4.1. Photospheric lines
Three of the five instruments used during the MUSICOS 96 campaign have a wide enough spectral coverage to give access to a large number of photospheric lines: these are the 2 MUSICOS spectrographs (Hawaii and the Canaries), and the OHP Elodie spectrograph. The time coverage provided by these 3 sites is only partial, due to their longitude distribution, and to the bad weather conditions at CFHT and OHP during the campaign. However, some very important conclusions can be drawn from these data.
4.1.1. Mean photospheric profile
We used the Least-Square Deconvolution (LSD) technique (Donati et al. 1997) to analyze the variations of a "mean" photospheric line. In this method, a line pattern function is constructed, containing all the lines supposedly present in the spectrum as Dirac functions, with heights set to the central line depths as calculated by Kurucz's (1979) "SYNTHE" program. The observed spectrum is then deconvolved with this line pattern function, yielding a "mean" photospheric line profile. With this technique, line blends are automatically taken into account when all the lines present in the spectrum are considered. Note that the depth of the resulting "mean" line has no physical meaning, but that time variations of this depth, as well as line profiles can be accurately analyzed with this technique. We used a Kurucz model for K and log , appropriate for AB Aur, for constructing the line pattern function. However, the photospheric spectrum of this star being somewhat peculiar (Böhm & Catala, 1993), we chose to include in the line pattern function only those lines that appear clearly in the high signal-to-noise spectra obtained during this campaign. In addition, we considered only the lines belonging to the common spectral domain of the three cross-dispersed echelle spectrographs used in the campaign. A total of 75 lines were finally used in this analysis. In the following of this work, we shall call "mean" photospheric line, the line constructed with the LSD technique, while the adjective "average" will be reserved for time averages of line profiles.
The shape and variability of the mean photospheric line profile, revealed for the first time with this level of precision thanks to the LSD analysis, is amazing. Fig. 1 displays the mean photospheric profile, additionally averaged in time over the whole MUSICOS 96 series, compared to two examples of individual profiles taken from the campaign. All the mean photospheric profiles obtained during the campaign are highly asymmetric, with a blue side deeper than the red one. We note that the red edge of the line is roughly constant over the whole series, while the blue edge moves back and forth, and the shape of the central parts of the line is highly variable.
The basic characteristics of the photospheric lines are the same in the 1991-1992-1994 spectra, a sample of which is displayed in Fig. 2, and compared with one typical spectrum of the 1996 series, computed using the same lines in the LSD procedure. The same kind of asymmetry is present. We also note that the red edge of the line tends to vary less than the blue and the central parts of the line, not only in the sample spectra presented in Fig. 2, but in all spectra of the 1991-1992-1994 series. Finally, we see that the short-term variations of the line in 1996 are at least as large as the long-term variations between 1991 and 1996.
As a first tentative conclusion of these basic characteristics of the photospheric line, we suggest that the blue side and central parts of the line are formed in regions of the photosphere that have peculiar temperature and/or velocity patterns. The red edge of the line, essentially constant throughout the MUSICOS 96 campaign, and on the longer term between 1991 and 1996, either reveals unperturbed regions of the photosphere, or corresponds to a quasi-permanent and constant accretion onto the stellar photosphere.
Using a sample of the strongest photospheric lines from spectra obtained in Dec. 91, Böhm & Catala (1993) concluded that AB Aur's photospheric spectrum was that of a normal A0V star, and determined a projected rotation velocity of km s-1.
Catala et al. (1993) also presented one spectrum, obtained on Jan. 21, 1991, where the Fe II 5018 Å line (multiplet 42) appears in absorption and symmetrical, while in all other spectra of AB Aur this line is purely in emission and highly asymmetric. On that occasion the profile of this line was quite compatible with a normal A0V photosphere with a projected rotation velocity of 80 km s-1, in agreement with the value determined by Böhm & Catala (1993), and a stellar heliocentric radial velocity of 21 km s-1, confirming the value measured by Finkenzeller & Jankovics (1984) using the higher Balmer lines.
We must then conclude that the vsini value of AB Aur cannot be much higher than 80 km s-1.
On the other hand, with the LSD method used here, we find that the mean photospheric line is always asymmetric and wider than a standard rotational profile of 80 km s-1, not only in our data of the 1996 MUSICOS campaign, but also in the re-analyzed data of 1991, 1992 and 1994. A projected rotational velocity of the order of 140 km s-1 would be necessary to reproduce the red edge of the line. Such a high value for vsini would be inconsistent with many of the spectra obtained during the 1991, 1992 and 1996 campaigns, which are narrower, with the profiles of the strong photospheric lines used in Böhm & Catala's analysis, and with the profile of the Fe II 5018 Å line of Jan. 21, 1991.
We cannot invoke a value of the star's radial velocity much higher than 21 km s-1, as it would be inconsistent with both Finkenzeller & Jankovics's measurements, and the Fe II 5018 Å line of Jan. 21, 1991. We also note that Finkenzeller (1983) concluded that AB Aur was not a member of a multiple system, so that we do not expect significant changes in its radial velocity.
Thus, in order to make the red edge of the mean photospheric line in our 1991, 1992, 1994 and 1996 data consistent with the previously determined values of vsini and the radial velocity, we need to assume
Turbulent velocities as high as 45 km s-1 were deduced in the wind of AB Aur (Catala & Kunasz 1987), and even higher values of the turbulent velocities in the wind are needed to account for the shape of the newly discovered N V resonance lines (Bouret et al. 1997). It seems clear that the variable parts of the photospheric lines (central and blue parts), which are likely to be partly formed at the base of the wind, may also be affected by strong turbulent motions. In the hypothesis discussed here, we also need strong turbulent motions in the unperturbed layers of the photosphere. However, we do not have any other independent evidence for such motions in the unperturbed photosphere, except that a high level of turbulence in the photosphere is predicted by the model of Lignières et al. (1996), which includes a mixing layer.
For the following analysis, we will therefore pursue the former hypothesis, assuming a non-turbulent photosphere with a projected rotation velocity of 85 km s-1, which is the upper limit derived by Böhm & Catala (1993). We note however that most of the conclusions presented below concerning the time variability of the photospheric lines do not depend much on the choice of the basic photospheric profile.
A photospheric rotational profile computed with the assumption cited above is also shown in Fig. 1, plotted in the IS Na I D line reference frame (i.e. shifted by 5 km s-1 with respect to the stellar rest frame). The depth of the computed profile was adjusted in order to reproduce correctly the shape of the mean photospheric line of a reference A0V star, HR 3314, calculated with the LSD method using the same lines as for AB Aur.
We also find that the shape of the mean photospheric line depends significantly upon the choice of lines used in the averaging procedure of the LSD method. We have computed the average photospheric line, using lines of Fe II, Fe I, Cr II, Mg II, O I, Ti II, and Si II, as well as choosing only lines from neutral species, or only lines of singly ionized species, or only strong lines. These various mean lines differ from each other. Fig. 3 presents an example of this behavior, and displays the mean photospheric profiles obtained with all lines, with only those of Fe II, and with only the strong lines of Si II, Mg II and O I. These results must be manipulated with care, as the LSD analysis is subject to line blending when only line subsets are considered. Before applying the LSD algorithm to particular line subsets, we carefully examined the potential presence of strong blends, and removed from the list lines which could suffer from such blends. Also, in order to optimize the final number of lines considered, we used only the spectra obtained at CFHT for this part of the analysis, as they provide the widest spectral domain. The results show that the mean profile obtained with only strong photospheric lines is much more symmetrical than that computed with all lines, probably because of the saturation of strong lines. On the contrary, Fe II lines seem to depart much more from a symmetrical profile. In particular, they show a very prominent emission in the central and red parts of the line, at the same velocity where we see a flux plateau in the mean photospheric profile computed with all available lines. We have checked that the behavior of the Fe II lines described above is representative of that of all moderate and weak lines in the spectrum of AB Aur. It seems clear that the phenomenon responsible for the departures from pure unperturbed photospheric profile affects the formation of most lines, rather than simply the adjacent continuum, as in the case of dark spots. It is quite reasonable to conclude from the remarks above that at least part of the observed line asymmetries are due to the presence in the stellar atmosphere of hot gas overlying some areas of the photosphere.
4.1.2. Photospheric line dynamical spectrum
From this point on, we will refer only to the mean photospheric line obtained with the LSD technique using all the photospheric lines present in the spectrum.
Fig. 4 presents the standard deviation across the line profile for the whole series, divided at each velocity by the mean photospheric line profile, averaged over the whole series. This procedure allows us to examine the relative variations of the line profile and to compare these relative variations in the different parts of the line. It can be readily checked that the line variations are real, with a main peak near zero velocity, and a secondary peak near -80 km s-1, corresponding to the blue side of the line. The maximum standard deviation in the line reaches 1.3%, while the standard deviation in the continuum adjacent to the line, giving a measure of the noise level, is only of the order of 0.3%.
We then subtracted, from each profile in the series, the rotational profile computed with the assumption of a non-turbulent phostosphere with vsini=85 km s-1. The resulting dynamic image is displayed in Fig. 5. The residuals show almost systematically a blue absorption component, an emission component slightly redward of line center, and a red absorption component. In the following, we shall call red component the emission component which is slightly redshifted, and red edge component the redmost component appearing in absorption. A modulation of the blue component is clearly suggested, with a period near 30 hours, while no clear modulated displacement of the red and redmost components can be seen. The red component is redshifted by 25 km s-1 on average, while the redmost component shows a mean redshift of 95 km s-1. We detect strong changes in the intensity of the red emission component, which are primarily responsible for the variations in the equivalent width reported in Sect. 4.1.4.
The redmost absorption component, which is constant in velocity, may reflect the fact that the baseline rotational profile that was subtracted from each one of the spectra does not correspond to the real underlying photospheric spectrum. This would be the case for instance if some high level of photospheric turbulence was present in AB Aur. Although, as argued earlier, we have no decisive evidence for such supersonic turbulence in the photosphere, we consider that the existence of this redmost absorption component is doubtful, and do not discuss it in detail in the rest of this paper.
In order to quantify the modulations displayed by the dynamical spectrum displayed in Fig. 5, and as a first step of a systematic analysis of these variations, we performed a period search across the photospheric line, in the following way. The line was divided into velocity bins 25 km s-1 wide, and a periodogram of the line intensity averaged over 25 km s-1 was calculated for each one of these bins. The periodogram is defined as:
where the are the data points, i.e. the intensity in the line after subtraction of the rotating profile, averaged over bins 25 km s-1 wide, and the the values of a fitting sine wave, computed for each trial period with the amplitude and the phase as free parameters. Such a periodogram is equivalent to the more classical Scargle (1982) periodogram, and has the additional advantage of providing an estimate of the phase and amplitude of the potential periodic variations. We constructed this periodogram, calculating for a set of 500 trial periods, ranging from 20 to 70 hrs.
Fig. 6 presents the result of this period analysis. If we except the peaks occurring at periods shorter than 28 hrs, dominated by the aliases produced by the gaps in the photospheric data, and the signal arising at very long periods, whose origin is uncertain, we find that the most prominent features in the part of this periodogram corresponding to the velocity range spanned by the photospheric line, are three major peaks, located at the velocities of the three components identified on the dynamical spectrum of Fig. 5: the slightly redshifted component and the red edge of the line exhibit two distinct peaks, both with a period near 43 hrs, while the third peak in the periodogram, corresponding to the blueshifted component, appears near 37 hrs. The two periods detected in this periodogram (37 and 43 hrs) appear significantly different. These results indicate that the variations of the photospheric line contain significant periodic components, but also suggest clearly different behaviors of the blue and red sides of the line. This first and natural attempt at analyzing the periodicity present in the mean photospheric profile motivated a more detailed investigation, which is presented below.
4.1.3. Components of the mean photospheric line
Guided by these results, we studied separately the three components appearing in the mean photospheric line after subtraction of the computed rotational profile introduced earlier. In each spectrum of the series, we identified "by eye" each one of these components, fitted it by a gaussian, and determined its centroid, intensity and width. The most interesting result of this analysis is displayed in Fig. 7, which shows the centroid of the blue component as a function of time. This component has a radial velocity modulated between -100 km s-1 and -40 km s-1.
The centroid of the blue component indeed seems periodically modulated. In order to quantitatively test this hypothesis, we computed a periodogram for the time series corresponding to this centroid, with the same method as described earlier Eq. (1). The best result is obtained when the last few data points (after t=160 hrs) are given a zero weight in the fitting, as there appears to be a strong change in the variability pattern around that time. The periodogram is displayed in Fig. 8, and indicates that the variations of this centroid are modulated with a period near 34 hrs. Using the definition of the periodogram given by Scargle (1982) instead of Eq. (1) (Scargle has demonstrated that the two approaches are equivalent), and applying the statistical analysis presented in that paper, we find a false alarm probability (i.e. the probability that this 34 hr peak is due to noise) of the order of for our 500 trial period periodogram. The width of the corresponding maximum in the periodogram gives us a rough estimate of the error bar on this period. We find hrs. This period is close to, and in any case within the error bar of, that found by Catala et al. (1986b) for the Ca II K line of AB Aur ( hrs). It is also consistent with the peaks corresponding to the blue component in the two-dimensional periodogram of the intensity in the photospheric mean profile, presented in Fig. 6. The sine wave fit corresponding to the highest peak in the periodogram (period: 33.7 hrs) is plotted in Fig. 7, to be compared to the time variation of the centroid of the blue photospheric component. The data, rephased with the 33.7 hr period, are shown in Fig. 9. The modulation is obvious on this figure, the data points at km s-1 near phases 0 and 1, which are the only ones departing from the modulation, corresponding to intervals of time after t=160 hrs, when the variability pattern obviously has changed. We also checked that replacing our data by a random time series produces no signal in the periodogram near 34 hrs. Figs. 6, 7, 8 & 9 convincingly demonstrate that the blue component velocity is indeed modulated with this period, although a certain level of additional variability is present on top of the strictly periodic modulation. The other two peaks appearing in the periodogram (near 22 and 26 hrs) are not real, but related to the gaps in our data set. We find that they still appear in the periodogram of a pure sine wave with period 33.7 hrs, restricted to the same time coverage as our data.
Compared to the variations of the blue component velocity, the red emission and redmost absorption components do not appear variable in velocity. The red component radial velocity is constant except at times when the blue absorption component is near its reddest location. It is not clear if the small variations measured in the red component velocity, ranging between 10 and 30 km s-1 with an rms dispersion of 10 km s-1, are real or simply reflect the fact that the blue absorption component is sometimes eating up the blue edge of the red component, thus resulting in an apparent redshift of the latter. The redmost absorption component is even more constant in velocity, with an rms dispersion of only 4 km s-1.
4.1.4. Variations of the photospheric line equivalent width
The total equivalent width of the mean photospheric line is highly variable, with an rms variation of 1.5 km s-1, i.e. 38% of its average value. We find that these variations, displayed in Fig. 7, present some regularity, including a seemingly periodic modulation. We note that these variations are primarily due to the red component, but the measurement of the intensity of this component alone is extremely difficult, and we have preferred to present the time variation analysis of the total equivalent width instead, which on the contrary is straighforward to measure.
As shown in the 2D periodogram of the intensity in the mean photospheric line presented in Fig. 6, the modulated component of these variations is mainly due to the red components of the line. A periodogram analysis of the equivalent width variations, similar to that performed on the centroid velocities, indicates that a 43.1 hr period may be present in the modulated component of these variations. This periodogram is shown in Fig. 8. This period is significantly distinct from the one we see in the velocity variations of the blue component, in agreement with the results of the 2D intensity periodogram of Fig. 6. As expected, the additional spurious peaks near 22 and 26 hrs, due to data gaps, are also present in this periodogram. However, a close inspection of Fig. 7 reveals that the 43.1 hr period found in the analysis is primarily due to two deep minima in the equivalent width near t=65 hrs and t=150 hrs. The periodicity of the equivalent width variations is therefore unclear, and the peak in the corresponding periodogram, as well as the corresponding peaks displayed by the 2D intensity periodogram of Fig. 6, may be fortuitous. Clearly, additional data would be required to check this point.
4.2. Search for circular polarization in the photospheric lines
The MUSICOS spectropolarimeter was used on the 3.6m CFHT for this campaign, in the hope of detecting directly a surface magnetic field in AB Aur, through the measurement of circular polarization Zeeman signatures in the line profiles. However, the seeing and transparency conditions experienced at Mauna Kea during the campaign were particularly bad, and the signal-to-noise ratios obtained in our spectra did not meet our expectations, by a large factor.
No signal was detected in the spectra of the V Stokes parameter for the average photospheric line, in any of the AB Aur spectra obtained with the instrument. The final 1- upper-limit for the strength of a net radial field in a magnetic region covering 2.5% of the total stellar surface, and facing the observer, would be of the order of 300 G.
4.3. The He I D3 line
The He I 5876 Å line most often appears in emission in AB Aur, as shown by Böhm et al. (1996). The spectacular variability exhibited by this line during the MUSICOS 92 campaign, as well as the indication that this variability may have a strong periodic component, motivated us for choosing it as the center of the MUSICOS 96 campaign. Therefore, all instruments involved in the campaign observed this line in the best possible configuration. As a result, the time coverage obtained on this line is better than that of the photospheric lines.
As in 1992, the variability of this line is amazing. Fig. 10 shows the mean profile for this line, averaged over the whole series, with two individual mean profiles showing the high level of variability. A quantitative measure of variability across the line is obtained by computing the standard deviation in each velocity bin and dividing this by the average flux in this velocity bin. The result of this analysis is displayed in Fig. 11.
4.3.1. Dynamical spectrum of the He I D3 line
A dynamic spectrum of the He I D3 line is presented in Fig. 12. The variability of the central part and red side of the line is in great part due to a single dramatic event, around t=65 hrs, when a deep and broad absorption appears on the red side and in the central part of the line. One of the spectra obtained during this event is presented in dashed line in Fig. 10. This phenomenon repeats itself, although with lower amplitude, near t=170 hrs. This event is reminiscent of a similar one, observed during the MUSICOS 92 campaign, with the same characteristics.
In addition to these strong events, we also note a strong variability present all along the series, as in 1992, with the following characteristics:
We calculated a 2D periodogram of the He I line intensity averaged over velocity bins of 25 km s-1, with the same period analysis method as applied to the photospheric lines, exploring 500 trial periods between 20 and 70 hrs. The result is displayed in Fig. 13, and clearly shows two separate peaks, corresponding respectively to the blue and the red components described above, appearing at the same period near 43 hrs. Note that, the time coverage of this series being much better than that of the photospheric lines analyzed earlier, this periodogram does not show any feature at short periods as the one of photospheric lines did.
We have therefore separated the analysis of both components, and this further investigation is presented below.
4.3.2. Components of the He I D3 line
Each profile in the series was fitted by a sum of 2 gaussians. This fitting procedure is very satisfactory, and the line can be successfully fitted by these 2 independent gaussians at most times during the campaign. However, the solution of the fitting is definitely not unique when the 2 components occur at similar velocities (leading to an almost symmetrical broad profile for the He I D3 line). Our automatic procedure will systematically choose two separate components, with moderate widths, rather than two wider components at the same velocity. We must keep this caveat in mind when discussing the results, and consider that the blue component can be somewhat redder than determined when it approaches the line center, while the red component can be bluer. Furthermore, the automatic procedure also leads to spurious results concerning the centroid of the blue component during phases when the red component appears strongly in absorption, i.e. near t=65 hrs and t=170 hrs.
Fig. 14 presents the centroids of these 2 gaussians as a function of time. Note that we have omitted the data points corresponding to phases when the red component is in absorption, near t=65 hrs and t=170 hrs, because of the problems mentioned above. It can be verified that the centroid of the red component is much less variable than the blue one. In fact, the centroid of the red component is more or less constant all through the campaign, except near t=180 hrs. The strong variation of this component is therefore primarily due to dramatic changes in intensity without much velocity variation.
The variations of the blue component centroid occur on a shorter time scale between t=0 and t=65 hrs than after t=65 hrs. A periodic modulation with a period near 45 hrs is suggested in the series after t=65 hrs, whereas a period twice as short seems to prevail between t=0 and t=65 hrs. Note that the moment when the double wave is changed into a simple wave coincides with the dramatic absorption event described earlier.
We applied to these data the same period search method as used in the case of photospheric lines. The results are presented in Fig. 15. We find a clear maximum of the periodogram near 45 hrs. Applying the same statistical analysis as for the photospheric line variations, derived from Scargle (1982), we find a false alarm probability near for this period in our 500 trial period periodogram. The corresponding peak in the periodogram is rather wide, and indicates a period hrs. This error bar, although it is wide, does not include the 34 hr period derived for the variations of the blue component of the mean photospheric line.
Fig. 16 presents the data rephased with a period of 45.1 hrs (period giving the best sine wave fit), plotting separately for the first and second halves of the data. We note that the data of the second part of the campaign (after t=65 hrs) are indeed consistent with a simple periodicity with P=45.1 hrs, while those of the first part of the campaign (before t=65 hrs) are distributed on a double wave in the phase diagram calculated with the same period. However, this conclusion is weakened by the fact that we see the double-wave for only 1.5 times the period. It may be worth to note that this double wave is not symmetric, which could be interpreted in terms of the presence of two different structured areas almost opposite to each other, one of which would disappear after 65 hrs.
Finally, we find no correlation between the variations of the centroid of the blue component of the He I line and those of the blue component of the mean photospheric line.
4.3.3. Variations of the equivalent width of the He I D3 line
In addition to the results presented above, we measure strong variations in the total equivalent width of the He I D3 line, with an rms dispersion of 20 km s-1, i.e. 56% of its mean value. The variations are dominated by the red half of the line, from line center redward, for which the rms dispersion is 100% of the average equivalent width. However, the measurements of the equivalent width of the He I line blue and red components are made difficult by the cross-talk between the two components that we have mentioned previously, and we have preferred to analyze the total equivalent width of the line, which is a more reliable and easy to measure quantity.
The variation of the total equivalent width reaches 200% during the absorption event at t=65 hrs. Fig. 17 displays these variations, compared to those of the equivalent width of the mean photospheric line during the campaign. We notice that the two data sets are correlated, with a correlation coefficient of 0.70. This correlation is shown in Fig. 18.
The variations of the equivalent width of the He I D3 line seem to be modulated with the same period as those of the centroid of the blue component, as shown in Fig. 15. The false alarm probability of the corresponding peak in the periodogram is . This period, near 45 hrs, is also close to that displayed by the equivalent width of the photospheric lines, whose variation is also dominated by the red half of the line. However, while the periodic appearance of the equivalent width variations in the photospheric lines may be fortuitous, due to moderate time coverage of our observations, there seems to be no doubt about the periodicity of the equivalent width of the He I line, which was much better covered by our observations (see Fig. 17). The correlation between these two quantities, shown in Fig. 18, therefore adds some credibility to the 43 hr periodicity found in the photospheric line equivalent width.
We detect a strong correlation between the centroid of the blue component of the He I D3 line and the total equivalent width of this line, with a correlation coefficient of 0.74. It seems that this correlation is driven by the intensity variations of the red component, which tends to be stronger when the blue component is near zero velocity. However, because of the cross-talk between the two He I components, we cannot exclude that it simply reflects variations of the blue component intensity accompanying its velocity variations. In this case, both the correlation discussed here and the 45 hr period detected in the total equivalent width variations would be attributable to the blue component alone.
4.4. The H line
The H line of AB Aur has been observed repeatedly in the past. It appears most often as a type II P Cygni profile, that is with an intense redshifted emission and a blueshifted absorption component. Occasionally, this line exhibits a single component emission and no absorption, or a type III P Cygni profile, i.e. with an additional blueshifted emission component on the blue side of the absorption component (Beskrovnaya et al. 1991, 1995).
We observed the three types of profiles during the campaign. Figs. 19 and 20 give an illustration of the high level of variability of this line. We note that the relative variation of the emission component is much smaller than that of the absorption component. Although a few of our spectra are saturated near the top of the H emission component, this strong difference between the levels of variability of the blue and red component of the H P Cygni profile is real. In the following, we will discuss mainly the absorption component of the line.
Fig. 21 shows a dynamic spectrum of the line, focussed on the absorption component (the emission component is "saturated" on the scale used for the figure). It can be noted that this component varies in intensity, width, and shape. Fig. 21 suggests no obvious strictly periodic modulation of any part of the H line, although dramatic variability is exhibited by its absorption component. The blue edge of the H absorption component appears much variable, exhibiting what appears to be a set of partial sinusoids in velocity space, with amplitudes in the range 100-150 km s-1, and with timescales of the order of 40-50 hours.
We have systematically looked for periodicity in this line, using a similar 2D periodogram as in the case of photospheric and He I lines, exploring 500 trial periods between 20 and 120 hrs. The result is shown in Fig. 22, and indicates a very complex behavior of this line. We do see some power at a period near 50 hrs, and at a velocity of -300 km s-1, which may be due to a real periodic behavior, although we cannot attach a great level of confidence to this result. The other strong peaks appearing in this 2-D periodogram do not correspond to periodic phenomena, but simply to the occurrence of only two events. In that case the period of a given peak in the periodogram simply measures the time separation between the two occurrences of the event. Thus, the feature seen in Fig. 22 near -400 km s-1 and 70 hrs is related to the widening of the absorption component near t=58 hrs and near t=130 hrs (see Fig. 21). The other strong peak near -100 km s-1 and 90 hrs corresponds to the apparent widening of the emission component near t=40 and t=135 hrs.
We have therefore no strong evidence for periodic modulation of H in our data. Clearly, a more sophisticated analysis, such as tomographic back-projection (Horne, 1991), would probably constitute a better approach to this particular set of data, and give us quantitative information about the peculiar behavior of the H absorption component, but this further analysis of our data is deferred to a subsequent paper.
It is interesting to notice that a strong widening of the H absorption component appears near t=65 hrs, at the same time as the same kind of phenomenon is seen in the He I D3 line.
Finally, we note that a blue emission component, typical of type III P Cygni profiles, appears in spectra between t=80 and t=95 hrs, i.e., while both the blue edge velocity and the equivalent width of the blue component are smallest. This small emission is at a velocity which varies from -330 km s-1 to -310 km s-1 between t=80 and t=95 hrs. It does not re-appear later in the series, most likely because the absorption component always extends bluer than -310 km s-1 after t=95 hrs.
© European Southern Observatory (ESO) 1999
Online publication: April 28, 1999