5 The long-period orbit

The period is known from previous studies with an accuracy of about 0.02 days. The mean time interval between the MC spectra and our data is 6200 days (300 periods; 17 years), so there is no problem with determining the number of intermediate epochs. As a matter of fact, periods corresponding to actual cycles $\pm$1 can already be ruled out by the MC data. Therefore, a definitely more precise period could be determined using the broad time base between the MC and our data sets without any bias.

We first tried to obtain a solution of our radial velocity data alone. However, due to a gap in the phase coverage of the velocity curve, the correlation among spectroscopic elements turned out to be strong, and solutions tended to be non-unique: e.g., possible solutions implied a relatively broad parameter range for K₁. More decisive results were expected when our data were combined with the older published data. However, a combined solution of different data sets requires the assumption of different $\gamma$ velocities for widely separated epochs. In view of the long-term radial velocity changes due to the mutual orbit of the two binary systems A and B, different values of $\gamma$ velocities are to be expected for data with long time separations. MC also noted a discrepancy between $\gamma$ velocities obtained from different lines, so the intended combination of the MC and LMS data with our measurements required some caution. Therefore, we only considered He I measurements. Since LMS give velocities calculated from a combination of He I and Si IV lines, we restricted our data sample to the rather homogeneous set of He I velocities by MC and ourselves.

Our results are listed in Table 2 and compared with the original MC results. The solutions somewhat depend on the assumed data weighting. Since the rms values for MC velocities and for our data were about 12 and 7 kms^-1, respectively, we gave the MC velocities a weight of 1, to ECHELEC velocities a weight of 2, and to CAT velocities a weight of 3. An independent solution of the MC data confirmed the original results by MC.

The large rms value of our high resolution data is somewhat unexpected and must be due to intrinsic variability of unknown nature. The error of fitting the Gaussians to line profiles is not larger than 2 kms^-1. The deviations of the velocities from the anticipated orbital curve therefore represent real shifts of line positions.

In Fig. 5, velocities published by Feast et al. (1956) and Buscombe & Kennedy (1966) are also shown for comparison. Mean velocities per plate are used here in the case of Feast et al. data. They follow our A1 radial velocity curve reasonably, and their $\gamma$ velocity can be calculated as $-39.5 \pm 8.3$ kms^-1. It is however difficult to comment on the data by Buscombe & Kennedy. The apparent difference in the $\gamma$ velocities will be discussed in Sect. 7.

Figure 5: Radial velocities of the A1 component derived from He I lines. The theoretical curves are calculated according to elements in the last column of Table 2 for MC (full line) and present (dashed line) $\gamma$ velocities. Triangles correspond to MC data, plus signs to the present data, open circles to data by Feast et al., asterisks to those of Buscombe & Kennedy, and crosses are values by Stickland