Machine Learning of First-Principles Force-Fields for Alkane and Polyene Hydrocarbons.

Area of Science:

• Computational Chemistry
• Materials Science
• Chemical Physics

Background:

• Developing accurate interatomic potentials is crucial for molecular simulations.
• Traditional force fields often struggle with transferability and accuracy for diverse molecular systems.
• Machine learning offers a promising avenue for creating high-fidelity potential energy surfaces.

Purpose of the Study:

• To generate machine learning interatomic potentials (ML-IAPs) for alkane and polyene hydrocarbons.
• To achieve ab initio quality accuracy for hydrocarbon simulations.
• To develop a potentially universal force field for organic molecules.

Main Methods:

• Utilized on-the-fly adaptive sampling and sparse Gaussian process regression (SGPR).
• Trained ML models on density functional theory (DFT) data at the PBE+D3 level.
• Incorporated molecular dynamics (MD) simulations for small molecules, clusters, and condensed phases.

Main Results:

• Demonstrated excellent transferability of ML-IAPs to longer alkanes.
• Accurately described the ab initio potential energy surface for polyenes.
• Achieved reasonable agreement with experimental data for liquid ethane simulations.

Conclusions:

• The developed ML-IAPs show significant promise for simulating hydrocarbon systems.
• This work represents a key step towards a universal, ab initio-quality force field for organic molecules.
• The methodology is applicable to a broader range of organic compounds.