Joshua Zhou, Bhavnesh Jangid, Matthew R Hermes, Laura Gagliardi
Lanthanide molecular magnets are promising candidates for quantum information technologies, but their discovery is limited by the computational cost of accurately predicting their electronic and magnetic properties. Multireference methods provide the required accuracy, yet their application at scale has been hindered by the need to manually select the model space (the states considered in the spin-orbit treatment). We present an automated multireference workflow for single-lanthanide complexes, based on complete active space wave functions followed by state-interaction spin-orbit (SISO) coupling. Its central feature is an iterative protocol for constructing the SISO model space, which enables reliable and reproducible calculations across diverse coordination environments. Benchmark calculations on lanthanide hexachlorides show that including spin-orbit-free states up to approximately 5 eV above the ground state reproduces experimental ligand-field splittings with mean absolute errors of 100-200 cm-1 across the lanthanide series. We further examine how the model-space size affects magnetic anisotropy and g-tensors in representative dysprosium- and erbium-based single-ion magnets and in the [Ln(DOTA)(H2O)]- family. Finally, we apply the protocol to 299 lanthanide complexes to generate a consistent data set of ligand-field splittings, g-tensors, magnetic axes, and approximate barriers to magnetization reversal. This work establishes a practical framework for high-throughput multireference calculations and data generation for lanthanide molecular magnets.