Berlin 2024 – wissenschaftliches Programm

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

DY: Fachverband Dynamik und Statistische Physik

DY 59: Brownian Motion and Anomalous Diffusion

DY 59.9: Vortrag

Freitag, 22. März 2024, 11:45–12:00, BH-N 334

Iterative parameterization of Markovian embedded generalized Langevin equations for molecular dynamics — •Viktor Klippenstein, Niklas Wolf, and Nico F. A. Van Der Vegt — Technical University of Darmstadt, Darmstadt, Germany

In molecular dynamics, coarse-grained (CG) models which aim to describe dynamic properties consistently with the underlying fine-grained (FG) system, typically introduce some dissipative thermostat to account for friction and fluctuations due to removed degrees of freedom. In many cases, the time scales of CG and FG degrees of freedom are not separated which necessitates a non-Markovian (NM) description typically based on a generalized Langevin equation (GLE).

To keep the CG models tractable, we augment the Hamiltonian equation of motion by individually coupling every coarse-grained particle to an isotropic GLE thermostat, where NM friction is fully characterized by a single scalar function termed memory kernel. For computational efficiency, the NM GLE thermostat can be mapped on a Markovian auxiliary variable thermostat (aux-GLE). While our recently introduced method (iterative optimization of memory kernels (IOMK)[1]) allows for efficient optimization of the GLE the parameterization of the aux-GLE is by itself non-trivial, and potentially both error-prone and computationally expensive. To sidestep this problem, we propose a Gauss-Newton type method (IOMK-GN), which allows us to directly optimize the aux-GLE parameters.
[1] Klippenstein V., van der Vegt N. F. A., J. Chem. Theory Comput. 2023, 19, 4, 1099-1110

Keywords: Molecular Dynamics; Coarse-Graining; Modeling; Non-Markovian; Stochastic Dynamics