Abstract text (incl. figure legends and references)

The development of high energy resolution EELS [1] has enabled the study of phonon properties of materials at the atomic scale, see for example Refs. [2,3]. One thereby exploits so-called impact scattering of electrons on the atomic nuclei itself, which is similar to the scattering of neutrons on phonons, albeit with a different atomic scattering factor [4,5]. With these tools it is now possible to locally probe the behaviour of phonons and thereby the flow of heat near interfaces [6-9], which is of large practical interest in the design of nano-scale devices and a unique capability of monochromated ultra high energy resolution STEM.

From a theoretical point of view, impact scattering on thicker specimen is difficult to model in a single inelastic scattering picture considering every possible transition due to the large amount of final states of the phonon system as well as the strong elastic interaction of the electron beam and the sample, the so-called dynamical diffraction. If one wants to consider a system containing an interface in such simulations, computational requirements may become prohibitive due to the large unit cells of suitable structure models. However topics such as electron channeling effects and delocalization of the vibrational (impact scattering) signal have not been fully addressed in literature yet, but are of high relevance for the correct interpretation of experiments.

We have introduced a qualitatively motivated extension of the Frozen Phonon Multislice (FPMS) method, the Frequency-Resolved FPMS (FRFPMS) method, which relies on molecular dynamics simulations using a frequency-selective thermostat rather than an explicit knowledge of phonon modes. The methodside-steps in this waysome of the computational complexity of simulations of inelastic phonon scattering in the STEM[10,11], but includes the effects of dynamical diffraction and is able to consider different scattering geometries. In this contribution, we first review the FRFPMS method and showcase some of the results obtained so far. We then consider a new sampling paradigm for the FRFPMS method, which relies on Fourier filtering of MD trajectories rather than a frequency-selective thermostat, and discuss it critically in comparison with the old scheme. This new sampling scheme promises a technicalsimplification of the calculation as well as a reduction of computational requirements.