Exploring Melting Phenomena in Self-Organized Magnetic Structures through Variational Autoencoder Analysis

The investigation of phase transition phenomena is a crucial aspect of various physics studies. However, defining order parameters in complex systems with self-organized structures presents challenges. This work introduces a method employing a variational autoencoder network for the definition of order parameters.

The effectiveness of our approach is demonstrated through the training of a deep learning network on a dataset containing spin configurations within a chiral magnetic system across diverse temperature conditions. By mitigating thermal fluctuations in the input data, the network preserves essential structural information, focusing on spin magnitudes. We utilize this information to establish an order parameter based on magnitude and validate our results against conventional analyses, revealing consistent outcomes.

Using the order parameter, we examine the thermal properties of the chiral magnetic system. Through systematic variations of physical parameters and data sizes, our investigation provides insights into the system's response to changing conditions, contributing to a nuanced understanding of its thermal behavior.