(162b) Exploiting Electronic Structure and Machine Learning Models for Discovery in Transition Metal Chemistry
Although the discovery and synthesis of new materials, catalysts, and functional molecules represents the foremost effort that unifies thousands of researchers in the chemical science community, presently characterized compounds represent a minute fraction of chemical space. The highly tunable electronic structure properties of inorganic complexes (i.e., variable spin, oxidation state, and coordination number) make them attractive targets for applications in energy storage, functional materials, and catalysis but present a daunting combinatorial challenge. This vast transition metal compound space cannot be fully enumerated by any traditional Edisonian approach. In order to advance quantitative structure-activity relationships, reveal emergent phenomena, and accelerate design of materials and catalysts, smarter and faster computational approaches are needed. In this talk, I will outline our efforts to accelerate first-principles (i.e., with density functional theory, or DFT) screening of inorganic complexes for catalysis and materials science: i) We automate and simplify simulation, eliminating the need for tedious preparation of input files or commands by the user, ii) We develop machine learning (ML) models (e.g., artificial neural networks) that predict the outcomes of simulations prior to their completion, and iii) We integrate these tools into an automated design workflow that can make essential decisions on which simulations are best to carry out and why, with awareness of ML model and DFT model uncertainty. Time permitting, I will describe applications of these tools for advancing understanding in catalysis and functional spin crossover materials.