StriderNET: A Graph Reinforcement Learning Approach to Optimize Atomic Structures on Rough Energy Landscapes
This work addresses a challenging problem in fields like drug design and materials discovery, offering a novel method that shows strong performance gains, though it is incremental in applying reinforcement learning to a known bottleneck.
The paper tackled the problem of optimizing atomic structures on rough energy landscapes by introducing StriderNET, a graph reinforcement learning approach that learns to displace atoms towards low-energy configurations, and demonstrated that it outperforms classical optimization algorithms by discovering lower energy minima and achieving a higher rate of reaching minima across three complex atomic systems.
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, StriderNET, that learns a policy to displace the atoms towards low energy configurations. We evaluate the performance of StriderNET on three complex atomic systems, namely, binary Lennard-Jones particles, calcium silicate hydrates gel, and disordered silicon. We show that StriderNET outperforms all classical optimization algorithms and enables the discovery of a lower energy minimum. In addition, StriderNET exhibits a higher rate of reaching minima with energies, as confirmed by the average over multiple realizations. Finally, we show that StriderNET exhibits inductivity to unseen system sizes that are an order of magnitude different from the training system.