High-throughput materials screening algorithm based on first-principles density functional theory and artificial neural network for high-entropy alloys

This work introduced the high-throughput phase prediction of PtPd-based high-entropy alloys via the algorithm based on a combined Korringa-Kohn-Rostoker coherent potential approximation (KKR-CPA) and artificial neural network (ANN) technique. As the first step, the KKR-CPA was employed to generate 2,720 data of formation energy and lattice parameters in the framework of the first-principles density functional theory. Following the data generation, 15 features were selected and verified for all HEA systems in each phase (FCC and BCC) via ANN. The algorithm exhibited high accuracy for all four prediction models on 36,556 data from 9139 HEA systems with 137,085 features, verified by R2 closed to unity and the mean relative error (MRE) within 5%. From this dataset comprising 5002 and 4137 systems of FCC and BCC phases, it can be realized based on the highest tendency of HEA phase formation that (1) Sc, Co, Cu, Zn, Y, Ru, Cd, Os, Ir, Hg, Al, Si, P, As, and Tl favor FCC phase, (2) Hf, Ga, In, Sn, Pb, and Bi favor BCC phase, and (3) Ti, V, Cr, Mn, Fe, Ni, Zr, Nb, Mo, Tc, Rh, Ag, Ta, W, Re, Au, Ge, and Sb can be found in both FCC and BCC phases with comparable tendency, where all predictions are in good agreement with the data from the literature. Thus, the combination of KKR-CPA and ANN can reduce the computational cost for the screening of PtPd-based HEA and accurately predict the structure, i.e., FCC, BCC, etc.