Back to search

University of Southampton

Machine learning of quantum mechanical lattice energies for molecular crystal structure prediction

Abstract

dc:description.abstract

It is essential to study the crystal phases, i.e. polymorphs, of an organic molecule, under different conditions for the discovery and safety of new functional materials, and the effectiveness and toxicity of pharmaceuticals. This can be avoided by understanding the differences and likely transitions between polymorphs. Crystal Structure Prediction (CSP) is the process of generating crystal structures of a molecule(s). One way of doing this is to apply symmetry operations to a unit cell containing a molecule(s) representing the periodic structure, and optimising the geometries to local energy minima on the potential energy surface (PES) for the lattice. A local optimisation algorithm can be more useful than finding the global minimum of the energy landscape since several low-lying energy polymorphs can be observed.<br/><br/>Due to the number of atoms involved in organic molecular crystals, classical force fields (FFs) are typically used for calculating the lattice energy, and only the most likely, low energy candidates, are re-optimised using a higher level of the hierarchy of electronic structure methods.<br/><br/>The final re-optimisation step using a suitable electronic structure method is crucial since the lattice energy of polymorphs can differ by sub kJ/mol (Nyman et al. (2016); Nyman and Day (2015)) so method errors can significantly change the ranking of the lowest energy polymorphs, particularly if free energies are calculated. This becomes even more problematic due to the thousands of polymorphs in the low energy window of the energy landscape, for example 20 kJ/mol or below.<br/><br/>The aim is to reduce the enormous computational cost of this using Machine Learning (ML). This work will begin by reviewing current CSP methods and ML techniques. A set of thiophenes will become the focus to demonstrate the usefulness of CSP in the discovery of organic semiconductor materials. ML techniques will be applied to the PES to investigate an inexpensive geometry optimisation, where poorly predicted structures may be limiting the accuracy of the energy ranking of the CSP landscape.

Degree

thesis:*
Name dc:type.qualificationname
Ph.D.
Level dc:type.qualificationlevel
doctoral
Grantor dc:publisher.institution
University of Southampton
Year dc:date.issued
2024

Author and committee

dc:creator, dc:contributor.*
Author dc:creator
  • Clements, Rebecca Jane
Advisors dc:contributor.advisor
  • Day, Graeme
  • Woods, Dave
ORCID person:identifier.orcid

Rights

Language dc:language
English, eng

Chain of custody

source
Harvested from
University of Southampton
Base URL
eprints.soton.ac.uk/cgi/oai2
Last updated
2026-11-19
Source record
OAI-PMH GetRecord
related terms
citation

Clements, Rebecca Jane. Machine learning of quantum mechanical lattice energies for molecular crystal structure prediction. doctoral thesis, University of Southampton, 2024.