Critical Composition Window for Co–P Bulk Metallic Glasses Revealed by Machine-Learned Interatomic Potentials

Bulk metallic glasses are a distinct class of alloys whose disordered atomic structure imparts exceptional mechanical strength, corrosion resistance, and catalytic activity. Nevertheless, the optimization of their compositional ratios remains predominantly empirical, relying largely on qualitative and nonpredictive computational models. In this work, we developed a machine interatomic potential for Co–P binary alloys, trained on approximately 32,000 density functional theory-labeled configurations spanning a wide range of compositions (5–40% P) and temperatures (300–1800 K). The model accurately reproduced reference energies and forces and remained reliable in high-temperature molecular dynamics simulations. Analysis of structural motifs revealed that high phosphorus content promotes medium-range ordering, while very low phosphorus content favors partial crystallinity. Intermediate compositions (20, 15, and 10% P) exhibited structural signatures of bulk metallic glasses, including pronounced icosahedral clustering, split second-neighbor peaks in radial distribution functions, and coordination environments resembling distorted bicapped square antiprisms. Thermal and dynamical assessments showed that these compositions form stable amorphous phases with favorable glass transition temperatures. The developed potential enables large-scale simulations and offers guidance for optimizing Co–P-based metallic glasses for structural and energy applications.