Challenges for AI in Discovery of new Crystalline Materials.

22^nd May 2025, 14:00
Robert Palgrave
University College London, UK

Abstract

Artificial Intelligence and Machine Learning (AI + ML) is making increasing impact in the physical sciences. New methods have appeared for prediction of crystalline compounds by combining AI+ML with computational methods like DFT, with the aim of accelerating crystal structure prediction and enabling targeted design of materials with specific properties. This talk will provide an overview of recent developments in this emerging field, and highlight early successes and failures. These include the huge databases of predicted crystalline materials produced by tech companies such as Google and Microsoft, as well as attempts at autonomous synthesis and characterisation of crystalline materials by 'self driving' laboratories. These early studies have highlighted the importance that will be placed on both crystallography and chemistry in an increasingly AI influenced field. The extremely high throughput of AI+ML crystal structure prediction magnifies some longstanding challenges in computational methods, but may also call for new perspectives on our descriptions of crystalline matter. How traditional materials science will change over the short and medium term will be discussed. Based in part on Leeman et al. PRX Energy 3, 011002.