Quantum-inspired inverse design of heterogeneous catalysts for hydrogen evolution reaction

Data-driven discovery of stable and functional materials remains a major challenge in chemistry and materials science, both from classical and quantum perspectives. Quantum machine learning provides new opportunities to optimize materials across vast chemical spaces, but hardware limitations restrict its practical application. Classical models can efficiently process large datasets, but their ability to explore unseen chemical spaces is fundamentally limited by the uncertainty of out-of-distribution predictions. Although transfer learning enables exploration beyond the training distribution, fine-tuning on the target domain often distorts the latent representations learned from the source domain. To address these challenges, we introduce a quantum-inspired generative framework for discovering catalytic materials for hydrogen evolution reactions (HER). Inspired by variational quantum circuits (VQCs), which outperform classical models in certain tasks, we hypothesized that the norm-conserving property of VQCs plays a key role in materials representation and optimization. Thus, we incorporated orthogonal transformations into the classical generative model to maintain latent space structure and enhance materials discovery. With additional data through transfer learning, the classical model with orthogonal transformations becomes useful. Our results provide practical guidelines for applying quantum machine learning to materials discovery and suggest the potential of quantum-inspired generative models to explore previously inaccessible chemical spaces.