Abstract
Utilizing computational frameworks that involve simulation, modeling, and machine learning has gained popularity in lubricant industries to speed up research development. The frameworks serve as digital twins to aid the design of new lubricants by allowing the study of molecular assembling processes, analyzing various candidate chemicals, and understanding their physical properties under different application conditions to complement laboratory experiments. This work aims to evaluate the performance of the Proximal Policy Optimization (PPO) deep reinforcement learning (RL) agent in describing long-chain folding hydrocarbons, compounds commonly used as a main ingredient in lubricant industries. The hexadecane structure is a suitable benchmark molecule for assessing the RL agent. The policy learned by the RL agent encodes the intramolecular characteristics required to dictate the activity of individual molecules. Once trained on ab initio molecular dynamics trajectories, the RL molecular agents act in a virtual environment. Observing the dynamics of topological shapes and their properties then demonstrates the agents’ ability.