Big Data Meets Quantum Chemistry Approximations: The Δ-Machine Learning Approach

Area of Science:

• Computational Chemistry
• Machine Learning in Chemistry
• Quantum Mechanics

Background:

• Accurate quantum chemistry studies are computationally expensive, limiting exploration of vast chemical spaces.
• Approximate quantum methods lack the necessary accuracy for comprehensive molecular studies.
• Predicting molecular properties for large datasets remains a significant challenge.

Purpose of the Study:

• To develop a cost-effective computational strategy for accurate chemical property prediction.
• To enable the study of significantly larger molecular sets than previously feasible.
• To bridge the accuracy gap between legacy quantum methods and high-level computational chemistry.

Main Methods:

• A composite strategy combining machine learning corrections with legacy quantum methods.
• Training machine learning models on smaller, computationally derived datasets.
• Applying trained models to predict enthalpies, free energies, entropies, and electron correlation energies.

Main Results:

• Achieved chemical accuracy for thermochemical properties of up to 16,000 isomers of C7H10O2.
• Enabled prediction of electron correlation energy at the cost of Hartree-Fock calculations.
• Demonstrated transferability of models trained on 1-10% of 134,000 molecules to predict properties of the remainder.
• Established a qualitative link between molecular entropy and electron correlation.

Conclusions:

• The proposed machine learning-enhanced quantum chemistry approach significantly reduces computational cost while achieving high accuracy.
• This method allows for the exploration of much larger chemical spaces than traditional quantum chemistry.
• The approach shows excellent transferability and potential for broad application in computational chemistry and materials science.