Phonon bottleneck in methane clathrate hydrate from machine learning molecular dynamics

The thermal conductivity of methane hydrate is a critical property governing its stability in natural environments and its potential as an energy resource. However, discrepancy exists between experimental measurements and molecular dynamics (MD) predictions based on empirical potentials. Herein, we develop a high-fidelity neuroevolution potential (NEP) for methane hydrate using a dataset from first-principles density functional theory (DFT) calculations. The trained NEP model demonstrates exceptional accuracy in energy, force, and virial predictions, and reproduces structural properties and anharmonic thermal expansion behavior. The thermal conductivity of methane hydrate, determined to be approximately 0.72–0.74 W·m −1 ·K −1 via complementary homogeneous non-equilibrium MD (HNEMD) and equilibrium MD (EMD) simulations, agrees closely with experimental measurements, validating the reliability of the NEP model. Detailed phonon analysis reveals the intrinsically low thermal conductivity originates from strong host–guest coupling. This interaction causes pronounced low-frequency phonon localization, dominant yet inefficient acoustic branches below 1.0 THz, and flattened optical branches. This work not only advances the understanding of heat conduction mechanisms in clathrate hydrates but also establishes an accurate and efficient machine-learning-accelerated framework for modeling gas hydrates, with significant implications for gas hydrate energy extraction technologies and environmental impact assessments.