Older GCE models predict [K/Fe] ratios as much as 1 dex lower than those inferred from stellar observations. Abundances of potassium are mainly based on analyses of the 7698 $\AA$ resonance line, and the discrepancy between models and observations is in part caused by the LTE assumption. We study the statistical equilibrium of KI, focusing on the non-LTE effects on the $7698 \ \AA$ line. We aim to determine how non-LTE abundances of K can improve the analysis of its chemical evolution, and help to constrain the yields of models. We construct a model atom that employs the most up-to-date data. In particular, we calculate and present inelastic e+K collisional excitation cross-sections from the convergent close-coupling and the $B$-Spline $R$-matrix methods, and H+K collisions from the two-electron model. We constructed a fine grid of non-LTE abundance corrections that span $4000<\teff / \rm{K}<8000$, $0.50<\lgg<5.00$, $-5.00<\feh<+0.50$, and applied the corrections to abundances from the literature. In concordance with previous studies, we find severe non-LTE effects in the $7698 \ \AA$ line, which is stronger in non-LTE with abundance corrections that can reach $\sim-0.7\,\dex$. We explore the effects of atmospheric inhomogeneity by computing a full 3D non-LTE stellar spectrum of KI for a test star. We find that 3D is necessary to predict a correct shape of the resonance 7698 $\AA$ line, but the line strength is similar to that found in 1D non-LTE. Our non-LTE abundance corrections reduce the scatter and change the cosmic trends of literature K abundances. In the regime [Fe/H]$\lesssim-1.0$ the non-LTE abundances show a good agreement with the GCE model with yields from rotating massive stars. The reduced scatter of the non-LTE corrected abundances of a sample of solar twins shows that line-by-line differential analysis techniques cannot fully compensate for systematic modelling errors.

The astrophysical S-factor and reaction rate of the direct capture process $\alpha+d$ $\rightarrow$ $^6$Li + $\gamma$, as well as the abundance of the $^6$Li element are estimated in a three-body model. The initial state is factorized into the deuteron bound state and the $\alpha+d$ scattering state. The final nucleus $^6$Li(1+) is described as a three-body bound state $\alpha+n+p$ in the hyperspherical Lagrange-mesh method. Corrections to the asymptotics of the overlap integral in the S- and D-waves have been done for the E2 S-factor. The isospin forbidden E1 S-factor is calculated from the initial isosinglet states to the small isotriplet components of the final $^6$Li(1+) bound state. It is shown that the three-body model is able to reproduce the newest experimental data of the LUNA collaboration for the astrophysical S-factor and the reaction rates within the experimental error bars. The estimated $^6$Li/H abundance ratio of $(0.67 \pm 0.01)\times 10^{-14}$ is in a very good agreement with the recent measurement $(0.80 \pm 0.18)\times 10^{-14}$ of the LUNA collaboration.

Results of calculations for inelastic e+Mg effective collision strengths for the lowest 25 physical states of Mg I (up to 3s6p 1P), and thus 300 transitions, from the convergent close-coupling (CCC) and the B-spline R-matrix (BSR) methods are presented. At temperatures of interest, ~5000 K, the results of the two calculations differ on average by only 4%, with a scatter of 27%. As the methods are independent, this suggests that the calculations provide datasets for e+Mg collisions accurate to this level. Comparison with the commonly used dataset compiled by Mauas et al. (1988), covering 25 transitions among 12 states, suggests the Mauas et al. data are on average ~57% too low, and with a very large scatter of a factor of ~6.5. In particular the collision strength for the transition corresponding to the Mg I intercombination line at 457 nm is significantly underestimated by Mauas et al., which has consequences for models that employ this dataset. In giant stars the new data leads to a stronger line compared to previous non-LTE calculations, and thus a reduction in the non-LTE abundance correction by ~0.1 dex (~25%). A non-LTE calculation in a supernova ejecta model shows this line becomes significantly stronger, by a factor of around two, alleviating the discrepancy where the 457 nm line in typical models with Mg/O ratios close to solar tended to be too weak compared to observations.

The astrophysical capture process $\alpha+d$ $\rightarrow$ $^6$Li + $\gamma$ is studied in a three-body model. The initial state is factorized into the deuteron bound state and the $\alpha+d$ scattering state. The final nucleus $^6$Li(1+) is described as a three-body bound state $\alpha+n+p$ in the hyperspherical Lagrange-mesh method. The contribution of the E1 transition operator from the initial isosinglet states to the isotriplet components of the final state is estimated to be negligible. An estimation of the forbidden E1 transition to the isosinglet components of the final state is comparable with the corresponding results of the two-body model. However, the contribution of the E2 transition operator is found to be much smaller than the corresponding estimations of the two-body model. The three-body model perfectly matches the new experimental data of the LUNA collaboration with the spectroscopic factor 2.586 estimated from the bound-state wave functions of $^6$Li and deuteron.