StriderNET: A Graph Reinforcement Learning Approach to Optimize Atomic Structures on Rough Energy Landscapes

Optimization of atomic structures presents a challenging problem, due to their highly rough and non-convex energy landscape, with wide applications in the fields of drug design, materials discovery, and mechanics. Here, we present a graph reinforcement learning approach, <PRE_TAG>StriderNET</POST_TAG>, that learns a policy to displace the atoms towards low energy configurations. We evaluate the performance of <PRE_TAG>StriderNET</POST_TAG> on three complex atomic systems, namely, binary Lennard-Jones particles, calcium silicate hydrates gel, and disordered silicon. We show that <PRE_TAG>StriderNET</POST_TAG> outperforms all classical optimization algorithms and enables the discovery of a lower energy minimum. In addition, <PRE_TAG>StriderNET</POST_TAG> exhibits a higher rate of reaching minima with energies, as confirmed by the average over multiple realizations. Finally, we show that <PRE_TAG>StriderNET</POST_TAG> exhibits inductivity to unseen system sizes that are an order of magnitude different from the training system.