A simple framework for contrastive learning phases of matter

TL;DR

This paper introduces SimCLP, a straightforward contrastive learning framework that effectively identifies and characterizes various phases of matter without manual feature engineering, applicable to classical and quantum systems.

Contribution

The paper presents a novel, flexible contrastive learning framework for phases of matter that requires only state configurations, eliminating the need for manual features or prior knowledge.

Findings

Successfully applied to classical and quantum systems

Able to generate representations and labels

Facilitates discovery of new phase transitions

Abstract

A main task in condensed-matter physics is to recognize, classify, and characterize phases of matter and the corresponding phase transitions, for which machine learning provides a new class of research tools due to the remarkable development in computing power and algorithms. Despite much exploration in this new field, usually different methods and techniques are needed for different scenarios. Here, we present SimCLP: a simple framework for contrastive learning phases of matter, which is inspired by the recent development in contrastive learning of visual representations. We demonstrate the success of this framework on several representative systems, including classical and quantum, single-particle and many-body, conventional and topological. SimCLP is flexible and free of usual burdens such as manual feature engineering and prior knowledge. The only prerequisite is to prepare enough state configurations. Furthermore, it can generate representation vectors and labels and hence help tackle other problems. SimCLP therefore paves an alternative way to the development of a generic tool for identifying unexplored phase transitions.