Session: COMP Poster Session:

Nowadays, QM/MM free energy are performed regularly to accurately predict reaction free energy profiles of solution-phase organic reactions. However, these calculations are rather compute-intensive, making them inaccessible to most research groups. In order to alleviate this problem, quantum-mechanics-based machine learning models have emerged recently to model the potential energy surface (PES) of large molecular systems and thus their reactivity at a lower cost.
In this presentation, we report our effort on the utilization of the deep learning package developed by Weinan E and co-workers¹to simulate two model organic reactions: (1) chloride exchange reaction of Cl^- + CH₃Cl; (2) Menshutkin reaction of NH₃ + CH₃Cl. The machine learning parameters, such as numbers of layers/neurons as well as the size of training dat were optimized. The accuracy of our ML-based free energy simulations was assessed by comparing the resultant free energy profiles against theoretical benchmark values.

Erick Calderon Leon

Appearances