Interplay of large materials databases, semi-empirical methods, neuro-computing and ®rst principle calculations for ternary compound former/nonformer prediction

A signi®cant breakthrough has been achieved using materials databases, semi-empirical methods and neural networks to aid in the design of new materials. A collaborative, international, team discovered that a non-linear expression involving one elemental property parameter could be used to predict, with 99+% accuracy, the occurrence of a compound for any ternary materials system. This elemental property parameter, referred to as the Mendeleev Number, was conceived by D.G. Pettifor in 1984 to group binary compounds by structure type. The near term signi®cance of this discovery is the obvious savings, in time and resources, relative to assessing the merits of future, yet-to-be-realized, materials systems. In longer term this breakthrough is the basis for both narrowing the search space for potentially bene®cial new materials and enabeling the prediction of even more speci®c materials information such as stoichiometries, crystal structures and intrinsic properties. 7 2000 Elsevier Science Ltd. All rights reserved.