Geometry-Based Neural-Network Prediction of Electron Localization Function Topology in Dense Hydrogen

Area of Science:

• Computational materials science
• Quantum chemistry
• Machine learning applications

Background:

• Predicting electron localization function (ELF) in dense hydrogen is crucial for understanding its unique properties.
• Traditional electronic-structure calculations are computationally expensive, limiting high-throughput studies.

Purpose of the Study:

• To develop a machine-learning framework for predicting ELF directly from atomic geometry.
• To bypass computationally intensive electronic-structure calculations for dense hydrogen.

Main Methods:

• Developed a machine-learning model trained on first-principles data of dense fluid hydrogen.
• Utilized a combined real- and reciprocal-space analysis to evaluate model accuracy and error sources.
• Tested the model's transferability to crystalline hydrogen configurations.

Main Results:

• The machine-learning model achieved high accuracy (R² > 0.99) in predicting ELF.
• Residual errors were analyzed, revealing long-wavelength components that increase with pressure.
• The model demonstrated robust transferability to crystalline hydrogen, preserving topological features.

Conclusions:

• The developed framework offers a computationally efficient route for evaluating hydrogen-networking characteristics.
• This approach enables high-throughput analysis of both fluid and crystalline hydrogen.
• The study highlights the potential of machine learning in accelerating materials science research.