Chemical Bonding in Metallic Glasses from Machine Learning and Crystal Orbital Hamilton Population
arXiv:2003.02722 · doi:10.1103/PhysRevMaterials.4.113603
Abstract
The chemistry (composition and bonding information) of metallic glasses (MGs) is at least as important as structural topology for understanding their properties and production/processing peculiarities. This article reports a machine learning (ML)-based approach that brings an unprecedented "big picture" view of chemical bond strengths in MGs of a prototypical alloy system. The connection between electronic structure and chemical bonding is given by crystal orbital Hamilton population (COHP) analysis; within the framework of density functional theory (DFT). The stated comprehensive overview is made possible through a combination of: efficient quantitative estimate of bond strengths supplied by COHP analysis; representative statistics regarding structure in terms of atomic configurations achieved with classical molecular dynamics simulations; and the smooth overlap of atomic positions (SOAP) descriptor. The study is supplemented by an application of that ML model under the scope of mechanical loading; in which the resulting overview of chemical bond strengths revealed a chemical/structural heterogeneity that is in line with the tendency to bond exchange verified for atomic local environments. The encouraging results pave the way towards alternative approaches applicable in plenty of other contexts in which atom categorization (from the perspective of chemical bonds) plays a key role.
arXiv commentThis update contains a number of enhancements. The manuscript has been updated after an anonymous review process by a journal whose policies discourage posting material from this process online. Few changes have been done in Sections I, II and III-A; however, Section III-B and IV have undergone significant changes
References in corpus (5)
- Quantum ESPRESSO: a modular and open-source software project for quantum simulations of materials
- Advanced capabilities for materials modelling with Quantum ESPRESSO
- On representing chemical environments
- Comparing molecules and solids across structural and alchemical space
- Machine Learning Unifies the Modelling of Materials and Molecules