Publication date: 15th December 2025
Metal halide perovskite quantum dots (PQDs) continue to challenge computational chemistry due to their structural softness and highly dynamic surface chemistry. In our group, we combine high-level DFT with molecular dynamics to generate next-generation machine-learning force fields (MLFFs) capable of capturing these complexities across realistic time and length scales. By curating cross-material datasets and integrating active learning, we train transferable MLFFs that describe ligand binding, surface reconstruction, and defect dynamics with near-DFT accuracy. These models unlock nanosecond molecular dynamics for CsPbBr₃ PQDs, providing statistically meaningful insight into the mechanistic processes that occur at PQD surfaces and their associated energy landscapes. We complement this with interpretable electronic-structure workflows that directly link structural motifs to emergent optoelectronic properties. This talk will show how machine learning enhanced simulations are reshaping our understanding of halide-perovskite QDs and highlight the methodological advances needed to achieve predictive, chemically aware MLFFs for complex, technologically relevant nanomaterials.
