MeGen - generation of gallium metal clusters using reinforcement learning

Abstract The generation of low-energy 3D structures metal clusters depends on the efficiency search algorithm and accuracy inter-atomic interaction description. In this work, we formulate as a reinforcement learning (RL) problem. Concisely, propose novel actor-critic architecture that generates low-lying isomers at fraction computational cost than conventional methods. Our RL-based uses previously developed DART model reward function to describe interactions validate predicted structures. Using incentivizes RL generate helps valid We demonstrate advantages our approach over methods for scanning local minima potential energy surface. not only isomer gallium minimal but also predicts families were discovered through previous density-functional theory (DFT)-based approaches.