Lutetium incorporation in magmas at depth: Changes in melt local environment and the influence on partitioning behaviour

- Coordination and bond length increase of Lu-O at 5 GPa in basalt analogue melts.
- Different structural incorporation of Lu in granitic and basaltic melts observed.
- First determination of trace element bond distance in silicate melts at pressure.
- Implications for partitioning of Lu at depth within magmas.

The structure of two Lu doped (4000 ppm) model end member silicate liquids, a highly polymerised haplogranite (Si-Al-Na-K-O) and a less polymerised anorthite-diopside (Si-Al-Mg-Ca-O), have been studied up to 8 GPa using in situ x-ray diffraction techniques. The results are the first to identify trace rare Earth element incorporation in silicate melts at high pressure. At pressures below 5 GPa, the bonding environment of Lu-O was found to be dependent on composition with coordination number CNLu-O=8 and bond distance rLu-O=2.36 Å in the haplogranite melt, decreasing to CNLu-O=6 and rLu-O=2.29 Å in the anorthite-diopside melt. This compositional variance in coordination number at low pressure is consistent with observations made for Y-O in glasses at ambient conditions and is coincident with a dramatic increase in the partition coefficients previously observed for rare Earth elements with increasing melt polymerisation. With increasing pressure we find that CNLu-O and rLu-O remain constant in the haplogranite melt. However, an abrupt change in both Lu-O coordination and bond distance is observed at 5 GPa in the anorthite-diopside melt, with CNLu-O increasing from 6 to 8-fold and rLu-O from 2.29 to 2.39 Å. This occurs over a similar pressure range where a change in the P-dependence in the reported rare Earth element partition coefficients is observed for garnet-, clinopyroxene-, and olivine-melt systems. This work shows that standard models for predicting trace elements at depth must incorporate the effect of pressure-induced structural transformations in the melt in order to realistically predict partitioning behaviour.