Deciphering nonlinear optical properties in functionalized hexaphyrins via explainable machine learning

Area of Science:

• * Molecular Photonics
• * Computational Chemistry
• * Materials Science

Background:

• * Understanding molecular nonlinear optical (NLO) properties is crucial for designing advanced NLO switches.
• * Previous research identified orbital contributions, aromaticity, planarity, and intramolecular charge transfer as key factors.
• * Hexaphyrin-based redox switches have shown promise for tunable NLO behavior.

Purpose of the Study:

• * To identify the driving forces behind the first hyperpolarizability (βHRS) in meso-substituted and/or core-modified [26]- and [30]hexaphyrins.
• * To utilize explainable machine learning (ML) for elucidating these structure-property relationships.
• * To develop a predictive model for βHRS applicable to diverse hexaphyrin systems.

Main Methods:

• * Kernel ridge regression (KRR) model with 6-fold cross-validation.
• * Application of Shapley additive explanations (SHAP) for feature importance analysis.
• * Investigation of various hexaphyrin derivatives, including redox states, substitution patterns, and topologies.

Main Results:

• * A strong correlation between βHRS and the HOMO-LUMO energy gap was observed.
• * Incorporating additional orbital information and charge-transfer features significantly improved the KRR model's accuracy.
• * SHAP analysis revealed that charge transfer excitation length is critical for 30R systems, while transition dipole moment is key for 26R systems.
• * The ML model effectively predicted βHRS for various hexaphyrin structures beyond the training set.

Conclusions:

• * Explainable ML provides deep insights into the molecular design principles for enhanced NLO properties.
• * Specific molecular descriptors, such as charge transfer and transition dipole moments, are critical for tuning βHRS in hexaphyrins.
• * The developed ML model offers a powerful tool for accelerating the discovery of novel NLO materials.