The complex interplay of chemical reactions, defect chemistry, solvation dynamics, transport phenomena, and structural evolution at reactive interfaces underpin much of energy capture, conversion and storage by materials. A fundamental understanding of these dynamical processes at angstrom-to-mesoscopic length scales over several nanoseconds is crucial, and urgently needed to enable development of novel functional materials for various energy applications. Despite recent advances in computations (hardware, and efficient algorithms), an accurate description of reactive interfaces remains challenging. In particular, there is a substantial gap between highly accurate <i>ab initio</i> models that are restricted to extremely small sizes, and the classical atomic/meso-scale models with limited accuracy but access to larger length/time scales. Here, we will use representative examples from some of our recent works to demonstrate how machine learning (ML) can be used to bridge this gap, and develop classical potentials that are orders of magnitude faster than ab initio calculations but retain the accuracy/predictive-power of quantum methods. Our ML framework employs a data-driven approach, including generation and manipulation of extensive training data sets through electronic structure calculations, defining novel potential functional forms, employing state-of-the-art ML algorithms to formulate highly optimized training procedures, and subsequently developing user-friendly workflow tools integrating these algorithms on high-performance computers.This framework shows remarkable success in developing accurate reactive interatomic potentials for wide range of materials at unprecedented pace (few weeks as opposed to years using conventional methods). These results will be discussed in the context of accelerating development of functional materials for various energy applications (e.g., catalysts, low-dimensional thermal management, and friction reduction).

Bridging electronic and atomistic scales in modeling energy-relevant reactive interfaces using machine learning

2019 Fall National Meeting

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