A comparison of data structures for the simulation of polydisperse particle packings

Simulation of particle packings is an important tool in material science. Polydisperse mixtures require huge sample sizes to be representative. Simulation, in particular with iterative packing algorithms, therefore requires highly efficient data structures to keep track of particles during the packing procedure. We introduce a new hybrid data structure for spherical particles consisting of a so-called loose octree for the global spatial indexing and Verlet lists for the local neighbourhood relations. It is particularly suited for relocation of spheres and contact searches. We compare it to classical data structures based on grids and (strict) octrees. It is shown both analytically and empirically that our data structure is highly superior for packing of large polydisperse samples. Copyright (C) 2010 John Wiley & Sons, Ltd.