Machine Learning Prediction for Fe(II) Spin-Crossover Complex in the Same Spin State Using Geometrical and...

Area of Science:

• Materials Science
• Computational Chemistry
• Solid-State Chemistry

Background:

• Spin-crossover (SCO) complexes are molecular materials exhibiting reversible switching between high-spin (HS) and low-spin (LS) states.
• Predicting SCO behavior from crystallographic data is crucial for designing novel SCO materials.
• Existing datasets lacked the diversity and annotation necessary for robust predictive modeling.

Purpose of the Study:

• To create a comprehensive dataset of iron(II) complexes with annotated SCO characteristics.
• To develop machine learning models for predicting SCO activity based on crystallographic data.
• To identify key factors influencing SCO behavior in different spin states.

Main Methods:

• Manually curated a dataset of 500 Fe(II)-N6 coordination complexes with explicit spin states and SCO potential (FeN6-SSD).
• Employed machine learning to classify SCO-active versus non-SCO complexes within the same spin state (HS or LS).
• Utilized the many-body tensor representation as a descriptor set for model training.

Main Results:

• Machine learning models achieved high prediction accuracy for SCO activity in both HS and LS states.
• Key predictors for SCO differed between spin states: local geometry in HS, ligand factors in LS.
• Environmental factors like solvents and counterions showed inconsistent influence on SCO classification.

Conclusions:

• The FeN6-SSD dataset and developed ML models provide a powerful tool for predicting SCO behavior.
• Understanding spin-state-specific predictors is vital for targeted SCO complex design.
• Further research is needed to fully elucidate the role of environmental factors in SCO phenomena.