Résumé
Crystallization experiments of a dacitic composition were conducted in a temperature range of 675–775 °C at 200 MPa in order to determine the evolution of the trace element partitioning between residual melts and coexisting mineral assemblages with ongoing crystallization. The starting composition was doped with ∼100 ppm of a series of incompatible trace elements. The adopted experimental approach consisted of performing two-step experiments. In a first step, long-time crystallization experiments were performed, starting with glass powder to obtain a mineral assemblage in equilibrium with residual melt. The phase assemblage was composed of melt, plagioclase, amphibole, biotite, and a few oxides. Because of the small size of crystal-free melt pools, a separation of glass from minerals was achieved by the use of the mineral grain trap technique in second-step experiments. The melts from first-step experiments were rhyolitic, allowing quartz to be used as a mineral trap. The following elements were analysed by means of laser ablation inductively coupled plasma mass spectrometry (LA-ICP-MS) analysis: P, Y, Zr, Nb, Cs, Ce, Yb, and U. Although the initial trace element concentrations were low, only Cs and U show a clearly incompatible behaviour, with a systematic enrichment in the melt with decreasing melt fraction from ∼85 area % to 65 area %. We demonstrate that the other elements are controlled either by the crystallization of accessory phases for P (apatite) and Zr (zircon) and/or by their incorporation into mineral phases (mainly amphibole, biotite, and oxide). In particular, changes in the relative proportions of mafic phases vs. plagioclase during ongoing crystallization may lead to changes in bulk mineral / melt partition coefficients from >1 to <1. At constant temperature (T), the proportions of mafic phases vs. plagioclase are strongly controlled by water activity, and the meltwater content is therefore a crucial parameter that controls the evolution of trace element concentrations in residual melts. The comparison of our data with published partitioning coefficients is in good agreement for Cs, U, and Y. However, Kdamph/melt and Kdbiot/melt may be higher than estimated previously for Yb, Ce, and Zr. Nb is less compatible than predicted, with a bulk partitioning value in the range of 1.5 to 2.