GULPS

GULPS is a two-qubit gate synthesis package for arbitrary native instruction sets. Given a set of native two-qubit gates and a cost for each, it selects the least-cost sentence, an ordered list of native gates with single-qubit gates before, between, and after them, that implements a target unitary. It then constructs those single-qubit gates with a dedicated solver, can_sandwich.

pip install gulps

The native gates can be any two-qubit gates with fixed parameters: Qiskit standard gates with bound parameters, or your own matrices wrapped in UnitaryGate.

from qiskit.circuit.library import CSXGate, iSwapGate
from qiskit.quantum_info import random_unitary
from gulps.decomposition import GulpsDecomposer

decomposer = GulpsDecomposer([CSXGate(), iSwapGate().power(1 / 3)], costs=[90, 120])
circuit = decomposer(random_unitary(4, seed=0))
print(circuit.draw("text"))

A cost can be a gate duration or any other quantity that adds up along a sentence.

• Compilation: charge for single-qubit layers, run GULPS in a Qiskit pass manager, and query synthesis costs from other transpiler passes.
• Pulse duration calibration: choose which native gate durations to calibrate for your workload.
• How synthesis works: how reachable regions guide sentence selection and circuit construction.

The paper describes the method, which builds on monodromy.