Quantum chemistry-machine learning approach for predicting and elucidating molecular hyperpolarizability: Application...

Area of Science:

• Computational Chemistry
• Materials Science
• Machine Learning Applications

Background:

• Organic chromophores with nonlinear optical properties are crucial for photonic and optoelectronic devices.
• Accurate and rapid computational methods are needed to accelerate the theoretical design of these molecules.
• Studying structure-property relationships in extended molecular systems requires efficient computational strategies.

Purpose of the Study:

• To develop a computational approach combining quantum chemistry and machine learning (ML) for predicting the first hyperpolarizability (β) of [2.2]paracyclophane-containing push-pull compounds.
• To investigate the influence of molecular structure, including terminal donor/acceptor pairs and molecular length, on β.
• To elucidate general trends and causal correlations between molecular descriptors and target properties using explainable ML techniques.

Main Methods:

• Ab initio elongation finite-field method was employed to generate accurate reference β values for long polymer chains with linear scaling efficiency.
• A neural network (NN) model was constructed for predicting β, utilizing quantum chemical properties of short systems and system length as input.
• Genetic algorithms were used for molecular descriptor selection, and partial dependence, accumulated local effects, and permutation feature importance were applied for ML model interpretability.

Main Results:

• The established NN model achieved high accuracy (R² > 0.99) in reproducing β values for long molecules.
• The model successfully predicted β based on properties of the shortest molecular segments and the overall system length.
• Analysis revealed that asymmetric extension of donor and acceptor regions resulted in unequal β responses, demonstrating how substituent electronic properties influence β.

Conclusions:

• The combined quantum chemistry and ML approach provides an efficient and accurate method for studying nonlinear optical properties in extended organic systems.
• The study highlights the importance of molecular structure, particularly donor/acceptor arrangement and length, in determining the first hyperpolarizability.
• The applied methodology facilitates conceptual discoveries in chemistry by enabling efficient data generation and providing insights into structure-property causal correlations.