Efficient Training of Machine-Learning Interatomic Potentials for Metallic Glasses
A workflow for fitting accurate, chemistry-specific machine-learning interatomic potentials to disordered metallic-glass alloys while minimizing costly first-principles data. A Lennard-Jones surrogate plus swap-Monte-Carlo sampling generates well-equilibrated amorphous configurations across many decades of supercooling, which single-point DFT corrections turn into a training set that yields near-DFT structural, dynamical, and mechanical accuracy.
2501.00589
Introduces a practical workflow for building chemistry-specific machine-learning interatomic potentials (MLIPs) for metallic glasses, a class of disordered metallic alloys whose rugged potential-ener…