Composable and adaptive design of machine learning interatomic potentials guided by Fisher-information analysis

Abstract

An adaptive physics-inspired model design strategy for machine-learning interatomic potentials (MLIPs) is proposed. This strategy relies on iterative reconfigurations of composite models from single-term models, followed by a unified training procedure. A model evaluation method based on the Fisher information matrix (FIM) and multiple-property error metrics is also proposed to guide the model reconfiguration and hyperparameter optimization. By combining the reconfiguration and the evaluation subroutines, we provide an adaptive MLIP design strategy that balances flexibility and extensibility. In a case study of designing models against a structurally diverse niobium dataset, we managed to obtain an optimal model configuration with 75 parameters generated by our framework that achieved a force RMSE of 0.172 eV/{\AA} and an energy RMSE of 0.013 eV/atom.