Composition and Temperature Dependence of Point-Defect and Elastic Properties in Multi-Component Alloys for Fusion Applications by Machine-Learning Simulations

In nuclear fusion technology, there is a need for microscopic models for defects as nonlinear sources of stresses and strains since the information about eigenstrains of radiation defects is crucial for the simulations using the finite element method. One of the methods that enable to compute the eigenstrains of defects effectively is molecular dynamics (MD). The primary challenge in atomistic simulations is the availability of realistic interatomic potentials for the alloy systems. An interesting alternative to the conventional interatomic potentials is the use of machine learning (ML) approaches.<br/>In this work, we demonstrate the advantages of ML interatomic potentials using two distinct groups of materials relevant for fusion applications: bcc Fe-Cr-He alloys and Ta-Ti-V-W high-entropy alloys. Accurate and fast ML potentials for both alloy systems were developed based on thousands of density functional theory (DFT) calculations performed on representative structures with varying alloy compositions and different atomic configurations. Various ML approaches, including linear ML and kernel models, along with different types of atomic descriptors, were systematically tested to achieve a good balance between accuracy, speed, and predictive power [1]. The root-mean-square errors between the forces computed using DFT and ML potentials are consistently below ~0.2 eV/Å.<br/>The developed ML potentials were applied in MD simulations to study selected properties, including elastic and point defect properties of Fe-Cr-He and Ta-Ti-V-W alloys as a function of composition. The results obtained were compared with available DFT results. Importantly, MD simulations with ML potentials enabled the study of alloy properties at elevated temperatures and their melting temperatures, which is nearly impossible using the DFT technique. Our study highlights the influence of temperature on the formation free energy of point defects, with vibrational entropy playing a significant role.<br/><br/>[1] A.M. Goryaeva, J. Dérès, C. Lapointe, P. Grigorev, T.D. Swinburne, J.R. Kermode, et al. Phys. Rev. Mater. 5 (2021) 103803.