Using Machine Learning Hamiltonians To Compute Molecular Motor Barrier Heights

Machine Learning Inter-atomic Potentials (MLIPs) have become a common tool in use by computational chemists due to their combination of accuracy and speed. Yet, it is still not clear how well these tools behave at or near transitions states found in complex molecules. Here we investigate the applicability of MLIPs in evaluating the transition barrier of two, complex, molecular motor systems: a 1st generation Feringa motor and the 9c alkene 2nd generation Feringa motor. We compared paths generated with the Hierarchically Interacting Particle Neural Network (HIP-NN), the PM3 semi-empirical quantum method (SEQM), PM3 interfaced with HIP-NN (SEQM+HIP-NN), and Density Functional Theory calculations. We found that using SEQM+HIP-NN to generate cheap, realistic pathway guesses then refining the intermediates with DFT allowed us to cheaply find realistic reaction paths and energy barriers matching experiment, providing evidence that deep learning can be used for high precision computational tasks such as transition path sampling while also suggesting potential application to high throughput screening.