Machine Learning Phases of Strongly Correlated Fermions

Kelvin Ch'ng,Juan Carrasquilla,Roger G. Melko,Ehsan Khatami

DOI: https://doi.org/10.1103/PhysRevX.7.031038

2017-09-11

Abstract:Machine learning offers an unprecedented perspective for the problem of classifying phases in condensed matter physics. We employ neural-network machine learning techniques to distinguish finite-temperature phases of the strongly correlated fermions on cubic lattices. We show that a three dimensional convolutional network trained on auxiliary field configurations produced by quantum Monte Carlo simulations of the Hubbard model can correctly predict the magnetic phase diagram of the model at the average density of one (half filling). We then use the network, trained at half filling, to explore the trend in the transition temperature as the system is doped away from half filling. This transfer learning approach predicts that the instability to the magnetic phase extends to at least 5% doping in this region. Our results pave the way for other machine learning applications in correlated quantum many-body systems.

Strongly Correlated Electrons

What problem does this paper attempt to address?

The problem that this paper attempts to solve is to classify the phases of matter at finite temperatures and estimate the phase - transition temperature in strongly - correlated fermion systems through machine - learning techniques. Specifically, the authors use a three - dimensional convolutional neural network (3D CNN) to train the auxiliary - field configurations generated by the quantum Monte Carlo simulation of the Hubbard model on a cubic lattice to distinguish different magnetic phases. The focus of the study is on the antiferromagnetic Neel phase transition under half - filling conditions, and further explores the trend of the phase - transition temperature with the doping level when the system deviates from half - filling.

The main contributions of the paper include:
1. **Proposing a new method**: Using 3D CNN to identify finite - temperature phase transitions in quantum many - body systems, which paves the way for machine - learning applications in other related fields.
2. **Verifying the effectiveness of the method**: By comparing with known theoretical results, it shows the accuracy of this method in predicting the Neel phase - transition temperature.
3. **Exploring the phase diagram**: Not limited to half - filling conditions, it also explores the changes in the phase diagram of the system under doping conditions and finds that magnetic instabilities can extend to at least a 5% doping region.

These achievements are of great significance for understanding the physical properties of strongly - correlated electron systems and for developing new materials and devices.