The Importance of Reaction Energy in Predicting Chemical Reaction Barriers with Machine Learning Models

Area of Science:

• Computational chemistry and catalysis
• Materials science
• Machine learning applications

Background:

• Understanding heterocatalytic processes requires electronic structure simulations and microkinetic models.
• Calculating activation barriers via saddle point searches is a bottleneck for complex reaction networks.
• Brønsted-Evans-Polyani (BEP) scaling models are limited in scope and still require reaction energy calculations.

Purpose of the Study:

• Investigate machine learning (ML) models for predicting activation barriers without prior knowledge of reaction energy.
• Assess the impact of including reaction energy on ML model performance and generalizability.
• Explore feasibility for complex reactions like electrochemical CO2 reduction and wastewater remediation.

Main Methods:

• Trained linear and nonlinear ML models on a database of over 500 dehydrogenation activation barriers.
• Evaluated model performance with and without reaction energy as a feature.
• Assessed generalizability to new catalytic systems beyond the training set.

Main Results:

• Nonlinear ML models achieved performance comparable to BEP scaling for predicting activation barriers on new systems, even without reaction energy.
• Inclusion of reaction energy significantly improved both linear and nonlinear ML models.
• Reaction energy inclusion substantially enhanced model generalizability to systems outside the training data.

Conclusions:

• Reaction energy is a critical feature for developing accurate predictive models of catalytic activation barriers.
• ML offers a promising route to overcome computational bottlenecks in catalysis research.
• Reliable prediction of reaction energies is crucial for advancing complex catalytic system modeling.