Equivariant electronic Hamiltonian prediction with many-body message passing

Area of Science:

• Computational materials science
• Quantum chemistry
• Machine learning applications

Background:

• Kohn-Sham Density Functional Theory (KS-DFT) is crucial for predicting material electronic properties.
• Current KS-DFT methods face computational challenges for large-scale applications.
• Developing efficient surrogate models is essential for accelerating materials discovery.

Purpose of the Study:

• Introduce MACE-H, a novel graph neural network for KS-DFT Hamiltonians.
• Achieve high accuracy and computational efficiency in predicting electronic properties.
• Enhance generalization ability for large-scale materials screening.

Main Methods:

• Developed MACE-H, a graph neural network incorporating high body-order message passing.
• Integrated node-order expansion for efficient O(3) irreducible representations.
• Captured f orbital matrix interaction blocks for comprehensive local chemical environments.

Main Results:

• MACE-H demonstrated high accuracy and computational efficiency.
• Achieved sub-meV prediction errors on matrix elements for benchmark datasets.
• Showcased excellent transferability on 2D materials and bulk gold datasets.

Conclusions:

• MACE-H effectively models KS-DFT Hamiltonians, accelerating electronic property predictions.
• The model's performance makes it suitable for high-throughput materials screening.
• Future work can leverage MACE-H for discovering novel materials with desired electronic properties.