New Insights into Thermal Transport in ε-CL-20 Revealed by Machine-Learned Potentials and Mode-Projection Analysis

The thermal conductivity of energetic molecular crystals is a critical safety parameter, yet its temperature dependence remains poorly understood from a mode-resolved perspective. Here, we elucidate the underlying mechanisms in ε-CL-20 by integrating a high-accuracy, machine-learned potential with comprehensive vibrational-dynamics analysis. Using transfer learning, we fine-tuned a neuroevolution potential that reproduces DFT-level accuracy. The anisotropic thermal conductivity, calculated with this potential, decreases significantly from 200 to 400 K. Spectral energy density analysis reveals that this decline originates from a universal shortening of phonon lifetimes and a crossover from particle-like propagation to wave-like tunneling. Full unit-cell mode-projection analysis demonstrates that temperature selectively strengthens anharmonic couplings between N-NO₂ bending vibrations and cage-skeleton deformation modes. Nonequilibrium mode-excitation simulations further reveal that the cage-deformation mode acts as an energy-flux hub, directing thermal energy toward nitro-group vibrations, with a temperature-gated redistribution of dominant receiving channels. Crucially, the wave-like tunneling channels and the mode-mode coupling pathways are two manifestations of the same anharmonic network. The degradation of heat conduction and the funneling of energy toward reactive trigger bonds are therefore two sides of the same coin. Under rapid external loading, the inward flux overwhelms dissipative backflow, driving vibrational-amplitude growth and bond cleavage. This work provides phonon-resolved dynamic evidence directly linking macroscopic heat conduction to the microscopic initiation of chemical reactions and paves the way toward a general framework for investigating energy-flux networks in complex molecular solids.