Machine Learned Fock Matrix.

Area of Science:

• Computational Chemistry
• Quantum Chemistry
• Materials Science

Background:

• Accurate prediction of molecular electronic structures is crucial for understanding chemical properties and reactions.
• Traditional quantum chemical methods can be computationally expensive, limiting their application to large systems or complex simulations.
• Developing efficient and accurate methods for electronic structure prediction remains an active area of research.

Purpose of the Study:

• To introduce a novel machine learning integral-based self-consistent field (miSCF) method for efficient molecular electronic structure prediction.
• To leverage atomic and geometric features for accurate prediction of the molecular Fock matrix.
• To enable precise electronic property predictions with reduced training data and computational cost.

Main Methods:

• Development of the miSCF method, integrating machine learning with integral-based self-consistent field calculations.
• Utilizing atomic information and geometric features as input for predicting the molecular Fock matrix.
• Training and testing the model on representative small molecules, H2 and H2O chains, and ice structures.

Main Results:

• The miSCF method demonstrated high precision and efficiency in predicting energies, wave functions, and electron densities.
• The approach showed good data sharing and transferability across chemically analogous systems.
• Accurate predictions were achieved with a small amount of training data, significantly reducing computational expenses.

Conclusions:

• The miSCF method provides an efficient and accurate tool for quantum chemical calculations.
• Its high precision, efficiency, and transferability make it suitable for various applications.
• This work lays a foundation for advanced applications such as potential energy surface construction, ab initio molecular dynamics, and chemical reaction simulations.