Unified deep learning framework for many-body quantum chemistry via Green's functions

Area of Science:

• Computational chemistry
• Quantum mechanics
• Machine learning

Background:

• Quantum many-body methods are crucial for electronic property calculations but are computationally expensive.
• Existing machine learning models struggle to capture complex many-electron wavefunctions and many-body physics.

Purpose of the Study:

• To develop a deep learning framework for predicting electronic properties by targeting the many-body Green's function.
• To unify predictions for ground and excited states and gain insights into many-electron correlation effects.

Main Methods:

• A graph neural network was employed to learn the many-body perturbation theory or coupled-cluster self-energy from mean-field features.
• The framework predicts one- and two-particle excitations and properties derived from the one-particle density matrix.

Main Results:

• The deep learning model demonstrated competitive performance in predicting electronic excitations and properties.
• The method showed high data efficiency and excellent transferability across diverse chemical systems and conditions.

Conclusions:

• This deep learning framework offers a computationally efficient and accurate approach to solving complex many-electron problems.
• The work paves the way for broader applications of machine learning in quantum chemistry and materials science.