Graph-component approach to defect identification in large atomistic simulations

The graph-theoretical concept of connected components is employed to extract the evolution of defect configurations in a polycrystalline aluminum structure containing -8.3 million atoms. This graph-component approach is applied to reveal details of defect formation, transport, and transformation in the polycrystalline Al under large shear deformation. Building upon standard nearest neighbor analysis, graph theory and associated tools are used to reduce the multi-million-atom system into discrete component subgraphs that represent distinct structural defects. This method allows the automated identification, characterization, and tracking of defective regions within large volumes of data representing atomic-scale processes. Such analysis elucidates relationships between external stimuli, such as strain, and defect distributions, which have a large influence on material properties. The Graph Analytics for Large Atomistic Simulations (GALAS) codebase that implements this analysis, together with user guidance, is openly available https://github.com/pnnl/galas.

Automated summary

This article introduces a method for analyzing the evolution of defect configurations in polycrystalline aluminum using graph theory. The method can automatically identify, track, and characterize defect regions in large-scale atomic data and elucidate the relationship between external stimuli and defect distributions.