Storage capacity and partitioning of sulfur during silicate melting of the Earth’s shallow upper mantle and the evolution of S/Dy during MORB-source melting

Abstract

Sulfur (S) in the mantle has conventionally been assumed to be stored exclusively in accessory sulfide or sulfate phases, with the S content of primary mantle melts sourced entirely from these S-rich minerals. In contrast, recent works have argued that nominally S-free, major silicate mantle minerals such as clinopyroxene (cpx), orthopyroxene (opx), plagioclase (pl), and garnet (gt) may contain 10′s of ppm sulfur. However, the experimentally determined mineral-melt partition coefficients for sulfur (D_S^min/melt) vary significantly between studies from as low as 10^-3 to as high as 10^-1. To contribute to this debate and to better understand the implications of the silicate mineral S-source for equilibrium mantle melting, here we present a series of new and existing silicate mineral-melt equilibria experiments with sulfur introduced to the system, and calculate the D_S^min/melt for these experiments. The reported experiments span hydrous and anhydrous basaltic starting compositions with 0.04–6.08 wt% S added as sulfide or sulfate at 1.0–3.0 GPa and 1200–1350 °C. The S content of partial melts was measured using electron microprobe, while that of silicate minerals was measured using NanoSIMS. The high spatial resolution and depth profiling enabled by NanoSIMS reveals abundant S-rich hotspots throughout samples, which increase in frequency with experimental S content; we interpret these as sulfide and sulfate nano-inclusions. When hotspots are avoided, the S-carrying capacities of both sulfide- and sulfate-saturated experiments are ≤2.44 ppm for calcic cpx, ≤1.13 ppm for gt, 0.80 ppm for pl, 0.48 ppm for pigeonite (pgt), and 0.55 ppm for opx. The sulfur capacity of the sulfide- and sulfate-saturated minerals exhibits an inverse correlation with experimental temperature, but does not appear to be affected by S speciation or pressure. At sulfide-added conditions, D_S^cpx/melt is ≤0.04190, D_S^gt/melt is ≤0.01081, D_S^pl/melt is 0.00074, D_S^opx/melt is 0.00039, and D_S^pgt/melt is 0.00034. At sulfate-added conditions, the increased solubility of S⁶⁺ in the coexisting silicate melt results in reduced partition coefficients of D_S^cpx/melt ≤0.00105 and D_S^gt/melt of 0.00019. These newly calculated values are orders of magnitude lower than previously published values, a consequence of the low mineral S content measured here. Finally, we apply our new partition coefficients and those of prior work to model the evolving S/Dy during adiabatic decompression melting of a sulfide-saturated MORB-source mantle. We find that minimal partitioning of S into silicate minerals, as observed in our experiments, has negligible impact on the S/Dy of MORBs. Importantly, we also find that the S/Dy ratio of peridotite partial melts does not remain constant as a function of extent of melting; it increases and then decreases from sulfide-present to sulfide-absent conditions.