Unsupervised classification of disordered patterns in an oppositely charged colloidal system

TL;DR

This paper presents an unsupervised machine learning method to classify disordered phases in a charged colloidal system, revealing key structural features and phase transition mechanisms.

Contribution

It introduces a novel unsupervised approach using PCA on local structural vectors to classify disordered phases in colloids, enhancing interpretability and understanding.

Findings

Successful classification of disordered phases

Consistent results with radial distribution functions

Insights into phase transition mechanisms

Abstract

We develop an unsupervised machine learning approach to classify disordered phases in a system of oppositely charged colloids. In this system, the interplay between Coulomb and van der Waals interactions leads to transitions in local structures, while the global structure remains disordered. Our method involves representing the local structures of the system as high-dimensional vectors and applying principal component analysis to identify distinct features of each phase. We demonstrate that our method results in a reasonable classification of disordered phases, which is consistent with that obtained from radial distribution functions. The interpretability of the method reveals the key characteristics of each phase and provides valuable insights into the mechanisms underlying the unconventional phase transitions.