Mechanistic understanding and prediction of vacancy formation energies in FCC HEAs from DFT and Machine Learning

Calculation of vacancy formation energies (VFE) in structural alloys is critical to understand high temperature deformation processes such as diffusional creep. Due to chemical and lattice symmetries, while pure metals have single values of VFE, concentrated alloys, such as high entropy alloys (HEAs), have a range of VFEs due to diverse nearest neighbor (NN) environments. While previous studies have illustrated a correlation between NN chemistry and VFE, here, using density functional theory (DFT) calculations, we show that atomic volume and electronegativity emerge as new parameters in HEAs that control VFEs. Specifically, larger atomic volume and charge gain due to higher electronegativity raise VFEs. We find that such mechanisms can lead to large VFE variations, at times over 1 eV that would impact deformation processes. Further, because DFT calculations are expensive, especially in the large compositional space of HEAs, we employ a graph neural network (GNN) machine learning model that enables VFE predictions in multi-elemental alloys from simpler binary alloys. In addition, we show that DFT calculations can be completely by-passed by carefully fine-tunning CHGNeT to predict atomic volume, charge transfer and atomic displacement in HEAs. While this model is developed for FCC alloys in Ni-Cu-Au-Pd, it enables accurate predictions in Ni-Co-Cr system with minor tuning indicating extrapolative nature of the model. The work is supported by NSF DMR CDS&E #2302763.