Prof. Aldo H. Romero - Systematic symmetry-based search from DFT to AI, methods to expand two-dimensional structural databases

In this talk, I will present a novel approach that combines an exhaustive symmetry-based
method with graph neural networks for structural search and design. I will provide a
comprehensive explanation of the methodology, focusing on the obtained geometries and
introducing new crystal families observed in two and three species compositions.
Additionally, I will delve into the concept of graph neural networks and their relevance in the
context of crystal structures. While highlighting the versatility of this method in materials
science, I will touch upon other potential applications beyond crystal structures. Regarding the
systematic symmetry-based approach, I will share details on the screening process and
introduce our transfer learning process aimed at optimizing crystal structures using the
proposed Universal potential. Furthermore, I will discuss previously unreported results and
explore how existing databases can assist theoreticians and experimentalists in their search for
intriguing materials.