A Machine Learning Approach for Prediction of Rate Constants

Area of Science:

• Computational Chemistry
• Chemical Kinetics
• Machine Learning

Background:

• Accurately predicting bimolecular thermal rate constants is crucial for chemical kinetics.
• Conventional transition state theory (TST) and Eckart-corrected TST (ECK) have limitations in accuracy, especially in tunneling and recrossing regions.

Purpose of the Study:

• To develop and validate a machine learning approach for predicting bimolecular thermal rate constants.
• To compare the accuracy of the machine learning method against TST and ECK.

Main Methods:

• Utilized Gaussian process (GP) regression to model the difference between accurate quantum calculations and TST/ECK.
• Trained the GP model on a database of 13 reaction/potential surface combinations.
• Tested the model on 39 reaction/potential surface combinations.

Main Results:

• The GP method achieved an average accuracy within 80% of accurate results.
• TST and ECK showed significantly lower accuracy, within 330% and 110% respectively.
• GP demonstrated superior accuracy in both the tunneling and high-temperature recrossing regions.

Conclusions:

• Machine learning, specifically GP regression, offers a highly accurate method for predicting thermal rate constants.
• The GP approach surpasses traditional methods like TST and ECK, particularly under challenging conditions.
• Further validation on 3D reactions indicates strong potential for this machine learning strategy.