Bayesian Optimisation of Photopolymer Formulations: Priors, Constraints, Mixed Variables and Experimental Efficiency

Bayesian optimisation can reduce experimental burden when each formulation experiment is costly, but photopolymer development combines continuous concentrations, discrete chemistry choices, conditional variables, multiple objectives and hard manufacturability constraints. This review maps Bayesian optimisation to that mixed design space without treating the optimiser as a substitute for formulation science. It distinguishes prior knowledge from unsupported bias, model uncertainty from measurement uncertainty, feasibility from performance, and exploratory efficiency from qualification evidence. The central recommendation is to optimise within explicitly versioned state and constraint domains, preserve unsuccessful experiments, and stop when the remaining decision value is low rather than when a model merely appears numerically stable. Published materials-discovery and adaptive-manufacturing studies demonstrate sample-efficient closed-loop search, but direct transfer to photopolymer formulation requires state-resolved validation. ER-402.