High-Throughput Calculation of Ab Initio Data for Automated Kinetic Model Development

University Of Georgia Research Foundation Inc, Athens GA

Investigators

Abstract

Several important scientific and engineering research areas depend on the ability to accurately predict phenomena that occur in gas-phase, chemically reacting systems. For example, high-fidelity simulation of the combustion of new fuel mixtures enables the co-design of fuels and engines for cleaner and more efficient use of transportation energy. Such simulations play an important role in atmospheric sciences and astrochemistry as well. The accuracy of these simulations depends on what is known as a “kinetic model.” Kinetic models are data files prescribing the chemical network of a simulation: the possible chemical species formed and consumed, and the rates at which they undergo various reactions as a function of temperature and pressure. The process of collating the thousands of reaction rates and other properties needed for such a model has been recently facilitated by machine-learning tools that extrapolate from a database of known values. However, such technologies suffer from a shortage of data needed to produce reliable predictive models that are applicable to a broad range of engineering applications. To address this shortage, this project transforms and scales the AutoMech software suite to enable both the generation of new, high-accuracy kinetic models from scratch and the population of databases to enhance the predictions made by machine learning tools. This project leverages and extends existing open-source software for parallel workflow orchestration, standardized quantum chemistry program operation, and molecular geometry optimization and commits to the continued maintenance of these essential tools. It also supports the FAIR (findable, accessible, interoperable, and reusable) data publication through a JSON data exchange file with a publicly available schema. This file includes all electronic structure and master equation results from an AutoMech mechanism development workflow, with sufficient provenance information for reproducibility. The software is designed to be highly adaptable to new applications and technologies: for parallel execution, workflow orchestration back-ends are hidden behind a common interface with flexibility to add new ones; for electronic structure calculations, users have the option to connect their own methods and programs as plug-ins and call external codes for alternate implementations of high-level routines; for data management, the database schema adheres to an objective representation of potential energy surface data adaptable to any application or range of conditions. As quantum chemistry and high-performance computing continue to advance, these design choices enable AutoMech to rapidly incorporate new technologies emerging from these two fields, benefiting its users and addressing scientific challenges. This award by the Office of Advanced Cyberinfrastructure is jointly supported by the Division of Chemistry in the Mathematical and Physical Sciences Directorate. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.