Subgraph Isomorphic Decision Tree to Predict Radical Thermochemistry with Bounded Uncertainty Estimation.

Area of Science:

• Computational Chemistry
• Chemical Kinetics
• Thermochemistry

Background:

• Accurate radical thermochemistry is crucial for detailed chemical kinetic models in industrial applications.
• Current methods for estimating hydrogen bond increment (HBI) corrections often rely on manual expert knowledge, limiting scalability and maintenance.
• Existing tree estimators for thermochemistry face challenges in data limitations and manual construction.

Purpose of the Study:

• To extend the subgraph isomorphic decision tree (SIDT) algorithm for automatic estimation of hydrogen bond increment (HBI) corrections in radical thermochemistry.
• To develop a more accurate, reliable, and scalable method for predicting thermochemical parameters of radicals.
• To enhance the interpretability and error cancellation in kinetic modeling through improved HBI estimations.

Main Methods:

• Extended the subgraph isomorphic decision tree (SIDT) algorithm to estimate HBI corrections.
• Introduced a physics-aware splitting criterion and explored uncertainty estimation methods.
• Compiled a dataset of thermochemical parameters for 2210 radicals (C, O, N, H) using quantum chemical calculations.
• Trained the SIDT model on the compiled dataset.

Main Results:

• The SIDT model offers an automated approach for generating and extending thermochemical tree estimators.
• Achieved improved accuracy and R-squared values compared to existing empirical tree estimators.
• Provided more realistic uncertainty estimates.
• Demonstrated a more advantageous tree structure for faster descent speed.

Conclusions:

• The developed SIDT estimator represents a significant advancement in kinetic modeling.
• Offers precise, reliable, and scalable predictions for radical thermochemistry.
• Facilitates the automatic generation and improvement of kinetic models by overcoming limitations of manual methods.