Discovery of highly fluorescent covalent organic frameworks through AI-assisted iterative experiment-learning cycles

The development of porous crystalline materials with targeted properties remains challenging owing to the vast chemical design space and the high cost of experimental screening. Here we develop an artificial-intelligence-assisted interactive experiment-learning evolution approach to accelerate the discovery of highly fluorescent covalent organic frameworks (COFs). This approach integrates model recommendation, experimental validation and active learning in an iterative refinement cycle, allowing the artificial intelligence model to evolve along the process. Among the 520 possible combinations derived from a library of 20 amine and 26 aldehyde building blocks, we needed to experimentally evaluate only 11 COFs to identify one with a remarkable photoluminescence quantum yield of 41.3%. By embedding electronic configuration and quantum-level insights into the learning process, this approach transcends intuition based on statistical analysis intuition to enable material discovery driven by chemical knowledge, enhancing prediction reliability and interpretability. We also reveal the fluorescence mechanism of these COFs and outline the critical role of HOMO-LUMO alignment and excited-state charge distribution.