Quotient Complex\n(QC)-Based Machine Learning for 2D\nHybrid Perovskite Design

With remarkable stability and exceptional optoelectronic\nproperties,\ntwo-dimensional (2D) halide layered perovskites hold immense promise\nfor revolutionizing photovoltaic technology. Effective data representations\nare key to the success of all learning models. Currently, the lack\nof comprehensive and accurate material representations has hindered\nAI-based design and discovery of 2D perovskites, limiting their potential\nfor advanced photovoltaic applications. In this context, this work\nintroduces a novel computational topology framework termed the quotient\ncomplex (QC), which serves as the foundation for the material representation.\nThe proposed QC-based features are seamlessly integrated with learning\nmodels for the advancement of 2D perovskite design. At the heart of\nthis framework lies the quotient complex descriptors (QCDs), representing\na quotient variation of simplicial complexes derived from materials’\nunit cell and periodic boundary conditions. Differing from prior material\nrepresentations, this approach encodes higher-order interactions and\nperiodicity information simultaneously. Based on the well-established\nnew materials for solar energetics (NMSE) databank, the proposed QC-based\nmachine learning models exhibit superior performance against all existing\ncounterparts. This underscores the paramount role of periodicity information\nin predicting material functionality, while also showcasing the remarkable\nefficiency of the QC-based model in characterizing materials’\nstructural attributes.