Empirical models relating viscosity and tracer diffusion in magmatic silicate melts

The Adam-Gibbs equations describing relaxation in silicate melts are applied to diffusion of trace components of multicomponent liquids. The Adam-Gibbs theory is used as a starting point to derive an explicit relation between viscosity and diffusion including non-Arrhenian temperature dependence. The general form of the equation is D_iη = A_iexp{Δ(s_cE_i)/TS_c}, where D is diffusivity, η is melt viscosity, T is absolute temperature, Δ(s_cE_i) is the difference between the products of activation energies and local configurational entropies for viscous and diffusive relaxation, A_i is a constant that depends on the characteristics of the diffusing solute particles, and S_c is configurational entropy of the melt. The general equation will be impractical for most predictive purposes due to the paucity of configurational entropy data for silicate melts. Under most magmatic conditions the proposed non-Arrhenian behaviour can be neglected, allowing the general equation to be simplified to a generalized form of the Eyring equation to describe diffusion of solutes that interact weakly with the melt structure: D_iη/T = Q_iexp{ΔE_i/RT}, where Q_i and ΔE_i depend on the characteristics of the solute and the melt structure. If the diffusing solute interacts strongly with the melt structure or is a network-forming cation itself, then ΔE_i = 0, and the relation between viscosity and diffusion has the functional form of the classic Eyring and Stokes-Einstein equations; D_iη/T = Q_i. If the diffusing solute can make diffusive jumps without requiring cooperative rearrangement of the melt structure, the diffusivity is entirely decoupled from melt viscosity and should be Arrhenian, i.e., D_i = Q_iexp{B_i/T}. A dataset of 594 published diffusivities in melts ranging from the system CAS through diopside, basalt, andesite, anhydrous rhyolite, hydrous rhyolite, and peralkaline rhyolite to albite, orthoclase, and jadeite is compared with the model equations. Alkali diffusion is completely decoupled from melt viscosity but is related to melt structure. Network-modifying cations with field strength Z_i²/r between 1 and 10 interact weakly with the melt network and can be modelled with the extended form of the Eyring equation. Diffusivities of cations with high field strength have activation energies essentially equal to that of viscous flow and can be modelled with a simple reciprocal Eyring-type dependence on viscosity. The values of Q_i, ΔE_i and B_i for each cation are different and can be related to the cation charge and radius as well as the composition of the melt through the parameters Z_i²/r, M/O, and Al/(Na + K + H). I present empirical fit parameters to the model equations that permit prediction of cation diffusivities given only charge and radius of the cation and temperature, composition and viscosity of the melt, for the entire range of temperatures accessible to magmas near to or above their liquidus, for magmas ranging in composition from basalt through andesite to hydrous or anhydrous rhyolite. Pressure effects are implicitly accounted for by corrections to melt viscosity. Ninety percent of diffusivities predicted by the models are within 0.6 log units of the measured values.