Identification of crystalline structures of clathrate hydrates during molecular simulations using machine learning

Abstract

Molecular simulation is a powerful tool that links a system’s microscopic behaviour to its macroscopic observable features. The data obtained from a molecular simulation act as a digital microscope, i.e., it contains information about the positions and velocities of all the atoms. This data contains all the necessary information to extract the local structure of the system being studied. However, it is necessary to develop additional tools in order to extract such useful information about system behaviour from this massive amount of simulation data. In this thesis, I present a general method for identification of crystalline structures within any system during a molecular simulation. The method presented here combines the information provided by order parameters quantifying crystalline environments with a suitable Machine Learning algorithm. The developed method is then successfully applied towards identifying crystalline structures of gas hydrates.