Bereiche | Tage | Auswahl | Suche | Aktualisierungen | Downloads | Hilfe

DY: Fachverband Dynamik und Statistische Physik

DY 39: Machine Learning in Dynamics and Statistical Physics II

DY 39.4: Vortrag

Donnerstag, 20. März 2025, 15:45–16:00, H47

Describing heat transport in crystalline polymers in real and reciprocal space — Lukas Reicht¹, Lukas Legenstein¹, Sandro Wieser², and •Egbert Zojer¹ — ¹Graz University of Technology, Austria — ²TU Wien, Austria

Heat transport modelling either relies on describing the propagation of phonons employing the Boltzmann transport equation or on simulating the real-space dynamics of atoms using (non)-equilibrium molecular dynamics techniques. Due to the structural complexity of crystalline polymers both approaches call for a highly accurate but at the same time numerically extremely efficient strategy for describing inter-atomic interactions. This is achieved via machine-learned potentials, where we combine an efficient active-learning strategy with moment-tensor potentials.[1,2] Additionally, real-space and reciprocal space approaches make fundamentally different approximations regarding anharmonicities and phonon occupations. Here, we show that for polymers of intermediate complexity, like crystalline polythiophene, real- and reciprocal space approaches yield consistent values of the thermal conductivities at least when using an accurate machine-learned potential. Interestingly, for the seemingly much simpler crystalline polyethylene such an agreement is only obtained when higher-order phonon scattering is considered. This can be traced back to a selection rule arising from the comparably simple phonon band structure of polyethylene. [1] npj Comput Mater 10, 18 (2024); [2] Molecules 29, 3724 (2024)

Keywords: machine-learned potential; moment-tensor potential; active learning; heat transport; crystalline polymer