Unsupervised local structure learning in elemental zirconium using Topological Data Analysis

Abstract

We discuss in this extended abstract a new application in materials science where topological data analysis (TDA) has proven to be relevant. TDA is used to analyze local atomic structures during homogeneous nucleation phenomena of pure zirconium in the undercooled liquid namely below its melting temperature. Persistent diagrams (PDs) are considered as descriptors to represent configurations of the physical system generated by large-scale molecular dynamics (MD) simulations up to one million atoms, and are evaluated by comparison to classical physical descriptors. This work is a first step towards a deeper understanding of the nucleation behavior of the system, where topological machine learning is a challenging but promising perspective.

Tasks

Reproductions