Solving Statistical Mechanics using Variational Autoregressive Networks

Paper link: arXiv:1809.10606 | Phys. Rev. Lett. 122, 080602 (2019)

The code requires Python >= 3.6 and PyTorch >= 1.0. If you have configured your PyTorch installation with a recent Nvidia GPU card, you can enjoy enormously acceleration (> 10x).

Directory src_ising contains code for 2D FM and AFM Ising model, and src_hop_sk contains code for Hopfield model, SK model and inverse Ising problem. Run python3 src_ising/main.py --help to see all configurations.

Script src_ising/reproduce.sh and src_hop_sk/reproduce.sh are commands to reproduce the results in Fig. 2~4. Directly running these scripts may take thousands of GPU hours, and produce hundreds GB of output data, most of which are network weights during training steps. In practice, you may run these commands in parallel on multiple GPUs, and set appropriate output path.