DFT-Based Permutationally Invariant Polynomial Potentials Capture the Twists and Turns of C₁₄H₃₀

Area of Science:

• Computational Chemistry
• Materials Science
• Chemical Physics

Background:

• Hydrocarbons are vital industrial materials, but predicting their dynamic properties is challenging.
• Current methods rely on force-field mechanics, which have limitations in accuracy.

Purpose of the Study:

• To develop accurate machine-learned potential energy surfaces (PESs) for hydrocarbons.
• To improve the prediction of dynamical properties for linear hydrocarbons.

Main Methods:

• Generated a large dataset of ~250,000 density functional theory (DFT) energies for C14H30.
• Developed two PESs using Permutationally Invariant Polynomials (PIPs) with many-body and fragmented-basis approaches.
• Validated PESs using molecular dynamics and direct DFT calculations.

Main Results:

• Achieved precise fits for energies and forces, with excellent out-of-sample agreement.
• Demonstrated the robustness of the PESs across a wide range of molecular conformations.
• The many-body PIPs PES showed direct transferability to other linear hydrocarbons.

Conclusions:

• Machine-learned PESs significantly enhance the prediction of hydrocarbon dynamical properties.
• The developed PESs offer a more accurate and robust alternative to traditional force fields.
• The many-body PIPs approach provides a transferable model for broader applications.