Mineral/melt partition coefficients of most elements are depend significantly on the composition of both phases. Relationships to crystal composition can be modeled quite precisely on the basis of crystal structure. Significant understanding of silicate melt structure is becoming available. The main compositional variables governing melt structure at ambient or near ambient pressure are the same variables as those affecting the partition coefficients. It is, therefore, feasible to characterize the solution behavior of major, minor and trace elements in terms of their structural behavior in silicate melts. This can be translated to a description of the role of energetically nonequivalent nonbridging oxygen in various Qn-species in the silicate melts because essentially all geochemically important trace and minor elements are network-modifiers. This non-equivalence exists because the next-nearest neighbor environment around nonbridging oxygen in the different Qn-species depends on the both the the type of Qn-species and on the distribution of tetrahedrally coordinated cations (in addition to Si4+) between the coexisting Qn-species. Characterization of how mineral/melt partition coefficients depend on melt compositions is, therefore, essentially a question of determination of Qn-species type and abundance and ordering of network-modifying cations among the nonbriging oxygens in these Qn-species.

The oft-used melt parameter, NBO/T (nonbridging oxygen per tetrahedrally coordinated cation) is often used to relate mineral/melt partitioning to melt structure. The NBO/T can also be calculated from melt composition provided that sufficient structural information exists from which to assign the proportion of tetrahedrally coordinated cations, T. However, because it is the XQn that governs the solution mechanisms and not NBO/T it is not surprising there is no general relationships between partition coefficients to NBO/T.

For example, the Fe2+/Mg exchange equilibrium coefficient between olivine melts, often central to assessment of the extent to which natural magmas have bee affected by crystal fractionation subsequent to partial melting in the earth’s upper mantle is not 0.3 as is often assumed, but varies between 0.17 and 0.45 depending on melt composition (structure). This variation can be understood via characterization of the the Fe2+-NBO and Mg-NBO bonding in the melts. Analogous relationships have been developed for other transition metals, such as, for example, Mn2+, Co2+, and Ni2+. Details of these relationships can be developed by considering the ionization potential of these and other elements. The effects become larger the greater the ionization potential of the cation of interest. Moreover, there is a competition between the element of of interest and other major elements that also form bonding with energetically non-equivalent nonbridging oxygens. This latter line of thought can be used, for example, the explain how major elements ratios such as alkalis/alkaline earths (or more specific elements, e.g., Na vs. Ca) affect solution behavior in melts and, therefore, mineral/melt partition coefficients.

Other compositional variables such as, for example Fe2+/Fe3+ ratio, Fe3+/(Fe3++Si) and Al/(Al+Si) also can be shown to govern the Qn-species distribution. Those latter variables do, therefore, have important effects on the behavior of mineral/melt partition coefficients. Therefore, the relationships between these variables and mineral/melt partition coefficients can be characterized via their influence on Qn-species type and abundance in silicate melts.

Variations in mineral/melt element partition coefficients in natural magmatic systems can be understood, therefore, via characterization of how melt structure reflects melt compositional variables. For example, the relations between melt structure and olivine/melt partitioning behavior lead to the suggestion that in natural magmatic systems mineral/melt partition coefficients are more dependent on melt composition and, therefore, melt structure the more alkali-rich and the more felsic the melt. Moreover, mineral/melt partition coefficients are more sensitive to melt composition the more highly charged or the smaller their ionic radius of the cation of interest.