Quotient-complex transformer for perovskite data analysis

The identification of advanced functional materials is critical for addressing challenges in sustainable energy and climate change. Hybrid organic-inorganic perovskites (HOIPs) have attracted significant interest due to their exceptional optoelectronic properties for photovoltaics. While geometric deep learning, particularly graph neural networks (GNNs), has shown promise for material property prediction, conventional GNNs struggle to capture periodicity and higher-order interactions inherent in crystalline materials. To overcome these limitations, we propose a quotient (cellular) complex (QC)-based structural representation and introduce the quotient-complex transformer (QCformer). By modeling materials as QCs, our approach naturally incorporates periodicity and many-body interactions through simplices of varying dimensions. Experiments on HOIP band-gap prediction demonstrate that QCformer outperforms state-of-the-art models, highlighting its effectiveness for perovskite materials.