Regensburg 2016 – wissenschaftliches Programm
Bereiche | Tage | Auswahl | Suche | Aktualisierungen | Downloads | Hilfe
CPP: Fachverband Chemische Physik und Polymerphysik
CPP 34: Crystallization, Nucleation, Self Assembly I (joint session CPP/DY, organized by CPP)
CPP 34.7: Vortrag
Mittwoch, 9. März 2016, 11:30–11:45, H42
Improved Transferability of Coarse Grained Models for Polymer Crystallization Using Machine Learning — •Chan Liu1, Christine Peter2, Kurt Kremer1, and Tristan Bereau1 — 1Max Planck Institute for Polymer Research, Ackermannweg 10, 55128 Mainz, Germany — 2Theoretical Chemistry, University of Konstanz, Konstanz, Germany
Coarse-grained (CG) models, which combine a number of atoms into superatoms or beads, can significantly speed up the simulations and provide reasonable resolution for studying polymer crystallization. One of the major challenges in CG modeling is that the reduction of the number of degrees of freedom makes the resulting coarse models state point dependent; that is, most CG force fields developed from the structures of an atomistic melt are not guaranteed to be transferred to crystalline structures. Thus deriving a transferable CG potential across different thermodynamic states is rather crucial if one want to study the phase behavior of polymeric systems. In this work, we introduce a Machine Learning approach to improve an existing CG model parametrized for a different phase by predicting the deviation between CG and atomistic forces, which can be seen as an external force added on the original CG force field. This model predicts a force on each bead based on the surrounding geometry without projecting it onto pairwise potentials such that it can potentially reproduce many-body contributions. This approach opens the perspective to modeling many-body interactions in CG simulations and thus improve the transferability and accuracy of its force field.