|
Prediction of vacancy defect diffusion paths in high entropy alloys via machine learning on molecular dynamics data
C. Reimer, P. Saidi, C. Casert, C. Beeler, C.G. Tetsassi Feugmo, S. Whitelam, E. Mansouri, A. Martinez, L. Beland, I. Tamblyn Journal of Applied Physics, 138, 7, 074306 (2025) Understanding how vacancy defects move through materials is essential for predicting how alloys evolve over time, particularly under extreme conditions like radiation exposure. High entropy alloys, which consist of five or more elements mixed in roughly equal proportions, present a unique challenge because the local chemical environment around each vacancy is different, making it difficult to predict which direction a vacancy will hop next. In this work, we train graph convolutional neural networks on molecular dynamics simulation data to learn the rules governing vacancy migration in a five-element (Fe, Ni, Cr, Co, Cu) alloy system. The resulting machine learning model can generate synthetic vacancy diffusion trajectories roughly 100 times faster than traditional molecular dynamics while maintaining comparable accuracy in predicting diffusion coefficients, providing a powerful tool for bridging atomic-scale simulations to the longer timescales relevant for understanding material degradation and design. |
