Novel materials for electrochemical energy storage and conversion, such as doped oxides or multi-component alloys, can have disordered or amorphous structures with highly complex chemical compositions. Such materials are challenging to model with first principles methods, as extensive sampling is typically required to determine the structure, atomic ordering, and composition simultaneously.<br/><br/>Here I will show how first-principles based machine-learning potentials (MLPs) constructed using the open-source aenet package (http://ann.atomistic.net) can be used to accelerate structural and compositional sampling. We employed specialized MLPs for the Monte-Carlo sampling of defect and dopant distributions in catalyst nanoparticles, and we developed an approach that couples MLPs with a genetic algorithm for constructing the phase diagram of amorphous materials. I will illustrate with practical examples how the thermodynamically accessible structure models identified by MLP-sampling can subsequently be used as input for accurate first-principles analyses.

Simulating complex inorganic materials for energy applications with machine-learning potentials

2019 Spring National Meeting

The official archive and repository for all ACS Meetings & Expositions abstracts and select posters or presentations.