Total: 1
Tungsten-tantalum alloys are promising plasma-facing materials for fusion reactors, but simulations of their coupled defect, mechanical, and microstructural evolution require an interatomic potential that remains accurate across diverse atomic environments. We develop a transferable W-Ta potential within the GRaph Atomic Cluster Expansion (GRACE) framework. The primary training dataset was generated automatically by maximizing information entropy in an atomically relevant feature space, avoiding manual enumeration of the structures and processes required for specific downstream applications. Extensive validation demonstrates a consistent description of unary and binary phase stability, elasticity, phonons, high-pressure response, surfaces, point defects, dislocation energetics, transformation pathways, alloy thermodynamics, and melting behavior. As a demonstration, the potential is applied to Ta segregation and shear-coupled migration of a $\Sigma5(120)$ W GB. Simulations reveal strongly site-selective segregation: Ta preferentially occupies a small subset of crystallographically distinct atomic rows. Under shear, segregated Ta impedes disconnection-mediated grain boundary migration. The de-pinning stress initially increases with Ta concentration but plateaus as the dominant pinning environment becomes saturated, while increasing temperature weakens both segregation and pinning. These results establish a broadly validated W-Ta potential and demonstrate how automated entropy-maximized data generation can produce transferable atomistic models capable of resolving mechanisms not explicitly targeted during training.