Cluster analysis of the microstructure of colloidal dispersions using the maximum entropy technique

Abstract A new method for identifying clusters in magnetic dispersion simulations is presented. The technique is based on maximisation of the entropy of the system. Clusters are generated which are maximally non-committal with respect to missing data resulting in assignments which closely reflect the nature of the association between the objects. By way of illustration, the method is applied to two-dimensional Monte-Carlo simulations of a ferrofluid where the particles are assumed to interact via a point dipole potential. The clusters determined by the algorithm would appear to be perfectly acceptable for this type of interaction. The method is also successfully applied to the more complex case of 3D structures in a strongly interacting particulate dispersion.