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 24: Poster Session I

CPP 24.20: Poster

Montag, 16. März 2020, 17:30–19:30, P3

Characterisation of the Free Energy Landscape of Syndiotatic Polysterene — •Atreyee Banerjee, Tristan Bereau, and Joseph F. Rudzinski — MPIP

Syndiotatic polysterene (sPS) is known to crystallise in distinct forms, commonly known as polymorphs [1]. Traditional molecular dynamics simulations are powerful tools for characterizing the molecular mechanism of transition between polymorphs, but require extreme computational resources due to the strong metastability of the polymorph states. Enhanced sampling methods have the potential to largely remedy this problem, but require prior knowledge of collective variables (CVs) that can resolve the relevant transition pathways, typically identified through physical or chemical expertise. CVs are also often used for constructing a kinetic model to better characterise the transition pathways of the characteristic long timescale processes of the system. A huge interest has grown to apply neural networks to automate the discovery of a low-dimensional representation [2]. Autoencoders are potentially powerful tools to identify good CVs, since the technique forces an information compression in the bottleneck region. A specialised autoencoder architecture, the Gaussian mixture variational autoencoder (GMVAE), performs dimensionality reduction and clustering within a single unified framework, and can identify the inherent dimensionality of the system by enforcing physical constraints in the latent space. In contrast to manually constructed CVs, we apply the GMVAE approach to accurately characterise the pathways of transition between polymorphs in sPS.