Taming Two-Dimensional Polymerization by a Machine-Learning Discovered Crystallization Model

Rapidly synthesizing high-quality two-dimensional covalent organic frameworks (2D COFs) is crucial for their practical applications. While strategies such as slow monomer addition have been developed based on an empirical understanding of their formation process, quantitative guidance remains absent, which prohibits precise optimizations of the experimental conditions. Here, we use a machine-learning approach that overcomes the challenges associated with bottom-up model derivation for the non-classical 2D COF crystallization processes. The resulting model, referred to as NEgen1, establishes correlations among the induction time, nucleation rate, growth rate, bond-forming rate constants, and common solution synthesis conditions for 2D COFs that grow by a nucleation-elongation mechanism. The results elucidate the detailed competition between the nucleation and growth dynamics in solution, which has been inappropriately described previously by classical, empirical models with assumptions invalid for 2D COF polymerization. By understanding the dynamic processes at play, the NEgen1 model reveals a simple strategy of gradually increasing monomer addition speed for growing large 2D COF crystals. This insight enables us to rapidly synthesize large COF-5 colloids, which could only be achieved previously by prolonged reaction times or by introducing chemical modulators. These results highlight the potential for systematically improving the crystal quality of 2D COFs, which has wide-reaching relevance for many of the applications where 2D COFs are speculated to be valuable.