Bayesian optimization for nanoscale design of functionally superior AlCoCrFeNi high-entropy alloys

The traditional approach to designing function-specific high-entropy alloys (HEAs) involves exploring vast compositional spaces through extensive experimental or computational characterization. This “Edisonian” trial-and-error method is often time-consuming and resource-intensive. To address these challenges, this study presents a novel integration of molecular dynamics (MD) simulations with the machine learning (ML)-driven active learning framework for the nanoscale design of AlCoCrFeNi HEAs with an improved stiffness-to-density ratio ([Formula: see text]/[Formula: see text]). The proposed computational framework operates using a minimal number of samples generated by performing a series of MD simulations. In these simulations, the concentration of individual constituent elements is systematically varied, and the corresponding composition-dependent Young’s modulus and density are recorded. This data is then used to drive an ML-driven Bayesian [Formula: see text] framework, which identifies the next most promising composition in the HEA design space — one likely to yield a maximized specific Young’s modulus. To guide this search efficiently, the framework employs the Expected Improvement (EI) score as the acquisition function, allowing for the ranking and selection of candidate compositions. The optimal composition predicted by the Bayesian approach is then further explored through MD simulations to understand the underlying deformation mechanisms responsible for the enhanced specific stiffness.