HlightReaxMD: A Machine Learning-Augmented Multiscale Analysis Framework for Radiation Chemistry Dynamics and Damage...

Area of Science:

• Materials Science
• Computational Chemistry
• Nuclear Engineering

Background:

• Molecular dynamics (MD) simulations are crucial for studying irradiation-induced material damage.
• Analyzing complex chemical reactions within MD trajectories presents significant challenges.
• Existing models like the Norgett-Robinson-Torrens (NRT) displacements per atom (dpa) model have limitations in predicting irradiation damage.

Purpose of the Study:

• To introduce HlightReaxMD, a novel cross-platform toolkit for analyzing MD simulations of irradiation damage.
• To enable direct extraction of chemical reaction and collision cascade information from MD trajectories.
• To develop a machine learning-driven model for predicting irradiation damage beyond traditional methods.

Main Methods:

• Developed HlightReaxMD, a toolkit supporting all elements in reactive force fields (ReaxFF).
• Implemented automated analysis of atomic-scale collision events using cascade trees and reaction network paths.
• Integrated a machine learning model for enhanced irradiation damage prediction.

Main Results:

• HlightReaxMD provides tools for chemical reaction analysis, kinetic parameter calculation, and collision cascade analysis.
• The toolkit automates the tracking of atomic events and analysis of reaction mechanisms.
• The machine learning model offers improved irradiation damage prediction by considering multiple factors.

Conclusions:

• HlightReaxMD offers a comprehensive solution for analyzing terabyte-level MD trajectory data.
• The toolkit facilitates systematic research into irradiation effects at atomic to microscale.
• HlightReaxMD advances the prediction of material irradiation damage, moving beyond the NRT-dpa model.