1 Introduction

LS I +61$^{\circ }$303 is an X-ray binary system associated with the galactic plane variable radio source GT 0236+610 discovered by Gregory & Taylor (1978). Optical observations (Hutchings & Crampton 1981) show that the system is composed of a neutron star and an early type, rapidly rotating B0V star with a stable equatorial disk and mass loss. Spectral line observations of the radio source give a distance of $2.0 \pm 0.2$ kpc (Frail & Hjellming 1991).

One of the most unusual aspects of its radio emission is the fact that it exhibits two periodicities: a 26.5 day periodic nonthermal outburst (Taylor & Gregory 1982, 1984) and a 1584 day ($\sim$4 years) modulation of the outburst peak flux (Gregory et al. 1999). The 26.5 day periodicity corresponds to the orbital period of the binary system (Hutchings & Crampton 1981). This periodicity has also been detected in UBVRI photometric observations (Mendelson & Mazeh 1989), in the infrared domain (Paredes et al. 1994), in soft X-rays (Paredes et al. 1997) and in the H$\alpha$ emission line (Zamanov et al. 1999). The 4 year modulation has been observed as well in the H$\alpha$ emission line (Zamanov et al. 1999).

Simultaneous X-ray and radio observations show that the X-ray outbursts occur at the periastron passage while, on the contrary, the strongest radio outbursts are always delayed with respect to (Taylor et al. 1996; Gregory et al. 1999). Both the presence of two periodicities (at 26.5 days and $\sim$4 years) and the delay between radio and X-ray outbursts are well explained in the framework of an accretion scenario of a precessing neutron star in a highly (e>0.4) eccentric orbit (Gregory et al. 1989; Gregory et al. 1999). The accretion rate in an eccentric orbit within the equatorial wind of the Be star has two peaks: the highest peak corresponds to the periastron passage and the second, lower amplitude peak occurs when the relative velocity of the neutron star and the Be star wind is at a minimum. For supercritical accretion, matter is ejected outwards in two jets perpendicularly to the accretion disk plane. Near periastron, inverse Compton losses are severe (due to the proximity to the Be star): X-ray outbursts are expected but not radio ones. For the second accretion peak, the neutron star is much farther from the Be star and both inverse Compton losses and wind opacity are lower, the electrons can propagate out of the orbital plane and we observe the radio outburst. The precession of the disk gives rise to the $\sim$4 year modulation (Gregory et al. 1989; Taylor et al. 1992; Massi et al. 1993; Martí & Paredes 1995; Gregory et al. 1999). The presence of an accretion disk is also invoked by Mendelson & Mazeh (1989) to explain details of the optical light curve. Liu et al. (2000) explain the variation of the H$\alpha$ emission with orbital phase as varying irradiation of the Be star's circumstellar disk by the X-ray emission from the neutron star's accretion disk.

However, the luminosity of LS I +61$^{\circ }$303 in the X-ray range (1-40 keV), is only $L_{\rm X}\simeq10^{35}$ erg s^-1 (Maraschi & Treves 1981), which is three orders of magnitude lower than the Eddington limit. Moreover, LS I +61$^{\circ }$303 is the most promising candidate for the optical counterpart of the $\gamma$-ray source 3EG J0241+6103 (Hartman et al. 1999), with a luminosity $L_{\rm \gamma}\simeq10^{37}$ ergs^-1. The fact that LS I +61$^{\circ }$303 has the bulk of its energy shifted from X-ray to $\gamma$-ray wavelengths is not understood in the context of the supercritical accretion model, showing the necessity for alternative models. These models suggest that the 26.5 d outburst events are produced by energetic electrons accelerated in the shock boundary between the relativistic wind of a young non-accreting pulsar and the wind of the Be star. In this case the 4 year modulation would be explained due to cyclic variations of the Be star envelope (Maraschi & Treves 1981; Tavani 1994; Tavani et al. 1998; Goldoni & Mereghetti 1995; Zamanov 1995; Taylor et al. 1996).

The recent discovery of the microquasar LS 5039 (Paredes et al. 2000) brings new credibility to the accretion model. LS 5039 is also subluminous in the X-ray range (even more than LS I +61$^{\circ }$303) and also shows the same puzzling behavior, having $L_{\gamma} > L_{\rm X}$. Therefore, LS 5039 and LS I +61$^{\circ }$303 could be the first two examples of a new class of X-ray binaries with powerful $\gamma$-ray emission. In the case of LS 5039, an ejection process fed by an accretion disk is clearly proved by a map obtained with the Very Long Baseline Array (VLBA), which shows bipolar jets emerging from a central core. Its high $L_{\gamma}$ is tentatively explained by inverse Compton scattering. In LS I +61$^{\circ }$303, although several VLBI observations show a complex source extending over a few milliarcseconds (Massi et al. 1993; Peracaula et al. 1998; Paredes et al. 1998; Taylor et al. 2000), such a clear jet structure has never been observed.

In this paper we report on EVN observations probing structures on scales of tens of milliarcseconds, larger than probed by previous observations. We clearly detect on this scale an elongation of the emission (Sect. 2), which we interpret in Sect. 3 as a one-sided jet. We determine its apparent expansion velocity and derive the intrinsic velocity considering Doppler boosting. Our conclusions are given in Sect. 4.