Local Structure Controlling the Glass Transition in a Prototype Metal-Metalloid Glass

We use a structure analysis technique that quantifies cluster geometries using machine learning to identify a bicapped square antiprism (BSAP) as the nearest-neighbor cluster involved in the glass transition in Pd82Si18, a prototype metal-metalloid metallic glass. BSAPs have slow dynamics and grow significantly in concentration during cooling, like icosahedra in other systems. These results are evidence that some preferred structures universally contribute to the glass transition and show that analysis of interatomic geometry can find important structures not revealed by topology.