Partitioning of rare earth and high field strength elements between titanite and phonolitic liquid

We present the results of a LA–ICPMS study of titanites and associated glasses from the mixed-magma phonolitic Fasnia Member of the Diego Hernández Formation, Tenerife, Canary Islands. We employ a method of identifying equilibrium mineral–melt pairs from natural samples using REE contents and a linear form of the lattice strain model equation (Blundy and Wood, 1994), where the Young's modulus (EM) for the 7-fold coordinated site is an output variable. For felsic magmas that contain crystals potentially derived from a variety of environments within the system, this approach is more rigorous than the use of solely textural criteria such as mineral–glass proximity. We then estimate titanite/melt partition coefficients for Y, Zr, Nb, REE, Hf, Ta, U and Th. In common with prior studies, we find that middle REE partition more strongly into titanite than either light or heavy REE, and that REE partitioning behavior in titanite is reasonably predicted by the lattice strain model. Titanite also fractionates Y from Ho, Zr from Hf, and Nb from Ta. Comparison with experimental data indicates that melt structure effects on partitioning are significant, most particularly in very highly polymerized melts. We use the data to estimate 7-fold coordination radii for trivalent Pr, Nd, Ho, Tm and Lu, and to make approximate predictions of titanite/melt partitioning of Ra, Ac and Pa. Interpolation of data for heavy REE does not predict the behavior of Y, indicating that factors other than charge and radius are involved in partitioning. Variations in Y/Ho induced by magmatic processes appear to be negatively correlated with temperature, and are expected to be greatest in near-minimum melts.

► We present the results of a LA–ICPMS study of titanites and associated glasses from phonolite.
► We identify equilibrium titanite–melt pairs and estimate D values for REE and HFSE.
► We find that REE partitioning behavior is predicted by the lattice strain model.
► We use the data to estimate 7-fold radii for trivalent Pr, Nd, Ho, Tm and Lu.
► We find from data that factors other than charge and radius are involved in partitioning.