Temperature-dependent and symmetry-enforced dynamical matrices of silicon by non-perturbative MD simulations

Within the framework of the phonon Boltzmann transport equation (phonon BTE), the accuracy of the predicted phonon-mediated lattice thermal conductivity (κ_Latt) relies on the fidelity of phonon spectra and phonon-phonon scattering matrices. In practice, the temperature independent, or bare, harmonic force constant matrices are often adopted. To quantitatively study phonon frequency renormalization up to the melting temperature, we have implemented a robust numerical algorithm to compute temperature-dependent and symmetry-enforced phonon dynamical matrices for all the q-points in the first Brillouin zone of reciprocal space. This method includes contributions of all orders of lattice anharmonicity without any perturbation approximations. Here, we will report our recent results of phonon dynamical matrices of silicon crystals from 300K to 1500K utilizing a machine learning interatomic potential of silicon. We will present a systematic comparison between the phonon BTE predicted temperature dependence in κ_Latt using either the original bare-phonon dynamical matrices or the current renormalized phonon dynamical matrices.