(443e) Using Data Driven Models to Gain Insight on Spin- and Oxidation-State Dependent Behavior of Reaction Energetics for Light Alkane Oxidation

Nandy, A., Massachusetts Institute of Technology
Kulik, H. J., Massachusetts Institute of Technology
Janet, J. P., Massachusetts Institute of Technology
Duan, C., Massachusetts Institute of Technology
Computational high-throughput virtual screening (HTVS) with first-principles density functional theory (DFT) can play a valuable role in unearthing design rules for scalable, industrially viable synthetic analogues that preserve selectivity and activity seen only in enzymes. Single-site catalysts represent the most promising analogues to these enzymes, often enabling atom-economy and selectivity not possible with bulk heterogeneous catalysts. Simultaneously, a wide chemical space must be explored in the realm of single-site catalysts, even beyond the reach of conventional DFT screening, in order to simultaneously address other constraints, such as high turnover number or robustness, earth abundance, and synthesizability. Single-site catalysts have the added dimensionality of spin- and oxidation-state, which can drastically impact the structure-property relationships of molecular complexes and remain unexplored for catalyst reaction energetics. We demonstrate our developments on data-driven models for key steps in single-site light alkane oxidation catalysts, which enable the prediction of reaction energetics in a spin- and oxidation-state dependent manner close to the accuracy of DFT. We first compare prediction of reaction energetics to other quantum mechanical properties that we have predicted, such as spin-splitting energetics, ionization potential, and frontier orbital energetics. We then discuss insights on representative catalytic reaction steps such as oxo formation and hydrogen atom transfer, including their dependence on spin- and oxidation-state. Lastly, we demonstrate the power of data-driven models for purposes of screening, enabling screens of large design spaces that would be infeasible to screen by DFT, even with the reduction in number of calculations through the use of descriptor energies. Having separate data driven models for different reaction energy steps allows predictions of reaction energies that contain weak or nonexistent linear free energy relationships (LFERs), removing dependence on LFERs that have been shown to be broken in single-site catalysis.