Hyperparameter\nOptimization for Atomic Cluster Expansion\nPotentials

Machine learning-based\ninteratomic\npotentials enable accurate materials simulations on extended time-\nand length scales. ML potentials based on the atomic cluster expansion\n(ACE) framework have recently shown promising performance for this\npurpose. Here, we describe a largely automated computational approach\nto optimizing hyperparameters for ACE potential models. We extend\nour openly available Python package, XPOT, to include an interface\nfor ACE fitting, and discuss the optimization of the functional form\nand complexity of these models based on systematic sweeps across relevant\nhyperparameters. We showcase the usefulness of the approach for two\nexample systems: the covalent network of silicon and the phase-change\nmaterial Sb<sub>2</sub>Te<sub>3</sub>. More generally, our work emphasizes\nthe importance of hyperparameter selection in the development of advanced\nML potential models.