Calculation of peridotite partial melting from thermodynamic models of minerals and melts. I. Review of methods and comparison with experiments

Thermodynamic calculation of partial melting of peridotite using the MELTS algorithm has the potential to aid understanding of a wide range of problems related to mantle melting. We review the methodology of MELTS calculations with special emphasis on the features that are relevant for evaluating the suitability of this thermodynamic model for simulations of mantle melting. Comparison of MELTS calculations with well-characterized peridotite partial melting experiments allows detailed evaluation of the strengths and weaknesses of the algorithm for application to peridotite melting problems. Calculated liquid compositions for partial melting of fertile and depleted peridotite show good agrement with experimental trends for all oxides; for some oxides the agreement between the calculated and experimental concentrations is almost perfect, whereas for others, the trends with melt fraction are comparable, but there is a systematic offset in absolute concentration. Of particular interest is the prediction by MELTS that at 1 GPa, near-solidus partial melts offset the peridotite have markedly higher SiO2 than higher melt fraction liquids, but that at similar melt fractions, calculated partial melts of deleted peridotites are not SiO2 enriched. Similarly, MELTS calculations suggest that near-solidus partial melts of fertile peridotite, but not those of depleted peridotite, have less TiO2 than would be anticipated from higher temperature experiments. Because both experiments and calculations suggest that these unusual near-solidus melt compositions occur for fertile peridotite but not for depleted peridotite, it is highly unlikely that these effects are the consequence of experimental or model artifacts. Despite these successes of the results of calculations of peridotite melting using MELTS, there are a number of shortcomings to application OS this thermodynamic model to calculations of mantle melting. In particular, calculated compositions of liquids produced by partial melting of peridotite derived more MgO and less SiO2 than equivalent experimentally derived liquids. This mismatch, which is caused by overprediction of the stability of orthopyroxene relative to olivine, muses a number of other problems, including calculated temperatures of melting that are too high. Secondarily, the calculated distribution of Na between pyroxenes and liquid does not match experimentally observed values, which leads to exaggerated calculated Na concentrations for near-solidus partial melts of peridotite. Calculations of small increments of batch melting followed by melt removal predict that fractional melting is less productive than batch melting near the solidus, when the composition of the liquid is changing rapidly but that once the composition of the liquid ceases to change rapidly, fractional and batch melting produce liquid at similar rates per increment of temperature increase until the exhaustion of clinopyroxene. This predicted effect is corroborated by sequential incremental batch melting experiments (Hirose & Kawamura, 1994, Geophysical Research Letters, 21, 2139-2142). For melting of peridotite in response to fluxing with water, the calculated effect is that melt fraction increases linearly with the amount of water added until exhaustion of clinopyroxene (cpx), at which point the production of melt created per increment of water added decreases. Between the solidus and exhaustion of cpx, the amount of melt generated per increment of water added increases with temperature. These trends are similar to those documented experimentally by Hirose & Kawamoto (1995, Earth and Planetary Science Letters, 133, 463-473).