Integrating machine learning interatomic potentials with hybrid reverse Monte Carlo structure refinements in...

Area of Science:

• Materials Science
• Computational Chemistry
• Crystallography

Background:

• Reverse Monte Carlo (RMC) is crucial for interpreting diffraction data.
• Hybrid RMC uses interatomic potentials for physically realistic results.
• Current hybrid RMC methods are limited by available interatomic potentials.

Purpose of the Study:

• To enhance the hybrid RMC method by integrating a wider range of interatomic potentials.
• To enable the study of materials lacking pre-existing interatomic potentials.
• To develop a methodology for training machine learning interatomic potentials for hybrid RMC.

Main Methods:

• Implemented a new interatomic potential constraint in RMCProfile.
• Integrated support for potentials from the Large-scale Atomic/Molecular Massively Parallel Simulator (LAMMPS).
• Developed a workflow to train machine learning interatomic potentials using RMC.

Main Results:

• Successfully extended hybrid RMC capabilities to include LAMMPS-supported potentials.
• Enabled the application of machine learning interatomic potentials within the hybrid RMC framework.
• Demonstrated a novel methodology for creating custom interatomic potentials for RMC.

Conclusions:

• The enhanced hybrid RMC method significantly broadens the scope of materials research.
• Machine learning interatomic potentials offer a flexible approach for RMC structure refinement.
• This work paves the way for applying hybrid RMC to a wider array of complex materials.