Leveraging transfer learning for accurate estimation of ionic migration barriers in solids
Machine learning for faster discovery of ion-conducting materials
Fast ion transport underpins key technologies such as batteries, fuel cells and sensors, but it is difficult to probe experimentally and expensive to model with conventional atomistic simulations.
In this collaborative work between UCL and the Indian Institute of Science (IISc) Bengaluru, we use machine learning to overcome this bottleneck. Because high-quality data for ionic migration barriers are scarce, we adopt a transfer-learning strategy: the model first learns general chemical trends from large datasets and is then fine-tuned using a curated set of several hundred high-accuracy calculations.
Using a specialised graph neural network, this approach delivers reliable migration-barrier predictions at a fraction of the usual computational cost, enabling large-scale screening of known and hypothetical materials for next-generation energy applications.
Authors: Reshma Devi, Keith T. Butler & Gopalakrishnan Sai Gautam