Machine-Learning-Potential-Driven Volcano Relationship in Grain Boundary Amorphicity and Ionic Transport of Antiperovskite Solid Electrolytes.

Ionic transport in solid-state electrolytes is governed by interfacial disorder and local chemical environments, yet their microscopic design principles remain challenging because large-scale interfacial kinetics are simulated using molecular dynamics (MD), where predictive design requires quantum-level accuracy. Here, we use an actively learned machine learning interatomic potential (MLIP) to bridge this scale-accuracy gap in halide-tunable antiperovskite solid electrolytes. By varying the local Cl-/Br- ratio and quantifying interfacial amorphicity and ion flux, we identify an MLIP-driven volcano relationship between structural disorder and ionic transport. Cl-enriched grain boundaries remain overly ordered, whereas Br- enrichment induces excessive disorder that disrupts migration. Intermediate halide compositions maximize ionic transport through moderately amorphous environments. Guided by this relationship, we design an anion-engineered grain boundary through selective halide tuning. These results establish controlled anion segregation as a strategy for reducing interfacial resistance and highlight MLIP-enhanced MD as a predictive framework for transport and disorder engineering in solid-state materials.