Simulation study of inelastic neutron-scattering spectra of Si crystals beyond the one-phonon Green's function approximation

Inelastic neutron-scattering (INS) is a powerful experimental probe to study thermal excitations in a vibrating lattice at finite temperatures. Yet, the majority of the current theoretical interpretations of these INS spectra have neglected multiphonon processes by using one-phonon's self-energy evaluated from perturbation theory. We have implemented a robust numeric algorithm to calculate the full scattering spectra based on large-scale molecular dynamics (MD) simulations. Using a silicon machine-learning interatomic potential, we have predicted a set of INS spectra of silicon crystals from 300K to 1500K. We demonstrate that a q-space symmetrization technique significantly reduces intrinsic numeric uncertainties associated with the MD simulations of thousands of atoms over hundreds of picoseconds. Our simulated temperature-dependent INS data are in excellent agreement with recent phonon dispersion curves. In addition, our data reveals many spectra details that are not yet observable even in the state-of-the-art experiment. Furthermore, we quantitatively analyze the individual contributions due to single-phonon processes, multiple phonon processes, and interference between the single phonon and multiple phonon processes, using a decomposition algorithm. We will discuss the temperature-dependent line shapes of the phonon peaks in the INS spectra and the implications for phonon lifetime calculations at high temperatures.