On the formation of C III 4647-50-51 and C III 5696 in O star atmospheres

Context. Surface abundances of OB stars provide important information on mixing processes in stellar interiors and on evolutionary status. Performing abundance determination from selected C, N and O lines in the optical spectra of OB stars is thus necessary to understand the evolution of massive stars. Aims. In this study, we investigate the formation of C III 4647-51-50 and C III 5696 in the atmosphere of O stars to see if they can be reliably used for abundance determinations. Methods. We used atmosphere models computed with the code CMFGEN. The key physical ingredients explaining the formation of the C III lines were extracted from comparisons of models with different stellar parameters and through examining rates controlling the level populations. Results. The strength of C III 5696 critically depends on UV C III lines at 386, 574 and 884 angstrom. These lines control the C III 5696 upper and lower level population. C III 884 plays a key role in late O stars where it drains the lower level of C III 5696. C III 386 and C III 574 are more important at early spectral types. The overlap of these UV lines with Fe IV 386.262, Fe IV 574.232 and S V 884.531 influences the radiative transfer at 386, 574 and 884 angstrom, and consequently affects the strength of C III 5696. C III 4650 is mainly controlled by the C III 538 line, which acts as a drain on its lower level. Fe IV 538.057 interacts with C III 538 and has an impact on the C III 4650 profile. Low temperature dielectronic recombinations have a negligible effect on the line profiles. The main effect of the wind on the C III lines is an increase of the emission as mass loss rate increases, due to a shift of the formation region toward high velocity. Such a shift also implies desaturation of C III 386, C III 538, C III 574 and C III 884, which in turn modify the C III 4650 and C III 5696 line profiles. Metallicity effects are complex and result from an interplay between ionization, metallic line interaction with UV C III lines, and wind effects. Conclusions. Given our current understanding of the stellar and wind properties of O stars, and in view of the present results, the determination of accurate carbon abundances from C III 4647-51-50 and C III 5696 is an extremely challenging task. Uncertainties lower than a factor of two on C/H determinations based only on these two sets of lines should be regarder as highly doubtful. Our results provide a possible explanation of the variability of C III 4650 in Of?p stars.