Thirty-year quest for structure–nucleation relationships in oxide glasses

Abstract

Over the past three decades, researchers have investigated the existence of possible relationships between crystal nucleation kinetics and the atomic-scale structure of silicate glasses. The main driving force for this quest was the fact that while the vast majority of glass-forming substances only undergo surface (heterogeneous) nucleation when sufficiently heated, a few systems also show the thermodynamically less favourable case of internal (homogeneous) nucleation on laboratory time/length scales. For such glass systems, various macroscopic properties, such as densities, configurational entropies and frozen-in birefringence, have suggested that the structure in the glassy state shows a closer resemblance to the structure of the phase formed upon crystallisation than in the case of glass systems only undergoing heterogeneous nucleation. However, the specific structural features and their length scales have remained uncertain. In this article, we review and discuss the research investigating relationships between the occurrence of internal nucleation and structural parameters related to various different length scales. The latter include (1) short-range order, concerning both network modifier cation-oxygen distances and coordination numbers as well as network former Qn distributions, (2) intermediate range order describing network former connectivities, network former/network modifier correlations, and network modifier distance distributions, and (3) medium range order as reflected by silicate tetrahedral ring-size statistics. Inspection of this data for several stoichiometric oxide glasses and their respective isochemical crystals suggests a positive correlation between homogeneous nucleation ability and structural similarity at the level of short- and intermediate-range order of the network modifier cations. In contrast, no correlation can be found with regard to any structural parameters describing the local structures of the network former species (Qn distributions). Based on the limited set of data available, we develop concrete recommendations for future experiments to test this hypothesis.