Dresden 2020 – wissenschaftliches Programm
Die DPG-Frühjahrstagung in Dresden musste abgesagt werden! Lesen Sie mehr ...
Bereiche | Tage | Auswahl | Suche | Aktualisierungen | Downloads | Hilfe
CPP: Fachverband Chemische Physik und Polymerphysik
CPP 32: Condensed-matter simulations augmented by advanced statistical methodologies (joint session DY/CPP)
CPP 32.4: Vortrag
Montag, 16. März 2020, 16:15–16:30, HÜL 186
Kernel-based machine learning for efficient molecular liquid simulations — •Christoph Scherer1, René Scheid1, Tristan Bereau1,2, and Denis Andrienko1 — 1Max-Planck-Institut für Polymerforschung, Mainz, Germany — 2University of Amsterdam, Netherlands
Most current force fields based on machine learning (ML) techniques result in high computational cost at every integration time step of an MD simulation. We describe a number of practical and computationally-efficient strategies to parametrize force fields for molecular liquids with kernel-based ML. We employ a particle decomposition ansatz to two- and three-body force fields and covariant kernels. Binning techniques allow to incorporate significantly more training data. Tabulation of the kernel predictions lead to MD simulations with the same computational cost than analytic three-body potentials. Results are presented for model molecular liquids: pairwise Lennard-Jones and three-body Stillinger-Weber systems, as well as an example from bottom-up coarse-graining of liquid water [1]. Many-body representations, decomposition, and kernel regression schemes are implemented in the open-source software package VOTCA [2].
[1] Scherer, Andrienko, PCCP, 20, 22387 (2018); [2] Rühle, Junghans, Lukyanov, Kremer, Andrienko, JCTC, 5, 3211 (2009)