Accelerating vacancy diffusion calculations by a DFT informed modified gaussian process regression method: A case study of austenitic 316 stainless steel

Advanced nuclear reactors have the potential to improve the efficiency and safety of nuclear energy generation. These reactors require advanced structural materials, such as austenitic stainless steels (SS), to withstand the inherently high-temperature and radiation environments. Qualification of these materials requires understanding time-dependent properties (i.e., creep and radiation induced segregation), which in turn depend on vacancy diffusivity (D_v) in the material. However, predicting D_v over large composition spaces, like that of austenitic SS, is prohibitively time intensive and expensive if conducted experimentally or using high-chemical-accuracy DFT calculations. To accelerate D_v prediction in nuclear structural alloys, this work constructed a modified Gaussian process regression (MGPR) based on select DFT calculations to examine the effects of atomic composition local to the vacancy on the vacancy migration energy barrier (E_v^m), using the 316 SS system as an initial case. A kinetic Monte Carlo (KMC) algorithm was employed, powered by the MGPR-predicted E_v^m, to predict D_v as a function of bulk composition and temperature within the austenitic SS bulk composition space. The E_v^m predicted by MGPR are utilized to determine that the average activated E_v^m is most sensitive to local Cr content. Effective activation energies inferred from Arrhenius analysis based on KMC D_v calculations show some agreement with those of experiment (' 0.76 eV or 23.24%) and KMC D_v values generally trend monotonically with bulk composition. Thus, this work provides a rapid and accurate pathway to understanding the bulk compositional dependence of D_v in austenitic SS. Further, this methodology may be extended to other solid solution ternary (and above) alloys for which D_v is of interest.