A Simple Approach to Atomic Structure Characterization for Machine Learning of Grain Boundary Structure-Property Models

Abstract

Grain boundaries (GBs) have a significant influence on the properties of crystalline materials. Machine learning approaches present an attractive route to develop atomic structure-property models for GBs because of the complexity of their structure. However, the application of such techniques requires an appropriate descriptor of the atomic structure. Unfortunately, common crystal structure identification techniques cannot be applied to characterize the structure of the vast majority of GB atoms (50-98% are classified as other). This suggests a critical need for atomic structure descriptors capable of identifying arbitrary atomic environments. In this work we present a simple procedure that facilitates the identification of arbitrary atomic structures present in GBs. We apply this approach to characterize the atomic structure of the 388 GBs from the Olmsted data set (Olmsted 2009). We show how this approach facilitates visualization of GB atomic structures in a way that reveals important structural information. We test the recently proposed hypothesis that Σ3 GBs contain facets of the GBs that form the corners of the corresponding GB plane fundamental zone. Finally, we briefly demonstrate how the structure descriptors resulting from our approach can be used as inputs to machine learning approaches for the development of atomic structure-property models for GBs.