Demystifying High Temperature/Pressure Material Synthesis through Physics-Informed Machine Learning

Design, discovery, and synthesis of new materials is a notoriously challenging problem, due largely to the massive associated design space and complex underlying phenomena. Simulations can provide a powerful means of navigating this problem space by providing both a capability for pre-screening and extracting otherwise inaccessible atomistically-resolved information on the underlying phenomena, but efforts are often limited by a lack of models exhibiting the necessary balance of accuracy and computational efficiency. In this presentation, we discuss recent efforts to overcome these challenges through development and targeted application of ChIMES, a physics-informed machine-learned interatomic model (ML-IAM) and supporting computational framework. We will present recent efforts to address grand challenges in ML-IAM development and application, e.g., toward reproducibility, reliability, and training efficiency as well as applications to high temperature/pressure nanocarbon synthesis.