tensor_network method powered by cuQuantum is not available

[HaoTy]

Informations

• Qiskit Aer version: 0.12.0
• Qiskit Aer GPU version: 0.11.2
• Python version: 3.9.16
• Operating system: Ubuntu 20.04.5 LTS

What is the current behavior?

Specifying method="tensor_network" ( #1637 ) with device="GPU" produces AerError: "Invalid simulation method tensor_network" for the latest release.

Steps to reproduce the problem

pip install qiskit
pip install cuquantum-python-cu11 cupy-cuda11x
pip install qiskit-aer-gpu

This should install qiskit-aer 0.12.0, qiskit-aer-gpu 0.11.2, and cuQuantum with cuTensorNet support. Then

from qiskit_aer import AerSimulator

backend = AerSimulator(method="tensor_network", device="GPU")

What is the expected behavior?

tensor_network should be available for AerSimulator when cuTensorNet is functional.

Suggested solutions

Not sure if it's related, but the latest version of qiskit-aer-gpu (available on PyPI) is 0.11.2.

[hhorii]

0.12.0 is not available through PyPI (though we will fix this issue soon). Could you compile and try Aer from codes?

[mtreinish]

Actually the qiskit-aer-gpu package does not include the cuQuantum support. If you'd like to use cuQuantum you will need to compile aer from source. There instructions on how to do this here: https://github.com/Qiskit/qiskit-aer/blob/main/CONTRIBUTING.md#building-with-gpu-support including the options needed to build the cuQuantum support.

[HaoTy]

Thanks for the clarification. I was aware that cuStateVec_enable needs Aer to be built from source with appropriate flags, but somehow forgot that this would likely be the case for tensor_network as well :/