Efficient exploration of potential energy surfaces (PES) is essential for predicting realistic material properties, since materials often exist as ensembles of metastable structures and transition pathways. Modern methods combine global optimization, molecular dynamics, metadynamics, and transition state search to sample both minima and their connections. Machine-learned interatomic potentials (MLIPs) now enable much faster and larger-scale PES exploration, but also introduce challenges in managing and comparing vast structural datasets. Improved representations and clustering methods are helping address this. Together, these advances are enabling deeper insight into materials such as batteries, catalysts, and functional solids, linking structure, dynamics, and experimental behavior more directly than before.
[more]
