SVGym (SciVerseGym): An Environment for Reinforcement Learning and Bayesian Optimization in Crystal Discovery

SciVerseGym is a new, Gymnasium-compatible environment designed for reinforcement learning and Bayesian optimization in crystal discovery, framing the problem as a Markov decision process. It allows agents to perform various chemically meaningful actions, such as elemental substitution and atomic displacement, and evaluates candidates using machine-learned interatomic potentials or ASE-compatible calculators, facilitating an open and extensible framework for researchers in materials science. This environment supports customizable chemical spaces and rewards, making it a valuable tool for practitioners aiming to streamline and enhance closed-loop crystal search methodologies.