MLCrystals

The detection and classification of local structural order is a problem which has been studied extensively in a small number of systems, namely spherical particles and water. These two systems offer a limited insight into the diversity of crystals structures available, particularly those found in molecular crystals.

This is a project to develop tools and approaches for detecting and classifying structure within simulation data. Where structure is any arrangement of molecules of interest, whether that be a highly ordered crystal or a particular disordered state.

The goals of the project are as follows:

• Open and reproducible
• Communal resource
• Develop a reference workflow for the creation of labelled training datasets
• Create machine learning algorithms for the classification of arbitrary crystal structures trained from labelled data
• Create tools for the unsupervised classification of local structure.