Closed-loop photoreactor design enabled by machine learning and digital twins

Abstract Designing efficient photoreactors remains challenging due to the complex interplay of reflection, scattering, and absorption. Here, we introduce a closed-loop workflow integrating ray-tracing digital twins with multi-objective Bayesian optimization to design photoreactors with high photon capture and uniform irradiation. A digital 3D reactor model is calibrated using measured emission characteristics, reflectivity data, and geometrical properties and validated against potassium ferrioxalate actinometry. The validated model explores reflector shape and vial positioning, identifying geometries that maximize incident photon flux while maintaining homogeneity across 16 vials. Experimental validation using a photocatalytic Suzuki–Miyaura cross-coupling shows that reaction yields track simulated photon flux. The optimized reactor is further applied to quantum-yield screening, enabling mechanistic insights across diverse conditions in a single experiment. This work establishes an experimentally validated virtual co-design framework in which machine learning and digital twins accelerate photoreactor development and support quantitative understanding of photochemical processes.