Ab Initio Accelerated Discovery of Mixed-Cation Perovskites for >33% Efficiency: A Hybrid DFT-MD Framework

The thermodynamic instability of organic-inorganic halide perovskites remains the singular bottleneck preventing the displacement of silicon photovoltaics. This research presents a "God-tier" computational framework unifying Density Functional Theory (DFT) with Machine Learning Force Fields (MLFF) to autonomously discover a hyper-stable perovskite stoichiometry with a theoretical efficiency exceeding 33% (Shockley-Queisser limit approach). By bridging the accuracy of the PBEsol functional with the temporal scale of Classical Molecular Dynamics (MD), we simulate degradation pathways over microseconds—a feat impossible with standard ab initio MD. The study identifies a triplecation lattice (Cs_{0.05}FA_{0.85}MA_{0.10}Pb(I_{0.97}Br_{0.03})_3) that suppresses the \alpha-to-\delta phase transition via entropy stabilization. This work leverages massively parallel Fortran kernels for electronic structure, Python for graph neural network potential training, and Java for real-time lattice visualization, offering a blueprint for the next generation of energy materials.