pytket-cutensornet

pytket-cutensornet is an extension to pytket that allows pytket circuits and expectation values to be simulated using cuTensorNet.

cuTensorNet is a high-performance library for tensor network computations, developed by NVIDIA. It is part of the cuQuantum SDK – a high-performance library aimed at quantum circuit simulations on the NVIDIA GPU chips.

We provide two core functionalities:

  • Exact value calculation: use GeneralState and GenearlBraOpKet to translate a pytket into a tensor network and obtain amplitudes and expectation values via full tensor network contraction using cuQuantum’s optimised contraction path.

  • Approximate state evolution: use simulate to simulate a pytket circuit, returning a tensor network representation of the approximate output state, from which you can query properties, such as amplitudes and expectation values.

pytket-cutensornet is available for Python 3.10, 3.11 and 3.12 on Linux. In order to use it, you need access to a Linux machine (or WSL) with an NVIDIA GPU of Compute Capability +7.0 (check it here). You will need to install cuda-toolkit and cuquantum-python before pytket-cutensornet; for instance, in Ubuntu 24.04:

sudo apt install cuda-toolkit
pip install cuquantum-python
pip install pytket-cutensornet

Alternatively, you may install cuQuantum Python following their instructions using conda-forge. This will include the necessary dependencies from CUDA toolkit. Then, you may install pytket-cutensornet using pip.

Example Notebooks

Changelog

Useful links