The results of our abundance analyses are given in Table 5, where we also give the results of Conlon et al. (1994) and Dixon & Hurwitz (1998) for Barnard 29 . A comparison of our results with RGB and cooler UV bright stars (only in Cen) is presented in Figs. 2 and 3. The M 13 data in Fig. 2 are from Smith et al. (1996) and Kraft et al. (1997). The Cen data for the giants in Fig. 3 were collected from Paltoglou & Norris (1989), Brown et al. (1991), Brown & Wallerstein (1993), and Smith et al. (1995), the data for the cool UVBS are from Gonzalez & Wallerstein (1994). All abundances were adjusted to a solar iron abundance of = 7.50.
As can be seen from Fig. 2 Barnard 29 shows similar abundance trends (except for iron) as the giant stars observed in M 13. This abundance pattern is likely caused by deep mixing and dredge-up of CNO-processed material on the first RGB (Pilachowski et al., 1996). However, the iron abundance we found for Barnard 29 is lower than the cluster value by 0.5 dex. This indicates that the atmosphere of Barnard 29 has become iron depleted during the star's evolution, most likely by the gas-dust separation proposed by Bond (1991) and discussed by Napiwotzki et al. (1994), which is also favoured by Dixon & Hurwitz (1998) for Barnard 29 . This scenario assumes that the dust formed in the cool and extended atmospheres of AGB stars is selectively removed by radiation pressure, while the gas remains at the stellar surface. Since metals with high condensation temperatures (e.g. iron) preferentially condense into dust grains and elements with lower condensation temperatures remain in the gas phase, the remaining gas forms a iron-poor atmosphere. Mathis & Lamers (1992) discussed a single-star and a binary scenario, which can both result in a selective removal of dust at the end of the AGB stage.
The interpretation of the abundance pattern of ROA 5701 is hampered by the complex patterns found in Cen stars in general (cf. Fig. 3). If we take as a metallicity tracer the sum of C 4, N, and O abundances (which remains unchanged by CNO processing), we find that the original metallicity of ROA 5701 is close to the median value determined for cluster stars (cf. Fig. 4). However, the iron abundance is 0.5 dex below the lowest value found for any RGB star plotted in Fig. 3. This points to an iron depletion similar to that detected in Barnard 29 , which also shows a similar behaviour in Fig. 4.
In summary, the C, N, O, and Si abundances of Barnard 29 and ROA 5701 are in agreement with that of the respective cluster red giant stars. No dredge-up during the AGB phase is necessary to explain these abundance patterns, although a moderate third dredge-up cannot be ruled out from our data alone. However, our low iron abundances point towards a significant iron depletion, most probably caused by a gas-dust separation during the late AGB stages.
© European Southern Observatory (ESO) 1998
Online publication: October 21, 1998