Large-scale chemical reaction networks are complicated nonlinear dynamical systems that are difficult to model and interpret. While <i>ab initio</i> molecular dynamics (MD) is a key computational method that can be used to interrogate reactive chemistry, many challenges remain due to its computational limitations in time and length scales. In this talk, we will first discuss how it is possible to use the wealth of data generated by a few expensive MD simulations to build fast kinetic Monte Carlo (KMC) models of chemical reaction networks. With this framework, one can imagine a future where MD data will be routinely archived to form a “chemical genome” that can be used to quickly simulate new chemical systems. Second, we will consider the question of how to build reduced models of these chemical reaction networks, both to build understanding and to accelerate simulations. We have developed a data-driven method utilizing L<sub>1</sub>-regularization for model reduction of nonlinear dynamical systems, which can handle large-scale systems with as many as thousands of reactions in a matter of minutes. Finally, we will conclude with a discussion of how machine learning can be used to adaptively reduce the learned reaction network under different initial conditions and enable truly long-time scale extrapolation of reduced-order models. Our software utilizes computationally efficient methods that can be interfaced with existing simulation software to enable interpretable and extrapolative reduced-order modeling of complex reactive chemistry.

Data-driven learning and interpretable model reduction of chemical reaction networks

Fall 2022 abstracts, posters, and presentations will launch in mid-September.

ACS Fall 2022

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