Capturing Short-Range Order in High-Entropy Alloys with Machine-Learning Potentials

Chemical short-range order (cSRO) is recently reported to strongly influence the mechanical properties of various high-entropy alloys (HEA). However, the intricate nature of cSRO has made it challenging for current machine-learning potentials (MLP) to capture this feature, and many proposed approaches lack quantitative analysis of MLP performance on this task. In this work, we propose a generalized strategy to construct first-principles training databases and effectively train MLPs capable of characterizing cSRO in HEAs. We demonstrate this strategy by quantitatively analyzing the MLP performances in reproducing cSRO effects in various properties of CrCoNi HEA, including defect properties and phase stability.