Physics-informed latent-space optimization for energy-aligned hole-collecting monolayers in inverted perovskite solar cells.

Precise energy-level alignment at buried interfaces is critical for high-performance inverted perovskite solar cells, where the highest occupied molecular orbital (HOMO) of hole-collecting monolayers (HCMs) must be tuned relative to the valence band maximum (VBM) to enable efficient hole extraction while minimizing energy loss. Systematic exploration of suitable molecular structures remains challenging because of the vast chemical design space and the computational cost of first-principles screening. A physics-informed generative molecular design framework is developed in which HCM discovery is formulated as an energy-alignment-constrained optimization problem. A high-accuracy HOMO prediction model trained on a large-scale first-principles dataset achieves a mean absolute error of 88 meV and demonstrates transferability to previously reported HCM molecules. The predictor is integrated with a variational autoencoder (VAE) to construct a continuous molecular latent space, where Bayesian optimization is performed using an objective function that explicitly encodes HOMO-VBM alignment, structural similarity to known HCM motifs, and anchoring-group requirements. Latent-space optimization selectively identifies candidate molecules that satisfy energy-level alignment while retaining key structural features of known HCMs and extending beyond existing chemical space. The resulting candidates exhibit both structural novelty and functional relevance, demonstrating that the framework enables directed and physics-guided exploration of molecular design space. This approach demonstrates the potential of combining molecular generation, HOMO prediction, and Bayesian optimization for HCM candidate discovery in inverted perovskite solar cells.