Phonon-Mediated Heat Transport in CO₂ Hydrate

Gas hydrates are a promising medium for CO₂ sequestration and eco-friendly refrigeration, yet the mechanisms governing their thermal transport properties remain poorly understood. Here, we develop a neuroevolution potential (NEP)-based machine learning force field for sI-type CO₂ hydrate, rigorously trained on first-principles data, to accurately predict its thermal transport properties. Our NEP model reproduces key structural characteristics, including radial distribution functions (RDFs) and temperature-dependent lattice expansion behavior, with near-density functional theory (DFT) accuracy. Using homogeneous nonequilibrium molecular dynamics (HNEMD) simulations, our NEP model reproduces key experimental trends, including the anomalous increase in thermal conductivity with temperature from 230 to 300 K, where the thermal conductivity rises nonlinearly from 0.58 to 0.72 W·m^-1·K^-1. Detailed phonon analysis reveals that heat transfer is dominated by low-frequency phonons (<10 THz), while strong host-guest scattering and localized vibrations above 20 THz suppress thermal conductivity. This work not only provides crucial mechanistic insights that resolve controversies in hydrate thermal physics but also establishes the reliable NEP framework as a robust tool for advancing hydrate-based technologies in carbon capture and thermal energy applications.