Fig. 6

Predicted \({\mathrm{K}}_{\mathrm{D},\mathrm{Fe}^{2+}-\mathrm{Mg}}^{\mathrm{ol}/\mathrm{liq}}\) values using Model 2 (function of T, \({\mathrm{X}}_{\mathrm{Si}}^{\mathrm{liq}}\), \({\mathrm{X}}_{\mathrm{Ti}}^{\mathrm{liq}}\), \({\mathrm{X}}_{\mathrm{Si}}^{\mathrm{liq}}({\mathrm{X}}_{\mathrm{Na}}^{\mathrm{liq}}+{\mathrm{X}}_{\mathrm{K}}^{\mathrm{liq}}\)), and olivine composition) vs. experimentally measured values for A the RKD experiments (n = 14); symbols as in Fig. 1, and B for the combined low-ƒO₂ data set (n = 187 literature (Lit.) experiments plus the n = 14 RKD experiments); symbols are the same as in Fig. 5B, and are given at the bottom of the figure. For all the experimental glasses, Fe³⁺/Fe²⁺ was calculated using Borisov et al. (2018) at the T and ƒO₂ of each experiment; a.a.d. is the average absolute deviation in measured \({\mathrm{K}}_{\mathrm{D},\mathrm{Fe}^{2+}-\mathrm{Mg}}^{\mathrm{ol}/\mathrm{liq}}\) – predicted \({\mathrm{K}}_{\mathrm{D},\mathrm{Fe}^{2+}-\mathrm{Mg}}^{\mathrm{ol}/\mathrm{liq}}\); in panel B, a.a.d. reflects only the literature data. The black diagonal line in each panel is a 1:1 line. Error bars are 1σ. Best-fit parameters for Model 2 are listed in Table 4