Unlocking chemical complexity in machine learning for battery materials

Machine learning (ML) can be regarded as a branch of a broader field of artificial intelligence that helps computers learn patterns from data. It builds on ideas from statistics and probability to find hidden relationships in complex systems - like understanding how human language works or how we recognise objects in photos.
An example of a complex mathematical relationship that chemists need to understand is that between the structure and composition of a chemical system and its physical properties such as energy, stability, and performance. For example, given a 3D model of a new molecule or material, chemists want to be able to predict whether it will be stable enough to synthesise. Being able to answer that question quickly and accurately would help scientists design new materials faster and understand what makes existing ones work so well.
Battery materials stand out as primary beneficiaries of such computational approaches since chemical compounds used as battery components are usually complex systems, often consisting of many unique elements from the periodic table. As electric vehicles attract increasingly more attention, we need to discover new materials to make batteries more efficient, more durable, and safer. Computer simulations using ML can speed up this discovery process and help us understand the materials we already use.
The research proposed herein aims to further the development of ML models that can accurately predict the physical properties of any chemical system based solely on the spatial arrangement of its atoms. Although similar models already exist, most are tailored to specific materials-meaning one needs to build a new model for every type
of substance they want to study. That approach is time-consuming and limits our ability to understand large systems, like batteries, which contain many different parts.
This project will explore emerging directions in the development of widely applicable interatomic potentials, with application to chemically complex battery material systems.
The work falls within the EPSRC Computational and theoretical chemistry research area and aims to push the boundaries of how we use machine learning to accelerate materials discovery.