Article révisé par les pairs
Résumé : Abstract The dominant site(s) of the r -process are a subject of current debate. Ejecta from r -process-enrichment events like kilonovae are difficult to directly measure, so we must instead probe abundances in metal-poor stars to constrain r -process models. This requires state-of-the-art non-LTE (NLTE) modeling, as LTE is a poor approximation for the low-opacity atmospheres of metal-poor giants. Neodymium (Nd) is a prominent r -process element detected in both near-infrared kilonova spectra and the spectra of metal-poor stars, so precise Nd stellar abundances are particularly needed to model kilonovae and constrain r -process sites. We thus constructed an Nd i /Nd ii model atom to compute NLTE abundances in FGK metal-poor stars. We obtain A (Nd) ⊙  = 1.44 ± 0.05, in agreement with the meteoritic value, when calibrating the model atom with a Drawin hydrogen collision factor of S H  = 0.1. For a sample of metal-poor r -process-enhanced stars with observed optical and near-infrared Nd ii lines, we find NLTE Nd corrections in the range −0.3 to 0.3 dex. Optical and UV lines have positive NLTE corrections, whereas H -band lines have negative corrections. Additionally, we compute a large grid of NLTE corrections for 122 Nd ii spectral lines ranging from the UV to the H band, for stellar parameters of typical metal-poor FGK dwarfs and giants with −3 ≤ [Fe/H] ≤ −1 and −2 ≤ A (Nd) ≤ 2. Within this grid, we find NLTE corrections ranging from −0.3 to +0.5 dex. Deviations from LTE are found to be strongest for blue lines with low excitation potentials in the most metal-poor giants.