Skip to content
A framework based on Tensorflow for running variational Monte-Carlo simulations of quantum many-body systems.
Branch: master
Clone or download
Latest commit 4aa9dc7 Jan 19, 2020
Type Name Latest commit message Commit time
Failed to load latest commit information.
.github/workflows support tf 1.14 Dec 7, 2019
examples fix bug in ising example Nov 27, 2019
src/flowket support tf 1.14 Dec 7, 2019
.gitignore add slurm logs to .gitignore May 19, 2019
.travis.yml remove requirements.txt May 21, 2019
LICENSE Initial commit May 9, 2019 Updated citation info and added link to PRL publication Jan 19, 2020
setup.cfg fix tf 1.14 competability Nov 18, 2019 Add longer description to PyPI Nov 21, 2019

FlowKet - A Python framework for variational Monte-Carlo simulations on top of Tensorflow

FlowKet is our framework for running variational Monte-Carlo simulations of quantum many-body systems. It supports any Keras model for representing a parameterized unnormalized wave-function, e.g., Restricted Boltzman Machines and ConvNets, with real or complex-valued parameters. We have implemented a standard Markov-Chain Monte-Carlo (MCMC) energy gradient estimator for this general case, which can be used to approximate the ground state of a quantum system according to a given Hamiltonian. The neural-network-based approach for representing wave-fucntions was shown to be a promising method for solving the many-body problem, often matching or even surpassing the precision of other competing methods.

In addition to an MCMC energy gradient estimator, we have also implemented our novel Neural Autoregressive Quantum State wave-function representation that supports efficient and exact sampling. By overcoming the reliance on MCMC, our models can converge much faster for models of same size, which allows us to scale them to millions of parameters, as opposed to just a few thousands for prior approaches. This leads to better precison and ability to invesitgate larger and more intricated systems. Please read our paper (arXiv version), cited below, for further details on this approach. We hope that users of our library will be able to take our method and apply to a variety of problems. If you use this codebase or apply our method, we would appreciate if you cite us as follows:

  title = {Deep Autoregressive Models for the Efficient Variational Simulation of Many-Body Quantum Systems},
  author = {Sharir, Or and Levine, Yoav and Wies, Noam and Carleo, Giuseppe and Shashua, Amnon},
  journal = {Phys. Rev. Lett.},
  volume = {124},
  issue = {2},
  pages = {020503},
  numpages = {6},
  year = {2020},
  month = {Jan},
  publisher = {American Physical Society},
  doi = {10.1103/PhysRevLett.124.020503},
  url = {}


FlowKet assumes Tensorflow is already part of the enviornment. We currently support Tensorflow 1.10-1.14, but plan to extend support to all >=1.10+ and 2.0.

The recommended way to intall FlowKet is via PyPI:

pip install flowket

Alternatively if you wish to work on extending our library, you can clone our project and instead run:

pip install -e /path/to/local/repo

Basic Tutorial

While we are working on writing a proper tutorial on using the framework, we suggest going through the example files.

You can’t perform that action at this time.