Transpile using the native gate set of Forte 1

[sebastianvromero]

Information

• Qiskit IonQ Provider version: 0.5.8

• Qiskit version: 1.1.0

• Python version: 3.9.6

• Operating system: macOS Sequoia v15.1.1

What is the current behavior?

Using transpile() from qiskit, I am trying to transpile a given circuit into the native set of gates of IonQ Forte 1, which considers ZZ gates instead of MS gates, as IonQ Aria platforms have.

Steps to reproduce the problem

from qiskit_ionq import IonQProvider
from qiskit import QuantumCircuit, transpile

provider = IonQProvider(token='your token')
native_simulator = provider.get_backend('ionq_simulator', gateset='native')
native_simulator.set_options(noise_model='forte-1')

qc = QuantumCircuit(2)
qc.rzz(1, 0, 1)

transpiled_qc = transpile(qc, backend=native_simulator, optimization_level=3)
print(transpiled_qc.count_ops()) # OrderedDict([('gpi2', 10), ('ms', 2), ('gpi', 1)])

What is the expected behavior?

Instead of MS gates, the transpiler should be aware of the backend chosen and transpile accordingly. If Forte backend is chosen, use ZZ as entangling gates instead.

Suggested solutions

The provider.get_backend() function could include a new input which is the device selected.

• If device='aria-1': use GPi, GPi2 and MS gates.
• If device='forte-1': use GPi, GPi2 and ZZ gates.

[splch]

Thank you for bringing this up! I have a PR up right now #231 that enables this (somewhat):

from qiskit_ionq import IonQProvider
from qiskit import QuantumCircuit, transpile

provider = IonQProvider()
forte_backend = provider.get_backend('qpu.forte-1', gateset='native')
native_simulator = provider.get_backend('ionq_simulator', gateset='native')
native_simulator.set_options(noise_model='forte-1')

qc = QuantumCircuit(2)
qc.rzz(1, 0, 1)

transpiled_qc = transpile(qc, backend=forte_backend, optimization_level=3)
print(transpiled_qc.count_ops()) # OrderedDict({'gpi2': 18, 'gpi': 9, 'zz': 2})

OrderedDict({'gpi2': 18, 'gpi': 9, 'zz': 2})

I like your usage of options and plan on adding a check for the noise model in the ionq_transpile function soon. If you need this functionality now, though, please use the above example!