Berlin 2024 – wissenschaftliches Programm

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

DY: Fachverband Dynamik und Statistische Physik

DY 5: Machine Learning in Dynamics and Statistical Physics I

DY 5.9: Vortrag

Montag, 18. März 2024, 11:45–12:00, BH-N 243

A Study of Quantum Non-Equilibrium Relations with Imaginary Path Integrals — •Jorge Castro, Eszter Pos, and Mariana Rossi — Max Planck Institute for the Structure and Dynamics of Matter, Hamburg, Germany

Several processes at the atomic scale occur in strongly out-of-equilibrium conditions. When such processes involve atomic motion at low temperatures or with light nuclei, it is expected that nuclear quantum effects play a pronounced role. However, theories that can rigorously treat quantum dynamics out of equilibrium are typically impossible to apply to explicit atomistic simulations of systems with many degrees of freedom. In this contribution, we study the protocols proposed in Ref.[1], where path-integral molecular dynamics simulations are used to evaluate quantum free energy differences through the Jarzynski and Crooks relations. We developed a code capable of generating non-equilibrium trajectories from path integral molecular dynamics. With this code, we investigated the efficiency and realm of validity of the work calculation employing driven path-integral molecular dynamics simulations. Our findings, based on 1D potentials, compare performance at various switching rates and varying coordinate shifts between the minima of starting and ending anharmonic potentials. We investigate the extend to which machine-learning methods targeting the inference of work distribution can speed up the statistical convergence of these protocols.

[1] R. van Zon, L. Hernández de la Peña, H. Peslherbe, and J. Schofield, Phys. Rev. E 78, 041103 (2008)

Keywords: Path Integral Molecular Dynamics; Non-equilibrium processes; Free-energy differences; Machine-learning