Reinforcement Learning Configuration Interaction

Selected configuration interaction (sCI) methods exploit the sparsity of full (FCI) wave function, yielding significant computational savings and function compression without sacrificing accuracy. Despite recent advances in sCI methods, selection important determinants remains an open problem. We explore possibility utilizing reinforcement learning approaches to solve By mapping problem onto a sequential decision-making process, agent learns on-the-fly which include ignore, compressed at near-FCI This method, we call reinforcement-learned interaction, adds another weapon arsenal highlights how can potentially help challenging problems electronic structure theory.