AI and the World's Hard Problems · Chemistry & materials · arXiv · 2026-10-22
SU(2)-Equivariant Spinor Message Passing for Molecular Magnetism
Because predicting molecular magnetism usually requires costly electronic-structure calculations with spin-orbit coupling, the authors developed eSpinor, an SU(2)-equivariant graph neural network that propagates spinor features initialized from an approximate SOC Hamiltonian, with gauge-covariant message passing and a gauge-invariant readout. On cobalt single-molecule magnet candidates it improves zero-field-splitting predictions over NequIP, including for unseen ligands. For a dysprosium MRI probe it improves real electron-spin correlations over a Hamiltonian-informed baseline with fewer parameters, and predicts low-field nuclear relaxation in water and after transfer to methanol without retraining.