Reverse Monte Carlo modelling of network glasses: useful or useless?

Abstract It has been ten years since the first reverse Monte Carlo (RMC) modelling study of a glass (vitreous silica) was published. The aim of that early work was twofold: firstly to assess whether the experimental data alone would support the continuous random network (CRN) model of Zachariasen (which they did), and secondly to show the possibilities for combination of neutron and X-ray diffraction data. The model itself had many faults, but these were significantly improved on in later work. Since then many RMC models of amorphous materials have been produced. Some results are still controversial, though it is generally accepted that the method is `here to stay'. But what has been learnt? Has RMC taught us anything about the structure of glasses that we did not know before? In general, though this may seem strange for a method which produces detailed structural models at the atomic level, those studies which have asked detailed questions about the structure have been less successful than those which have asked more general questions – `how does it work?' rather than `where exactly are the atoms?' This can be traced back to the fact that the data themselves do not contain sufficient information to answer the detailed questions, regardless of the method used to analyse them, but a model can nevertheless be sufficient to explain the properties of interest. The major advantage of RMC in this context is that models have to be physically possible, i.e., the composition and the density are correct and atoms have a finite size, and they agree with the data. In this paper we will review some of the existing RMC models of silicate glasses, including some previously unpublished results, and examine what contribution the method has made to our understanding of their structures and other properties.