A great deal of effort and innumerable theses and papers have gone into modelling the derivation of magmas from their parent rocks, especially the mantle, over the last three decades.  Most is based on the division of trace elements into “compatible” and “incompatible”, the first being those which tend to remain in minerals that make up the residuum during magmagenesis, and the second those that favour melts.  Most incompatible elements have large ionic radii. The modelling centres on the degree to which elements remain in solids, the appropriate parameter being an element’s mineral-melt partition coefficient (K_D).  Partition coefficients are usually deduced from an element’s abundance in phenocrysts that are in contact (and supposed equilibrium) with an igneous rock’s groundmass material, which is assumed to have formed from magma, and its concentration in that once liquid phase.  Models for partial melting and fractional crystallisation, plus several variants, all involve K_Ds, for olivines, pyroxenes, feldspars, garnet, amphiboles and so on.  For the generation of basaltic magmas, the first step is partial melting in the mantle itself, for which direct estimation of K_Ds is not possible.  Instead they are assumed from mineral-melt chemistries in crustal igneous rocks, with some allowance for elevated temperatures and pressures and other conditions.  Each mineral has its own distinctive suite of K_Ds for many elements, and the chemistry of an igneous rock has often been traced back to which suite of minerals was present in a residue, i.e. the source rock itself, as well as the degree to which one or other process proceeded.  The 19 February 2004 issue of Nature included an ominous article (Hiraga, T, et al. 2004.  Grain boundaries as reservoirs of incompatible elements in the Earth’s mantle.  Nature, v. 427, p. 699-703). 

The study by geochemists at the University of Minnesota and Oak Ridge National Laboratory, USA, concentrated only on the mineral olivine, and a few elements present at trace levels in it.  Their experiments simulated equilibrium conditions under mantle conditions.  Results showed that incompatible elements in olivine, such as Ca and Al, tend to concentrate mainly at boundaries between grains where they are readily available to any melt that starts to form, rather than uniformly throughout the mineral grain.  The finer the grain size of the rock, the greater the area of grain boundaries, and so the more incompatible elements tend to be concentrated at them  The tendency is predictable on thermodynamic grounds, but has only been studied previously in alloys and other artificial materials.  Geochemists have generally regarded grain boundaries as places where impurities in rocks gather.  If the same rock is analysed with and without the crushed powder having been washed in acid, different trace element concentrations result.  This has been attributed to secondary effects, such as the passage of hydrothermal fluids or groundwater.  Since K_Ds that are used widely involve concentrations in whole mineral grains, the basis of geochemical modelling might be compromised.  Melting begins at grain boundaries, so the low degrees involved in generating basalts could be biased by the effect.  Moreover, vapour phases moving through the mantle (supercritical water and CO₂), will follow grain boundaries too, and so may easily pick up and transport incompatible elements.  Their entry into the crust carrying mantle-derived incompatible elements, such as rare-earths, strontium and lead, would lead to metasomatic effects that could play havoc with interpretations of isotopic data based on these elements.  Carbonatites, probably formed from mantle-derived carbonic fluids, are enriched in many incompatible elements.  Similarly worrying data, such as estimates of the incompatible element partitioning into carbonic fluids, have emerged in the past, but so far have been notable only for the silence with which most geochemists greeted them.