Sr and Ba abundances: Comparing machine-learning with star-by-star analyses

Abstract

Context. A new large sample of 895 s-process-rich candidates out of 454 180 giant stars surveyed by LAMOST at a low spectral resolution (R ~ 1800) has been reported by Norfolk et al. (2019, MNRAS, 490, 2219; hereafter N19). Aims. This study is aimed at confirming the s-process enrichment at the higher resolution (R ~ 86 000) offered by the HERMES-Mercator spectrograph for the 15 brightest targets of the N19 sample, which consists of 13 Sr-only stars and two Ba-only stars (designating stars with only the Sr or only Ba lines strengthened). Methods. Abundances were derived for elements Li, C (including the 12C/13C isotopic ratio), N, O, Na, Mg, Fe, Rb, Sr, Y, Zr, Nb, Ba, La, and Ce, using the TURBOSPECTRUM radiative transfer LTE code with MARCS model atmospheres. Binarity has been tested by comparing the Gaia DR2 radial velocity (epoch 2015.5) with the HERMES velocity obtained 1600–1800 days (about 4.5 yr) later. Results. Among the 15 programme stars, 4 show no s-process overabundances ([X/Fe] < 0.2 dex), 8 show mild s-process overabundances (at least three heavy elements with 0.2 < [X∕Fe] < 0.8), and 3 have strong overabundances (at least three heavy elements with [X/Fe] ≥ 0.8). Among the 13 stars classified as Sr-only by the previous investigation, 4 have no s-process overabundances, 8 are mild barium stars, and 1 is a strong barium star. The two Ba-only stars turn out to be both strong barium stars. Especially noteworthy is the fact that these two are actually dwarf barium stars. Two among the three strong barium stars show clear evidence in support of their binary character, as expected for objects produced through mass-transfer. The results for the no s-process and mild barium stars are more surprising; namely, among the no-s stars, there are two binaries out of four, whereas only one out the eight diagnosed mild barium stars shows a clear signature of radial-velocity variations. Conclusions. Blending effects and saturated lines have to be considered very carefully when using machine-learning techniques, especially when applied to low-resolution spectra. Among the Sr-only stars from the previous study sample, about 60% (8/13) of them can be expected to be true mild barium stars and about 8% to be strong barium stars; this fraction is likely close to 100% for the N19 Ba-only stars (2/2). Therefore, we recommend to limit the sample to N19 Ba-only stars when one needs an unpolluted sample of mass-transfer (i.e., extrinsic) objects.