Title: Thermo-mechanical properties of nitrogenated holey graphene (C2N): A comparison of machine-learning-based and classical interatomic potentials

Abstract: Thermal and mechanical properties of two-dimensional nanomaterials are commonly studied by calculating force constants using the density functional theory (DFT) and classical molecular dynamics (MD) simulations. Although DFT simulations offer accurate estimations, the computational cost is high. On the other hand, MD simulations strongly depend on the accuracy of interatomic potentials. Here, we investigate thermal conductivity and elastic modulus of nitrogenated holey graphene (C2N) using passively fitted machine-learning interatomic potentials (MLIPs), which depend on computationally inexpensive ab-initio molecular dynamics trajectories. Thermal conductivity of C2N is investigated via MLIP-based non-equilibrium molecular dynamics simulations (NEMD). At room temperature, the lattice thermal conductivity of 85.5 W/m-K and effective phonon mean free path of 37.16 nm are found. By carrying out uniaxial tension simulations, the elastic modulus, ultimate strength, and fractural strain of C2N are predicted to be 390 GPa, 42 GPa, and 0.29, respectively. It is shown that the passively fitted MLIPs can be employed as an efficient interatomic potential to obtain the thermal conductivity and elastic modulus of C2N utilizing classical MD simulations. Moreover, the possibility of employing MLIPs to simulate C2N with point defects has been investigated. By training MLIP with point defect configurations, the mechanical properties of defective structures were studied. Although using the MLIP is more costly than classical interatomic potentials, it could efficiently predict the thermal and mechanical properties of 2D nanostructures.