Machine learning-based closed-loop for optimizing HOMO–LUMO gap in diarylethene

Abstract

Diarylethenes are a class of photochromic molecular switches whose performance in photoresponsive applications critically depends on the optimization and fine-tuning of the HOMO–LUMO gap. The ability to modulate this gap through rational structural modification has become a key factor in expanding the functionality of diarylethene-based systems. In this work, we introduce an automated, closed-loop optimization framework in which a machine learning model, trained on an existing dataset, serves as a surrogate model to predict costly-to-obtain measurements during the exploration of a vast diarylethene derivative chemical space. This approach enables the efficient identification of candidate molecules optimized for a target HOMO–LUMO gap without human intervention. The top-performing candidates predicted by the model are subsequently validated using density functional theory calculations. Comparisons with available benchmarks demonstrate that the proposed strategy outperforms existing approaches. Overall, this study provides a general methodology and practical tools for integrating molecular structure data with advanced machine learning techniques to accelerate the discovery and design of photoresponsive materials with tailored electronic properties.