QCut

QCut is a quantum circuit knitting package built on top of qiskit for performing gate cuts and resetless wire cuts allowing simulation of larger quantum circuits on smaller quantum devices or simulators at the cost of a circuit overhead. QCut has been designed and tested to work with IQM's qpus, and the Finnish Quantum Computing Infrastructure (FiQCI).

QCut has been built at CSC - IT Center for Science (Finnish IT Center for Science).

See the documentation for the full usage guide, every option, examples and the theory behind the decompositions.

Installation

A UNIX-like system is needed, since PyMetis is used for automatic cut finding. On Windows use WSL.

uv pip install QCut
#or
uv add QCut

QCut does not install a simulator. The examples run on Qiskit Aer: uv pip install qiskit-aer.

For IQM hardware and fake backends, install the extra, which holds Qiskit below 2.2:

uv pip install "QCut[iqm]"

See Getting started for installing from source and other details.

Quick start

Mark where to cut, split the circuit, build the experiment for the observables wanted, run it, and put the expectation values back together:

import QCut as ck
from QCut import cut, cutGate
from qiskit import QuantumCircuit
from qiskit.circuit.library import CXGate
from qiskit_aer import AerSimulator

circuit = QuantumCircuit(5)
for qubit in range(5):
    circuit.ry(0.4 + 0.3 * qubit, qubit)
circuit.cx(0, 1)
circuit.append(**cutGate(CXGate(), 1, 2))  # a gate cut
circuit.cx(2, 3)
circuit.append(cut(), [3])  # a wire cut
circuit.cx(3, 4)

observables = ["IIIIZ", "IIZII", "ZZIII"]

cut_circuit = ck.get_locations_and_subcircuits(circuit)
# On a device with a coupling map, transpile the subcircuits for it here:
# cut_circuit = ck.transpile_circuits(cut_circuit, backend)
experiment = ck.get_experiment_circuits(cut_circuit, observables)

job = ck.submit_experiments(experiment, backend=AerSimulator(), shots=4096)
print(ck.estimate_expectation_values(job.result()))

[0.9204 0.3866 -0.0315], against exact values of [0.9211 0.3806 -0.0292].

find_cuts() finds and places cuts automatically instead, and ck.estimate_run(experiment) calculates what running an experiment will cost before submission. The usage guide goes through each step.

Several circuits at once

CutEstimator and CutQuasiSampler can run a list of cut circuits in one call, shaped like FiQCI-EMS's estimator and sampler:

from QCut import CutEstimator, CutOptions, find_cuts

ghz = QuantumCircuit(6)
ghz.h(0)
for qubit in range(5):
    ghz.cx(qubit, qubit + 1)
found = find_cuts(ghz, options=CutOptions(finder_num_partitions=2))

estimator = CutEstimator(AerSimulator())
job = estimator.run([cut_circuit, found], [observables, "ZZZZZZ"], shots=4096)
print(job.result().expectation_values())

[array([0.9160, 0.3853, -0.0328]), array(0.9999)]

CutQuasiSampler reconstructs the distribution over the qubits a circuit measured instead. See Primitives.

Why QCut

Where one of QCut's joint decompositions applies, the same cuts cost less than with IBM's qiskit-addon-cutting:

	QCut	addon
three wires cut at the same point	γ 15, 240 subexperiments	γ 64, 1024
three rotations crossing one split	γ 7.16, 264 subexperiments	γ 10.5, 432
an rzz and an rxx on each of three crossing pairs	γ 115	γ 203

Where none applies the two agree exactly. benchmarks/QCutVsAddon.ipynb has the comparison, run with uv sync --group benchmark.

Documentation

fiqci.fi/QCut. To build it locally:

uv sync --group docs
cd docs
uv run sphinx-build -v -b html . build/sphinx/html -W

Acknowledgements

This project is built on top of Qiskit which is licensed under the Apache 2.0 license.

License

Apache 2.0 license