6 Line profile variability

As summarized for example in Andrillat et al. (1973), many line profiles of the spectrum of HD108 vary over the years, with some of them changing spectacularly from P Cygni profiles to pure absorption or emission lines. This is especially the case for the Balmer Hydrogen lines and some He I lines ( $\lambda \lambda$4388, 4471, 4713). In our data, He I $\lambda \lambda$4388, 4713 appeared only once, in 1987, with a P Cygni profile while these lines display pure absorptions since then.

To better quantify the line variability, we used the temporal variance spectrum (TVS), as defined by Fullerton et al. (1996). This TVS analysis, when applied on the common spectral range of the data collected in 1987 and from 1996 to 2000, yields interesting results (see Table 4): in addition to the He I and H I lines, the absorption lines of N III (from 4510 Å to 4534 Å) and of He II $\lambda$4542, as well as the emission lines of N III $\lambda \lambda$4634-41, He II $\lambda$4686, Si III $\lambda \lambda$4552, 4568 and C III $\lambda \lambda$4647-50, do all vary, though with a smaller amplitude. On the contrary, other emission lines (e.g. O II and the unidentified $\lambda \lambda$4486, 4504) were stable during the whole observing campaign.

To compare the variations of the different lines, we have used the local pattern cross-correlation technique discussed by Vreux et al. (1992) and by Gosset et al. (1994). But before that, we convolved the 1998 and 2000 data until they reached the same resolution as the 1996, 1997 and 1999 spectra. Only then have we cross-correlated the variation pattern of some unblended lines (e.g. He I $\lambda$4471, He II $\lambda$4686) with the whole spectra. Using the He II $\lambda$4686 emission line as a reference, a strong correlation appears between the deformation pattern of He II $\lambda$4686 and the ones of the N III $\lambda \lambda$4634, 4641 lines. On the contrary, the He I $\lambda$4471 pattern seems rather related to the variation of the N III absorptions between 4510 and 4534 Å, He II $\lambda$4542, C III $\lambda$4647-50, He I $\lambda$4713 and H$\beta$. The behaviour of the N III $\lambda \lambda$4634-41 lines suggests therefore that these lines form in the same physical region as the He II $\lambda$4686 emission, while the clearly distinct variations of the He I and Balmer emissions point towards a different origin and/or emission mechanism for the latter lines. The variability pattern of He II $\lambda$4686 and N III $\lambda \lambda$4634-41 between 1996 and 2000 reveals an enhancement of the core of the line while the red wing is progressively depleted.

These N III profile variations probably point towards a wind contribution to the emission. In this context, it is worth recalling that once that a velocity gradient exists, the fluorescence mechanism, suggested by Swings (1948) to account for the occurrence of the N III $\lambda \lambda$4634-41 emission lines in the spectra of Of stars, can become extremely effective (Mihalas 1973).

Figure 8: Aspect of N III $\lambda \lambda$4634-41, C III $\lambda \lambda$4647-50 and He II $\lambda$4686 for different years: the dispersion used in 1987, 1996 and 1999 is 33 Å mm-1, while it is 16 Å mm-1 in 1998 and 2000 (see Table 1). The C III lines show a very significant decrease in intensity between 1987 and the recent years, while all lines seem sharper in 2000. The heliocentric correction has not been applied to the spectra displayed in this figure.

The most impressive variability is unquestionably displayed by H$\alpha$, H$\beta$, H$\gamma$ and He I $\lambda$4471, as already quoted in Sect. 5. The long-term line profile variability of these lines is presented in Table 6 and Fig. 9: the changes are clearly seen. We do not have a lot of information about H$\alpha$, but the line is clearly variable. Our 1997 H$\alpha$ data yield a normalized peak intensity of 1.8, whereas in 1990 (Underhill 1994), it reached a value of 2.5. A similar behaviour was observed by Beals (1950): in 1938, the normalized intensity of H$\alpha$ was over 2, but it was only 1.3 in 1945. The H$\alpha$ line thus roughly behaves in the same way as H$\beta$, H$\gamma$ and He I $\lambda$4471. Other Balmer lines (e.g. H$\delta$, H$\epsilon$) and He I lines (e.g. $\lambda \lambda$4388, 4713) also display a similar behaviour, but we will now focus on the H$\gamma$, He I $\lambda$4471 and H$\beta$ lines, for which we have the most extensive data set.

These three lines displayed a P Cygni profile at the beginning of our observations in 1987. Then the emission component progressively weakened until the lines appeared completely in absorption. The first line that went into absorption is He I $\lambda$4471, followed by H$\gamma$ and finally H$\beta$. After this change, the depth of the He I absorption lines continued to increase gradually, as can be seen in Figs. 7 and 9. Since the change into absorption occurred only recently, we have no information yet about the behaviour of the H$\gamma$ and H$\beta$ lines after the emission has disappeared.

Interestingly, a similar transition was observed some 56 years ago (see Table 6): all three lines were seen as pure absorptions in 1944 and also in 2000. In 1934, He I $\lambda$4471 was already described as an absorption, while H$\gamma$ and H$\beta$ were still displaying P Cygni profiles. Then, H$\gamma$, followed very soon by H$\beta$, went into absorption. Finally, in 1954, a weak emission reappeared in H$\beta$. Therefore, the long term line profile variations of HD108 seem to be recurrent with a timescale of about 56 years.

However, the interpretation of the Balmer line profile variations is complicated by the fact that these lines are blended with He II Pickering lines. Following the same procedure as in Lamers & Leitherer (1993), we attempted to restore the H I line profiles by subtracting a fake He II line obtained by interpolation (H$\gamma$) and extrapolation (H$\beta$) from the unblended He II $\lambda$4200 and $\lambda$4542 lines. The reconstructed profiles are striking: the hydrogen lines no longer appear as P Cygni profiles, but as rather sharp and symmetrical emissions superimposed on broader absorptions (see the lower panels of Fig. 9). This emission fades with time, and finally nearly disappears in H$\gamma$ in 1999. The RVs of the "restored'' H$\beta$ and H$\gamma$ emission peaks become more negative during the last two years.

We have also tried to remove the photospheric He I absorption to recover the stellar wind line profile using two slightly different approaches. First, we assume that the emission has completely disappeared in the spectra observed in 2000, and use this profile for the He I photospheric contribution. Second, we evaluate the photospheric absorption by fitting a Gaussian profile to the blue wing of the $\lambda$4471 line, assuming it is not contaminated by emission. In this latter case, we set the RV of the photospheric He I component to be equal to the observed RV of the He II $\lambda \lambda$4200, 4542 lines. The resulting pure emission profiles are displayed in Fig. 9. We caution however that the reconstruction of the He I $\lambda$4471 wind profile is far more uncertain than the restoration of Balmer lines.

The rather narrow Balmer emission profiles without evidence for an associated P Cygni wind absorption are probably difficult to explain in a spherically symmetric model for the wind of HD108. In fact, the width of the lines indicates that they must be formed over a small range in radial velocity and hence the radial extent of the line-forming region in the expanding wind should be rather small. Under such circumstances, a spherically symmetric wind would most probably result in a pronounced P Cygni absorption component that is not observed.

On the other hand, the different behaviour of the He II $\lambda$4686 and N III emission lines on the one side and the He I and Balmer lines on the other side, clearly indicates that there must be at least two distinct regions in the atmosphere of HD108 where emission lines are formed.

Figure 9: Upper panels: aspect of H$\gamma$, He I $\lambda$4471 and H$\beta$ for some recent years. The lines gradually evolve from P Cygni profiles into pure absorption lines. Bottom panels: reconstructed profiles of the same lines (see text). The He II contribution has been subtracted from H$\gamma$ and H$\beta$ (bottom, left and right) while the He I photospheric component has been subtracted from the He I total profile (bottom, middle). The heliocentric correction has not been applied to the spectra displayed in this figure.

 

 

Table 6: Aspect of the H$\beta$, H$\gamma$ and He I $\lambda$4471 lines since 1919. The codes for the references are the following: Pl24 for Plaskett (1924), Me25 for Merrill et al. (1925), Sw42 for Swings & Struve (1942), Be50 for Beals (1950), Ma55 for Maninno & Humblet (1955), Ho68 for Houziaux & Ringuelet-Kaswalder (1968), An73 for Andrillat et al. (1973), Hu75 for Hutchings (1975), Un94 for Underhill (1994) and Ba99 for Barannikov (1999).

Date	H$\beta$	H$\gamma$	He I	reference
1919	P Cyg.	P Cyg.	P Cyg.	Me25
1921-1923	P Cyg.	P Cyg.		Pl24
1934-1938	P Cyg.	P Cyg.	Abs.	Be50
1941	P Cyg.	P Cyg.	Abs.	Ma55
1942	P Cyg.	P Cyg.		Sw42
				Ma55
1944-1945	Abs.	Abs.	Abs.	Be50
1950	Abs.	Abs.	Abs.	Ma55
1953	Abs.	Abs.	Abs.	Ma55
1954	P Cyg.	Abs.	Abs.	Ma55
	(weak)
1966	P Cyg.			Ho68
1968-1974	P Cyg.	P Cyg.	P Cyg.	An73
				Hu75
1976	P Cyg.	P Cyg.	P Cyg.	this work
1982-1985	P Cyg.	P Cyg.		Ba99
1986-1991	P Cyg.	P Cyg.	P Cyg.	Un94
1989-1991	P Cyg.	P Cyg.		Ba99
1991		P Cyg.	P Cyg.	this work
1993-1994		P Cyg.		this work
1996-1997	P Cyg.	P Cyg.	Abs.	this work
1998-1999	P Cyg.	Abs.	Abs.	this work
2000	Abs.		Abs.	this work