Prediction of plagioclase-melt equilibria in anhydrous silicate melts at 1-atm

Many models for plagioclase-melt equilibria have been proposed over the past 30 years, but the focus is increasingly on the effects of water content and pressure. However, many geological and petrological applications concern low pressure and low water systems, such as the differentiation of large terrestrial basaltic magma chambers, and lunar and asteroidal magmatism. There is, therefore, a justified need to quantify the influence of anhydrous liquid composition on the composition of equilibrium plagioclase at 1-atm. With this in mind, a database of over 500 experimentally determined plagioclase-liquid pairs has been created. The selected low pressure, anhydrous, experiments include both natural and synthetic liquids, whose compositions range from basalt to rhyolite. Four equations are proposed, derived from this data. The first is based on a thermodynamically inspired formalism, explicitly integrating the effect of temperature. This equation uses free energies and activities of crystalline anorthite available from the literature. For the activity of anorthite in the liquid phase, it is found that current models of the activity of individual oxides are insufficient to account for the experimental results. We have therefore derived an empirical expression for the variation of anorthite activity in the liquid as a function of melt composition, based upon inversion of the experimental data. Using this expression allows the calculation of plagioclase composition with a relative error less than 10%. However, in light of the fact that temperature is not necessarily known for many petrological applications, an alternative set of T-independent equations is also proposed. For this entirely empirical approach, the database has been divided into three compositional groups, treated independently for regression purposes: mafic–ultramafic, alkali-rich mafic–ultramafic, and intermediate-felsic. This separation into distinct subgroups was found to be necessary to maintain errors below acceptable limits, but results across group boundaries were found to be comparable. Overall, 50% of plagioclase compositions are predicted to within 2% of the experimentally derived value, and 90% to within 5%, representing a significant improvement over existing models.