Documentation

Homepage

Looking for the project overview? Head back to the tachys homepage.

Benchmarks

Reference variational energies for frustrated lattice models — pick a model, a lattice geometry and a size on the benchmarks page.

tachys is a JAX library for finding the ground states of quantum lattice models with neural-network wavefunctions. Build a Hamiltonian by composing spin and fermionic operators, use any Flax network as the wavefunction, and optimize it with variational Monte Carlo — the same code runs on a laptop and across a multi-GPU cluster. It also trains foundation models: a single network covering a whole family of Hamiltonians, rather than one run per coupling.

Installation

# CPU-only JAX
pip install jax

# GPU (CUDA 12)
pip install "jax[cuda12]"

# tachys
pip install tachys

Getting Started

• Quickstart

Documentation

• User Guide
• Core Concepts
• Foundation Models
• API Reference