

Figure 2: The dashed-dotted line shows the spectrum of the symbiotic star Y CrA in the region of two TiO band heads (dotted lines). The full lines below show the spectrum of Y CrA divided by the spectra of standard stars of type M5 (lowest curve), M6 (middle), and M7 (top). For the purpose of comparison, all curves are arbitrarily scaled. Evidently, the TiO bands of Y CrA are not satisfactorily compensated by division by a M5 spectrum, while they are over-compensated with the M7 standard, and the best result is reached with the M6 standard. The strong features above 8950 Å are due to telluric absorptions

The standard stars we used are listed in Table 2. We selected them from the catalogue of Keenan & McNeil (1989). A classification system restricted to "half'' subtypes (i.e. M0, M0.5, M1...) is sufficiently refined for our purposes. We therefore coarsened Keenan & McNeil's classification. The adopted subtypes are given in the third column of Table 2. Keenan & McNeil's classification did not yield a perfect sequence as far as the behaviour of the near IR TiO bands is concerned. This can be due to variability, especially at the latest types. It is, however, also possible that the IR features do not behave exactly as the features in the blue region that are used classically. We reached a continuous sequence by adopting a slightly different spectral type for VY Leo.

K subtypes are not used consistently in the literature (Jaschek & Jaschek 1987). We adopted the subtypes K0, K1, K2, K3, K4, K5, and K7. For HD 90362 which is classified K6 by Keenan & McNeil we adopted K7.

Strong nebular emission is present in many symbiotic systems and may contribute a noticeable fraction to the continuum radiation. This nebular radiation weakens the absorption of the TiO bands relative to the overall continuum and mimics an earlier spectral subtype. To avoid this source of error we had to correct the spectra of the symbiotic giants for the nebular contribution, before comparing them with the standard stars. The nebular continuum contribution, $F_{\rm neb}$,was determined from the equivalent width W(P14) of the HI Paschen emission line P14$\;\lambda8598$. According to photoionization models (see e.g. Osterbrock 1989) nebular emission alone produces an equivalent width $W({\rm P}14)\approx30$ Å relative to the nebular continuum. Thus, the nebular continuum flux around $\lambda8600$ is $F_{\rm neb}=F_{\rm tot}\cdot W({\rm P}14)/30~{\mbox \AA}$, where $F_{\rm tot}$ is the sum of the continuum fluxes of the nebula and the cool giant at this wavelength. In correcting the overall spectrum we neglected the spectral variation of the nebular continuum flux $F_{\rm neb}$,which is rather flat in the range of interest. In only a few cases, the correction for the nebular contribution altered the measured strength of the TiO bands noticeably.

  

Table 2: Standard stars used in our comparison

Table 3 summarizes the classifications derived from the selected 5 TiO band heads. The objects are ordered according to their right ascension, like in the catalogues of Allen (1984) and Kenyon (1986). A long dash "--'' means that the respective band is not visible because it is a) not present or too weak to be measured, b) not covered by the observations, c) evidently filled by nebular emission as the presence of bands at longer wavelengths indicates, or d) not measurable due to emission lines and insufficient resolution.

   

Table 3: Spectral classification of the cool components in symbiotic systems according to the strength of 5 TiO band heads. "<'' means "earlier than'', for example

 

Table 3: continued

 

Table 3: continued

 

Table 3: continued

None of the spectra of He2-104, He2-106, SSM1, and H1-36 shows absorption features that can be safely attributed to the cool star. Since He2-104, He2-106, and H1-36 show photometric variations and dust emission in the IR typical of mira type stars (Whitelock 1987) they should contain a rather late M-type giant. Therefore, we suppose that in these four systems the cool star is obscured by dust, so that its light cannot compete with the nebular emission. Very little is known about SSM1.

  

Figure 3: Comparison of the adopted classification and that suggested by the different TiO band heads

Spectral types from different TiO bands.

In Fig. 3 we compare the spectral types derived from the different TiO band heads. It is apparent that the spectral classification derived from the TiO band $\lambda7054$ is systematically shifted by about two spectral subclasses towards earlier types, when compared with the TiO bands at longer wavelengths. The bands $\lambda\lambda$7589, 8432, 8859, and 9209 correlate very well. Already Kenyon & Fernández-Castro (1987) had noticed that the molecular absorptions at shorter wavelengths mimic earlier spectral types for the giants in symbiotic systems. The observed inconsistency is possibly explained by assuming significant contamination of the $\lambda7054$ TiO band by nebular emission. The bands at longer wavelengths are expected to be much less affected due to the rapid increase in brightness of the red giant towards the IR.

For the final classification given in the last column of Table 3, we "averaged'' the results from the TiO bands $\lambda\lambda$7589, 8432, 8859, and 9209. Due to the above described effect, the band at $\lambda7054$ is only considered where the other bands are not available.

3.2 Comparison with published classifications

  

Figure 4: Comparing our spectral types with the five most extended classification works in the literature. Top panel: comparison with Kenyon & Fernández-Castro (1987, $\times$) and Schulte-Ladbeck (1988, +); bottom: comparison with Medina Tanco & Steiner (1995, $\square$), Harries & Howarth (1996, $\diamondsuit$), and Mikoajewska et al. (1997, $\triangle$)

In Fig. 4 we compare our classification with publications that contain a certain number of objects in common with us. The top panel shows that our results correlate well with those of Kenyon & Fernández-Castro (1987) and Schulte-Ladbeck (1988). Their classifications differ on average less than 0.8 spectral subtypes from ours. We take this as a measure for the reliability of our classification. Note that non-negligible differences have to be expected due to the intrinsic spectral variability of mira systems. The spectral types of a few systems may also be affected by outbursting hot components which may lower unexpectedly the absorption contrast of the molecular absorptions (see Sect. 3.3). Further, we cannot exclude the possibility of small variationes (<1 subclass) due to e.g. irradiation or tidal effects. The good agreement with Schulte-Ladbeck (1988) is not so surprising as she employed a similar technique and similar spectral data as in this work. Kenyon & Fernández-Castro (1987) employed for their classification work molecular bands between 6000 Å and 8000 Å. They noticed, as in this work, the veiling problem in the shorter wavelength bands. The good agreement with our work indicates that their careful consideration of this problem overcame the disadvantage of using a spectral range that does not reach very far into the IR.

The results of Medina Tanco & Steiner (1995), Harries & Howarth (1996), and Mikoajewska et al. (1997) differ significantly from ours in the average by roughly 2 subclasses (Fig. 4). These authors based their classification mainly on spectral features at shorter wavelengths and/or used insufficient spectral resolution. Presumably, the scatter in the bottom panel of Fig. 4 is at least partly caused by the veiling problem for red giants in symbiotic systems as discussed previously. The tendency of spectral types towards an earlier classification supports this interpretation.

3.3 Temporal variability

Several systems have been observed more than once. This allows to search for spectral variations. We can detect variability only in systems with strong changes exceeding two spectral subtypes. We find in Table 3 seven such variable systems, namely AX Per, He2-38, He2-127, BF Cyg, RR Tel, and HBV 475.

The systems He2-38, He2-127, and RR Tel contain mira variables as cool giants (see Whitelock 1987) and changes of the spectral type related to pulsation are therefore expected.

The spectra of AX Per and BF Cyg were taken during a Z And-type outburst phase. A comparison with V-band light curves of AX Per and BF Cyg from the AFOEV (Association Francaise des Observateurs d'Étoiles Variables) reveals a correlation between the spectral type and the system brightness. We obtained for AX Per a spectral type of M5 at magnitude $m_{V}=11\hbox{$.\!\!^{\rm m}$}5$, M6 at $m_{V}=10\hbox{$.\!\!^{\rm m}$}5$, but M3.5 and M4 during maximum brightness ($m_{V}=9\hbox{$.\!\!^{\rm m}$}5$). Similarly, we find M7 for BF Cyg at $m_{V}=11\hbox{$.\!\!^{\rm m}$}8$, M4.5 at $m_{V}=11\hbox{$.\!\!^{\rm m}$}0$, and M3.5 and M4.5 near maximum ($m_{V}=10\hbox{$.\!\!^{\rm m}$}4$). Thus, we registered weaker TiO band absorptions during outburst. This effect can be explained by the additional continuum contribution of the hot (outbursting) companion, which lowers the absorption contrast of the TiO bands. Another explanation is that the hot star heats up the red giant atmosphere. However, this effect cannot be very strong as no significant changes in the IR bands of the cool giant are seen during such outbursts (e.g. Munari et al. 1992). It is interesting to note that also Kenyon & Fernández-Castro (1987) found for a given symbiotic system weaker absorptions (i.e. earlier spectral types) when the target was bright.

 

Table 4: Systems with spectral types earlier than K5

The variations we find for HBV475 are probably not real. The much earlier spectral type $\rm (M1-M3)$ obtained for the run on JD 47754 (see Table 5) is most likely a spurious result. The reason is the limited resolution of that spectrum ($\Delta\lambda=15$ Å), which fails to resolve the stellar absorptions from the rich and strong emission line spectrum of HBV475. Note that we derived a spectral type of M7 from a higher resolution spectrum ($\Delta\lambda=2$ Å) taken just two days later. This example is a clear warning that spectral types from lower resolution spectra have to be considered cautiously. In our case, this concernes the run at JD 47754.

We conclude that significant and real changes in the cool giant's spectral type are seen only in systems containing mira variables. Our data set is, however, not sufficiently homogeneous and systematic to search for small spectral type variations ($\leq1$ spectral subtype) intrinsic to the cool giant, which may be caused by e.g. irradiation effects, tidal interaction or outburst activity.

3.4 Classification of the yellow symbiotics

The spectral classification based on near IR TiO bands fails for cool giants with spectral types earlier than K5. For these objects we analyze spectra taken in the blue wavelength region.

  

Table 5: Spectral types of the cool components of symbiotic stars

 

Table 5: Continued

  

Figure 5: The blue spectra of AS201 (bottom), BD-21.3873 (middle), and CD-43.14304 (top). The dashed lines mark the absorption lines SrII$\; \lambda4077$, CaI$\;\lambda4227$,CH$\;\lambda4300$, and FeI$\;\lambda4405$. In addition, the TiO$\;\lambda4955$band head is marked in the spectrum of CD-43.14304

Fortunately, the observed G and K giants are sufficiently bright in the blue wavelength region, such that the veiling by the nebular continuum emission can be neglected. Thus we can use the known line ratio criteria for the spectral classification (Keenan & McNeil 1976; Yamashita et al. 1977; Jaschek & Jaschek 1987). However, the blue spectra of the symbiotic giants are still contaminated by strong emission lines, mainly H and He recombination and nebular [OIII] emission lines (see Fig. 5). This prevents particularly the use of criteria based on hydrogen lines. Due to the moderate resolution of our spectra, we are additionally limited to strong spectral features. In view of these restrictions we confined our classification to the spectral types G0, G5, K0, and K2 similar to the system of the HD catalogue. As main classification criteria we use the G band, which is strongest for G5, CaI$\;\lambda4227$, which increases steadily from G0 to K7, and FeI$\;\lambda4405$ which is only strong in K types. In addition we consider also the absorption features of FeI, SrII, and other species as secondary indicators. Because our classification relies relatively strongly on the CH band we point out that our classification may be inaccurate for stars with peculiar abundance patterns (e.g. carbon rich or carbon deficient objects).

In our sample we identify a very homogeneous group consisting of V741 Per, Wray157, AS201 and S190. All of them exhibit a very strong CH-band and relatively weak lines from CaI and FeI. Therefore we assign to them a spectral type of G5. HD 330036 exhibits a very similar absorption spectrum except for the CH-band which is substantially weaker than in the previous group. From the FeI, CaI and SrII lines alone and assuming that the weak CH-band is due to an abundance anomaly we adopt for this star as well a spectral type of G5.

BD-21.3873 and AG Dra have a relatively weak CH-band and strong features of CaI and FeI$\;\lambda$4405 compatible with a K2 classification. In the spectrum of He2-467 the FeI lines are stronger than in AG Dra or BD-21.3873 while CaI$\;\lambda$4226 is rather weak. Again, this could be an abundance effect as both, BD-21.3873 and AG Dra are known to be metal poor objects (Smith et al. 1996, 1997). Based on these considerations we assign a spectral type K0 to He2-467. In the spectrum of CD-43.14304 we see clearly a weak TiO absorption band at $\lambda$4955. This suggests a spectral type of K7 in contradiction to the type listed in Table 3.

3.5 Symbiotic carbon stars

S32, AS210, and SS38 show strong CN absorptions bands which indicate unambiguously that they contain a cool giant of spectral type C. For carbon stars the strength of the molecular bands depends not only on the photospheric temperature but also strongly on the carbon abundance (e.g. Keenan & Morgan 1941; Yamashita 1972). Because we do not possess the respective standard star spectra we abstain from a detailed classification. Nonetheless, the overall spectra indicate that the giant in S32 is an early C star (see also Schmid 1994), while AS210 and SS38 contain late C stars, which are probably mira variables (Whitelock 1987).