Hidden Flaws in Orthopyroxene Experiments Skew Models of Mantle Melting

Deep beneath mid-ocean ridges, the mantle sometimes melts so thoroughly that clinopyroxene—the mineral that normally hoards most trace elements—is exhausted, leaving orthopyroxene in control of the chemical budget. Understanding how orthopyroxene partitions trace elements between crystal and melt is therefore essential for modeling the evolution of Earth’s mantle and crust, and for interpreting the composition of lavas erupted on Earth and other planetary bodies. A new study published in Frontiers in Geochemistry by Madison X. Betts, Gokce K. Ustunisik, and Roger L. Nielsen of South Dakota Mines and the American Museum of Natural History takes a hard look at the experimental database that underpins these models—and finds it riddled with sources of uncertainty that have gone largely unexamined.

The team compiled published experimental data on trace element partitioning between low-Ca pyroxene and melt from the traceDs database, hosted through the EarthChem Experimental Petrology community. Their dataset included only experiments with melt compositions containing less than 10 percent water and more than 35 percent silica, and excluded natural phenocryst-matrix mineral-melt pairs. For each experiment, they recorded the pressure, temperature, oxygen fugacity, run duration, capsule type, and the major and trace element compositions of the phases, along with the cited analytical uncertainties.

The first problem the researchers documented is that the calibration dataset is different for every element. The number of experiments reporting a specific trace element in orthopyroxene ranges from just 15 for thulium to 86 for scandium. More strikingly, the average pressure, temperature, and liquid composition of the experiments differ dramatically from element to element. Experiments where strontium was measured cluster around 1.5 gigapascals and 1,300 degrees Celsius, while those where thulium was determined average about 4 gigapascals and 1,600 degrees Celsius. The average silica content of the experimental liquids for thulium experiments is 7 weight percent higher than for ytterbium experiments, and the number of ytterbium determinations is nearly five times greater. Because predictive models are built by regression on these data, each element ends up with its own regression constants calibrated over a different slice of pressure-temperature-composition space—meaning a model that predicts a smooth rare earth element pattern may instead produce an irregular one.

The second issue concerns the identity of the mineral itself. Orthopyroxene has an orthorhombic crystal structure, while pigeonite, another low-Ca pyroxene with overlapping composition, is monoclinic. The two can have identical calcium contents—expressed as the wollastonite component—yet belong to different structural species. Pigeonite’s larger, more distorted M2 crystallographic site favors large ions like the light rare earth elements, so its partition coefficients tend to be higher than those of orthopyroxene. Yet most experimental studies did not determine the actual structure of the pyroxene in their charges. Instead, they relied on calcium content alone to label the phase as orthopyroxene or pigeonite—a practice the authors show to be unreliable. If pigeonite data are inadvertently folded into an orthopyroxene calibration dataset, the resulting model will overestimate trace element compatibility.

The third problem is analytical. The researchers compared the reported one-sigma uncertainties for trace element measurements made by laser ablation inductively coupled plasma mass spectrometry (LA ICP-MS) against those made by secondary ion mass spectrometry (SIMS). The LA ICP-MS measurements showed significantly higher relative uncertainty, even at concentrations where counting statistics should not be the limiting factor. The explanation lies in analytical volume: LA ICP-MS samples a spot roughly 50 micrometers across, while SIMS samples about 5 micrometers. In experiments where orthopyroxene crystals are skeletal or acicular—common habits in partitioning experiments—the larger beam is far more likely to include glass from the surrounding melt in the analysis, contaminating the mineral measurement.

The consequences of such contamination are not trivial. For an element with a partition coefficient of 0.1, just 1 percent melt contamination inflates the apparent partition coefficient by 9 percent. For an element with a partition coefficient of 0.001, the same 1 percent contamination pushes the apparent value up by an order of magnitude. The authors demonstrated this with a synthetic dataset: assuming an uncontaminated orthopyroxene containing 9 parts per million zirconium and a glass containing 1,500 parts per million, with randomized analytical error and up to 4 percent modeled contamination, the reported average of the mixed analyses would suggest a partition coefficient of 0.027—more than four times the true value. Without access to the individual analyses, there is no way to detect or correct this bias.

The problem is compounded by a publishing convention that the authors identify as perhaps the single greatest obstacle: experimental partitioning data are almost universally reported as averages of multiple analyses, with the individual mineral-melt pair measurements left unpublished. In no case in the orthopyroxene database were individual analyses reported. This makes it essentially impossible to quantitatively understand the magnitude and sources of error, to identify which analyses were contaminated, or to apply correction techniques that rely on regression through individual data points. The authors note that similar issues have been documented for plagioclase, clinopyroxene, amphibole, and garnet databases, but the orthopyroxene case is unique in the added complication of structural ambiguity between orthopyroxene and pigeonite.

The study also compared the trace element partitioning database against the Library of Experimental Phase Relations (LEPR), which calibrates major element models like the MELTS family of algorithms. The traceDs database contains fewer experiments, covers a more restricted range of major and minor elements such as sodium and titanium, and includes fewer experiments at low temperatures between 800 and 1,000 degrees Celsius. This gap exists largely because orthopyroxene is absent from experiments containing significant water—a fact that may compromise efforts to model the liquid lines of descent of subduction-related magmas. The partitioning behavior of titanium and sodium in the two databases overlaps at high temperature, which matters because many trace elements behave similarly to or correlate with titanium.

The authors examined the ratio of titanium to zirconium partition coefficients as a function of calcium content in both orthopyroxene and clinopyroxene. In clinopyroxene, the ratio is relatively constant above 9 weight percent calcium oxide, while in orthopyroxene it is consistent between 2 and 4 weight percent but changes more rapidly as calcium decreases. Understanding why these ratios differ between the two pyroxenes is fundamental to interpreting the trace element signatures of the mantle as the degree of melting increases and clinopyroxene is exhausted. The lack of a coherent trend for orthopyroxene between zirconium and titanium partition coefficients—unlike the tighter correlation seen in clinopyroxene—could be related to glass contamination or to uncertainty about whether the mineral is truly orthopyroxene.

To move the field forward, the authors recommend a set of best practices for future experimental studies: publication of individual analyses alongside averages and standard deviations; reporting of detection limits for the elements of interest under relevant conditions; repeat analyses of secondary standards to characterize instrument uncertainty; application of projection techniques to evaluate whether analyses represent more than one phase; and images of the experimental charges with analysis spots marked. They also recommend that future low-Ca pyroxene experiments include structural determination of the pyroxene species, using electron back-scattered diffraction or optical extinction angles rather than relying on calcium content alone. Until these changes are adopted, the gap between the cited uncertainty in published partition coefficients and the precision demonstrated by repeated standard measurements—a factor of roughly ten or more—will continue to limit the predictive power of trace element models for igneous processes on Earth and beyond.

Subject of Research: Sources of uncertainty in the experimental orthopyroxene/melt trace element partitioning database

Article Title: The landscape of the experimental orthopyroxene/melt partitioning database

Article References: Betts, M. X., Ustunisik, G. K., & Nielsen, R. L. (2025). The landscape of the experimental orthopyroxene/melt partitioning database. Frontiers in Geochemistry, 3, Article 1660826. https://doi.org/10.3389/fgeoc.2025.1660826

Image Credits: AI Generated

DOI: 10.3389/fgeoc.2025.1660826

Keywords: orthopyroxene, trace element partitioning, experimental petrology, pigeonite, LA ICP-MS, SIMS, mantle melting, geochemistry, glass contamination, crystal-melt partition coefficients, rare earth elements, mid-ocean ridge basalt