Abstract by Brayden Bekker
Physics and Astronomy
Transferable Machine-Learned Interatomic Potential for Materials Discovery
We use a machine learning method called Moment Tensor Potentials (MTPs) to explore Co-based alloys for high-temperature strength materials called superalloys. The MTP learns the atomic properties of reference calculations and extrapolates to create a predicted potential energy surface of the alloys system. We demonstrate that a ternary (three-element) MTP generalizes well to quaternary (four-element) systems by identifying important atomic configurations. We show the consistency of our results with published experimental results for the Al-Co-Ta-V system and make predictions for the Al-Co-Nb-V system.