Finding Needles in Haystacks: Deciphering a Structural Signature of Glass Dynamics by Machine Learning

The complex, disordered structure of glasses makes it challenging to elucidate how their atomic structure control their dynamics. Here, based on molecular dynamics simulations, we adopt machine learning (ML) to interrogate whether a structural signature governing the dynamics of atoms in a glass can be found. We find that the dynamics of glasses is encoded in their static structures. These results establish machine learning as a promising pathway to “find needles in Haystacks,” that is, to pinpoint important structural patterns in large complex datasets generated by atomistic simulations.