First-principles methods have become increasingly important in the quest for alloys with extraordinary properties. Given the extensive composition space and intricate atomic structure of high entropy alloys, sophisticated ab initio-based methods are paramount for their efficient and accurate characterization. Despite the absence of long-range order, high entropy alloys often exhibit local short-range ordering, exerting a significant influence on their properties. Hence, the accurate assessment of local order is crucial for obtaining a precise structural representation, estimating phase stability and predicting order-disorder transitions.

In this study, we present a novel methodology for exploring Pareto-optimal compositions by integrating density functional theory (DFT) computational workflows into a probabilistic Bayesian multi-objective optimization framework. Initially, a workflow is employed to evaluate short-range ordering and structural stability through effective interactions. Subsequently, this information is used to assess various mechanical and physical models, employing parameters using ab-initio accurate actively-learned interatomic potentials and efficient coherent potential approximation. The outputs of these models serve as optimization objectives.

To minimize the amount of queries during the search for optimal alloys, we adopt a two-stage approach. First, we build a surrogate model that evaluate our objectives in a cost-aware fashion. Then, the Pareto front is determined from the model posterior mean via genetic algorithms. This two-stage approach
not only reduces the overall computational effort but also enhances robustness against potential outliers and noise. Ultimately, we conduct a systematic exploration of the composition space to identify multi-component alloys that are Pareto optimal with respect to diverse properties, including strength, ductility, thermal stability, and more.