RDCI: A novel method of cluster analysis and applications thereof in sample molecular simulations

TL;DR

RDCI is a new cluster analysis method that identifies clusters in 3D space without preset criteria, demonstrated on molecular dynamics simulations to analyze clustering behavior.

Contribution

The paper introduces RDCI, a novel cluster analysis technique that requires no prior assumptions and effectively analyzes clustering in molecular simulations.

Findings

Successfully applied to Lennard-Jones fluids and polymer systems

Extracted quantitative clustering information

Analyzed clustering dynamics in simulations

Abstract

A novel method, termed Reduced Dimensionality Cluster Identification, RDCI, is presented, for the identification and quantitative description of clusters formed by N objects in three dimensional space. The method consists of finding a path, as short as possible, connecting the objects, and then tracking down the size s, of a subgroup i-n, i-n+1, ..., i+n, of 2n+1 < N particles for i varying from n+1 to N-n. Clusters are located where local minima of s(i) occur whereas local maxima serve as delimiters partitioning the path in subsets containing the clusters. Minimal post-processing allows for the removal of outliers on the basis of user-defined criteria and the identification of clearly defined clusters. The advantage of the method is that it requires no predetermined input or criteria of "clusterness" such as number of objects or size of aggregates. Among the numerous possible applications of the method, results are herein reported, from Molecular Dynamics simulations of a binary mixture of Lennard-Jones fluids and of a model polymeric system with a gas-like substance dispersed in it. The method is shown to allow the extraction of meaningful quantitative information regarding the tendency of molecules to cluster or de-cluster and the dynamics of the clustering processes.