Computational design of compositionally-complex structural disordered refractory alloys

New alloys exhibiting both high-temperature strength and room-temperature ductility are urgently needed for structural applications. Multi-component (4+ elements) refractory high-entropy alloys (RHEAs) have emerged as promising candidates for overcoming this strength-ductility tradeoff. However, exploring the vast compositional space of RHEAs requires new computational techniques, since processing challenges limit availability of experimental data [1]. We are developing chemical context-dependent machine learning techniques based on high-quality electronic structure calculations to efficiently explore this combinatorically-large design space. We report DFT and density functional perturbation theory (DFPT) calculations of structural refractory unary and disordered binary BCC alloys of Mo, Nb, Ta, and W across systematically-varied compositions [2]. The results have been validated against available experimental data, yielding excellent agreement for bulk properties, binary alloy lattice and elastic constants, and formation enthalpies. Based on this computed electronic structure data we have constructed novel quantum-scale descriptors to serve as the latent space foundation for machine learning models of RHEA properties. The descriptors include lattice-dependent elastic property nuclear relaxation fields (C₁₁, C₁₂, C₄₄; Young’s, bulk, and shear moduli; Poisson’s ratio), and unfolded band structures with associated spectral weights.