A Bayesian Committee Machine-Based Force Field for Organic Nitrogen Compounds.

Area of Science:

• Computational Chemistry
• Materials Science
• Chemical Physics

Background:

• Large-scale atomic simulations are crucial but computationally expensive.
• Machine learning force fields (MLFFs) offer a cost-effective, accurate alternative to traditional methods like density functional theory (DFT).
• Kernel-based MLFFs face challenges in generalizing across diverse atomic environments and compounds.

Purpose of the Study:

• To develop a robust machine learning force field (MLFF) applicable to various carbon-nitrogen-hydrogen (C-N-H) compounds.
• To overcome the limitations of existing kernel-based models in predicting potential energies across different molecular structures.
• To enable efficient and accurate simulations of C-N-H systems.

Main Methods:

• Developed a novel MLFF utilizing the Robust Bayesian Committee Machine (RBCM) framework.
• Trained the RBCM-based MLFF using first-principles calculations and molecular dynamics simulations.
• Employed diverse C-N-H molecules for comprehensive training data generation.

Main Results:

• The RBCM-based MLFF demonstrated excellent agreement with DFT results for longer amine structures.
• Accurate predictions were achieved for two Diels-Alder reactions, validating the model's performance.
• The developed MLFF effectively captures the potential energy surfaces of organic molecules.

Conclusions:

• Machine learning models, specifically the RBCM-based MLFF, can accurately predict potential energy surfaces for C-N-H organic molecules.
• This approach significantly enhances simulation efficiency compared to traditional methods.
• The developed MLFF provides a powerful tool for studying a wide array of C-N-H compounds.