Predicting mechanical properties from electronic structure in refractory high-entropy alloys using machine learning and DFT

The electronic density of states (DOS) at the Fermi level, N( E f ) is emerging as a key descriptor of bonding, stiffness, and ductility in the analysis of mechanical properties of alloys, yet its direct evaluation requires computationally expensive density-functional theory (DFT) calculations. We develop a machine-learning (ML) model that accurately predicts N( E f ) from composition-derived chemical and thermodynamic descriptors, enabling rapid estimation of N( E f ) in refractory high entropy alloys (RHEAs). The DFT-calculated N( E f ) is then incorporated as an input feature in the ML models for predicting Young's modulus ( E VRH ) and Pugh ratio (G/B). The proposed framework achieves excellent agreement with DFT, with R 2 = 0.973 for N( E f ), R 2 = 0.925 for E VRH and R 2 = 0.968 for Pugh ratio. The models capture systematic variations associated with d-band filling and bond stiffness rather than relying on purely empirical correlations. By identifying N( E f ) as a transferable descriptor, this work directly links composition, electronic structure and mechanical properties, thereby providing efficient exploration of RHEAs for strength-ductility relationships by circumventing expensive DFT calculations yet including their information.