Comparison of intermediate-range order in GeO2 glass: Molecular dynamics using machine-learning interatomic potential...

Area of Science:

• Materials Science
• Condensed Matter Physics
• Computational Chemistry

Background:

• Understanding the atomic-scale structure of amorphous materials like GeO2 glass is crucial for predicting their properties.
• Short-range order (SRO) and intermediate-range order (IRO) significantly influence the macroscopic behavior of glasses.
• Distinguishing between different structural models derived from experimental data and simulations remains a challenge.

Purpose of the Study:

• To investigate and compare the short-range and intermediate-range order in GeO2 glass using two distinct computational approaches.
• To characterize the structural differences between models obtained from machine learning-based molecular dynamics and reverse Monte Carlo fitting.
• To evaluate the influence of ab initio calculations and diffraction data on the resulting structural models.

Main Methods:

• Molecular dynamics (MD) simulations employing a machine-learning interatomic potential trained on ab initio data.
• Reverse Monte Carlo (RMC) fitting of neutron diffraction data.
• Analysis of total/partial structure factors, coordination numbers, ring size/shape distributions, and persistent homology.

Main Results:

• Both MD and RMC methods produced similar two-body correlations but differed in their short- and intermediate-range ordering.
• Significant discrepancies were observed in ring size distributions: RMC models showed broad distributions, while MD models exhibited narrower ones.
• Ab initio calculations, via the machine learning potential, imposed stricter network assembly preferences compared to RMC with coordination constraints.

Conclusions:

• The choice of simulation method and underlying theoretical approximations (ab initio vs. RMC constraints) critically impacts the predicted three-dimensional structure of GeO2 glass.
• Machine learning potentials trained on ab initio data provide a more constrained and potentially more accurate representation of network assembly in GeO2 glass.
• These findings highlight the importance of considering higher-order structural correlations beyond simple pair distributions for accurate glass modeling.