Machine Learning Materials Modelling

Large-scale machine learning (ML) models are transforming materials modelling by dramatically increasing the scale and speed at which materials can be explored. A central example of this development is the Alexandria database, which we created through high-throughput density functional theory (DFT) searches accelerated by crystal-graph-attention networks. With millions of DFT-relaxed crystal structures spanning one-, two-, and three-dimensional materials, Alexandria multiplies the number of known stable materials and provides a foundation for training next-generation universal ML force fields.
The talk will further explore how ML methods help bridge the gap between ab initio simulations and real experimental conditions. Examples include interpretable ML models that correct systematic errors in density functional theory and ML force-field–accelerated simulations that enable the study of materials questions previously beyond computational reach. Together, these advances highlight the transformative impact of ML on modern materials science.