Variational Monte Carlo for quantum many-body systems

tachys is a variational Monte Carlo library developed in JAX with one central aim: bringing neural-network quantum states (NQS) to realistic quantum materials. To reach the system sizes that real materials demand, the code is engineered to scale across large GPU clusters and has been benchmarked on systems with up to 1,000 electrons, a scale already meaningful for studying unconventional superconductors. With the same focus on realistic materials, tachys can treat itinerant fermions and localized spins within a single Hamiltonian, as required by Kondo lattice models of heavy-fermion compounds. It also supports Foundation NQS, in which one network is trained across an entire family of Hamiltonians, allowing the parameter space relevant to a given material to be explored in a single optimization.

Documentation: https://tachys.org/quickstart.html

Installation

CPU:

pip install tachys

CUDA (GPU):

pip install "tachys[cuda13]"    # CUDA 13
pip install "tachys[cuda12]"    # CUDA 12

These install JAX together with the CUDA 13 or CUDA 12 libraries, so the machine needs only the NVIDIA driver. Choose the version that the driver supports: the header of nvidia-smi shows the highest one.

Install from source

git clone https://github.com/tachys-dev/tachys.git
cd tachys
pip install .

Development install

To install in editable mode (changes to the source are reflected immediately without reinstalling):

pip install -e ".[test]"            # CPU
pip install -e ".[cuda13,test]"     # CUDA 13
pip install -e ".[cuda12,test]"     # CUDA 12
pytest tests/

Authors

• Riccardo Rende
• Luciano Loris Viteritti

License

Copyright 2026 The Simons Foundation, Inc.

Licensed under the Apache License, Version 2.0.