The orbits of subdwarf-B plus main-sequence binaries II. Three eccentric systems; BD+29°3070, BD+34°1543 and Feige 87

Context. The predicted orbital-period distribution of the subdwarf-B (sdB) population is bi-modal with a peak at short (<10 days) and long (>250 days) periods. Observationally, many short-period sdB systems are known, but the predicted long period peak is missing as orbits have only been determined for a few long-period systems. As these predictions are based on poorly understood binary-interaction processes, it is of prime importance to confront the predictions with reliable observational data. We therefore initiated a monitoring program to find and characterize long-period sdB stars. Aims. In this paper we aim to determine the orbital parameters of the three long-period sdB+MS binaries BD+29 degrees 3070, BD+34 degrees 1543 and Feige 87, to constrain their absolute dimensions and the physical parameters of the components. Methods. High-resolution spectroscopic time series were obtained with HERMES at the Mercator telescope on La Palma, and analyzed to determine the radial velocities of both the sdB and MS components. Photometry from the literature was used to construct the spectral-energy distribution (SED) of the binaries. Atmosphere models were used to fit these SEDs and to determine the surface gravities and temperatures of both components of all systems. Spectral analysis was used to check the results of the SEDs. Results. An orbital period of 1283 +/- 63 d, a mass ratio of q = 0.39 +/- 0.04 and a significant non-zero eccentricity of e = 0.15 +/- 0.01 were found for BD+29 degrees 3070. For BD+34 degrees 1543 we determined P-orb = 972 +/- 2 d, q = 0.57 +/- 0.01 and again a clear non-zero eccentricity of e = 0.16 +/- 0.01. Last, for Feige 87 we found P-orb = 936 +/- 2 d, q = 0.55 +/- 0.01 and e = 0.11 +/- 0.01. Conclusions. BD+29 3070, BD+34 1543 and Feige 87 are long period sdB + MS binaries on clearly eccentric orbits. These results are in conflict with the predictions of stable Roche-lobe overflow models.