Systematics of water partitioning in damp mantle melting models

Abstract

In regions of the mantle where water is a trace species in the nominally anhydrous minerals (rather than a free vapor phase or a hydrous mineral), melting is controlled by water partitioning into melt and the resulting freezing-point depression. Modeling undersaturated melting therefore requires a well-constrained formalism for the bulk partition coefficient (D_(H_2O)) and its variations, as well as an accurate model for hydrous melt-mineral equilibria. Hirth and Kohlstedt [1] developed a water partitioning model that leads to a solely pressure-dependent D_(H_2O). Their model expresses the solubility of H_2O in minerals and melts as a function of the standard state f°_(H_2O) (linear for minerals, square root for melts). In the absence of data on undersaturated systems, they assume Henry's law, such that D_(H_2O) is the ratio of solubilities, at any water content. Thermodynamically, however, one expects a square root dependence of H_2O content in the melt on the actual water fugacity (f°_(H_2O) = a_(H_2O)f°_(H_2O)). This implies that D_(H_2O) decreases with decreasing water content as a^(1/2) H_2O, approaching zero as the system dries out. The coupling of this partitioning model to pMELTS is the basis of one approach to damp mantle melting [2]. The prediction of a concentration-dependent D_(H_2O) has some important but difficult-to-observe consequences (e.g., the removal of H_2O from the residue would be an accelerating process, leading to a more abrupt viscosity transition) but also makes the direct prediction that water should behave more compatibly in more-water rich systems. There has been considerable argument as to which is the most appropriate analogue trace element for water partitioning. Estimates have ranged from (as incompatible as) La in depleted MORB to (as compatible as) Nd in BABB [3], though Ce is presently a popular choice [4]. This variation can be a natural and systematic consequence of partitioning rather than a complex combination of source and process effects, and is due to the non-Henrian behavior of water in the melt rather than a pressure dependence. Direct mineral-melt water partitioning data from undersaturated experiments are needed to test this prediction.

Additional details