Molecular-orbital-based machine learning for open-shell and multi-reference systems with kernel addition Gaussian...

Area of Science:

• Computational Chemistry
• Machine Learning in Quantum Chemistry
• Electronic Structure Theory

Background:

• Accurate prediction of total correlation energies is crucial for understanding molecular properties.
• Existing molecular-orbital-based machine learning (MOB-ML) methods face challenges with complex systems.
• The need for efficient and accurate machine learning strategies in quantum chemistry is growing.

Purpose of the Study:

• To introduce a novel machine learning strategy, kernel addition Gaussian process regression (KA-GPR), within MOB-ML.
• To enhance the learning efficiency and prediction accuracy for total correlation energies of general electronic structure theories.
• To validate KA-GPR's performance on both closed- and open-shell molecular systems.

Main Methods:

• Implementation of kernel addition Gaussian process regression (KA-GPR) into the MOB-ML framework.
• Training and testing MOB-ML(KA-GPR) on the smallest criegee molecule (closed-shell, multi-reference).
• Application to predict properties of small free radicals, H10 chain, and water OH bond dissociation.
• Validation against large benchmark datasets: QM9, QM7b-T, GDB-13-T (closed-shell), and QMSpin (open-shell).

Main Results:

• MOB-ML(KA-GPR) demonstrated learning efficiency comparable to the original MOB-ML for the smallest criegee molecule.
• Prediction accuracies for small free radicals reached chemical accuracy (1 kcal/mol) after training on a single structure.
• Accurate potential energy surfaces were generated for the H10 chain and water OH bond dissociation.
• Successful prediction of large benchmark datasets for both closed- and open-shell molecules.

Conclusions:

• KA-GPR is an effective extension of MOB-ML for learning total correlation energies.
• The method shows high accuracy and efficiency across diverse chemical systems, including challenging multi-reference and open-shell cases.
• MOB-ML(KA-GPR) offers a promising approach for accurate and efficient electronic structure calculations.