Memory-Augmented Generative Design of Octahedral Co(II) Complexes with Targeted Zero-Field Splitting
Abstract
The discovery of coordination complexes with targeted magnetic anisotropy remains challenging because the enormous chemical space of ligands and coordination environments makes propertydirected molecular design difficult to achieve by conventional screening. Here, we introduce a memoryaugmented generative framework for the inverse design of octahedral Co(II) complexes with targeted zero-field splitting (ZFS). A chemically curated dataset of ~6000 Co(II) complexes, was used to establish a broad electronic-structure landscape spanning positive and negative ZFS regimes. Building on this chemical space, a generative strategy was developed in which ligand structures are iteratively mutated, recombined, and selected according to their predicted magnetic properties, while an elitememory mechanism preserves high-performing molecular motifs and guides subsequent generations toward increasingly favorable candidates. This enables targeted exploration of ligand combinations and coordination environments associated with large negative ZFS while simultaneously controlling rhombicity. The generated complexes were subjected to independent CASSCF calculations, including spin–orbit coupling, to directly validate their magnetic properties. Notably, the generative framework identified candidates with strongly enhanced axial anisotropy, including a representative complex exhibiting D ≈ −223 cm-1 after ab initio validation. The validated candidates reproduced the targeted ZFS with deviations of approximately 15–18 cm-1. These results establish a generative, memory-guided route from ligand design to electronically validated magnetic complexes, providing a practical strategy for discovering previously unexplored Co(II) molecular magnets and advancing molecular-magnet design from screening toward property-directed generation.
Related
- A targeted ionic iron/platinum-based precatalyst for bias-dependent reconstruction in alkaline water splitting
- CrystalGRW: generative modeling of crystal structures with targeted crystallographic properties via geodesic random walks
- Chiral Octahedral Metallomesogens with Δ/Λ Core Chirality: Extending Wallach's Rule from Crystals to Liquid Crystals
- Octahedral Ligand Field as Magnetic Coordinate Frame in a Luminescent DyIII Single Molecule Magnet
- Octahedral Molybdenum and Vanadium (Oxo)clusters as Dual Photoinitiators and Crosslinkers for PEGDA Hydrogels with Antibacterial Activity
BibTeX
@misc{rana2026memoryaugmented,
title = {{Memory-Augmented Generative Design of Octahedral Co(II) Complexes with Targeted Zero-Field Splitting}},
author = {Rajanikanta Rana and Aditya Choudhary and Abhijeet Biswas and Subodh Bhalchandra Rawool and Gopalan Rajaraman},
howpublished = {ChemRxiv},
year = {2026},
month = oct,
doi = {10.26434/chemrxiv.15010177/v1},
url = {https://chemrxiv.org/doi/full/10.26434/chemrxiv.15010177/v1},
note = {CC BY},
}
RIS
TY - UNPB
TI - Memory-Augmented Generative Design of Octahedral Co(II) Complexes with Targeted Zero-Field Splitting
AU - Rajanikanta Rana
AU - Aditya Choudhary
AU - Abhijeet Biswas
AU - Subodh Bhalchandra Rawool
AU - Gopalan Rajaraman
PY - 2026
DA - 2026/10/09
PB - ChemRxiv
DO - 10.26434/chemrxiv.15010177/v1
UR - https://chemrxiv.org/doi/full/10.26434/chemrxiv.15010177/v1
AB - The discovery of coordination complexes with targeted magnetic anisotropy remains challenging because the enormous chemical space of ligands and coordination environments makes propertydirected molecular design difficult to achieve by conventional screening. Here, we introduce a memoryaugmented generative framework for the inverse design of octahedral Co(II) complexes with targeted zero-field splitting (ZFS). A chemically curated dataset of ~6000 Co(II) complexes, was used to establish a broad electronic-structure landscape spanning positive and negative ZFS regimes. Building on this chemical space, a generative strategy was developed in which ligand structures are iteratively mutated, recombined, and selected according to their predicted magnetic properties, while an elitememory mechanism preserves high-performing molecular motifs and guides subsequent generations toward increasingly favorable candidates. This enables targeted exploration of ligand combinations and coordination environments associated with large negative ZFS while simultaneously controlling rhombicity. The generated complexes were subjected to independent CASSCF calculations, including spin–orbit coupling, to directly validate their magnetic properties. Notably, the generative framework identified candidates with strongly enhanced axial anisotropy, including a representative complex exhibiting D ≈ −223 cm-1 after ab initio validation. The validated candidates reproduced the targeted ZFS with deviations of approximately 15–18 cm-1. These results establish a generative, memory-guided route from ligand design to electronically validated magnetic complexes, providing a practical strategy for discovering previously unexplored Co(II) molecular magnets and advancing molecular-magnet design from screening toward property-directed generation.
ER -