Machine-learned approximations to Density Functional Theory Hamiltonians.

Area of Science:

• Computational Materials Science
• Quantum Chemistry
• Machine Learning Applications

Background:

• Density Functional Theory (DFT) calculations are computationally intensive and do not scale well with system size.
• Current semi-empirical DFT approximations require manual intervention, hindering high-throughput screening.
• Efficient methods are needed to predict electronic structures for large material systems.

Purpose of the Study:

• To develop a machine-learning approach for predicting DFT Hamiltonians.
• To enable faster and more scalable electronic structure calculations.
• To facilitate high-throughput screening of materials.

Main Methods:

• Utilized machine learning, specifically Kernel Ridge Regression.
• Employed representations of atomic neighborhoods for feature extraction.
• Focused on predicting DFT Hamiltonians without basis set or material system specificity.

Main Results:

• Achieved accurate and transferable prediction of DFT Hamiltonians across diverse material environments.
• Electronic structure properties (ballistic transmission, band structure) computed with predicted Hamiltonians closely matched DFT results.
• Demonstrated method's independence from DFT basis sets and material systems.

Conclusions:

• Machine learning offers an efficient and automatable solution for predicting DFT Hamiltonians.
• The proposed method significantly reduces computational cost for electronic structure calculations.
• This approach is scalable for predicting Hamiltonians of any material system, advancing materials discovery.