Machine learning structural phases of a flexible Stockmayer polymer

We use machine learning methods to investigate structural phases and transitions in a chain of magnetic colloidal nanoparticles characterized by dipolar interactions and short-range Lennard-Jones attractive interactions (i.e., Stockmayer polymer). In particular, we utilize the neural network-based, semi-supervised confusion method, which does not require any prior knowledge about the existence of a transition. Through this method, we were able to locate boundaries between multiple structural phases as a function of energy. By applying dimensionality reduction techniques, we further explore two-dimensional representations of the configuration space, in which we can visually distinguish different structural phases in the system.