Unsupervised phase discovery with deep anomaly detection

This repository contains the code to our paper Unsupervised phase discovery via anomaly detection, where we use deep neural networks auto encoders (AE) to find phase transitions in the one dimensional extended Bose-Hubbard model.

The repository provides the code to generate the training data with DMRG and the whole training pipeline for the AE.

Run the code

You will have to install the following packages:

• tensorflow (v1.12.1)
• numpy (1.17.3)
• matplotlib (3.1.1)
• TenPy (> v0.4.0) (pip install physics-tenpy)

The versions in brackets indicate with which version the code was tested. To check if the TenPy installation was succesful run AD_tools.py for a test run.

The Jupyter Notebook Bose_Hubbard.ipynb contains all the necessary elements to draw the phase diagram along

.

The Jupyter Notebook Bose_Hubbard_precalc.ipynb allows to load data and plot the whole phase diagram in 2D with

.