Developing and applying machine learning approaches to understand water at interfaces

Brief description of the context of the research including potential impact;
This project will focus on the development of highly accurate water potentials at interfaces and on surfaces using machine learning approaches. Research is continually hampered by having to make the tradeoff between accuracy and speed, should an investigation into a particular problem make use of classical force field molecular dynamics, or does the nature of the problem warrant the greater accuracy of ab initio methods?

Aims and objectives
Despite the familiarity of this question to many researchers, there is no fundamental reason why classical MD cannot attain the same accuracy that ab initio methods have; it is the aim of this project to design a force field capable of providing this accuracy. A highly accurate molecular dynamics potential would facilitate the study of large ensembles of molecules while maintaining the accuracy of the underlying ab initio training data. The usefulness of such a capability would be immediately obvious, for example in understanding the behavior of water under confinement, in developing nanoporous water filters and in understanding the processes of ice nucleation and crystal growth.

Novelty of the research methodology
Machine learning techniques have grown in popularity in recent years in a wide variety of fields due to their unique power in unravelling complex multidimensional problems. While they have been applied to the development of molecular force fields before, previous research has focused largely upon the bulk material. In applying this methodology to an interfacial system (e.g. water on carbon) this research will both push the boundaries of the current state-of-the art and produce an end 'product' which has a vastly more widespread applicability; interfaces being generally more interesting from a research perspective than the bulk.

Alignment to EPSRC's strategies and research areas
The wider impact of this research would be felt across many branches of the EPSRCs portfolio, it is challenging to conceive of a field which would not benefit hugely from development of highly accurate molecular dynamics force fields. There would be applications in medicine (biosensor development, understanding of fundamental biological processes, drug development), engineering (highly selective filtration, corrosion processes, materials development) etc.

Any companies or collaborators involved
This research is currently being conducted in collaboration with Gabór Csányi at the university of Cambridge, and Laurent Joly of Université Lyon.