Confusion-driven machine learning of structural phases of a flexible, magnetic Stockmayer polymer

TL;DR

This paper employs a semi-supervised neural network approach to identify structural phase transitions in magnetic polymers, revealing new low-temperature phases and demonstrating the method's effectiveness without prior phase knowledge.

Contribution

It introduces a confusion-based machine learning technique to detect phase transitions in magnetic polymers using simulation data, including previously unexplored low-temperature regions.

Findings

Effective identification of transition points via neural network accuracy

Discovery of new structural phases at low temperatures

Validation with conventional order parameters

Abstract

We use a semi-supervised, neural-network based machine learning technique, the confusion method, to investigate structural transitions in magnetic polymers, which we model as chains of magnetic colloidal nanoparticles characterized by dipole-dipole and Lennard-Jones interactions. As input for the neural network we use the particle positions and magnetic dipole moments of equilibrium polymer configurations, which we generate via replica-exchange Wang--Landau simulations. We demonstrate that by measuring the classification accuracy of neural networks, we can effectively identify transition points between multiple structural phases without any prior knowledge of their existence or location. We corroborate our findings by investigating relevant, conventional order parameters. Our study furthermore examines previously unexplored low-temperature regions of the phase diagram, where we find new structural transitions between highly ordered helicoidal polymer configurations.