Astron. Astrophys. 363, 995-1004 (2000)

3. Data analysis

3.1. H line-profile variations

That the morphology of the H line of HD199478 is temporily variable becomes clear from a simple qualitative comparison of profiles arising from "snapshot" observations separated by months or years. For example, Rosendhal (1973) observed a double-peaked emission with a central absorption at +68km/s while Denizman & Hack (1988) detected profiles consisting of single blue-shifted emission and a weak absorption superimposed on the red emission wing at about +60km s^-1.

During our observations a large variety of H profile shapes were seen: emission which was almost symmetric and unshifted, with respect to the stellar rest frame; blue- or red-shifted asymmetric emission; double-peaked emission with different values of the V/R intensity ratio; triple-peaked emission and even a reverse P Cygni-type profile. In addition to the well-developed emission core, broad emission wings, extending to more than 1000 km s^-1 from the line center, are clearly discernible. Examples of representative profiles are displayed in Fig. 1. The spectra are shifted vertically by an amount that ensures better visibility. Two features on the right of H belong to the CII resonance doublet. Vertical lines represent the laboratory position of the lines.

Fig. 1. Examples of differently shaped profiles of HD199478 obtained during our observations. On the right of the profiles of the CII resonance doublet are seen. Vertical lines represent the laboratory position of each of the three lines.

To localise the variability in an objective and statistically rigorous manner, a simplified version, first reported by Prinja et al. (1996), of the so-called "Time Variance Spectrum" (TVS) analysis (Fullerton et al. 1996) was used. The method consists of a simple computation of the rms deviations, with respect to the mean for a given time-series line profile, as a function of wavelength, , under the additional assumption that the noise is dominated by photon noise and is nearly the same for each spectrum in the time series. The quantity was calculated using the following expression

where is the normalised intensity in the i-th of N spectra, and F() is the mean spectrum. In our case N=41. The averaged profile and the corresponding TVS are displayed in Fig. 2 as a function of velocity. The rms deviations averaged within the continuum windows is indicated by a dashed line. Statistical significance for variability is represented by a dotted line at a confidence level of 99% probability. Deviations above this level must be regarded as genuine variability. Fig. 2 indicates that the region of significant variability is concentrated within the line center and the blue extension of the profile, i.e. between -280 and +150 km s^-1. No significant variability in the emission wings of the line is indicated. The fact that the averaged profile is blue-shifted and single-peaked, although on several dates during the runs we have detected well-developed double-peaked emission, is certainly due to an observational sampling effect.

Fig. 2. The averaged profile and the rms deviations, , for the entire data set as a function of velocity across the line. The threshold for variability (p = 1%) is indicated by a dotted line. The dashed line represents the level of the deviations in the continuum.

The spectra obtained during the June-July 1998 and January-February 1999 observations are presented in Fig. 3 in the form of two-dimensional grey scale images ("dynamical spectra"). The spectra are rebinned to velocities within an interval of 48Å with the zero point set at the laboratory wavelength of the line. The intensities are converted into levels of grey according to the scales shown at the right-hand side of the top panels. Gaps between observations if equal or larger than 0.5 days are represented in black. Panels on the top show all spectra plotted within an appropriate intensity interval to display clearly the fluctuation at a specific velocity. The image portions of Fig. 3 indicate that the variability manifests itself by variations in position and intensity of emission peak(s). It appears that twice in 1998 and once in 1999 similar patterns of variations were recorded implying that some recurrent phenomenon is likely at work. The recurrent appearance of well-resolved, almost undisplaced absorption, with respect to the systemic velocity, that fades away without changing its position makes an impression. Note also the sudden appearance of weak red-shifted absorption in the second run of the 1998 campaign. The last persists for at least a few days at almost the same velocity, of about +50km s^-1, and seems to be similar to that observed earlier by Rosendhal (1973) and Denizman & Hack (1988). The observations do not give evidence for propagating blue-shifted absorption components similar to DACs observed in the UV (Bates & Gilheany 1990).

Fig. 3. Dynamical spectra of in June-July 1998 (left image) and January-February 1999 (right image). The top panels show an overplot of all profiles from the relevant time series. Time is measured in days. The grey-scale bar on the right of the top panels displays the intensity scaling.

Since the dynamical spectra provided clear evidence for periodic variability, we performed a period analysis, based on the Discrete Fourier Transform and the iterative CLEAN algorithm originally developed by Roberts et al. (1987) in FORTRAN and subsequently reproduced by one of us (T.V) in IDL, for each wavelength bin in the 1998 and 1999 series. The obtained results are shown in Fig. 4. The image portions of the figure display a gray-scale representation of the power at a given frequency as a function of position in the line. For the 1998 dataset (left panels of Fig. 4) the periodogram exhibits maximum power at a frequency of 0.0361 day^-1, which is concentrated at the center and red extension of the profile. This frequency is equivalent to a period of 27.70 days. The second highest peak is at 0.018 day^-1 (i.e. 55.5 days), which is longer than the observing period (42 days) and therefore unreliable. The other peaks are probably not significant. For the 1999 dataset (right panel of Fig. 4) the period analysis revealed the presence of periodic variation with a frequency of 0.026 day^-1, i.e. 38 days, which is concentrated at the center and blue extension of the profile.

Fig. 4. Temporal variance spectrum and 2d-Fourier transform created from the time-series in 1998 (left panels) and 1999 (right panels) The top panels show the average profile and the corresponding TVS. The middle panels display the power at a given frequency as a function of velocity (image portion) and the power summed over the line (right-hand panels).

To ascertain which of the line parameters is responsible for the periodic variations, detected through Fourier analysis, we measured the velocity of the emission peak(s) and the total equivalent width (EW) of the line. The obtained data are partially shown in Fig. 5 and Fig. 6 as a function of JD. Fig. 5 displays variations in velocity of the blue- and red-shifted emission peaks of over the June-July 1998 (upper panel) and January-February 1999 (lower panel) observations. The velocity of each feature was measured by bisecting the upper half of its profile. The measurements are corrected for the systemic velocity. The dashed and the solid lines shown in Fig. 5 represent sine curves:

with period P taken from the corresponding Fourier analysis and a, b and chosen (i.e. not fitted) such that the functions overlays the relevant data reasonably well. The intention of these curves is to guide the eye and to emphasise the periodicity detected by the Fourier analysis. The use of two sine curves (with the same period but of different scaling) is required by the fact that we have two sets of datapoints for each time-series: one for the blue and the other for the red emission peak of the line. The obtained results indicate that: (i) the variations are not symmetric with respect to the line center; (ii) within each observational series the variations in velocity are consistent with the periodicity detected by the corresponding Fourier analysis, i.e. 27.7 days for the 1998 time-series and 38 days for the 1999 time-series. The pattern of variability can be described in the following qualitative way: blue-shifted, asymmetric, single-peaked emission slowly evolves into a double-peaked emission, which in turns evolves into unshifted, almost symmetric single-peaked emission, which changes into asymmetric red-shifted single-peaked emission, which evolves into double-peaked emission and later into blue-shifted, asymmetric, single-peaked emission. We emphasize that the last two phases are not well documented by observations and must therefore be considered as suggestive.

Fig. 5. Variations in velocity of blue- and red-shifted emission peaks of as a function of JD for the 1998 (upper panel) and 1999 (lower panel) observations. The measurements are corrected for the systemic velocity. The velocity of blue- and red-shifted emission peaks is represented by dots and open circles, respectively. Sine curves with periods, derived by Fourier analysis, are plotted as a dashed and a solid line to indicate the periodic behaviour in velocity of the two emission peaks. The measurements from the first observational series are consistent with a period of 27.7 days while those from the second series - 38 days.

Fig. 6. Variations in EW of as a function of JD for the 1998 observational campaigns. Asine curve with a period of 55.5 days, derived by the corresponding Fourier analysis, is plotted as a solid line to guide the eye and to emphasise the periodicity detected.

The equivalent width of was estimated through line-flux integrating between 6556 and 6568Å. Following the analysis of Chalabaev & Maillard (1983) we estimated the internal precision of individual EW measurements by means of the formula:

where is the dispersion per pixel, M is the number of pixels including in the wavelength interval within which the integration has been done, is the measured equivalent width. The first term in this equation represents the contribution of the photometric uncertainty to the error, while the second represents the cumulative effect of the uncertainty in the continuum placement over the extent of the profile. In our case the latter effect turned out to be much stronger than the former, due to the large breadth of the wings, thus dominating the internal uncertainty of the EW determinations, which in turn was estimated to be always smaller than 10%.

The measurements indicate that the EW of varies within 50% of its mean value of -1.24 0.37Å, obtained as an overall data average. The variations must be genuine since their amplitude exceeds the uncertainty of the individual determinations. There appears to be no clear correlation between variations in EW and changes in position of the emission peaks of . Fourier analysis performed for the entire dataset did not give evidence for periodic variations in EW. However, an analysis of the data obtained during the 1998 observational campaign showed that the pattern of variability detected is consistent with a periodic variation with P=55.5 days (i.e. a period corresponding to the second highest peak in the relevant periodogram (see Fig. 4, left panel)). This finding is illustrated in Fig. 6 where the solid line represents a sine curve (scaled in an appropriate way) with P=55.5 days. This result implies that the emissivity may still vary in some regular way but on a time-scale that is either close to or larger than the length of the observing window and thus is difficult to detect.

Summarising, we conclude that the variability in HD199478 is expressed by strong variations in the shape of the profile due to cyclic variation in velocity and intensity of emission peaks accompanied by (cyclic?) variations in total equivalent width of the line. The time-scale of velocity variability is not constant with time. In particular, it equals 27.7 and 38 days for the 1998 and 1999 observations, respectively. It is not clear at present what the pattern of EW variability is and whether this variability is related to cyclic variations in the shape of the profile or not. Additional observations on a much longer time-scale are needed to resolve this problem adequately.

3.2. Behaviour of the line and the CII resonance doublet

To gain more information about the depth structure of the transition zone between the photosphere and the wind of HD199478 we examined the lpv of a number of absorption lines situated in the vicinity of , such as , and .

In contrast to Rosendhal (1973), we were always able to detect the CII resonance lines as weak, relatively narrow ( 70 km s^-1) absorption features superimposed on the red emission wing of (between 600 and 1000 km s^-1). The averaged CII profiles and the TVS for the entire dataset are shown in Fig. 7 as a function of wavelength. The level of variability in the continuum is indicated by a dashed line. The dotted line represents the threshold for variability (p = 1%). The shape of the TVS appears to be double-peaked, suggesting existence of radial velocity variability for these lines (Fullerton et al. 1996). The velocity width over which significant variability occurs, V, is equal to 120 and 90 km s^-1 for the and lines, respectively.

Fig. 7. The averaged profiles of the and lines and the TVS spectrum for the entire dataset as a function of wavelength. The level of variability in the continuum and the threshold for significant variability (p = 1%) are indicated by a dashed and a dotted lines, respectively.

The measurements show that the lpv detected manifests itself by variations in velocity and line-strength, i.e. EW. In particular, we found that the EW, determined by means of line-flux integration within the profiles, varies within 13% around a mean value of 0.22 (for ) and 0.16Å (for ). The variations must be genuine since their standard deviations exceed the accuracy of individual determinations (0.01Å), calculated by means of Eq. 3. To determine the velocity of the lines, we fitted a parabola to the lower half of their profiles and used the minimum of the fit as a measure of line position. The mean velocity, derived as an overall data average (N=41), equals -20.64.1 km s^-1) and -21.24.5 km s^-1 for and 6583, respectively. The rms deviations are larger than the internal uncertainty of our velocity determinations (1.6 km s^-1), specified by the rms deviation in velocity of the IS band at 6613, thus indicating significant variability in velocity of the CII resonance lines, in agreement with the result derived from the analysis of the TVS. The 2d Fourier analysis performed for the 1998 and the 1999 spectroscopic series did not detect any periodicity for the and lines. It may well be, however, that this is a result of observational selection. For example, cyclic variations on a time-scale close to or larger than the length of the observing window are difficult to detect.

In the part of our spectra (N=13) covering a wavelength range of 204Å, the line is observed as a relatively strong, narrow ( 80 km s^-1) symmetric absorption. The averaged profile and the TVS spectrum are displayed in Fig. 8 as a function of velocity. The rms deviations averaged within the adjacent continuum is indicated by a dashed line. The dotted line represents the threshold for variability (p = 1%). The distribution of reveals the presence of significant variability concentrated between -100 and +90km s^-1 (V =190km s^-1) within the profile, i.e symmetric about the systemic velocity.

Fig. 8. The averaged profile and the rms deviations, , for a subset of 13 spectra, as a function of velocity. The mean level of the deviations in the adjacent continuum is indicated by a dashed line. The dotted line represents the threshold for variability (p = 1%).

The measurements, performed as described above, showed that the velocity and the strength (i.e. EW) of the line are both variable: the former varies within 25% around a mean value of - 21.3 km s^-1 while the latter deviates within 7% around a mean value of 0.618Å. The variations must be real since their amplitudes exceed the uncertainty in the relevant determinations, which is equal to 10% in velocity and to 2% in EW. Due to the limiting number of available data no time-series analysis was performed for this line.

It is worth noting in addition that the absorption lines of our sample show almost the same values of radial velocity, which are furthermore equal (within the error) to the systemic velocity of -16 km s^-1 (Denizman & Hack 1988). This finding confirms the photospheric origin expected for these lines. However, the breadth of the lines (i.e. the FWHM) is larger than that of a photospheric line with no other broadening than stellar rotation (i.e. sini=45 km s^-1). This result indicates that one or more mechanisms exist that additionally broadens the lines. Recent studies, using both LTE and non-LTE techniques (Smartt et al. 1997; Gies & Lambert 1992), have shown that microturbulences of 15 to 30 km s^-1 appear to be required for B-type supergiants. Thus, we conclude that the line-width excess is likely due (at least partially) to microturbulence in the stellar photosphere.

Summarising, we conclude that absorption lpv consisting of continuous radial-velocity and line-strength (i.e. EW) variations certainly occurs in the spectrum of HD199478. The properties of the studied lines (e.g. radial-velocity and FWHM) and the parameters of their TVS (e.g. V) indicate that this variability is more likely coupled to processes in the stellar photosphere. The phenomenon could not be completely interpreted in terms of possible redistribution of a fixed amount of line absorption, as would occur if, e.g., the variability was caused by a macroscopic velocity field alone. Because variations in absorption EW do occur, the process or processes responsible for absorption lpv must also alter the number of absorbers, e.g. through changes in the ionization/excitation in the deepest layers of the atmosphere. All this indicates that absorption lpv in HD199478 is more likely caused by variations in velocity and temperature structures of the stellar photosphere.

The presence of significant absorption lpv in HD199478 implies that the variability of observed might not be due solely to variations in the physical properties of the wind but also might reflect changes in the underlying photospheric component. To test this possibility, we compared the behaviour of (in terms of EW variations) with that of the photospheric line. The result, illustrated in Fig. 9, shows that the two sets of data tend to anti-correlate such that an increase in the CII photospheric absorption appears to be accompanied by a decrease in the emission. This finding suggests that the variability could at least partially be assigned to variations in the underlying photospheric profile.

Fig. 9. Variations in the equivalent width compared to similar variations in . The two sets of data show a tendency to anticorrelate.

© European Southern Observatory (ESO) 2000

Online publication: December 5, 2000
helpdesk@link.springer.de