Bayesian optimization of optical materials with tunable bandgaps

Bayesian optimization methods allow for efficient exploration of stable solar cell materials with desired values of bandgap energies. quantum-chemical methods to prospect the best candidate from a pool of thousands compounds can take more than 150 experiments. Conventional global optimization methods require over 2,000 iterations, which renders high-accuracy calculations of candidate materials computationally intractable. Scientific-of-science: Many optimization techniques are very sensitive to properties of the optimization problem, a situation that scientific-of-science study would aim to reveal. Bayesian symbolic regression reveals a procedure for the eigensolver to converge from a random wave function in 94 % of the cases. These human insights on the shape of the band structure are ingrained in the Bayesian optimization approaches, and in our work, the band gaps of the best performing hybrid organic— perovskite materials are found within 40 iterations for a single detail of the band structure and within 80 iterations for discriminating between multiple crystal structures. The methodology is embodied as an interactive design framework that connects atomic and device level features. This approach not only presents the Bayesian optimization on tunable bandgap optical materials as a new and powerful paradigm for materials development but also is predicted to be widely applicable given the progress of materials chemistries in recent decades.

Перевод пока недоступен