1 Introduction

Subluminous B (sdB) stars dominate the populations of faint blue stars of our own Galaxy and are found in both the disk (field sdBs) and globular clusters (Moehler et al. 1997). Observations of elliptical galaxies with the Ultraviolet Imaging Telescope (Brown et al. 1997) and the Hubble Space Telescope (Brown et al. 2000) have shown that these stars are sufficiently common to be the dominant source for the "UV upturn phenomenon'' observed in elliptical galaxies and galaxy bulges (see also Greggio & Renzini 1990, 1999). Their space distribution and kinematical properties indicate that the field stars belong to the intermediate to old disk population (de Boer et al. 1997; Altmann & de Boer 2000).

However, important questions remain concerning their formation process and the appropriate evolutionary timescales. This is a major drawback for the calibration of the observed ultraviolet upturn in elliptical galaxies as an age indicator.

It is now generally accepted that the sdB stars can be identified with models for Extreme Horizontal Branch (EHB) stars burning He in their core, but with a very tiny (<2% by mass) inert hydrogen envelope (Heber 1986; Saffer et al. 1994). An EHB star bears great resemblance to a helium main-sequence star of half a solar mass and its further evolution should proceed similarly (i.e. directly to the white dwarf graveyard) as confirmed by evolutionary calculations (Dorman et al. 1993).

How stars evolve to the EHB configuration is controversial. The problem is how the mass loss mechanism in the progenitor manages to remove all but a tiny fraction of the hydrogen envelope at precisely the same time as the He core has attained the minimum mass ($\approx$ $0.5~M_\odot$) required for the He flash.

Both non-interacting (scenario i), and interacting (scenarios ii and iii) evolutionary scenarios have been proposed to explain the origin of the sdB stars (see Bailyn et al. 1992).

(i) Enhanced mass loss on the red giant branch (RGB) before or during the core helium flash may remove almost the entire hydrogen-rich envelope. This is usually modelled by increasing the $\eta$ factor in the Reimers (1975) formula to estimate mass loss rates for RGB stars. It has been conjectured that the mass loss rates increase with increasing metallicity, implying that metal rich populations should produce more sdB stars than metal poor ones. Birthrate estimates for sdB stars indicate that only 2% (Heber 1986) or even less (0.25% to 1%, Saffer & Liebert 1995) of the RGB stars need to experience such enhanced mass loss. Evidence that this is possible comes from the existence of RR Lyrae stars of population I which must also have lost half of their mass during evolution. In both cases the physical reason for such strong mass loss is not yet understood.

(ii) Mengel et al. (1976) suggest that sdBs could be formed from binaries in which mass transfer starts on the red giant branch and results in a reduction of the hydrogen envelope prior to the helium core flash. Hence all sdBs star are predicted to be found in close binary systems.

(iii) An alternative scenario was proposed by Iben (1990), who pointed out that sdBs can be formed from mergers of helium white dwarf binary systems. Iben & Tutukov (1992) estimate that 80% of the sdBs could have been formed by mergers. Hence the frequency of sdBs still being in binaries should be at most 20%.

Several dozens of objects with composite spectra consisting of an sdB and a dwarf G-K star have been discovered (e.g. Ferguson et al. 1984; Theissen et al. 1993, 1995; Allard et al. 1994) which implies that the binary frequency of sdBs is 50% or more (Allard et al. 1994). The observed large binary frequency rules out the merger scenario (iii) and we are left with scenarios (i) and (ii), i.e. either the sdB binaries are mostly wide systems that did not interact so that the sdB precursors have evolved independently from the companion (i), or they are close systems formed by interaction of the sdB precursor with the companion star (mass exchange, ii).

The high spatial resolution of the Planetary Camera (PC) on board the Hubble Space Telescope (HST) allows to perform a crucial test. As we will show in this paper, it should be possible to resolve a significant fraction of the known composite spectrum systems containing an sdB star if scenario (i) is correct, i.e. if the systems have a distribution of separations like normal main sequence binaries (Duquennoy & Mayor 1991). The interacting scenario (ii), however, predicts that all sdB stars reside in short period ($P \le 100$ d) binaries and consequently none of the systems should be resolvable even with the PC. In order to measure their distribution of separations we have imaged 23 sdB binary candidates with the PC by taking advantage of the snap shot mode of HST observations.