Multi-objective Bayesian Optimization for Experimental Design in Copolymerization and Revealing Chemical Mechanism of Pareto Fronts

The chemical properties of copolymers are strongly influenced by a number of intrinsic characteristics, such as their molecular weight and their monomer composition ratio. Identifying the optimal conditions for the copolymerization process that result in a synthesized copolymer with the desired characteristics is a major challenge. Optimization of the copolymerization process has traditionally been based on trial-and-error approaches by humans, relying on empirical rules. Thus, the design space that can be explored experimentally is severely limited by time and economic constraints. In addition, solving problems such as nonuniformity of both the temperature and the concentration of chemicals in the reaction field is also challenging. In this study, we established multiobjective Bayesian optimization and flow copolymerization systems to explore optimal copolymerization conditions for synthesizing copolymers that simultaneously exhibit multiple target characteristics. Finally, we visualized Pareto fronts representing trade-offs between polymer characteristics and employed quantum chemical calculations to reveal the chemical origins of the Pareto fronts.