Deep Learning Order Parameter for Polymer Phase Transition

TL;DR

This paper introduces a deep autoencoder framework that automatically discovers an effective order parameter from molecular dynamics data to characterize polymer phase transitions, offering a generic and efficient alternative to traditional methods.

Contribution

The study presents a novel deep learning approach that autonomously identifies order parameters from MD data, improving characterization of polymer phase transitions.

Findings

Accurately characterizes coil to globule transition as a function of temperature.

Provides a generic method applicable to various phase transitions in soft materials.

Offers computational efficiency over traditional MD-based characterization.

Abstract

We report a deep learning (DL) framework viz. deep autoencoder that autonomously discovers an appropriate order parameter from molecular dynamics (MD) simulation data to characterize the coil to globule phase transition of a polymer. The deep autoencoder encodes the 3N dimensional MD trajectory of a polymer in a one-dimensional feature space and, subsequently, decodes the one-dimensional feature to its original 3N dimensional polymer trajectory. The feature space representation of a polymer provides a new order parameter that accurately describes the coil to globule phase transition as a function of temperature. This method is very generic and extensible to identify flexible order parameters to characterize wide range of phase transitions that take place in polymers and other soft materials. Moreover, this MD-DL approach is computational very efficient than a pure MD based characterization of phase transition, and has potential implications in accelerating phase prediction.