Unlocking the power of π-π stacking in mesogenic lanthanide complexes: a machine learning enhanced approach to...

The rational design of highly luminescent lanthanide(III) complexes has long been hampered by the inability to simultaneously capture the subtle interplay of electronic structure, supramolecular packing, and thermally driven dynamics within these systems. Here, we shatter this barrier by introducing a multiscale computational framework that seamlessly integrates density functional theory, semiempirical methods, molecular dynamics, and state-of-the-art machine learning interatomic potentials - to achieve ab initio accuracy at scales previously inaccessible for lanthanide complexes. This approach resolves a long-standing puzzle: how π-π stacking between phenyl substituents in β-diketonate ligands governs triplet excited states and the delicate balance between nonradiative relaxation, and intramolecular forward and back energy transfer processes. Our simulations identify π-π stacking as a hidden structural switch that regulates molecular packing and supramolecular organization in condensed phases. Depending on the stacking geometry, it can suppress nonradiative decay and quenching pathways by rigidifying the ligand environment. By quantifying these effects through Judd-Ofelt theory and Voronoi-Dirichlet polyhedra analysis, we establish clear, predictive structure-property relationships that link short-range coordination geometry to long-range luminescence efficiency and liquid-crystalline behavior. This work not only provides a blueprint for the rational design of next-generation lanthanide-based luminophores but also demonstrates that machine-learning-accelerated simulations are an essential tool for uncovering the fundamental mechanistic principles that govern functional optoelectronic materials, bridging molecular design with macroscopic performance.