Comment on Reverse Monte Carlo Simulation

In a recent paper McGreevy and Pusztai [1] have described an interesting simulation technique for determining the structure of disordered systems (liquids and glasses) that uses as input the experimentally measured radial distribution function g E(r 12), or equivalently, the structure factor a E(k). The essence of their procedure is to generate, via Monte Carlo, a set of particle configurations that yield a g(r 12), where r 12 ≡ |r 2—r 1| is the distance between particles, that is consistent with g E(r 12). From these configurations further information about the structure can be extracted, including higher-order correlation functions such as the 3-body function ρ(3) (r 1, r 2, r 3) or bond-angle distributions. Although the idea of reverse Monte Carlo (RMC) is not new (see the reference in [1]), McGreevy and co-workers have shown convincingly that it is computationally feasible and have produced results for a variety of liquids, including multicomponent systems. The RMC procedure is certainly appealing; it suggests that the data which is obtained from an X-ray or neutron diffraction experiment can be used to infer more structural information than that which is inherent in the pairwise function g/(r 12). Since, in certain quarters, this is regarded as heresay it is of some interest to ask what is the formal status of the RMC procedure.