A universal density matrix functional from molecular orbital-based machine learning: Transferability across organic...

Area of Science:

• Computational chemistry
• Quantum chemistry
• Machine learning applications

Background:

• Predicting post-Hartree-Fock correlation energies is crucial for accurate molecular modeling.
• Existing methods require significant computational resources for high accuracy.

Purpose of the Study:

• To assess the accuracy and transferability of machine learning for predicting post-Hartree-Fock correlation energies.
• To develop and apply the molecular-orbital-based machine learning (MOB-ML) method.

Main Methods:

• Refined feature design and selection strategies.
• Application of the MOB-ML method to various molecular systems.
• Training models with limited reference calculations.

Main Results:

• MOB-ML accurately describes potential energy surfaces (within 1 mhartree) for water molecules using single reference calculations.
• Achieves chemical accuracy with threefold fewer training geometries compared to Δ-ML for organic molecules.
• Demonstrates 36-fold fewer training calculations for transferability to larger systems.

Conclusions:

• MOB-ML offers a highly efficient and accurate approach for predicting electronic correlation energies.
• The method shows remarkable accuracy and transferability across diverse chemical spaces.
• MOB-ML significantly reduces the computational cost associated with high-accuracy quantum chemical calculations.