Developing a neural network machine learning interatomic potential for molecular dynamics simulations of La-Si-P...

Area of Science:

• Materials Science
• Computational Chemistry
• Machine Learning

Background:

• Molecular dynamics (MD) simulations are crucial for atomistic studies of materials.
• Accurate modeling of interatomic interactions is essential for reliable MD simulations.
• Artificial neural network machine learning (ANN-ML) offers a new paradigm for developing interatomic potentials.

Purpose of the Study:

• To develop an accurate and transferable ANN-ML interatomic potential for the La-Si-P system.
• To investigate the role of training data in ML potential development.
• To apply the developed potential to study material properties and phase behavior.

Main Methods:

• Development of an ANN-ML interatomic potential using a comprehensive training dataset.
• Validation of the potential against known crystalline structures and liquid states in the La-Si-P system.
• MD simulations to predict melting temperatures and study nucleation and growth phenomena.

Main Results:

• The ANN-ML potential accurately describes energy-volume relationships for La-Si-P crystalline structures.
• The potential successfully models La-Si-P liquid structures across various compositions.
• MD simulations predict melting temperatures that, while underestimated, show agreement with experimental trends.
• The potential aids in studying LaP nucleation and growth, consistent with experimental observations.

Conclusions:

• ANN-ML potentials provide an accurate and efficient method for simulating the La-Si-P system.
• The developed potential is transferable and reliable for studying thermodynamics and kinetics.
• This work highlights the importance of training data quality in ML potential development.
• The findings support the application of ANN-ML potentials in materials discovery and simulation.