Cheap Turns Superior: A Linear Regression-Based Correction Method to Reaction Energy from the DFT

Area of Science:

• Computational Chemistry
• Quantum Chemistry
• Materials Science

Background:

• Density functional theory (DFT) workflows for chemical reaction networks often suffer from systematic reaction energy errors due to approximations in exchange-correlation functionals.
• Machine learning (ML) models can mitigate these errors but struggle with out-of-distribution systems not present in their training data.

Purpose of the Study:

• To develop a simple, bond-based correction method to enhance the accuracy of DFT-derived reaction energies.
• To assess the method's effectiveness across different DFT functionals and its applicability to out-of-distribution systems.

Main Methods:

• A linear regression model was employed to derive bond-specific correction terms.
• Correction terms were trained using the QM9 dataset.
• The method was tested using three DFT functionals across different rungs of Jacob's ladder.

Main Results:

• The bond-based correction method significantly improved the accuracy of DFT reaction energies for all tested functionals, particularly the PBE functional.
• The corrected DFT reaction energies for out-of-distribution molecules were within 0.05 eV of the high-accuracy G4MP2 method.

Conclusions:

• The proposed bond-based correction offers a computationally inexpensive yet effective way to improve DFT reaction energy predictions.
• This method shows promise for enhancing the reliability of DFT in predicting chemical reaction networks, even for novel molecular systems.