<ul><li>A new multi-agent reinforcement learning method called Multi-Agent Crystal Structure optimization (MACS) has been proposed for periodic crystal structure optimization in computational chemistry and materials design.</li><li>MACS treats geometry optimization as a partially observable Markov game where atoms adjust their positions collectively to discover a stable configuration.</li><li>The method has been trained across various compositions of reported crystalline materials and shows scalability and zero-shot transferability, successfully optimizing structures of larger sizes and unseen compositions.</li><li>Benchmarking against state-of-the-art optimization methods, MACS optimizes periodic crystal structures faster, with fewer energy calculations, and lower failure rates.</li></ul>

MACS: Multi-Agent Reinforcement Learning for Optimization of Crystal Structures

Discover more