Chemical reactions are rarely performed in isolation. Most real-world applications involve chemical reaction networks in which many simultaneous chemical reactions of many species occur. Advances in high-performance computer simulations and laboratory automation provide an increasingly detailed picture of these chemical reaction networks, but the underlying representation in terms of a graph of chemical species and reactions quickly becomes intractable to analyze as the complexity of the network grows. Network or graph models are often used to represent macroscopic dynamics over processes whose microscopic dynamics do not occur on practical temporal or spatial scales. Chemical reaction networks follow this pattern and represent first principles traversal of a high dimensions potential energy surface. Methods such as DFT can capture these paths, but cannot scale to useful macroscopic dimensions. Moreover, common graph or network representation of chemical reactions does not faithfully represent common higher-order or n-ary relationships. The approach described in this proposal uses approximations of these first principles to provide a fine enough selection method to find high quality paths from higher-order generalizations of graphs known as hypergraphs and thereby reduce overall DFT evaluation. The methods we use are built on novel geometric and topological tools which find mathematical meaningful global structures in practically applicable to chemical reaction data for Li-ion batteries and solid-state chemistry. The initial effort will focus on the extraction, representation, and analysis of Li-Ion Battery Electrolyte (LIBE) dataset. This dataset contains complete data related to the reaction chemistry present at the Surface Electrode Interphase (SEI). Our core objective is to provide principled exploration of large and applicable complex chemical reaction networks and thereby reduce the number of validation cycles by DFT. This is divided into three tasks: 1. Develop ETL software pipelines to build novel data representations of chemical reaction networks; 2. Develop computational reaction synthesis tools from topological data analysis and spectral theory for these novel data representations; 3. Demonstrate these new tools and showcase flexibility of data representation in the context of SEI chemical reactions. The new representation and methods can be compared to existing work on the same dataset which use a kinetic Monte Carlo inspired search over graphs. The phase I project will focus on chemical reaction networks related to the synthesis of lithium ion battery electrodes (inorganic chemistry) and chemical degradation processes of electrolytes in lithium ion batteries (organic chemistry). This has a direct commercial application to the development of improved lithium ion batteries for transportation and electronics, in alignment with the goals of the 2022 CHIPS and Science act. If successful, our approach can be applied to many other problems involving chemical reaction networks in Phase II, such as the development of new extreme-UV photoresists for the manufacture of next-generation microchips and âone-potâ chemical synthesis reactions that reduce the time, cost, and environmental impact of pharmaceutical synthesis and biopolyme
