Skip to content

dinesh110598/ising-autoregressive

Repository files navigation

Autoregressive network on 2D Ising model

Aims to construct autoregressive networks (PixelCNN and similar) that outputs the log probabilities of configurations (and samples from it), given by Boltzmann distribution for Ising model to arbitrary accuracy. This uses Tensorflow 2 on a Jupyter notebook.

Open main.ipynb in Colab

In the graphs for Free energy predictions in various VarPCNN models, the variance is very small but the bias due to imperfect training seems to be larger. One very interesting observation is that the variance seems to reduce with increased lattice size, indicating a convergence to the thermodynamic limt.

About

No description, website, or topics provided.

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published