Our new orbital solution confirms the conclusion, drawn earlier on the basis of the interferometric data, that $\delta$ Sco is a binary system in a high-eccentricity orbit. The presence of an additional stellar component, which may be responsible for the short-term RV variations reported by van Hoof et al. (1963) and Levato et al. (1987), would result in a more complicated RV curve than the observed one. The short-term RV amplitude is well explained by the periastron acceleration. Perturbations in the velocity curve near periastron due to a third body would be easily detected in the Ritter data due to their high accuracy and dense coverage around this orbital phase.

Another important and not easily resolvable problem concerns the mass loss and the disk formation mechanism. All the spectroscopic data available show that the emission-line spectrum of the primary was stronger during the current (2000) periastron passage than during the previous one. This suggests that the amount of circumstellar matter, and hence the mass loss, grows with time. One possibility to induce the mass loss is the mass transfer through the inner Lagrangian point L₁, if at least one of the components fills its Roche lobe. According to our orbital solution, the smallest distance between the stars at periastron is d=0.06a, where a is the orbital semi-major axis. Using the components' mass ratio and the rotational parameter of the system F₁=0.12 (the ratio of the primary's angular rotation velocity and the mean orbital angular velocity), one can calculate that the equilibrium point at periastron is located at x_L₁= 0.61d from the primary's center (Kallrath & Milone 1998). This point has the same meaning as L₁ for close binaries in circular orbits. Since $d \sim 24~R_1$ , where R₁ is the primary's radius calculated from the bolometric luminosity and the spectroscopic $T_{\rm eff}$ presented in Sect. 1, and x_L₁=14.6 R₁, the primary is unlikely to fill its Roche lobe during a relatively fast periastron passage. The situation is almost the same for the secondary, which is supposed to be even smaller than the primary.

The presence of a cool secondary, filling its Roche lobe, can be ruled out by the following arguments. As we will show below, the line emission appeared before the optical brightening, while one would expect the opposite in the case of the Roche lobe filling by the secondary. Furthermore, it is easy to show that even at periastron, when an upper limit for the secondary's radius is $\sim$ 10 R₁, the system would brighten by $\ge$ 0.7 mag in the V-band if the secondary's $T_{\rm eff} \ge 5000$ K. If the secondary is cooler, then the overall brightness increase would be smaller, but the absorption-line spectrum of the secondary would be easily detectable. None of these phenomena was observed. Thus, direct mass transfer is not expected in the present binary orbit.

As mentioned above, Smith (1986) found $\delta$ Sco to show nonradial pulsations, which may result in mass loss. However, the UV spectra (Snow 1981) indicate that the mass loss from the star is extremely small. Nevertheless, $\delta$ Sco displays a detectable emission-line spectrum starting from 1990. The data from Coté & van Kerkwijk (1993) and our campaign show that the emission was observed near periastron, and that there was no noticeable increase of the emission between the last 2 periastron passages. Hence, one may argue that the emission strength increase correlates with the binary period and with the periastron passage time in particular. Since the components are very close to each other at periastron, tidal interaction between them may become significant enough to amplify the primary's pulsational instability and trigger an enhanced mass loss. This is a working hypothesis which needs to be verified by modelling. Alternatively, an unrelated growth of the disk mass (such as is seen in other Be stars) may have been coincidentally underway at the time of this periastron passage.

Let us now study the combined photometric and spectroscopic behaviour of $\delta$ Sco. The photometric data obtained before 1990 reveal an almost constant visual brightness, $V=2.32\pm0.01$ mag. The only different result (V=2.21 mag) was published by Hogg (1958). However, comparison of his photometry ( $V_{\rm H}$ ) with the modern Johnson system data ( $V_{\rm J}$ ) shows that the colour difference was not properly taken into account in Hogg's catalogue. For example, for blue stars $V_{\rm H} \le V_{\rm J}$ , while for red stars the situation is reversed. As a result, Hogg's measurement does not differ from other values within the inter-system translation errors.

Between 1990 January 22 and 1992 July 24 the star was monitored by the HIPPARCOS satellite (ESA 1997). This period covered the previous periastron passage, and the data show no noticeable variations. The mean brightness registered by HIPPARCOS, ${\overline V}=2.29\pm0.01$ mag (translated into the Johnson system using a formula by Harmanec 1998), is very close to the normal brightness of the star. The HIPPARCOS value is 3 per cent brighter which is not significant due to statistical errors of the translation. Thus, there is no evidence for any photometric variation of $\delta$ Sco before the year 2000.

The 2000 photometric observations began visually on June 26 (Otero et al. 2001). By July 4 the star was marginally brighter than usual ( $m_{\rm vis}=2.24$ mag). The first spectroscopic observations of $\delta$ Sco known to us in 2000 were obtained by the French amateurs D. and S. Morata on June 2. They show that the H $\alpha$ line was nearly as strong as in our first spectra. Between July 16 and 28 the star brightened up to $m_{\rm vis}=1.9$ mag and then experienced a $\sim$ 2-month brightness minimum (hereafter referred to as "the dip''), which coincided in time with the H $\alpha$ EW and intensity maximum and the line RV minimum (see Fig. 4). The dip ended in the beginning of October, and the brightness remained stable at $m_{\rm vis}=1.89\pm0.02$ mag until October 18, when the star became inaccessible. The new observing season started in late December at a similar brightness level. Although the results of visual observers differ up to 30 per cent, the mean star's magnitude through March 2001 is $m_{\rm vis}=1.95\pm0.15$ .

The disk's growing size and optical depth resulted in two major effects. The disk contribution to the overall system brightness through free-free emission and a partial attenuation of the primary's surface both increased with time. An interplay between these factors was responsible for the dip's shape. In June 2000 the disk was apparently small. It produced the H $\alpha$ emission, while its continuum emission was roughly compensated by the attenuation so that the overall brightening of the system was small. This also suggests that the starting point of the observed enhanced mass loss took place not long before the first spectroscopic observation of $\delta$ Sco in 2000.

In July the disk emission grew faster than the attenuation. At some threshold point, the decreasing separation between the stellar components limited the disk's ability to grow further before periastron because it reaches the primary's Roche lobe size. This implied a density increase on the side of the disk facing the secondary which, however, smeared out quickly because of the disk's rotation, with a period of a few days. The disk's enhanced density caused its larger optical depth, which resulted in the overall brightness fading. After periastron the components' separation began to increase, allowing the disk to grow freely again. As a result, the disk's density decreased, and the dip ended. The disk's density variations can be seen in Fig. 7, where larger values of the relative peak separation correspond to larger densities (see Hanuschik et al. 1988). It is also seen that the density dropped noticeably 5 months after periastron (triangles in Fig. 7). Thus, a combination of the primary's attenuation and the disk optical brightness, which depends on both its optical depth and its spatial extent, is capable of a qualitative explanation of the dip occurrence.

$\begin{figure} \par\includegraphics[width=6.5cm,clip]{aa1284f5.eps}\end{figure}$

Figure 5: The binary orbit in the plane of the sky. The speckle interferometry data are shown by filled circles. The primary component is placed at the origin of the coordinate system, and is marked with a cross. The dashed line represents the orbital solution from Bedding (1993), the dashed-dotted line shows the solution from Hartkopf et al. (1996), and the solid line shows our solution.

$\begin{figure} \par\includegraphics[width=7cm,clip]{aa1284f6.eps}\end{figure}$

Figure 6: The mean RV of the H $\alpha$ profile from the Ritter (filled circles), CAO (open squares), and ESO (filled square) data. The measurements close in time to each other are averaged. Their uncertainties are of the order of the point size. The observational data are shifted by +6 kms^-1 to account for the systemic velocity. The line types correspond to those in Fig. 5.

$\begin{figure} \par\includegraphics[width=6.8cm,clip]{aa1284f7.eps}\end{figure}$

Figure 7: The relative peak separation of the H $\alpha$ line versus its EW. The data obtained before periastron are shown by filled circles, those obtained after periastron in 2000 by open circles, and the data obtained in 2001 by triangles. The arrow-head lines show temporal trends before (solid) and after (dashed) periastron in 2000, and in 2001 (dotted). The solid line represents the average relationship between the displayed parameters for Be stars found by Hanuschik (1989).

This scenario explains most details of the system behaviour in 2000. The observed behaviour of $\delta$ Sco during the last binary cycle suggests that close periastron passages play an important role in the mass loss process observed in this binary system. However, there are a number of problems to be solved. These include the mass loss mechanism, the nature of the secondary, and the temporal evolution of the Fe II 5317 Å line (see Fig. 2d). They will require follow up observations and modelling.

The overall behaviour of $\delta$ Sco during this period of outburst provides a chance to see the onset of the emission line activity, which can reveal fundamental information about the building of stellar envelopes and the causes of the Be phenomenon. In this sense, it is interesting to compare the present $\delta$ Sco episode with the onset of activity in other Be stars. One remarkable and well-studied example is $\mu$ Cen, which in the 1980's presented several activity episodes arising from a quiescent status characterized by an absorption-line spectrum (Peters 1986; Hanuschik et al. 1993). A first comparison shows that the time scales involved are very different. Outbursts in $\mu$ Cen had a fast rise lasting 2-10 days, and a slower decay during one to two months. Conversely, the current $\delta$ Sco outburst is still in the rising phase (by June 2001), one year after the first detected activity. The smooth increase of emission line strength and visual magnitude in $\delta$ Sco is much more like the onset of activity in $\gamma$ Cas in the 1950's after several years of quiescence, leading to its current active phase which has lasted more than 50 years by now (Cowley & Marlborough 1968; Doazan et al. 1983). $\mu$ Cen also experienced a long lived outburst which started in 1989 and led into an active phase lasting to the present time. Superimposed on this long term active phase, several short lived outbursts also occurred (Rivinius et al. 1998). From such comparisons we might differentiate two types of outbursts in Be stars: 1) short outbursts, produced by episodic mass ejection, which give rise to a non-stable circumstellar envelope that dissipates in few weeks; 2) long outbursts, produced by continuous mass ejection, which allow the formation of stable circumstellar envelopes that last several decades. $\mu$ Cen presents both types of outbursts, while $\gamma$ Cas, and so far $\delta$ Sco, present only the long lived ones.

5 Discussion