This paper presents a physics-informed Bayesian optimization (BO) method that improves the efficiency of optimizing expensive-to-evaluate functions, such as those from physical simulations. The proposed method uses multi-output Gaussian processes to model the full vector of physical observables before mapping to a scalar objective, retaining more information and accelerating convergence. The performance of this method, implemented within JCMwave's JCMoptimizer suite, is benchmarked against standard BO and heuristic methods on real-world problems, including the inverse design of a nanophotonic beam-splitter simulated with JCMsuite.
I. Sekulic, et al. Physics-informed Bayesian optimization of expensive-to-evaluate black-box functions. Mach. Learn.: Sci. Technol. 6, 040503 (2025).