Machine-Learning for Aqueous Interfaces

Supervisors: David Wilkins and Gareth Tribello


The interface between water and inorganic surfaces is the site of many key processes in materials science and chemistry, from geochemical processes through electrochemical applications to industrial uses. While a large number of experimental techniques exist for the surface-sensitive study of these systems, the complex nature of the interactions between water and the surface means that to fully understand these systems we must combine experiments with theory, particularly atomistic simulations. A concrete example is the dissolution of minerals, which begins with the adsorption of water on the surface, and which we need an accurate description of the water-surface interaction to understand. The complexity of these interactions means, however, that the electronic-structure theory required for simulations can be computationally very expensive, which has traditionally limited the kinds of systems that can be studied, and hindered direct comparisons between computation and experiment.

Aims and Objectives

Machine learning methods are becoming a popular means to sidestep expensive electronic-structure theory calculations, achieving accurate results with a fraction of the effort required. Recent developments in the field mean that it is possible to predict a wide variety of properties that were previously unobtainable, allowing not only the efficient simulation of interfacial systems, such as minerals, but also a high-quality prediction of experimental properties, including the results of spectroscopic experiments. The goal of this project is to develop the methods and models required for a highly accurate description of the interface between water and inorganic surfaces and to apply these tools to understand the interaction of water with mineral surfaces in great detail, and to rationalize the results of experiments on these systems.

The nature of this project means that the student will be trained in a variety of techniques used in modern computational chemistry, including atomistic simulations and electronic structure theory, allowing them to tackle problems at several scales, as well as the handling of large datasets. Further, the expertise they will gain in machine learning will put them at the forefront of this increasingly popular field.