| Abstract Scope |
Liquid and glass lack lattice periodicity, and yet their atomic structures are not totally random, characterized by strong short- and medium-range correlations. Describing and understanding such a state is a highly non-trivial task. Despite the long history of effort, the subject is still poorly resolved. Thus, machine-learning appears to be a powerful tool to unravel such a complex state. However, here the distinction between describing and understanding becomes clear. We show an example of the principal component analysis that gives a good description of the structure. But understanding requires conceptual development, which ML does not do a good job. The concepts, such as the medium-range order, are less accurate but give more useful idea. To achieve better understanding, an approximation, which gives a poorer description but clear idea, often is better suited. This work is supported by the US Department of Energy. |