anomaly-detection-PEPS
Deep anomaly detection for mapping out phase diagrams with PEPS
These are the complete notebooks to reproduce the plots in "Unsupervised mapping of phase diagrams of 2D systems from infinite projected entangled-pair states via deep anomaly detection" by Korbinian Kottmann, Philippe Corboz, Maciej Lewenstein and Antonio Acín (arxiv link tba).
The notebooks are nummerically ordered to reproduce Figs. 1-3.
The data from the simulated PEPS (bond singular values and reduced density matrices) are in data
Intermediate results are saved for convenience in data_results.
AD_tools.py contains some functions that are used for the anomaly detection in all notebooks.
plots contains the resulting figures as well as additional images like training convergences.
CNN_data contains the parameters of the neural networks involved. Training is very fast here, so there is not really a need for it.