Crystal structure prediction of energetic materials

Abstract

Crystal Structure Prediction (CSP) is a revolutionary tool for the field of materials design. CSP can aid in the characterisation and determination of experimental crystal structures. This is particularly useful for the field of Energetic Materials (EMs), where synthesising and handling these materials can carry extreme risks. Within this thesis, CSP is examined for its use on small organic EMs and is validated on a test set of 10 molecules.

CSP is used in conjunction with experimental synthesis throughout this research, with one prediction being made prior to synthesis. Flexible-molecule approaches are introduced in addition to rigid-molecule approaches and the results suggest both approaches should be used in tandem where possible.

A new technique for the CSP of copper(i) - organic extended network salts is introduced and a new force field, FIT-Cu, is parameterised for this purpose. This is shown to be very successful but also possess limitations for one copper(i) - organic extended network salt, CuADNP. Active Learned Neural Networks (ALNNs) are used to learn periodic DFT level single point total energies for structures at the force field level and predicted energy corrections are obtained for roughly 2 million crystal structures. The MACE foundation model is shown to be an effective intermediate optimisation stage for CSP, where re-optimisation of roughly half a million crystal structures can be performed at a reasonable computational expense.

The overarching aim of this thesis is to verify that CSP can be performed successfully for a range of different EMs, so that effective guidance can be provided for experimental determination and characterisation of such materials.

More information

Identifiers

ORCID for Graeme Day: orcid.org/0000-0001-8396-2771

Catalogue record

Export record