Analyzing molecular dynamics scattered data for large atomic movements

Molecular dynamics (MD) simulations generally produce massive amounts of data, which contain key structural and dynamical information about the material system under consideration. Information extraction from atomic position-time series is, however, a non-trivial task. It perhaps requires examining the entire data set since it is not known a priori in which parts of the data and in what form the desired information resides. A manual analysis involving all atomic positions at each and every time step is infeasible and ineffective even for moderate-sized data. Here, we develop a system to perform an automatic analysis of MD simulation data sets. Our system focuses on how constituent atoms move and provides an algorithm to automatically identify their movements that are larger than usual oscillating (wiggling) motions in terms of the distance covered within a short time window. The algorithm employs methods from signal processing and uses a regression model to distinguish different types of motions. The system also explores local bonding and structural changes occurring in the time intervals of large displacements, which can perhaps be responsible for structural rearrangement as well as transport phenomena. For illustrations, we analyze the first-principles molecular dynamics simulations of liquid silica (SiO2) from 2800 to 4000 K at zero pressure. Our system detects, in an automated fashion, numerous significant movements of Si and O atoms, and then associate them to Si-O and O-Si coordination states. The silica liquid is known to form a tetrahedral network mainly consisting of four oxygen coordinated Si atoms and doubly silicon coordinated O atoms (i.e., bridging oxygen). Our analysis suggests that only small fractions of odd coordination states account for the detected large atomic movements. Moreover, threefold coordinated Si atoms and singly coordinated O atoms (non-bridging oxygen) appear to contribute to atomic-diffusion more than pentahedral Si and threefold O states. (C) 2014 Elsevier B.V. All rights reserved.