Dresden 2026 – scientific programme
Parts | Days | Selection | Search | Updates | Downloads | Help
DY: Fachverband Dynamik und Statistische Physik
DY 14: Machine Learning in Dynamics and Statistical Physics II
DY 14.10: Talk
Monday, March 9, 2026, 17:30–17:45, HÜL/S186
Spin-phonon systems in the age of modern atomistic simulations — •Ilija Srpak1,2, Michael J. Willatt2, Stuart C. Althorpe1, and Ali Alavi1,2 — 1Yusuf Hamied Department of Chemistry, University of Cambrige, Cambridge, United Kingdom — 2Max Planck Institute for Solid State Research, Stuttgart, Germany
Spin-phonon systems are molecules or crystals containing open-shell atoms whose spin-spin interaction is significantly affected by lattice displacements, sometimes leading to spin-Peierls phase transition. They typically inherit some of the most challenging aspects of statistical physics where many configurations across the phase space may contribute to its properties, and of ``strongly-correlated" physics where mean-field methods such as density functional theory and self-consistent field approaches break down.
Over the years a plethora of Monte Carlo based techniques was developed to tackle this problem with some success, but not without (sometimes significant) limitations. Approaching this problem from atomistic simulations background, we developed a path integral molecular dynamics framework which doesn't require any Monte Carlo during the simulation runtime to sample the phase space, it can take arbitrary system parametrizations or even ab-initio description and simulate the system at an arbitrary temperature as well as include nuclear quantum effect. Using neural networks we have developed this framework further. We achieved a speed up of 2-3 orders of magnitude and are able to treat higher dimensional systems.
Keywords: Heisenberg antiferromagnet; Spin-Peierls; Spin-phonon; Neural networks; Molecular dynamics