Order-Parameter-Free Analysis of Soft Matter: Applications of Machine Learning via Image Recognition

Abstract

Various characteristic structures, with no long-range spatial order, have often been observed in studies on the structural formation of soft materials. The order parameters, used to date, are not promising for computer detection of these types of structures. In this previous study, it is shown that machine-learning analysis using convolutional neural networks is very effective for the structural formation of spherical colloidal particles. This method is applied to non-spherical inverse patchy colloids and demonstrated that this order-parameter-free analysis method is effective for non-spherical soft matter, which often exhibits complex structures. A recent development in the structural formation of colloidal particle systems corresponds to the problem of monolayers of core-corona particle systems that exhibit a variety of structures. Monte Carlo simulations are performed for core-corona particles, confined between parallel plates, to clarify the conditions for the appearance of the bilayer and its in-plane structure formation. Parameter-free analysis is performed using image-based machine learning. The bilayer formation of the Jagla fluids is performed, and the phase diagram is clarified.