The landscape of the experimental orthopyroxene/melt partitioning database

Quantitative modeling is a powerful tool that allows us to develop an understanding of identity and extent of igneous processes on Earth and other planetary bodies. Such models are built upon data from experiments that describe the major and trace element behavior among the phases in a system of interest. Previous work on the analysis of calibration datasets has been done for many of the mafic rock forming minerals such as plagioclase, clinopyroxene, amphibole, olivine, and garnet. What is missing is an evaluation of the sources of uncertainty in the trace element orthopyroxene/melt partitioning experiments. The goal of this investigation is to document the identity and magnitude of a number of sources of uncertainty that were observed in the experimental database for low-Ca pyroxene/melt trace element partitioning. These include multiphase analyses (e.g., mineral and glass analysis in the same analytical volume) as well as uncertainty with regards to the specific structural state of the low-Ca pyroxene in the experiments (e.g., is the mineral reported orthopyroxene or pigeonite?). Our findings confirm earlier work documenting the dependence of the orthopyroxene/melt partition coefficients for the high field strength and rare earth elements on the Ca content of pyroxene. However, the development of predictive expressions for trace element partitioning between orthopyroxene and melt is compromised by a number of database characteristics related to commonly applied analytical methods and phase determinations. Specifically, the average composition of the experimental orthopyroxene and liquids overlap, but is different for each element (e.g., the average composition of all experiments where Sm was determined is different but overlap with those where Zr was determined). Second, the reported analytical uncertainty is significantly higher for experiments where the trace element determinations were done by laser inductively coupled plasma mass spectroscopy compared to secondary ion mass spectroscopy - an observation attributable to the differences in analytical volume between two analytical techniques. Lastly, most experimental studies did not determine the specific low-Ca pyroxene present but rather used the calcium content alone to identify orthopyroxene vs. pigeonite which we showed as unreliable. Taken together, these database characteristics impact any regression analysis and result in an internal bias in the predicted behavior of trace elements. Our ability to “fix” the database is inhibited by the current convention of publishing experimental partitioning data as averages and not publishing individual analyses - making essentially impossible to quantitatively understand the magnitude and sources of error.